Bevacizumab promotes active biological behaviors of human umbilical vein endothelial cells by activating TGFβ1 pathways via off-VEGF signaling

Objective: Bevacizumab is a recombinant humanized monoclonal antibody that blocks vascular endothelial growth factor (VEGF) with clear clinical benefits. However, overall survival of some cancer types remains low owing to resistance to bevacizumab therapy. While resistance is commonly ascribed to tumor cell invasion induced by hypoxia-inducible factor (HIF), less attention has been paid to the potential involvement of endothelial cells (ECs) in vasculature activated by anti-angiogenic drugs. Methods: Human umbilical vein ECs (HUVECs), bEnd.3 cells, and mouse retinal microvascular ECs (MRMECs) were treated with bevacizumab under conditions of hypoxia and effects on biological behaviors, such as migration and tube formation, examined. Regulatory effects on TGFβ1 and CD105 (endoglin) were established via determination of protein and mRNA levels. We further investigated whether the effects of bevacizumab could be reversed using the receptor tyrosine kinase inhibitor anlotinib. Results: Bevacizumab upregulated TGFβ1 as well as CD105, a component of the TGFβ receptor complex and an angiogenesis promoter. Elevated CD105 induced activation of Smad1/5, the inflammatory pathway and endothelial–mesenchymal transition. The migration ability of HUVECs was enhanced by bevacizumab under hypoxia. Upregulation of CD105 was abrogated by anlotinib, which targets multiple receptor tyrosine kinases including VEGFR2/3, FGFR1-4, PDGFRα/β, C-Kit, and RET. Conclusions: Bevacizumab promotes migration and tube formation of HUVECs via activation of the TGFβ1 pathway and upregulation of CD105 expression. Anlotinib reverses the effects of bevacizumab by inhibiting the above signals.


Introduction
Bevacizumab is a recombinant humanized monoclonal antibody targeting VEGF-A that has beneficial clinical effects 1,2 . However, improvements in progression-free survival (PFS) but not overall survival (OS) have been reported in a number of clinical trials [3][4][5] , for instance, in patients with progressive glioblastoma 6 . In another study, bevacizumab failed to reduce tumor growth, instead exacerbating brain tumor invasion in mice bearing glioma 7 , indicating a capacity to stimulate malignant behavior of tumor cells. Glioblastoma and colorectal cancer recurrence, characterized by highly infiltrative behavior, have additionally been documented after bevacizumab treatment [8][9][10] . These results highlight the urgent need to identify "high-risk individuals" prone to progressive disease induced by unregulated anti-angiogenic treatment to avoid unproductive therapy.
Besides malignant cells, another non-negligible factor in resistance to anti-angiogenesis therapy is potentially the endothelial cells (ECs) of micrangium around the tumors. However, it remains to be established whether endothelial cells of vessels undergo a similar phenomenon as neoplastic cells. The potential malignant behavior of ECs has rarely been explored as they are considered mature, gene-stable cells that lack features of malignancy. However, in response to pro-angiogenic signals, ECs become active and motile with protruding filopodia and retain high plasticity, responding to angiogenic tumor expansion 11 . Additionally, tumors can "hijack" pre-existing blood vessels into the vasculature whereby ECs acquire "motivated status" with endothelialmesenchymal transition (Endo-MT) associated with TGFβ-CD105-Smad and Notch signaling to boost neointima formation and leukocyte transmigration [12][13][14][15] . In the current study, treatment of human umbilical vein EC (HUVEC) cells with bevacizumab under hypoxia led to increased migration and tube formation, similar to activated ECs with EMT phenotype in previous reports 16,17 . Therefore, the issue of whether normal vessel ECs can be activated by aberrant stimulation and the underlying mechanisms need further exploration.
TGFβ1 in ECs activates Sma-and Mad-related (Smad) proteins. Smad and JNK signaling in the TGFβ1 pathway promote Endo-MT 13,15 and inflammation in rat peritoneal mesothelial cells 18 . The extracellular and cytoplasmic domains of the auxiliary TGFβ receptor CD105 (endoglin) interact with alk1 in ECs. Alk1 and alk5 (different TGFβ superfamily receptor I types) in ECs are proposed to regulate the balance between proliferation and quiescence 19 , whereby binding of CD105 with alk1-Smad1/5/8 activates ECs in association with Endo-MT 13 . Additionally, elevated CD105 is associated with inflammatory infiltration in vivo 20 and endogenous secreted CCL20 levels are increased in oral cancer cells due to CD105 stimulation 21 . Here, we further focused on elucidating the mechanisms underlying the impact of bevacizumab on downstream inflammatory factors of the TGFβ-CD105 pathway in HUVECs.

In vitro angiogenesis assay
HUVECs were treated with various concentrations of bevacizumab for 24 h under hypoxia conditions. Next, cells were seeded in a 48-well plate pre-coated with 150 μL matrigel (BD Biosciences, Bradford, MA, USA) at a density of 4 × 10 4 cells/ well. After 5 h, images of enclosed tubes were obtained with an inverted phase-contrast microscope (Leica DMI6000B, 50× magnification).

In vivo angiogenesis assay
Female BALB/c-nu mice 6-8 weeks old were purchased from the Model Animal Center of Nanjing University (Nanjing, China). In keeping with a previous protocol 22 , HUVECs (2 × 10 7 cells/mL) were resuspended on ice in phenol redfree matrigel solution, mixed with different doses of bevacizumab (0, 10, and 100 μg/mL) together with 1 μg/mL VEGF (PeproTech), and implanted subcutaneously into BALB/c-nu mice. Mice were divided into three groups intraperitoneally injected with 0, 5, and 50 mg/kg bevacizumab twice a week for 1 month. Images of the matrigel were obtained and fixed with 4% paraformaldehyde for CD105 immunohistochemistry (ab137389, anti-human CD105 antibody does not crossreact with mouse CD105). Experiments were replicated using 4 mice per group. To confirm the efficacy of bevacizumab on endothelial cells, experiments on bEnd.3 cells were additionally performed.

RT-PCR
Total RNA was extracted with TRIzol (Qiagen, Valencia, CA, USA) and cDNA generated by reverse transcription using a first-strand cDNA synthesis kit (TransGen Biotech, Beijing, China), RT-PCR was performed using the TransScript ® RT/ RI Enzyme Mix, 2×TS ReactionMix. After that, quantitative real-time PCR was performed using the TransStart Top Green qPCR SuperMix (TransGen Biotech, Beijing, China). The reactions were performed under the following conditions as suggested by the manufacturer: 94 °C for 30 s, followed by 40 cycles of 94 °C for 5 s and 60 °C for 30 s, followed by a dissociation protocol. Single peaks in the melting curve analysis indicated specific amplicons. Results were expressed as relative fold change calculated using the delta CT method. The primers used in this study are listed in Supplementary  Table S1.

MTT assay
HUVECs were transfected with CD105 siRNA for 24 h, plated into 96-well plates (2,000 cells/well), and incubated overnight with bevacizumab (100 μg/L). Next, 20 μL/well MTT (Solarbio Beijing, 5 mg/mL dissolved in PBS pH 7.4) was added to the plates. After 4 h, 150 μL DMSO was added, followed by shaking for 20 min. The plate was read using a Microplate Reader (Bio-Rad Laboratories, Hercules, CA, USA) at a wavelength of 490 nm.

Statistical analysis
Results are expressed as means ± SD on the basis of triplicate experiments. ANOVA and Student's t-test (two-tailed) were employed for statistical analysis of significant differences between groups. For immunochemistry score analysis, the non-parametric test was used. P values < 0.05 were considered statistically significant.

Bevacizumab promotes HUVEC migration and tube formation in vivo and in vitro
HUVECs were treated with 0-160 μg/mL bevacizumab under hypoxia or normal oxygen conditions for 24 h. Following treatment with 80 or 160 μg/mL bevacizumab in normoxia conditions (Supplementary Figure S1A and S1B), migration of HUVECs was decreased relative to the control group. However, under hypoxia, both concentrations of bevacizumab promoted cell migration to a significant extent ( Figure 1A and 1C, P < 0.05). In view of the finding that migration of HUVECs was activated by both 80 and 160 μg/mL bevacizumab, we employed a fixed concentration of 100 μg/mL for subsequent experiments 23 .
Tube length was significantly greater in the 100 μg/mL bevacizumab group than in the control group ( Figure 1B  vessel formation in matrigel (400 μL) plugs in female nude mice (n = 4 per group) by bev (bevacizumab: 0 μg/mL, 10 μg/mL and 100 μg/mL in matrigel). Mixed matrigel containing HUVECs, bevacizumab, and VEGFA was subcutaneously injected into mice. Mice were intraperitoneally injected with 0, 5, and 50 mg/kg bevacizumab twice a week for 1 month. The image shows matrigel separated from mice, with darker red indicative of higher blood content in vasculature in the gel. CD105 expression (brown: CD105 + , the antibody was only reactive to human endothelial cells) determined via immunochemical assay. The CD105 + stain was stronger in HUVECs treated with high concentrations of bevacizumab than those treated with low concentrations of bevacizumab. (B) Comparison of blood vessel formation in matrigel (400 μL) plugs in female nude mice (n = 4 per group) from control (bevacizumab: 0 μg/mL in matrigel), bev10 (bevacizumab: 10 μg/mL in matrigel), and bev100 (bevacizumab: 100 μg/mL in matrigel) groups. Mixed matrigel containing bEnd.3 cells, bevacizumab, and VEGFA was subcutaneously injected into mice, followed by intraperitoneal injection with 0, 5, or 50 mg/kg bevacizumab twice a week for 9 days. The image shows matrigels separated from mice, with darker red indicative of higher blood content in vasculature in the gel. (C) Histogram displaying immunochemistry scores of CD105 in matrigel containing HUVECs (n = 22 per group, data represent mean ± SD, **P < 0.01, ****P < 0.0001; non-parametric test). 100 μg/mL bevacizumab promoted angiogenesis of HUVECs (Figure 2A) as well as bEnd.3 cells (Figure 2B). To determine the effects of bevacizumab on CD105 expression, immunohistochemical staining was conducted, which revealed upregulation of CD105 in the 100 μg/mL treatment group, compared to the control and 10 μg/mL treatment groups (Figure 2A and 2C, P < 0.01).
In HUVECs, MRMECs and bEnd.3 cells treated with anlotinib for 6 h, followed by bevacizumab for 18 h, upregulation of CD105 was also suppressed (Supplementary Figure S3A-S3C, P < 0.01), indicating that the sequence of anlotinib treatment does not influence its inhibitory effect. Following treatment of HUVECs with anlotinib (5 μM) in hypoxia conditions, migration (Figure 6A and 6C, P < 0.001) and tube formation abilities (Figure 6B and 6D, P < 0.001) were significantly decreased. Sequential treatment of HUVECs with bevacizumab and anlotinib resulted in marked downregulation of tube formation and migration abilities, compared to cells treated with bevacizumab alone (Figure 6B and 6D, P < 0.001). Consistent with our in vitro findings, anlotinib reversed the pro-angiogenic effects of bevacizumab in bEnd.3 cells in vivo (Figure 6E). The blood content was higher in more dense vessel structures in bEnd.3 cell matrigel plugs treated with 100 μg/mL bevacizumab and significantly decreased in vessel structures in matrigel plugs treated with anlotinib.

siRNA targeting CD105 suppresses migration and proliferation of HUVECs and downregulates downstream factors
CD105 was depleted using siRNA, even with bevacizumab stimulation, as evident from western blot analysis (Figure 7A-7C). Downstream factors of CD105 and inflammatory factors, such as CCL20 and IL1B, were additionally decreased (Supplementary Figure S4A-S4C, P < 0.01), along with Endo-MT-related factors, such as Twist, N-cadherin, and Snail (Supplementary Figure S4D-S4F, P < 0.01). Migration of HUVECs treated with 80 μg/mL bevacizumab and positive siRNA under hypoxia conditions was markedly decreased, compared with that in the negative siRNA group (Figure 7D and 7E, P < 0.01). Proliferation of HUVECs was additionally suppressed in both normal and hypoxia conditions (Figure 7F), clearly supporting stimulatory effects of CD105 on both cell migration and proliferation.

Discussion
Bevacizumab has been developed as a key anti-angiogenic agent to reinforce the efficacy of chemotherapy with recognized benefits in the clinic 25,26 . However, a number of studies have reported negative results without significant prolongation of OS 5 . These findings highlight a common problem of single-target anti-angiogenic drugs, i.e., triggering of hypoxia and subsequent activation of tumor cells, initiating interstitial-epithelial transformation and vascular mimicry 27,28 , leading to therapeutic failure. Bevacizumab neither decreased tumor growth nor improved survival of mice bearing orthotopic or endogenous glioma while exacerbating brain tumor invasion 7,8,10,29 . Some tumors develop resistance, even after adequate anti-angiogenic therapy 9,10,27 . In many reports 30 ,  resistance is ascribed to augmentation of HIF, which promotes tumor cell invasion. However, few investigations to date have focused on the potential role of vascular endothelial cells (ECs) in resistance. Tumors can "hijack" and remodel normal vessels, convert normal ECs to "aberrant" cells 31 , and mobilize precursor ECs in the circulation to form vasculatures 32 . Accordingly, enhanced circulating CD105+-activated ECs are indicative of resistance to anti-vascular drugs and tumor development 33 . These findings strongly suggest that ECs can be activated, which contribute to resistance to anti-angiogenic therapy. Results from current study demonstrate that bevacizumab promotes tube formation and migration by HUVECs in hypoxia, and activates the ECs through VEGF-independent pathways. The optimal concentration of bevacizumab determined from the cell experiments was approximately similar to the clinical dose. In the clinic, the plasma concentration of bevacizumab is reported to reach 136.3 μg/mL after treatment with 7.5-15 mg/kg bevacizumab 23 . The conversion is calculated as follows: at a standard patient weight and blood volume of 60 kg and 4 L, respectively, plasma concentration was 60*(7.5 to 15)/4 mg/L = 112.5-225.0 mg/L (i.e., 112.5-225.0 μg/mL). However, increased invasion and metastases for some malignancies have been documented upon administration of the standard dose 34 , suggesting that the standard dose may not benefit every individual and the dose for the whole body may not be always proper rather than excessive concentration in all tumor lesions. We believe that the "relatively excessive concentration" in local lesions due to heterogeneity of tumors activates ECs. However, reduction of the therapeutic dose is not a feasible option. The main purpose of this study was to explore whether "improper treatment" could activate HUVECs, determine the underlying pathways, and identify potential markers on vascular endothelial cells that could be effectively utilized to indicate anti-angiogenic drug resistance.
VEGF-A signaling is the established canonical pathway of angiogenesis. However, other alternative mechanisms exist 35,36 , including PDGF, TGFβ, and FGF pathways. Blockage of VEGF-A in glioblastoma has been shown to increase MET activity in a hypoxia-independent manner, in turn, enhancing tumor invasion 37 . In a murine glioma model, TGFβ activation mediated escape from VEGF inhibition 38 . Experiments from the current study mainly focused on whether HUVECs can be stimulated independently of VEGF via the TGFβ-CD105-Smad pathway.
Migration of HUVECs treated with high concentrations of bevacizumab in normoxia conditions was not increased (Supplementary Figure S1A and S1B). In HUVECs subjected to hypoxia only, migration was not significantly increased ( Figure 1A and 1C) either, clearly indicating that either hypoxia or bevacizumab alone is not sufficient to activate HUVECs, while hypoxia and bevacizumab acted synergistically to promote migration of HUVECs through upregulation of TGFβ1-CD105 (Figures 3A and 5A). Canonically, tumor cell invasion is often attributed to stimulation of hypoxia 39,40 . However, in our experiments, when HUVECs were treated with a high concentration of bevacizumab in normoxia conditions, TGFβ1 was also elevated (Supplementary Figure S1C and S1D), consistent with results obtained under hypoxia ( Figure 5A) and our recent report 41 . Hypoxia alone failed to increase TGFβ1 (Figure 5A), suggesting that signaling pathways other than HIF-α activate ECs. However, we hypothesize that activation of HUVECs is a consequence of synergistic effects based on earlier evidence of elevated TGFβ by hypoxia. TGFβ activation was additionally confirmed in MRMECs and bEnd.3 cells.
CD105, an angiogenesis marker, was elevated by high concentrations of bevacizumab, both in vivo and in vitro, which was validated in MRMECs and bEnd.3 cells (Figure 4C and 4D). We further confirmed that this effect was specifically caused by bevacizumab but not control protein IgG1 (Figure 3F). Knockdown of CD105 was closely associated with downregulation of downstream inflammatory (CCL20 and IL1B) and Endo-MT-related factors (Snail, N-cadherin, Twist) (Supplementary Figure S4), which are responsible for cell migration, adhesion, and vessel formation [42][43][44] . Transfection with siCD105 suppressed HUVEC migration (Figure 7D-7E), suggesting that CD105 is a key contributory factor in endothelial cell activation. Since CD105 plays significant roles in angiogenesis, inflammation, and cancer development 45,46 , our results may partially explain the mechanism underlying vasculature endothelial cell resistance to anti-angiogenesis agents. Figure S2A), termed Endo-MT 12 , play important roles in neointima formation. Endo-MT-derived cells promote tumor development by secreting specific proteins 47 . Moreover, ECs migrate to tumor sites and form vasculatures that favorably promote tumor growth 48,49 . In the present study, in HUVECs treated with high concentrations of bevacizumab, the vessel-like structure became dispersed and discontinuous, termed "co-opted vasculature". This co-opted vasculature has been shown to exacerbate hypoxia to stimulate tumor cells to further release VEGF and enhance resistance to anti-angiogenesis agents 14,50 .

ECs with properties of mesenchymal cells (Supplementary
In our experiments, VEGF-A was completely blocked using bevacizumab (Figure 5G), signifying that activation of HUVECs is not VEGF-A-dependent. To our knowledge, this is the first study to demonstrate that HUVECs can be activated through TGFβ1-CD105-Smad signaling triggered by bevacizumab independently of the VEGF pathway. These results highlight the common shortcomings of single-target drugs, i.e., when one signaling pathway of VEGF was shut down the alternative byways may be stimulated. TGFβ1 can activate downstream CD105 51 and the classical Smad cascade 52 . Concordantly, elevation of activated circulating endothelial cells (aCEC) positive for CD105 is an indicator of NSCLC resistance to anti-angiogenesis and poor prognosis 53 . Our findings are consistent with a previous report that CD105 is upregulated in hypoxia conditions via activation of the MAPK pathway (including p38 and JNK) 54 .
Upregulation of CD105 by bevacizumab was reversed by anlotinib, in keeping with recent results obtained by our group 41 . These results indicate that multi-target drugs can attenuate resistance through suppressing multiple byway signaling pathways initiated by a single-target inhibitor 24,[55][56][57] . Although the mechanisms by which bevacizumab and anlotinib exert their activities on TGFβ remain to be established, previous studies suggest that FGF signaling is activated when VEGF signaling is blocked 58 and that under conditions of bevacizumab-induced decrease in VEGF, angiogenin and bFGF levels are significantly increased 59 . Meanwhile, FGF2 cooperates with TGFβ to promote motility and proliferation of endothelial cells 60 , which may explain the observed bevacizumab-mediated activation. Blockage of FGF signaling by anlotinib 56 could underlie inactivation of HUVECs. Furthermore, anlotinib inhibited tube formation and migration of HUVECs in our experiments.
Our study has a number of limitations that should be taken into consideration. Firstly, the precise mechanisms by which bevacizumab and anlotinib affect TGFβ1 signaling in cells remain unclear. Moreover, expression changes in CD105, TGFβ1, VEGF, HIF-1a and tumor vasculature in the presence of low and high doses of bevacizumab in nude mouse models and clinical specimens require further elucidation. Our future studies will investigate whether bevacizumab interacts directly with TGFβ1.

Conclusions
In conclusion, higher concentrations of bevacizumab (80-160 μg/mL) can activate the TGFβ1-CD105-Smad pathway, promoting migration and tube formation of HUVECs under hypoxia (Supplementary Figure S5). CD105 may serve as a potential marker of resistance to anti-angiogenesis drugs. Anlotinib effectively reverses the effects of bevacizumab.