Activated FGF2 signaling pathway in tumor vasculature is essential for acquired resistance to anti-VEGF therapy

Anti–vascular endothelial growth factor (VEGF) therapy shows antitumor activity against various types of solid cancers. Several resistance mechanisms against anti-VEGF therapy have been elucidated; however, little is known about the mechanisms by which the acquired resistance arises. Here, we developed new anti-VEGF therapy–resistant models driven by chronic expression of the mouse VEGFR2 extracellular domain fused with the human IgG4 fragment crystallizable (Fc) region (VEGFR2-Fc). In the VEGFR2-Fc–expressing resistant tumors, we demonstrated that the FGFR2 signaling pathway was activated, and pericytes expressing high levels of FGF2 were co-localized with endothelial cells. Lenvatinib, a multiple tyrosine kinase inhibitor including VEGFR and FGFR inhibition, showed marked antitumor activity against VEGFR2-Fc–expressing resistant tumors accompanied with a decrease in the area of tumor vessels and suppression of phospho-FGFR2 in tumors. Our findings reveal the key role that intercellular FGF2 signaling between pericytes and endothelial cells plays in maintaining the tumor vasculature in anti-VEGF therapy–resistant tumors.


Identification of activated cell signaling pathways in VEGFR2-Fc-expressing tumors.
To identify activated cell signaling pathways in tumors after chronic exposure to VEGFR2-Fc, RNA-Seq analysis was performed by using Mock and VEGFR2-Fc-expressing tumors. We identified 1351 genes (gene set A) in the Renca model and 950 genes (gene set B) in the B16F10 model that were highly expressed in VEGFR2-Fc-expressing tumors compared with Mock tumors (Fig. 2a). The overlap genes between gene set A and B comprised 293 genes (A ∩ B). Pathway analysis of these 293 genes identified 'FGFR2 ligand binding and activation' pathway (with the lowest p value) (Fig. 2b). In addition, this pathway ranked as the top signaling pathway in each separate gene set ( Supplementary Fig. 3a,b). These results suggest that the FGFR2 signaling pathway was activated in Renca VEGFR2-Fc and B16F10 VEGFR2-Fc tumors as a common acquired resistance mechanism against anti-VEGF therapy.
We then examined mRNA levels of FGF ligands and FGF receptors in the RNA-Seq data. Consistent with the result of the pathway analysis, expression levels of Fgf2 and Fgfr2 mRNA were commonly increased in VEGFR2-Fc-expressing tumors compared with Mock tumors in both the Renca and B16F10 models ( Supplementary Fig. 3c,d). When we examined other genes related to tumor angiogenesis ( Supplementary  Fig. 3e,f), we found no pro-angiogenic receptors and ligands that were consistently upregulated in both models. To confirm the increased mRNA levels of FGF ligands and FGF receptors, we conducted real-time quantitative reverse transcription PCR (RT-qPCR) analysis. Consistent with the RNA-Seq analysis, Fgf2 and Fgfr2 mRNA levels were significantly upregulated in VEGFR2-Fc-expressing tumors compared with Mock tumors in both models. The mRNA levels of some of other FGF ligands and FGF receptors were also significantly upregulated in either the Renca or B16F10 model but not both (Fig. 2c,d). To examine the activation status of FGFR2 and the levels of FGF2 and FGFR2 in Mock and VEGFR2-Fc-expressing tumors, we conducted Western blot analysis of tumor lysates from tumors resected at a volume of ~500 mm 3 . FGFR2 was highly phosphorylated, and FGF2 levels were upregulated in VEGFR2-Fc-expressing tumors, compared with Mock tumors in both the Renca and B16F10 models (Fig. 2e,f), although FGFR2 protein levels were not markedly altered. Thus, the FGF2 signaling pathway is a candidate for involvement in acquired resistance to anti-VEGF therapy.
In contrast, Fgf2 mRNA expression level was not significantly different between Mock and VEGFR2-Fcexpressing tumor cells in vitro, and Fgfr2 mRNA expression was not detected ( Supplementary Fig. 3g,h). This result suggests that the observed increase of Fgf2 and Fgfr2 mRNA in VEGFR2-Fc-expressing tumors in vivo (Fig. 2c,d) may be due to stromal cells which include endothelial cells, pericytes, cancer-associated fibroblasts, and tumor infiltrated lymphocytes rather than cancer cells.

Upregulation of FGF2 in pericytes covering endothelial cells in VEGFR2-Fc-expressing tumors.
To investigate the mechanism of activation of the FGFR2 signaling pathway by upregulated FGF2, we conducted immunofluorescence staining using anti-FGF2 Ab, anti-CD31 Ab (for endothelial cells), and anti-SMA Ab (for pericytes) in the Mock and VEGFR2-Fc expressing tumors. CD31 staining was observed in both Mock and VEGFR2-Fc-expressing tumors (Fig. 3a,b). In contrast, SMA staining and FGF2 staining were barely observed in Renca Mock or B16F10 Mock tumors, but were clear in Renca VEGFR2-Fc and B16F10 VEGFR2-Fc tumors (Fig. 3a,b). The strong FGF2 staining in VEGFR2-Fc-expressing tumors was consistent with the results of Western blot analysis (Fig. 2e,f). Interestingly, SMA-stained cells were adjacent to CD31-stained cells and thus seemed to be  aligned to tumor endothelial cells in the VEGFR2-Fc-expressing tumors (Fig. 3c,d). Moreover, FGF2-positive areas were also adjacent to CD31-positive (CD31 + ) cells, and SMA-positive (SMA + ) cells might be co-localized with FGF2-positive areas in the VEGFR2-Fc-expressing tumors (Fig. 3c,d). These results suggest that the tumor microenvironment was changed by chronic treatment with VEGFR2-Fc and that anti-VEGF therapy resistant tumor vessels were covered with pericytes, which produce FGF2 to activate the FGFR2 signaling pathway.
To further examine which cell types expressed FGF2 and FGFR2 in tumor microenvironments, we isolated CD31 + cells (as endothelial cells), Thy-1-positive (Thy-1 + ) cells (as pericytes), and CD45-positive (CD45 + ) cells (as tumor-infiltrating lymphocytes) that are major components in tumor stroma cells by using microbeads coated with each respective Ab. Total RNA was then collected from the isolated cells, and mRNA levels of Fgf2, Fgfr2, Pecam1 (CD31), Acta2 (SMA), and Ptprc (CD45) were analyzed by RT-qPCR (Supplementary Fig. 4a-d). Although a small portion of CD45 + cells were nonspecfically included by isolation of the anti-CD31 and anti-Thy-1 microbeads, the results confirmed that the microbeads mainly enriched the expected cell population. The isolated Thy-1 + cells showed high expression of Acta2 mRNA as expected for pericytes 20   Fig. 4a-d). Because SMA + cells were mainly adjacent to CD31 + cells in VEGFR2-Fc expressing tumors (Fig. 3c,d), we regarded Thy-1 + cells as pericytes. The isolated CD31 + cells showed significantly higher Fgfr2 mRNA expression in the VEGFR2-Fc-expressing tumors than in the Mock tumors (Fig. 3e,f), whereas the isolated Thy-1 + cells showed significantly higher expression of Fgf2 mRNA in the VEGFR2-Fc-expressing tumors (Fig. 3g,h). For CD45 + cells, there was no significant difference in the levels of Fgf2 and Fgfr2 mRNA between Mock and VEGFR2-Fc-expressing tumors in the Renca tumor model (Fig. 3i). In the B16F10 model, for CD45 + cells, Fgf2 mRNA levels were significantly upregulated, and Fgfr2 mRNA levels were decreased in   www.nature.com/scientificreports www.nature.com/scientificreports/ the VEGFR2-Fc-expressing tumors compared with in the Mock tumors, although the level changes were small (Fig. 3j). These results show that pericytes were the main cell type that produced FGF2 in the VEGFR2-Fcexpressing tumors. Therefore, FGF2 supplied by pericytes may effectively activate FGFR2 in endothelial cells in a juxtacrine manner, and induce an escape pathway from anti-VEGF therapy in the acquired resistant models.

Effects of simultaneous expression of VEGFR2-Fc and FGFR2-Fc on in vivo tumor growth and tumor angiogenesis in the Renca and B16F10 models.
To examine the roles of the FGF signaling in in vivo tumor growth of VEGFR2-Fc-expressing tumors, we established Renca and B16F10 tumor cells overexpressing FGFR2-Fc (the extracellular domain of mouse FGFR2 fused to the Fc region of human IgG4) ( Supplementary  Fig. 5a). Cell culture supernatants of the Mock and FGFR2-Fc-expressing tumor cells were collected to assess the specific binding of FGFR2-Fc to FGF1 and FGF2. FGFR2-Fc protein specifically interacted with FGF1 and FGF2, but not with VEGF ( Supplementary Fig. 5b,c). Furthermore, FGFR2-Fc significantly suppressed FGF2-induced tube formation in a HUVEC sandwich tube formation assay in which HUVECs were co-cultured with FGFR2-Fc-expressing tumor cells ( Supplementary Fig. 5d-g). In a similar assay where tube formation of HUVECs was induced by VEGF plus FGF2, tube formation was partially disrupted by co-culturing with either VEGFR2-Fc-or FGFR2-Fc-expressing tumor cells (Fig. 4a,b), and completely disrupted by co-culturing with a mixture of VEGFR2-Fc-and FGFR2-Fc-expressing tumor cells (Fig. 4a,b; Supplementary Fig. 6a,b). These results indicate that combination of VEGFR2-Fc and FGFR2-Fc inhibited the induction of HUVEC tube formation by VEGF plus FGF2 in vitro.
Next, we examined in vivo tumor growth and tumor angiogenesis by inoculating various combinations of Mock, VEGFR2-Fc-expressing, or FGFR2-Fc-expressing tumor cells into syngeneic mice (Fig. 4c,d) in Renca and B16F10 models. Inoculation with mixtures of Mock and VEGFR2-Fc-expressing tumor cells produced significantly slower tumor growth in vivo than inoculation with Mock cells alone in both models. In vivo tumor growth after the inoculation of mixtures of Mock and FGFR2-Fc-expressing tumor cells was similar to that of Mock cells alone in the Renca model, but was slower than that of Mock cells alone in the B16F10 model. This result suggests that in vivo tumor growth of the B16F10 tumors was partially dependent on the FGF signaling pathway. Inoculation with a mixture of VEGFR2-Fc-and FGFR2-Fc-expressing tumor cells showed further significant delay of in vivo tumor growth compared with inoculation with a mixture of Mock and VEGFR2-Fcexpressing tumor cells (Fig. 4c,d). Consistent with the results of the HUVEC tube formation assay in vitro, tumor angiogenesis was strongly suppressed in tumors formed from a mixture of VEGFR2-Fc-and FGFR2-Fc-expressing tumor cells compared with the other combinations above (Fig. 4e,f). The 2D in vitro tumor growth rates of Mock and FGFR2-Fc-expressing tumor cells were similar to each other ( Supplementary Fig. 6c,d). VEGFR2-Fc specifically interacted with VEGF, and FGFR2-Fc selectively interacted with FGF1 and FGF2, in tumors expressing VEGFR2-Fc or FGFR2-Fc, respectively ( Supplementary Fig. 6e,f). Taken together, these results indicate that simultaneous inhibition of both the VEGF and FGF signaling pathways in tumors caused further delay of in vivo tumor growth by enhancing tumor angiogenesis inhibition compared with inhibition of either VEGF or FGF signaling alone in both the Renca and B16F10 models.

Lenvatinib inhibits tumor growth and angiogenesis in VEGFR2-Fc-expressing tumor cells.
Lenvatinib is a multiple tyrosine kinase inhibitor that shows dual inhibition against VEGFR and FGFR 21,22 . In contrast, sorafenib is a tyrosine kinase inhibitor that targets VEGFR, but not FGFR 23 . In a sandwich tube formation assay with HUVECs induced by VEGF plus FGF2, lenvatinib inhibited tube formation (IC 50 , 18.7 nM), and sorafenib needed higher concentrations to suppress tube formation (IC 50 , 1280 nM) (Fig. 5a). Both lenvatinib and sorafenib needed a much higher concentration (lenvatinib; IC50 > 5 μM, sorafenib; IC50 > 5 μM) to inhibit in vitro proliferation of Renca Mock , Renca VEGFR2-Fc , B16F10 Mock , and B16F10 VEGFR2-Fc tumor cells than to suppress tube formation. Therefore, it is unlikely the in vivo antitumor activity of lenvatinib and sorafenib arises through direct inhibition of proliferation of cancer cells.
We next evaluated the antitumor activity of lenvatinib, aflibercept (a VEGF trap) 24,25 , and sorafenib against Mock or VEGFR2-Fc-expressing tumors in the Renca and B16F10 models. All three compounds significantly inhibited Mock tumor growth in both models (Fig. 5b,c; Supplementary Fig. 7a,b). In contrast, only lenvatinib significantly inhibited the growth of VEGFR2-Fc-expressing tumors (Fig. 5d,e and Supplementary Fig. 7c,d). To evaluate the antiangiogenic activity of lenvatinib, aflibercept, and sorafenib, resected tumors were stained with anti-CD31 Ab. Tumor angiogenesis in Mock tumors was inhibited by all three compounds in both models. In contrast, tumor angiogenesis in VEGFR2-Fc-expressing tumors was clearly inhibited by lenvatinib, although aflibercept and sorafenib appeared to slightly suppress tumor angiogenesis (Fig. 5f,g). To examine the effects that lenvatinib, aflibercept, and sorafenib on phosphorylation of FGFR2 in VEGFR-Fc-expressing tumors, we performed Western blot analysis using lysates of resected tumors. Although aflibercept and sorafenib did not decrease phosphorylation of FGFR2, lenvatinib inhibited phosphorylation of FGFR2 in both models (Fig. 5h,i). These results suggest that activation of the FGF2 signaling pathway plays a role in the escape mechanism that underlies acquired resistance to anti-VEGF therapy.

Discussion
In this study, we demonstrated that activation of the FGF2 signaling pathway is a conserved mechanism for acquired resistance to anti-VEGF therapy in two mouse tumor models. In the VEGFR2-Fc-expressing tumors, FGF2 was expressed in pericytes immediately adjacent to the endothelial cells that expressed FGFR2. Therefore, there might be spatially effective activation of the FGF2 signaling pathway between endothelial cells and pericytes in tumor vasculatures that are resistant to anti-VEGF therapy. Because we observed that Fgfr2 mRNA was significantly and consistently induced in VEGFR2-Fc-expressing tumors in the two tumor models, we focused on this receptor. However, we also found that the mRNA levels of Fgfr3 in the Renca model and Fgfr1 and Fgfr4 Scientific RepoRtS | (2020) 10:2939 | https://doi.org/10.1038/s41598-020-59853-z www.nature.com/scientificreports www.nature.com/scientificreports/ in the B16F10 model were significantly upregulated by chronic expression of VEGFR2-Fc (Fig. 2c,d), consistent with previous reports that FGFR1, 3, and 4 also play a critical role in tumor angiogenesis 26,27 . Lenvatinib inhibits multiple tyrosine kinases, including VEGFR and FGFR, with similar inhibitory activity against several FGFR   www.nature.com/scientificreports www.nature.com/scientificreports/ subtypes (FGFR1-4) 28,29 . Therefore, further analysis is required to determine the role of specific FGFR subtypes in the resistance of endothelial cells to anti-VEGF therapy.
The number of pericyte-covered vessels has been reported to increase following anti-VEGF therapy 30,31 , although the mechanism by which pericytes cover tumor vessels after such therapy remains unknown. Consistent with these reports, after long-term exposure to VEGFR2-Fc, we consistently observed that SMA + cells were co-localized with CD31 + cells in both the Renca and B16F10 models. Because we observed fewer SMA + cells in Mock tumors than VEGFR2-Fc-expressing tumors in both models, these cells, during chronic exposure to anti-VEGF therapy, might be recruited from blood into the tumor microenvironment by secreted factors or might be differentiated from other cell types in the tumor microenvironment. Indeed, mesenchymal stem cells, which derive from bone marrow, are a known source of pericytes 32,33 . Myofibroblasts 34,35 and endothelial cells 36,37 can also differentiate into pericytes. Therefore, to improve anti-VEGF therapy, further analyses are required to understand the underlying mechanism of the recruitment of pericytes into the tumor microenvironment or the differentiation of other cell types into pericytes, and the role of FGF expression in pericytes.
In the differential gene expression analysis, no other receptors or related ligands were consistently upregulated in both Renca and B16F10 models, but Pdgfra mRNA levels were significantly upregulated in both models. Furthermore, Angpt1 and Tek mRNA levels were increased in B16F10 VEGFR2-Fc tumors compared with B16F10 Mock tumors ( Supplementary Fig. 3f). PDGF-PDGFR signaling pathways contribute to the migration of pericytes 38,39 , and the ANGPT1-TEK signaling pathway regulates maturation of blood vessels through crosstalk between pericytes and endothelial cells 40,41 . Therefore, our results suggest that tumor vessels in anti-VEGF therapy-resistant tumors are more mature than those in pre-treatment tumors. The frequency of pericyte-covered vessels increases with anti-VEGF Ab and anti-VEGFR1 Ab therapy, and it has been hypothesized that pericyte-covered vessels are resistant to anti-VEGF therapy [42][43][44] . The crosstalk between pericytes and endothelial cells is regulated by multiple signaling pathways, such as the EFNB2-EPHB4 45-48 , JAG1-NOTCH3 49,50 pathways, as well as the ANGPT-TEK and PDGF -PDGFR 51,52 pathways mentioned above. But, it is still unknown whether selective inhibition of these signaling pathways in either pericytes or endothelial cells can overcome the resistance of pericyte-covered vessels to anti-VEGF therapy. Our study reveals a new role for FGF2 in pericytes, and activation of the FGF signaling pathway in tumor vessels in anti-VEGF therapy-resistant tumors. Even after dual inhibition of VEGFR and FGFR, complete tumor growth inhibition was not achieved. To further combat resistance against antiangiogenic therapy, we might need to suppress other pro-angiogenic pathways as well: e.g., TEK, PDGFR, EPHB, and NOTCH signaling pathways. Our decoy system would be useful for evaluating the contributions of these angiogenic receptors to tumor angiogenesis.
In summary, tumor angiogenesis mainly depends on the VEGF signaling pathway (Fig. 6a). During chronic exposure to anti-VEGF therapy, VEGF-dependent vessels are suppressed. Dependency on the FGF2 signaling pathway for tumor angiogenesis then emerges as the number of tumor vessels covered with pericytes, which overexpress FGF2, increases (Fig. 6b). Therefore, suppression of FGFR in addition to VEGFR inhibition by agents such as lenvatinib would be needed to maximize anti-VEGF therapy (Fig. 6c). Dual inhibition of VEGFR and FGFR is a mode of action unique to lenvatinib: i.e., it is different from VEGF or VEGFR selective inhibitors, antibodies against VEGF or VEGFR2, and VEGF traps. Our findings provide insight into the importance of the FGF2 signaling pathway inhibition for enhancing anti-VEGF therapy, thus improving antiangiogenic therapy.
Immunoprecipitation assay. Cell culture supernatants of Mock, VEGFR2-Fc, FGFR2-Fc, and soluble Fc transfectants were collected in separate Eppendorf tubes. Protein A beads and recombinant proteins (VEGF, FGF1, or FGF2 [R&D systems]) were added to the tubes and rotated overnight at 4 °C. For in vivo immunoprecipitation assays, tumor lysates were prepared from resected tumor tissues. To immunoprecipitate endogenous B16F10 model. Scale bars represent 100 μm. (h,i) Western blot analysis of phosphorylation status of FGFR2 in VEGFR2-Fc-expressing tumors in mice that were non-treated (control) or treated with lenvatinib, aflibercept, or sorafenib (same concentrations as in (b,c)); tumors were resected at ~500 mm 3 Figure 6. Schema of mechanisms of FGF2-driven acquired resistance to anti-VEGF therapy. (a) VEGF induces tumor angiogenesis by activating the VEGF signaling pathway in endothelial cells. Tumor vessels are abnormal in that they are mixtures of large and small tortuous vessels, have impaired function, and are hardly covered with pericytes. (b) VEGF-dependent vessel formation is suppressed by chronic expression of VEGFR2-Fc. However, the activated FGF2 signaling pathway maintains tumor vessels despite chronic inhibition of the VEGF signaling pathway. Tumor vessels covered by pericytes become resistant to VEGFR2-Fc. c Simultaneous expression of FGFR2-Fc and VEGFR2-Fc inhibits activation of the FGF signaling between pericytes and endothelial cells. This further decreases the number of tumor vessels and enhances tumor growth suppression. Thus, dual inhibition of FGFR and VEGFR, such as in lenvatinib treatment, enhances antitumor activity.