Angiogenic redundancy is one of the earliest mechanisms leading to refractoriness or acquired resistance to anti-angiogenic therapies mainly targeting VEGFA and its receptors. Although VEGFA is the best known angiostimulatory protein, angiogenesis can be triggered by several growth factors including angiopoietins (ANGs), epidermal growth factor (EGFs), fibroblast growth factors (FGFs), hepatocyte growth factor (HGF), transforming growth factors (TGFs), placental growth factor (PlGF) or stromal cell-derived factor 1 (SDF1) (Xin et al., 2001; Bergers and Hanahan, 2008; Pardali et al., 2010; Brooks et al., 2012; Fagiani and Christofori, 2013). Except PlGF, which binds to VEGF receptors, all these angiogenic factors signal through different receptors expressed at the membrane of ECs (van Beijnum et al., 2015). This diversity of growth factors extents the toolbox of tumors to create blood vessels. Breast or pancreatic cancers for example rely on these angiogenic factors rather than on VEGFA and are poor responders to bevacizumab (Casanovas et al., 2005). Moreover, preclinical and clinical studies showed that anti-VEGFA antibodies and tyrosine-kinase inhibitors of VEGF receptors stimulate the production of these different growth factors (van Beijnum et al., 2015; Falcon et al., 2016; Haibe et al., 2020).

Preclinical studies for instance demonstrated an increase of SDF1 and PlGF in mice treated with anti-VEGFR2 compounds (Ebos et al., 2007; Fischer et al., 2007). The prolonged use of anti-VEGFR2 antibodies in transgenic mice models of pancreatic cancer stimulated the expression of ANG1 and FGFs. This increase correlated with a shorter survival (Casanovas et al., 2005). Comparable results were reported in head and neck squamous cell carcinoma (HNSCC) xenografts models. Indeed, microarray analysis, showed increased levels of FGF2 and its receptor (FGFR3) in bevacizumab-resistant tumors (Gyanchandani et al., 2013). These results were extended to lung cancer models resistant to angiogenesis-inhibitors, overexpressing EGFRs and FGFRs (Cascone et al., 2011). Clinical studies conducted on bevacizumab-treated colorectal cancer patients evidenced an increase of circulating PlGF, SDF1 and HGF levels (Willett et al., 2005; Kopetz et al., 2010). Equivalent results were obtained for glioblastoma patients treated with cediranib/Recentin^®, a tyrosine kinase inhibitor of the VEGFR1, VEGFR2 and VEGFR3 (Batchelor et al., 2007, 2010). FGF and SDF1 increased expression was correlated with tumor relapse in cediranib-treated glioblastoma patients.

Angiopoietins belong to a family of protein controlling vascular maturation during developmental and pathophysiological angiogenesis (Stratmann et al., 1998; Jeansson et al., 2011; Thurston and Daly, 2012). The predominant angiopoietins are ANG1 and ANG2. ANG1 mediates migration and survival of endothelial cells through binding to Tie2 receptor found on ECs of blood vessels and monocytes, whereas ANG2 promotes cell death and vascular regression (Hanahan, 1997; Maisonpierre et al., 1997). VEGFA and ANG2 promote neovascularization and ANG2 plays a key role in tumor relapse following anti-VEGFA treatment (Asahara et al., 1998). In preclinical models of anti-VEGFR-treated tumors, upregulation of ANG2 stimulates vascular remodeling and sprouting (Crawford and Ferrara, 2009). This observation was supported by clinical studies showing that patients suffering of colorectal cancers who are poor responders to bevacizumab, exhibit high serum levels of ANG2 (Ogawa et al., 2004; Goede et al., 2010). Equivalent results were obtained for melanoma and breast cancer patients treated with anti-angiogenic therapies. Increased serum level of ANG2 correlated with disease progression (Sfiligoi et al., 2003; Helfrich et al., 2009). Preclinical studies recently showed that simultaneous blockade of VEGFA and ANG2 inhibits angiogenesis and tumor growth (Brown et al., 2010; Kienast et al., 2013; Schmittnaegel et al., 2017; Wolf and Langmann, 2019). Clinical trials using such combination are ongoing for the treatment of metastatic colorectal cancers (NCT01688206, NCT02141295). Recently, the vanucizumab, a bispecific anti-ANG2/anti-VEGFA antibody has been evaluated in a phase I study. Vanucizumab displayed acceptable safety profile and encouraging anti-tumor activity (Hidalgo et al., 2018).

Fibroblast-growth factors belongs to a family of 22 cell-signaling proteins involved in a broad variety of processes. FGF binds to tyrosine kinase receptors (FGFRs), expressed on tumor and stromal cells including endothelial cells, cancer-associated-fibroblasts or myeloid cells infiltrating tumors (Beenken and Mohammadi, 2009; Ornitz and Itoh, 2015). The FGF pathway promotes cancer progression and angiogenesis by activating RAS/RAF/MEK//ERK and PI3K/AKT/mTOR pathways (LaVallee et al., 1998; Hart et al., 2000). FGFs and FGFRs up-regulation, are involved in mechanisms of resistance to anti-VEGFA therapy (Casanovas et al., 2005; Kopetz et al., 2010). VEGFR2 inhibitors induce FGF2 expression and accelerate the growth murine pancreatic neuroendocrine tumors (Casanovas et al., 2005). Clinical studies on glioblastoma furthers confirmed this observation (Batchelor et al., 2007; Lee et al., 2019).

The proangiogenic role of FGF and its involvement in resistance to VEGFA inhibitors constitute a strong rationale for the development of inhibitors targeting the FGF and VEGFA pathways. The combined inhibition of FGF2 and VEGFA was highly efficient in preclinical models of head and neck carcinoma or pancreatic tumors (Casanovas et al., 2005; Gyanchandani et al., 2013). FGFR inhibitors notably restore the sensibility to bevacizumab in experimental models in mice suggesting a promising therapeutic combination (Gyanchandani et al., 2013). However, clinical investigations failed to demonstrate the relevance of this association (Norden et al., 2015; Semrad et al., 2017). Lenvatinib, a multiple receptor tyrosine kinase inhibiting the VEGFRs, FGFRs, and PDGFRs has shown promising therapeutic effects against various solid tumors and should be considered for counteracting resistance to anti-angiogenic agents (Suyama and Iwase, 2018).

In the 1970’s several groups demonstrated the existence of growth factors for fibroblasts and smooth muscle cells derived from platelets (Paul et al., 1971; Bowen-Pope and Ross, 1982). These factors were named platelet-derived-growth factors (PDGF) and were one of the first growth factors to be characterized. By binding to their receptors PDGFRs, PDGFs are major mitogens for many cell types and actively participate in angiogenesis (Papadopoulos and Lennartsson, 2018). In cancer, PDGFs exert autocrine loops that stimulate tumor cell proliferation, and paracrine signaling for angiogenesis (Liu et al., 2011; Manzat Saplacan et al., 2017). Upregulation of PDGF was evidenced in glioblastoma patients following anti-angiogenic therapy (Liu T. et al., 2018). The blockade of PDGFR pathway increases the sensibility to VEGFA-neutralizing treatment, giving the rationale for new therapeutic opportunities. However, imatinib, a PDGFR inhibitor, in combination with bevacizumab, failed to demonstrate efficacy in renal cell carcinoma (RCC) patients (Hainsworth et al., 2007; Rock et al., 2007). Despite an increase of PFS, the VEGFRs and PDGFR inhibitor sunitinib is not curative for RCC patients (Motzer et al., 2009).

Based on the redundancy in angiogenic pathways, limited benefits to patients were observed by targeting a single angiogenic growth factor or its receptor. This redundancy is at the origin of innate or acquired resistance, by activation of alternative proliferation/survival pathways. Inhibition of ANG2-, FGF- or PlGF-mediated signaling pathways with those of VEGFA overcomes aspects of resistance to VEGFA blockade, but a sustained inhibition remains to be demonstrated.

The Transforming Growth factor-β (TGF-β) family regulates cell proliferation, differentiation and apoptosis (Massagué, 2000). In tumors, the role of TGF-β is ambivalent with tumor suppressive effects in early stage, thereafter switching toward tumor progression at later stages (Derynck et al., 2001). TGF-β induces the production of extracellular matrix and stimulates tube formation by ECs therefore inducing angiogenesis (Ferrari et al., 2009). Upregulation of TGF-β expression was reported in mice models of glioma resistant to anti-VEGF therapy (Park et al., 2016). Inhibition of TGF-β in hepatocellular carcinoma (HCC) and glioblastoma revealed anti-angiogenic benefit offering the rational to combine anti-TGF-β agents with anti-VEGF (Fransvea et al., 2009; Comunanza and Bussolino, 2017). The combination of galunisertib, a small inhibitor of TGF-β with sorafinib led to durable response in mice models of breast cancer (Holmgaard et al., 2018). TGF-β is also a major inducer of cancer associated fibroblast (CAF) development and fibrosis that are determinant in tumor aggressiveness. Targeting two hallmarks of cancer with one molecule probably explain the therapeutic response.

Combining anti-VEGF/VEGFR therapies to inhibitors of alternative angiogenic pathways appears relevant. However, the toxicity of such approach is an important issue. Treatment targeting concomitantly VEGFR and receptors involved in relapse is another option. One of the best example was the approval of cabozantinib (Cabometyx^®) an inhibitor of VEGFR but also of cMET and AXL to actors involved in relapses after sunitinib treatment. Cabozantinib was approved as a second line treatment for RCC patient experiencing progression on sunitinib (Choueiri et al., 2016). It showed also a better efficacy as compared to sunitinib for RCC patients with poor or intermediate risk (Choueiri et al., 2017).

Hypoxia arises from the combination of high proliferative and metabolic rates with abberant tumor vascularisation with poor oxygen delivery (Semenza, 2014). Beside redundant pro-angiogenic pathways, tumor hypoxia is considered as an “immediate early” response to anti-angiogenic therapy. Although anti-angiogenic therapies reduce and normalize tumor vasculature, limiting tumor hypoxia, alternative theory defends an increased intra-tumor hypoxia (Kerbel and Folkman, 2002; Jain, 2005). Hypoxia plays an important role in resistance to conventional therapies leading to the selection of more aggressive stem cells and a shorter survival (Harris, 2002; Wilson and Hay, 2011; Chen et al., 2018). Indeed, anti-angiogenic agents induce intra-tumoral hypoxia and a concomitant stabilization of the hypoxia-inducible factors 1 and 2 alpha (HIF1/2α). HIF1 is considered as a tumor suppressor whereas HIF2 is considered as an oncogene. HIF1α is a major transcriptional regulator of angiogenic factors. It transactivates hundreds of pro-angiogenic genes, including growth factors (VEGFA, PlFG, FGF-2, PDGF) and their receptors (VEGFRs) (Hirota and Semenza, 2006; Rapisarda and Melillo, 2009). Moreover, HIF1 inhibits the production of anti-angiogenic factors, exacerbating angiogenesis (Hanahan and Folkman, 1996; Laderoute et al., 2000). Hence, HIF1 exerts also potent transcriptional inhibition especially following a long exposure in hypoxic conditions (Hantelys et al., 2019).

Hypoxia and HIF1 activation also trigger EMT and metastasis by regulating the expression of key genes such as c-MET, CXCR4, and lysyl oxidase (LOX), events occurring later as discussed above (Joseph et al., 2018). Moreover, the hypoxic microenvironment generated following anti-angiogenic therapy stimulates β1-integrin expression, a well-known marker of resistance to cancer treatments (Foubert and Varner, 2012) which is consistent with its upregulation in clinical specimens of bevacizumab-resistant glioblastoma (Cordes and Park, 2007). Preclinical studies in mice models of glioblastoma demonstrated also the implication of β1-integrin in resistance to angiogenic therapies (Sidorov et al., 2016). The tumor microenvironment is hypoxic and the active metabolism of tumor cells induces the release of CO2 and lactate (Parks et al., 2013). The effect of hypoxia on tumor metabolism is detailled in the tumor metabolic adaptation part below.

The important role played by HIF in tumor aggressiveness stimulated the development of HIF inhibitors especially HIF2 that has oncogenic properties. Such treatments dissociate the HIF2α HIF1β dimer and consequently inhibit the transcriptional activation of HIF2. This treatment was successfully used in a multi-treated RCC patients (Chen W. et al., 2016). Hence, this treatment combined with classical anti-angiogenic drugs or immunotherapies (see below) is promising and should be further validated (Martínez-Sáez et al., 2017).

Autophagy is a physiological process involving the sequestration of unnecessary or dysfunctional cell components and their degradation in lysosomes (Mizushima, 2007; Janku et al., 2011; David, 2012). In pathophysiological conditions, autophagy is an adaptative response to stress. In cancer, autophagy acts as a double-edged sword by serving as a pro-survival or pro-death process (Mathew et al., 2007). Autophagy plays an important role in enabling tumor cells to overcome harsh conditions arising from the microenvironment following treatment (Chandra et al., 2019). By enhancing the survival of tumor cells, it is indeed now considered as an important mechanism of resistance to cancer drugs (Li et al., 2017; Desantis et al., 2018). Hypoxia-induced autophagy favor the survival of hypoxic tumor cells (Brahimi-Horn et al., 2011). Two mechanisms drive hypoxia-dependent autophagy; the non-selective and the selective autophagy extensively reviewed (Chandra et al., 2019).

A cytoprotective role of autophagy was supported by several preclinical studies using radiation or imatinib as anti-cancer strategies (Miyazawa et al., 2010; Gewirtz, 2014). Resistance to sorafenib in hepatocellular carcinoma was attributed to increased activation of mTOR or Akt pathway triggering autophagy and cell survival (Zhai et al., 2014; Luan et al., 2019). The pro-tumoral role of autophagic processes in mediating resistance to anti-cancer treatments in HCC was highlighted by combining sorafenib to autophagy inhibitors (Shimizu et al., 2012; Lin et al., 2013; Hwang et al., 2015). These preclinical studies gave the proof of concept to initiate clinical trials combining inhibitors of autophagy to sorafenib.

In addition to tumor cells, stromal cells use autophagy as a mechanism of resistance to anti-angiogenic drugs. ECs, the direct targets on anti-angiogenic therapies, are inevitably exposed to drugs via the blood stream. Hence, resistance to sunitinib depends at least, on autophagy processes in ECs (Wu et al., 2020). Sunitinib-resistant RCC display an increased number of lysosomes allowing an enhanced sequestration of the drug which limits its therapeutic activity by isolating the drug from its cytoplasmic targets (Giuliano et al., 2015). The basic pKa of sunitinib induces its lysosomal sequestration., It prevents its accessibility to the tyrosine kinase domains of the receptors targeted by the drug (VEGFR1, 2, 3, PDGFR, CSF1R and cKIT), limiting the efficacy of the treatment.

The updated “Hallmarks of cancer: The Next Generation” includes the deregulation of cellular energetics as a key actor of tumor progression (Hanahan and Weinberg, 2011). Over the last decades, tumor hypoxia, by shaping cell metabolism was demonstrated as a key actor of tumor adaptation to anti-angiogenic therapies. Tumor cell metabolism and angiogenesis are tightly regulated by hypoxia (Semenza, 2014). Several genes involved in glycolysis are under HIF1 control, such as GLUT1, GLUT3, PDK1 or LDHA (Favaro et al., 2011). The more hypoxic the cell, the more glycolysis is used, leading to pyruvate production. Instead of entering the tricarboxylic acid cycle, most of pyruvate is converted to lactate. This excess of lactate diffuses in the extracellular environment and is picked up by oxygenated cells, that revert the lactate to pyruvate and enhance their oxidative phosphorylation (Cassim et al., 2020; Parks et al., 2020). Consequently, their need for glucose decreased, and more glucose is available for the more hypoxic area of tumors (Nakajima and Van Houten, 2013). Following sunitinib treatment, the establishment of this symbiotic loop allows the proliferation of the remaining viable cells despite the dramatic increase of hypoxia following angiogenesis inhibition (Pisarsky et al., 2016).

In addition to low oxygen, increased acidification is also a hallmark of hypoxic tumors. It plays a key role in resistance to anti-cancer therapy (Erra Díaz et al., 2018). While mammalian cells protect their cytosol from acidification through expression of membrane transporters and exchangers such as the Na⁺/H⁺ exchanger (L’Allemain et al., 1985) and the monocarboxylate transporter 1 (Halestrap and Price, 1999), hypoxic tumors have developed additional mechanisms to regulate their pH. In solid tumors, the transcription of carbonic anhydrase (CA) IX is controled by HIF1. CAIX catalyzes the hydration of carbon dioxide (CO₂) into H⁺ and bicarbonate (HCO3^–) which is rapidly uptaken into cell by Na⁺-HCO3^– transporters sustaining alkaline pHi compatible with cell survival (Parks et al., 2013). In bevacizumab-resistant glioblastomas, increased levels of CAIX and of c-MET were observed (Jahangiri et al., 2013). Analysis of bevacizumab-resistant glioblastoma further revealed modifications in the expression of genes regulating cell metabolism, with (i) an increase of glycolysis-involved genes and (ii) a decrease of genes regulating oxidative phosphorylation (Kumar et al., 2013). Soluble CAIX is also correlated with a poor response to bevacizumab in breast cancers (Janning et al., 2019). Moreover, hypoxia leads to AMPK activation, inducing the metabolic switch from glycolysis to oxidative phosphorylation (McIntyre and Harris, 2015). Following anti-angiogenic therapy, tumor metabolism shifts from glycolysis to lipid consumption allowing tumor relapse (Sounni et al., 2014). Several clinical trials combining metabolism-targeting or hypoxia-targeting drugs with anti-angiogenics are ongoing (McIntyre and Harris, 2015). Recently, exciting novel concepts involving dual blockade of angiogenesis and metabolic adaptation have emerged and could revert the resistance to anti-angiogenic drugs (Jiménez-Valerio and Casanovas, 2017).

Recent findings demonstrated that metabolic reprogramming also occurs in TECs. TECs display upregulation of anabolic pathways in comparison to normal ECs. Unbiased meta-analysis revealed that Aldh18a1 and Sqle were consistently induced in TECs raising the possibility to identify specific targetable TECs markers (Rohlenova et al., 2020).

Activation of angiogenic receptors also occurs independently of ligand binding, therefore constituting another mechanism of insensitivity to cancer therapies. This process depends in part on galectins. They belong to a family of carbohydrate-binding proteins displaying high affinity for beta-galactoside (Camby et al., 2006). Galectin-1 is overexpressed in tumors and its expression correlated with metastatic dissemination and immune-escape (Hsu et al., 2013). Tumors refractory to anti-VEGFA treatments exhibit enhanced angiogenesis. Anti-VEGFA treatment and hypoxia increased galectin-1 production. Galectin-1 binds to N-glycans glycoproteins on endothelial cells, including VEGFR2. This binding prolongs the presence of VEGFR2 at the cell surface and promotes angiogenesis without VEGFA binding (Croci et al., 2014; Stanley, 2014). Further clinical investigations are needed to consider galectins as relevant targets for anti-angiogenic therapies.

Matrix metalloproteinases (MMPs) play a key role in angiogenesis and in tumor progression (Deryugina and Quigley, 2010). MMPs can be pro- or anti-angiogenic depending of their categories. On the one hand, MMP-3 and MMP-7 display anti-angiogenic properties (Deryugina and Quigley, 2010). On the other hand, MMP-2 and MMP-9 promote the release of VEGFA from the ECM sustaining angiogenesis (Bergers et al., 2000). MMP-1 induces matrix remodeling and migration of ECs (Chun et al., 2004). Hence, it is now established that MMP inhibitors can induce tumor progression by favoring tumor angiogenesis. Therefore, MMP inhibitors combined to inhibitors of angiogenesis should be considered as a therapeutic option (Winer et al., 2018).

Tumors are a complex association of cancer cells as well as a stromal compartment with cellular and noncellular components. Tumor stroma plays crucial roles in tumor progression and in resistance to treatments. The dense tumor stroma can limit the access of therapeutic agents to their target due to fibrosis, high interstitial pressure and degradation of drugs by stromal enzymes (Valkenburg et al., 2018). The rigid extracellular matrix can reduce blood vessel density, creating a barrier that drugs cannot perfuse (Olive et al., 2009). In parallell, the high interstitial pressure in tumor microenvironment affects drug delivery (Provenzano and Hingorani, 2013). Beside these effects, the cytochrome P450, expressed by fibroblasts, metabolizes toxic molecules including therapeutic drugs contributing to agressive behaviors of tumors (Hirth et al., 2000). In mice models of lung cancer, treatment with bevacizumab led to acquired resistance via upregulation of VEGFA, FGF2 and its receptor FGFR2 and PDGFR in stromal cells (Mitsuhashi et al., 2015). It now becomes evident that cancer therapies should include strategies to target and constrain the tumor stroma. Some agents targeting CXCR4, TGF-β or hyaluronic acid are currently under clinical consideration (Valkenburg et al., 2018).

Cells constituting the tumor environment play a key role in the resistance to angiogenesis inhibitors, especially cancer-associated fibroblasts (CAFs) and pericytes.

Anti-angiogenic therapies normalize vessels but also increase intra-tumoral hypoxia leading to the recruitment of bone marrow-derived cells (BMDCs) (Jain and Duda, 2003). Infiltration of BMDCs in tumors has been linked to tumor progression and angiogenesis for several years (Jain and Duda, 2003). As above-mentioned, anti-angiogenic therapies stimulate the production of pro-angiogenic factors (VEGFA, Angiopoietins, FGFs). However, the stress induced by the treatment stimulates inflammatory pathways involved in the production of cytokines such as SDF1, IL-8 or granulocyte colony-stimulating factor (G-CSF). These cytokines trigger the recruitment of BMDCs that exhibit high plasticity and pro-angiogenic potential limiting the efficacy of anti-angiogenic drugs (van Beijnum et al., 2015).

Metastases can remain for months or years in a quiescent, dormant state, in the tissue they colonized. These micro-metastases constitute a residual disease characterized by the persistence of tumor cells, undetectable by conventional diagnostic techniques. The tumor dormancy can be defined as the lag in tumor growth occurring between primary tumor formation and the appearance of clinically detectable metastases (Yadav et al., 2018). The presence of disseminated tumor cells (DTCs) in bone marrow of prostate-cancer and breast-cancer patients have been reported before the development of overt metastases (Banys et al., 2012; Lam et al., 2014). Three molecular mechanisms characterize tumor dormancy: mitotic arrest, immunological and angiogenic dormancy (Senft and Ronai, 2016). The angiogenic dormancy may explain the reasons why angiogenic therapies simply delay tumor progression. More than 20 years ago, pharmacological inhibition of angiogenesis was found to induce dormancy in several mouse models (Holmgren et al., 1995; O’Reilly et al., 1996). The supposed but unproven “angiogenic switch” is supposed to play a key role in the maintenance of the dormancy, since dormant cells upregulate angiogenesis inhibitors such as thrombospondin-1 (TSP-1) (Senft and Ronai, 2016). Despite the lack of clinical evidences, the “angiogenic switch” of dormant cells has to be considered in cancer relapse following treatment arrest.

Cancer stem cells (CSCs) constitute a small population of cells within tumor exhibiting abilities of self-renewal, differentiation and high tumorigenicity potential. They play a key role in the initiation of cancer and in the metastatic cascade. In 2003, CSCs were first identified in human breast and brain cancers (Al-Hajj et al., 2003; Singh et al., 2003). CSCs express CD44, CD24, CD29, CD90, CD133 and aldehyde deshydrogenase (ALDH1) allowing their identification (Yu et al., 2012). CSCs drive angiogenesis in hypoxia and HIF mediates CSCs proliferation and self-renewal (Tong et al., 2018). CSCs was suggested to give rise to endothelial cells and thus neovascularization processes (Fujita and Akita, 2017). Moreover, CSCs can differentiate in pericytes, supporting tumor vessel function (Cheng et al., 2013).

Their tumor initiating properties and their metastatic potential suggest that CSCs are involved in resistance to therapies. Conventional treatments including chemo- and radiation therapies generate the production of CSCs promoting tumor escape (Chen X. et al., 2016; Li et al., 2016; Liu L. et al., 2018). CSCs are actors of anti-angiogenic resistance. Preclinical studies on experimental models of breast cancers showed that sunitinib and bevacizumab increase the CSCs populations through HIF1 activation (Conley et al., 2012). These results indicate that administration of anti-angiogenic agents accelerate tumor growth by increasing CSCs population. Several CSCs-targeting therapies are currently under development. Inhibition of ALDH1 prevents CSCs enrichment and reduces tumor formation of experimental triple-negative breast cancer and NSCLC in mice (Schech et al., 2015; MacDonagh et al., 2017). Evaluation of CD44, CD133 or Hedgehog inhibitors are currently under considerations for further clinical developments (Shibata and Hoque, 2019).

Epithelial-mesenchymal transition (EMT) defines the acquisition of characteristics of invasive mesenchymal cells by epithelial cells. EMT is implicated in tumor invasion and metastasis and correlates with poor clinical outcome in several solid tumors (Mittal, 2018). During the EMT process, epithelial cells lose their phenotypes, with a downregulation of E-cadherin and α-catenin and acquire mesenchymal markers (N-cadherin, vimentin, fibronectin) leading to cell mobility and invasiveness (Zeisberg and Neilson, 2009). Several signaling pathways induce EMT (TGF-β, Wnt, Notch), by controlling the transcription factors Snail, Slug, ZEB1/2 and Twist (Garg, 2013). Hypoxia and HIF1α are also well-known drivers of EMT. The expression of Twist and Snail, the downregulation of E-cadherin and the induction of vimentin promoting tumor invasiveness, have been reported following anti-angiogenic treatments (Cooke et al., 2012; Maione et al., 2012). Similarly, enhanced invasiveness and growth capacity of glioblastoma and RCC cells have been demonstrated following VEGFA inhibition (Grepin et al., 2012; Lu et al., 2012). This enhanced invasion abilities led to metastatic dissemination, a later step discussed in part 3.

Several studies highlighted the role of the tyrosine kinase receptor c-MET in promoting tumor invasiveness and metastasis in response to anti-angiogenic therapies (Pàez-Ribes et al., 2009; Lu et al., 2012; Sennino et al., 2012).

Although sunitinib and anti-VEGFA decreased tumor volume, invasiveness, hypoxia and EMT markers are increased (Ebos et al., 2009; Mizumoto et al., 2015). In addition, c-MET and the phosphorylated active forms of c-MET also increased as a consequence of treatment-induced hypoxia. The c-MET pathway is one of the most investigated pathways in the field of resistance to anti-angiogenic therapies. Its stimulation through HGF binding, triggers the activation of the RAS/RAF/MEK/ERK, PI3K/AKT/mTOR, and STAT3 pathways promoting tumor growth and invasiveness (Jeon and Lee, 2017). Bevacizumab-treated glioblastoma patients have increased relapse in comparison to bevacizumab-untreated patients. This clinical observation was recently linked to the upregulation of c-MET and phospho-c-MET (Jahangiri et al., 2013). Hence, c-MET is a robust actor of anti-angiogenic resistance by promoting EMT-like phenotype and invasiveness in glioblastoma. This observation has subsequently led to development of c-MET inhibitors. Cabozantinib, a promising multi-kinase inhibitor of c-MET, VEGFR2, and AXL, improves overall survival of RCC patients with bone metastases (Motzer et al., 2018).

Historically, lymphatic vessels were considered as passive participants in metastatic dissemination, only acting as channels for tumor cells transit. Nowadays, it becomes evident that lymphatic vessels have an active role in promoting metastasis. The first pro-lymphangiogenic factors identified more than 20 years ago were the VEGFC and VEGFD that bind to VEGFR3 expressed on lymphatic endothelial cells triggering lymphangiogenesis (Joukov et al., 1996; Yamada et al., 1997). Overexpression of VEGFC and VEGFD increases the number of tumor-associated lymphatic vessels and the incidence of lymph node metastases (Christiansen and Detmar, 2011). Moreover, overexpression of VEGFC and VEGFD is correlated to intra-tumoral lymphatic vessel density, lymph node metastasis and poor outcome in patients with melanoma and breast cancers (Mohammed et al., 2007; van der Schaft et al., 2007). More recently, HGF, c-MET, Tie-2, PDGF and FGF were also identified as pro-lymphangiogenic factors (Christiansen and Detmar, 2011). Immunohistochemical analysis of tumor samples showed that lymphatic vessel invasion (LVI) correlated with lymph node metastasis (Christiansen and Detmar, 2011). Moreover, tumor cells through the expression of chemokine receptors exploit the lymphatic network to form metastases. Indeed, CXCR4 and CCR7 expressed on human breast cancer cells promote metastasis to organs expressing their respective ligands, SDF1 and CCL21 (Müller et al., 2001). CXCR4 is upregulated by hypoxia. Since dissemination to distant organs is governed by the SDF1 gradient, CXCR4/SDF1 antagonists inhibited lymph nodes spreading of cancer cells in experimental tumors in mice (Müller et al., 2001).

Drugs destroying blood vessels stimulate the development of tumor lymphatic vessels contributing to treatment failure. Tumors from sunitinib-treated RCC patients in a neoadjuvant setting exhibit increased lymphatic vessels and increased lymph node invasion. This detrimental effect is explained at least by the stimulation of VEGFC expression following sunitinib administration (Dufies et al., 2017a). Indeed, sunitinib stimulate vegfc gene transcription, mRNA stability and protein production and the subsequent VEGFC-dependent development of lymphatic vessels. Moreover, hypoxia upregulated VEGFC expression (Morfoisse et al., 2014; Ndiaye et al., 2019). Lymphangiogenesis participates in treatment failure and its targeting can be considered in the therapeutic arsenal but only for advanced tumors.

The central role played by VEGFA plus ELR+CXCL cytokines and especially CXCXL8/IL-8 was first documented by Sparmann and Bar-Sagi (2004) in colon cancers. The role of ELR+CXCL and their receptors-CXCR1/2 on tumor cell proliferation, angiogenesis and microenvironment adaptation following anti-angiogenic therapies was highly documented (Vandercappellen et al., 2008). The pro-inflammatory interleukin (IL-1 β stimulates CXCL7 production in RCC models resulting in tumor growth (Grepin et al., 2012; Grépin et al., 2014). CXCL7 is a predictive marker of sunitinib efficacy in RCC (Dufies et al., 2017b). CXCL5 in response to lysosomal sequestration of anti-angiogenic drugs plays also a key role in resistance to anti-angiogenic in renal and breast cancers (Giuliano et al., 2019). Inhibitors of CXCR1 and CXCR2 efficiently inhibit the growth of experimental HNSCC and RCC by decreasing tumor cell proliferation, angiogenesis and inflammation (Dufies et al., 2019).

Beside angiogenesis, new vascular networks are generated by the attraction of endothelial progenitor cells, intussusseptive angiogenesis, vessel co-option and vasculogenic mimicry.

The ultimate consequence of the resistance of anti-angiogenic therapies is the increased rate of metastasis. As developed in the previous parts, anti-angiogenesis therapies lead to (i) intrinsic reprogramming of tumor cells with upregulation of alternative pro-angiogenic pathways, increased of lymphangiogenesis-related genes and processes and initiation of EMT (ii) conditioning the microenvironment, with the recruitment of local- and bone marrow-derived cells or used novel neovascularization modalities. All of these mechanisms lead to increased metastatic rate. Ten years ago, Ebos et al. (2009); Pàez-Ribes et al. (2009) were the first to describe the association between anti-angiogenesis drugs and increased distant metastases. Preclinical models of breast cancers showed that sunitinib enhance lung and liver metastasis (Ebos et al., 2009). Anti-angiogenic treatments can make the host more permissive for metastatic seeding. Sunitinib-treated mice exhibit vascular changes such as reduced pericyte coverage and increased leakiness of normal vessels (Chung et al., 2012; Maione et al., 2012; Singh et al., 2012). Therefore, these systemic actions facilitate the creation of a metastatic niche at distance from the primary tumor.

Increased metastasis rate following anti-angiogenic therapies are highly variable and depends on several parameters such as the type of treatment, the dose and the schedule. Singh et al. (2012) showed that sunitinib enhanced the agressiveness of tumor cells whereas the use of an anti-VEGF antibody did not. Chung et al. (2012) further demonstrated that inhibition of VEGF signaling by antibodies does not promote metastasis, in contrast to small molecule RTK inhibitors at elevated-therapeutic drug dosages. Dosing and scheduling of anti-angiogenic administration can also induce resistance. Short-term and high dose of sunitinib increased growth of breast cancer and enhance liver and lung metastasis (Ebos et al., 2009). In contrast, treatment with low dose of sunitinib did not induce metastasis (Welti et al., 2012).