TGF-β induces oncofetal fibronectin that, in turn, modulates TGF-β superfamily signaling in endothelial cells

Fibronectins (FNs) are high-molecular-mass adhesive glycoproteins of the extracellular matrix (ECM) that are involved in several cellular processes, including cell growth and differentiation, cell adhesion and migration (Hynes, 1990; Pankov and Yamada, 2002). FN is encoded by a single gene (FN1) (Zardi et al., 1982), but various isoforms are generated by alternative splicing mechanisms and post-translational modifications (Schwarzbauer, 1991). Three FN regions can be affected by alternative splicing i.e. the extra-domain A (EDA), the extra-domain B (EDB) and the type III connecting sequence (IIICS) (Pankov and Yamada, 2002; Schwarzbauer, 1991). Specifically, EDA and EDB can be included or excluded when using or skipping over an exon (Pankov and Yamada, 2002). Both the EDA- and EDB-containing isoforms, EDA+FN and EDB+FN, preferentially appear in neo-vessels during embryogenesis and physiological angiogenesis as well as under pathological conditions, including wound healing, atherosclerosis and tumorigenesis (Carnemolla et al., 1989; Castellani et al., 1994; Ffrench-Constant et al., 1989; Rybak et al., 2007; Zardi et al., 1987). Therefore, EDA+FN and EDB+FN are also termed oncofetal FN isoforms. EDB+FN is expressed surrounding the vasculature exclusively during neo-angiogenesis (Castellani et al., 1994), so that antibodies that specifically recognize EDB+FN injected in tumor-bearing mice or cancer patients specifically accumulate in the tumor vasculature (Neri and Bicknell, 2005; Santimaria et al., 2003). The EDB-specific antibody L19 is presently under investigation in clinical trials in cancer patients as one component of fusion proteins with cytokines, such as interleukin-2 (NCT02957019) (Carnemolla et al., 2002; Zegers et al., 2015) or tumor necrosis factor (TNF)-α (NCT02076620) (Borsi et al., 2003; Danielli et al., 2015), and in its radiolabeled form (Borsi et al., 2002; Erba et al., 2012; Poli et al., 2013).

Several genetic studies ascribed a fundamental role of FN in the morphogenesis and physiology of blood vessels (Astrof and Hynes, 2009). Transforming growth factor β (TGF-β) superfamily members are fundamental regulators of endothelial cell physiology, too (Pardali and ten Dijke, 2009). The TGF-β superfamily consists of 33 members, including the three TGF-β isoforms, TGF-β1, TGF-β2 and TGF-β3, and ten bone morphogenetic proteins (BMPs). TGF-β superfamily signaling is triggered by the binding of TGF-β superfamily ligands to members of two functional classes of transmembrane receptors, the type 1 and type 2 receptor Ser/Thr kinases. Upon ligand binding, a heterocomplex between type I and type II receptors is formed, and the type I receptor in turn, phosphorylates receptor-regulated SMAD (R-SMAD) proteins. The latter, together with SMAD4, translocate into the nucleus, where they modulate gene expression depending on the type and state of the cell (Massagué, 2012). The type I receptors ALK4, ALK5 and ALK7 (officially known as ACVR1B, TGFBR1 and ACVR1C, respectively) phosphorylate SMAD2 and SMAD3, whereas ALK1, ALK2, ALK3 and ALK6 (officially known as ACVRL1, ACVR1, BMPR1A and BMPR1B, respectively) phosphorylate SMAD1, SMAD5 and SMAD8. TGF-β signaling is commonly mediated by the ALK5/SMAD2/SMAD3 axis. However, in endothelial cells, TGF-β may also promote activation of SMAD1, SMAD5 and SMAD8 by the endothelium-specific receptor ALK1 (Goumans et al., 2002). Beyond, ALK5 has also been shown to regulate signaling of SMAD1 and SMAD5 (hereafter referred to as SMAD1,5) (Liu et al., 2009). Importantly, TGF-β responses also occur independently of SMAD through the activation of TGF-β-dependent non-canonical signaling pathways, including the phospho-inositide 3′ kinase (PI3K)/Akt pathway (Zhang et al., 2013), the mitogen-activated protein kinase MAPK/ERK pathway, and the p38 and c-Jun N-terminal kinase (JNK) pathways (Moustakas and Heldin, 2005).

Functional interactions between TGF-β and FN have been reported. TGF-β induces FN expression in different cell types (Ignotz and Massagué, 1986). Also, TGF-β has been identified as one of the most important regulators of alternative splicing affecting the EDA+FN and EDB+FN, and leading to an increase in EDA+FN and EDB+FN in several cell types (Balza et al., 1988; Borsi et al., 1990; Viedt et al., 1995). However, FN and its receptor integrin α5β1 positively regulate TGF-β- and BMP-induced SMAD1,5,8 phosphorylation by promoting the formation of a complex between ALK1 and the TGF-β superfamily co-receptor endoglin in human microvascular endothelial cells (Tian et al., 2012). Beyond, in fibroblasts, a role for EDA+FN in influencing TGF-β activity has been suggested (Muro et al., 2008; Serini et al., 1998; White et al., 2008).

TGF-β is a central regulator of the malignant phenotype of glioblastoma, the most common and aggressive malignant primary brain tumor in adults (Rodon et al., 2014; Wang et al., 2016). In glioblastoma, TGF-β has a central role in the induction of angiogenesis and the promotion of vessel remodeling (Dieterich et al., 2012). FN is upregulated within blood vessels of glioblastoma (Dieterich et al., 2012), with EDB+FN expression being abundant and correlating positively with the tumor grade in glioma (Castellani et al., 2002). Here, we extend these investigations on FN in glioblastoma by investigating EDA+FN as well. We also aim at better understanding the interplay of glioblastoma-derived oncogenic TGF-β and oncofetal FN at the tumor-vasculature interface.

We first investigated the expression of total FN and of the EDA+FN and EDB+FN isoforms in ten human glioblastoma sections, and stained non-tumor human brain tissues derived from four patients with epilepsy as controls. We used antibodies IST-5 that reacts with a FN domain common to all FN isoforms, IST-9 that reacts with EDA, and C6 that specifically recognizes EDB+FN (Fig. 1A). In all sections investigated, tumor-associated vessels − identified by staining serial sections for the endothelial cell marker CD31 − showed high levels of FN, EDA+FN and EDB+FN, whereas the tumor tissue itself was negative for EDA+FN and EDB+FN, and showed only low or very low levels of total FN (Fig. 1B, representative sections of two glioblastoma patients). In all non-tumor tissues analyzed, blood vessels stained strongly positive for total FN, but negative or weakly positive for EDA+FN and negative for EDB+FN (Fig. 1B, representative sections of one patient with epilepsy). Accordingly, mRNA analysis of freshly dissociated glioblastoma samples divided into CD31+ and CD31− fractions showed that FN, EDA+FN and EDB+FN were preferentially expressed in the CD31+ fraction containing the endothelial cells (Fig. 1C). Pericytes and vascular smooth muscle cells (VSMCs) express α-smooth muscle actin (αSMA), whose expression is restricted to the tumor blood vessels in glioblastomas (Fig. 1D). Given the vascular distribution of FN, EDA+FN and EDB+FN in glioblastomas we asked whether pericytes/VSMCs are also involved in FN expression in glioblastomas. We, therefore, aimed at a more-detailed expression analysis by performing single-cell real-time polymerase chain reaction (scRT-PCR) of FN, EDA+FN and EDB+FN, and using CD31 and αSMA as markers of endothelial cells, and pericytes and VSMCs, respectively, in cells derived from six different, freshly dissociated human glioblastoma tissues. Of 465 single cells analyzed, 35 were selected as CD31-positive and 164 were selected as αSMA-positive (Fig. S1). Indeed, FN, EDA+FN, and EDB+FN were preferentially expressed in CD31+ vs CD31− cells (Fig. 1E, upper panels) and in αSMA+ vs αSMA− cells (Fig. 1E, lower panels), suggesting that both endothelial cells and perivascular cells contribute to the expression of EDA+FN and EDB+FN in glioblastoma blood vessels. With around half of CD31+ cells also showing αSMA mRNA expression (Fig. S1), we also confirmed that endothelial cells can express αSMA in glioblastoma (Huang et al., 2016). In line, investigating CD31+ cells isolated from freshly dissociated tumor tissues from four other patients, we confirmed expression of αSMA in this ex vivo model. Since it has been reported that EDA+FN controls αSMA expression in fibroblasts (Serini et al., 1998), we performed selective gene silencing, reaching >90% in all models for both targets (data not shown), of EDA+FN or EDB+FN in four CD31+ cell lines derived from human glioblastoma tissue (ZH-464, ZH-483-2, ZH-613, ZH-616). EDA+FN and EDB+FN gene silencing reduced αSMA mRNA levels in all the four cell lines (Fig. 1F), suggesting a role for EDA+FN and EDB+FN in the regulation of αSMA mRNA expression in glioblastoma-derived endothelial cells.

Since TGF-β induces FN (Ignotz and Massagué, 1986) and since alternative splicing of FN, including the generation of EDA+FN and EDB+FN, is mainly regulated by TGF-β in several cell types, we also considered the three TGF-β isoforms, TGF-β1, TGF-β2 and TGF-β3 in our single-cell analysis of the six glioblastoma patients for a correlation analysis in the CD31+ and αSMA+ cell subpopulations. With regard to the CD31+ subpopulation, we observed a positive correlation in the expression of TGF-β1 and FN (r²=0.13, P=3.5e-02), TGF-β1 and EDA+FN (r²=0.22, P=4.3e-03) and of TGF-β3 and FN (r²=0.12, P=3.8e-02) and EDB+FN (r²=0.14, P=2.5e-02) (Fig. 2, upper panels). In the αSMA+ cell fraction, all TGF-β isoforms showed a significant but, overall, less-pronounced positive correlation with the expression of FN, EDA+FN and EDB+FN (Fig. 2, lower panels). Of note, in both CD31+ and αSMA+ cells, we also observed a positive correlation in the expression of FN and of transforming growth factor β-induced protein ig-h3 (TGFBI) (Fig. 2), an extracellular matrix protein that is associated with malignancy and is inducible by TGF-β (Gopal et al., 2017). Relevant proteins for TGF-β storage and activation in the ECM also include the latent TGF-β-binding proteins (LTBPs), comprising isoforms LTBP1, LTBP2, LTBP3 and LTBP4 (Robertson et al., 2015). Therefore, we also included these targets into the analysis, which, in the CD31+ cell fraction, revealed a significant positive correlation of LTBP1 and FN, EDA+FN, and EDB+FN, of LTPB2 and FN and EDA+FN, and of LTBP3 and FN (Fig. 2, top three rows of panel). In the αSMA+ cell fraction, all LTBPs significantly positively correlated with FN, EDA+FN and EDB+FN, except for LTBP3 and EDB+FN (Fig. 2, bottom three rows of panel).

To investigate the potential regulation of EDA+FN and EDB+FN expression by TGF-β isoforms in endothelial cells in vitro, we used human cerebral microvascular endothelial cells (hCMECs). Exposure of these cells to TGF-β1, TGF-β2 or TGF-β3 increased mRNA expression of total FN, EDA+FN and EDB+FN, as well as protein levels of soluble FN, EDA+FN and EDB+FN, as measured in the respective conditioned culture medium of hCMECs (Fig. 3A). We also measured the levels of FN incorporated in the insoluble extracellular matrix and found that the levels of insoluble total FN and EDA+FN were increased, whereas the levels of insoluble EDB+FN were not significantly changed upon TGF-β treatment, suggesting that the EDB+FN isoform is mainly released into solution (Fig. S2). To explore the molecular mechanism(s) by which TGF-β isoforms induce FN expression in hCMECs, we first used the TGF-βRI-specific kinase activity inhibitor SD-208 (Uhl et al., 2004). SD-208 abolished the TGF-β1-, TGF-β2- and TGF-β3-mediated induction of FN, EDA+FN and EDB+FN on both mRNA and protein levels (Fig. 3A), indicating that the process is TGF-βRI-dependent and equally valid for all three TGF-β isoforms. To further specify the signal transduction pathway downstream of the TGF-β receptors, we first investigated the canonical SMAD-dependent TGF-β signaling pathway. Gene silencing of SMAD4, with an efficacy on mRNA levels of ∼88% (data not shown), inhibited TGF-β1- and TGF-β2-induced increases in FN, EDA+FN and EDB+FN on mRNA and protein levels, indicating involvement of the SMAD signaling pathway (Fig. 3B, data for TGF-β1 not shown). Also, gene silencing of SMAD3 but not of SMAD2 (Fig. 3C, left panel) abolished TGF-β1- and TGF-β2-dependent induction of FN, EDA+FN and EDB+FN mRNA expressions (Fig. 3C, data for TGF-β1 not shown). The efficacy of the gene silencing on mRNA levels was ∼90% and ∼60% for SMAD2 and SMAD3, respectively; SMAD2 and SMAD3 gene silencing was also verified on protein levels when the cells were treated with TGF-β (data not shown). Notably, gene silencing of SMAD3 reduced the relative fraction of EDB+FN two-fold, whereas the relative fraction of EDA+FN remained unchanged (Fig. 3C, right panel), pointing towards a specific role for SMAD3 in the control of the splicing of EDB. The inhibitory effect of SMAD3 gene silencing on TGF-β1- and TGF-β2-mediated induction of FN, EDA+FN and EDB+FN also translated into reduced protein levels (Fig. 3C, central panels, data for TGF-β1 not shown). Since a role for JNK in the TGF-β-dependent induction of FN expression has been reported in human fibrosarcoma cells (Hocevar et al., 1999), and in light of the known crosstalk between canonical/SMAD-dependent and non-canonical/SMAD-independent TGF-β signaling pathways, we also explored the role for JNK in our model. The JNK-specific inhibitor SP600125 had only minor inhibitory effects on the TGF-β2-dependent induction of total FN or EDA+FN but abolished the TGF-β2-dependent induction of EDB+FN on mRNA and protein levels (Fig. S3). These data suggest JNK signaling as a potential additional pathway controlling EDB+FN splicing in hCMECs.

To investigate this process in a model more closely reflecting the in vivo situation, we analyzed the effect of TGF-β1 and TGF-β2 on FN in human glioblastoma-derived CD31+ cells. TGF-β1 and TGF-β2 induced the expression of FN, EDA+FN and EDB+FN in the two patient-derived CD31+ cell lines ZH-459 (Fig. 4A) and ZH-613 (Fig. 4B). In line with our results for hCMECs (Fig. 3), TGF-β2-induced expression of FN, EDA+FN and EDB+FN was TGF-βRI- (Fig. 4C) and SMAD3- and SMAD4-dependent (Fig. 4D). In CD31+ cells derived from another glioblastoma patient (ZH-616), TGF-β2 did not induce FN, EDA+FN and EDB+FB, but gene silencing of SMAD3 and SMAD4 even reduced basal levels of EDA+FN and EDB+FN (Fig. 4E).

In endothelial cells, TGF-β may activate signal transduction pathways depending on both SMAD1,5,8 and SMAD2,3, involving ALK1 and ALK5, respectively (Goumans et al., 2002). Further, FN and integrin α5β1 may control phosphorylation of SMAD1,5,8 by promoting the formation of ALK1−endoglin complexes (Tian et al., 2012). To decipher the situation in hCMECs with very low levels of phosphorylated SMAD2 (pSMAD2) under basal conditions, we first performed ALK1 and ALK5 gene silencing. ALK1 gene silencing affected neither the phosphorylation of SMAD1,5 nor that of SMAD2,3 under basal conditions or upon TGF-β stimulation (Fig. S4A). Gene-specific silencing of ALK5 (Fig. S4A), treatment with the TGF-βRI-specific kinase activity inhibitor SD-208 (Fig. S4B) or gene-specific silencing of TGF-βRII (Fig. S4C) inhibited the TGF-β-induced increase in SMAD1,5 and SMAD2,3 phosphorylation levels, and reduced basal levels of phosphorylated SMAD3 (pSMAD3) without affecting basal levels of phosphorylated SMAD1,5 (pSMAD1,5). This implies that, in hCMECs, TGF-β-induced phosphorylation of SMAD1,2,3,5 is TGF-βRII- and ALK5-dependent, and that endogenous TGF-β rather controls basal pSMAD3 levels. Basal pSMAD1,5 levels, however, might be controlled by other members of the TGF-β superfamily, such as BMPs. With these data in mind, we further investigated the potential control of TGF-β superfamily signaling by EDA+FN and EDB+FN in endothelial cells, by performing transient gene silencing of EDA+FN and EDB+FN in hCMECs, with EDA-specific knockdown also affecting the levels of total FN, which points towards the presence of EDA in the majority of the FN molecules (Fig. 5A, left panel). The levels of pSMAD1,5 were reduced in both siEDA+FN and siEDB+FN gene-silenced cells. Cells in which EDA+FN was silenced showed also reduced levels of pSMAD3, whereas the levels of pSMAD2 were unaffected (Fig. 5A). To evaluate the mechanism of the reduction of pSMAD levels by EDA+FN and EDB+FN gene-silencing, we analyzed the levels of TGF-β1 in the conditioned medium of the respective cells. We did not look at the levels of the other TGF-β isoforms because hCMECs show low levels of TGF-β2 and undetectable levels of TGF-β3 (data not shown). TGF-β1 is synthesized as a pro-protein (pro-TGF-β1) of 55 kDa and its processing results in the generation of two molecules; the pro-domain, known as latency-associated peptide (LAP) with a molecular mass of 37 kDa, and the mature TGF-β1 with a molecular mass of 12.5 kDa (Constam, 2014). After pro-TGF-β has been processed, a covalent dimer of the LAP and a covalent dimer of mature TGF-β1 remain non-covalently associated, forming the small latent complex (SLC) with a molecular mass of ∼110 kDa. Both the SLC and the secreted unprocessed pro-TGF-β1 covalently bind through LAP to molecules that belong to the LTBP family, forming the large latent complex (LLC) with a molecular mass of ∼270 kDa. In conditioned medium derived from siEDA+FN cells – and, to a less pronounced extent, also in that from siEDB+FN cells − analyzed by immunoblot under reducing conditions, the levels of pro-TGF-β1 as well as LAP and mature TGF-β1 were reduced (Fig. 5B, left panel; the conditioned medium of hCMECs with the gene silencing of TGF-β1 is shown to identify the correct bands). The analysis of the same conditioned medium under non-reducing conditions revealed that the reduction in TGF-β1 levels in the conditioned medium of siEDA+FN and siEDB+FN cells coincided with reduced levels of TGF-β1/LLC (Fig. 5B, central panel). We also analyzed the respective lysate to detect intracellular TGF-β1 levels and found that, upon EDA+FN and EDB+FN gene silencing, the intracellular levels of pro-TGFβ1 and LAP were unchanged (Fig. 5B, right panel). In agreement, mRNA levels of TGF-β1 were not significantly changed in siEDA+FN or siEDB+FN cells when compared to control cells (data not shown). Taken together, this analysis − differentiating TGF-β1 levels in the lysates and supernatants − suggests that FN and oncofetal FN do not affect TGF-β1 protein secretion but, rather, control the levels of TGF-β1 that are available extracellularly. Since the gene silencing of EDA+FN also affected total protein levels of FN and EDB+FN (see Fig. 5A, left panel), we extended our analysis by using a different approach to confirm the specific involvement of EDA and EDB in the control of SMAD phosphorylation. This was based on the use of recombinant FN fragments including or excluding EDA and EDB (Fig. 5C). Indeed, hCMECs treated with a recombinant FN fragment that contained the EDB and the flanking FN type III domains 7, 8 and 9 (7-EDB-8-9) showed reduced levels of pSMAD1,5 compared with untreated cells or cells treated with the corresponding FN fragment that lacks EDB (7-8-9). The levels of pSMAD2 and 3 were unchanged (Fig. 5D). Treatment with recombinant EDB alone (without the flanking domains 7, 8 and 9) had only minor effects on the levels of phosphorylated SMAD1,5 (pSMAD1,5), indicating the involvement of flanking domains in mediating the EDB effect. Compared to untreated cells or cells treated with the recombinant FN fragments that included the two flanking EDA domains 11 and 12 and lacked EDA (FN fragment 11-12), hCMECs treated with recombinant EDA alone showed decreased levels of pSMAD1,5 whereas the levels of pSMAD2 and pSMAD3 were unaffected. pSMAD1,5 levels were reduced in cells treated with the FN fragment 11-12 to a lesser extent than those in cells treated with EDA alone (Fig. 5D). Upon stimulation with recombinant TGF-β2, FN fragment 7-EDB-8-9 and EDA both inhibited the increase in phosphorylation of SMAD1,5 and, to a lesser extent, that of SMAD2,3 when compared to untreated cells or to cells treated with FN fragments lacking the EDA and EDB (Fig. 5E), confirming that EDA and the EDB are both involved in the modulation of TGF-β superfamily signaling in hCMECs.

The FN splice isoforms EDA+FN and EDB+FN, are strongly expressed in the vasculature and the heart of the developing embryo (Astrof and Hynes, 2009). After completion of developmental processes, both EDA+FN and EDB+FN are downregulated, and are almost undetectable in mature adult blood vessels and tissues, except in tissues undergoing physiological remodeling and angiogenesis, such as uterus and ovaries. During pathological tissue remodeling in chronically inflamed tissues and tumors these oncofetal FN isoforms are re-expressed and accumulate around newly forming vessels (Astrof and Hynes, 2009). In particular, EDB+FN is considered to be a marker of angiogenesis because its expression in the adult blood vasculature is restricted to vessels that undergo neo-angiogenesis (Castellani et al., 1994).

Previous reports have shown that FN is upregulated in glioblastoma vasculature (Dieterich et al., 2012) and that in gliomas the percentage of vessels expressing EDB+FN correlates with the tumor grade (Castellani et al., 2002). Here, we show that in glioblastoma the expression of not only isoform EDB+FN but also of isoform EDA+FN is abundant and restricted to tumor vasculature (Fig. 1B). In addition, we show that both CD31+ and αSMA+ cells derived from freshly dissociated glioblastomas − representing endothelial cells and pericytes/VSMCs, respectively − express FN, EDA+FN and EDB+FN, showing the contribution of both endothelial and perivascular cells to the production of EDA+FN and EDB+FN in glioblastoma blood vessels (Fig. 1C,E). Fibroblasts, which are characterized by expression of αSMA, are rare within the brain. However, a subpopulation of cells that show properties of cancer-associated fibroblasts, expressing αSMA and localizing around blood vessels, has been isolated in glioblastoma (Clavreul et al., 2012). Thus, there might be a minor contribution of αSMA+ cells derived from fibroblast-like cells in addition to the one of pericytes/VSMCs in expressing (oncofetal) FN. Overall (as shown in Fig. 1D and as previously reported by Takeuchi et al., 2010), αSMA shows a vascular distribution pattern in glioblastoma, so that the vast majority of the cells included in the αSMA+ cell subpopulation is blood vessel-associated. Endothelial cells in glioblastoma may undergo endothelial-to-mesenchymal transition and express αSMA (Huang et al., 2016). In line with this, we observed that CD31+ cells derived from freshly dissociated human glioblastoma express αSMA ex vivo. In these cells, gene silencing of EDA+FN and EDB+FN strongly reduced expression of αSMA, suggesting a role for oncofetal FN in promoting the expression of αSMA, thereby favoring endothelial-to-mesenchymal transition in glioblastoma-derived endothelial cells (Fig. 1F). Similarly, a role for EDA+FN in promoting the conversion of fibroblastic precursors into myofibroblasts expressing αSMA has been reported (Muro et al., 2008; Serini et al., 1998; White et al., 2008).

Glioblastomas are characterized by high vascular density and vascular abnormalization with the typical formation of glomeruloid vascular structures (Wen and Kesari, 2008). All TGF-β isoforms and, most abundantly, TGF-β1 and TGF-β2 are expressed in glioblastoma (Frei et al., 2015), and have been attributed a central role in both tumor angiogenesis and vessel remodeling. A direct role for TGF-β2 in regulating the phenotype of glioblastoma vessels, including the expression of components of the extracellular matrix, has been proposed (Dieterich et al., 2012). Indeed, SMAD2−SMAD4 and SMAD3−SMAD4 complexes localize mainly to the vascular and perivascular areas of glioblastomas (Dieterich et al., 2012). TGF-β induces FN in several cell types (Ignotz and Massagué, 1986) and is one of the best-characterized modulators of the alternative splicing affecting EDA and EDB (Balza et al., 1988; Borsi et al., 1990; Viedt et al., 1995). Here, we show a trend of positive correlation in the expression of the three TGF-β isoforms, and EDA+FN and EDB+FN in CD31+ cells derived from freshly dissociated human glioblastomas (Fig. 2). Also, we observed positive correlations regarding the expression of all TGF-β isoforms, and FN, EDA+FN and EDB+FN in the αSMA+ cells fraction (Fig. 2). Notably, our studies reveal that TGF-β1, TGF-β2 and TGF-β3 induce FN, EDA+FN, and EDB+FN in hCMECs (Fig. 3). Here, induction of FN, EDA+FN and EDB+FN by TGF-β involves the SMAD-dependent canonical TGF-β signaling pathway and, specifically, SMAD3 (Fig. 3). In vitro experiments performed in CD31+ cells derived from freshly dissociated human glioblastomas revealed inter-patient variability, since TGF-β induced FN, EDA+FN and EDB+FN only in two out of the three model cell lines derived from these patients (Fig. 4). We did not observe inducibility of FN and/or oncofetal FN by TGF-β in the third primary cell line tested (ZH-616), but endogenous expression levels of oncofetal FN were dependent on SMAD3 and SMAD4. Thus, our data support a role for TGF-β/SMAD3/SMAD4-signaling in (oncofetal) FN expression within endothelial cells (Figs 3 and 4).

In both CD31+ and αSMA+ cells, we also observed a positive correlation in the expression of TGFBI and FN, EDA+FN and EDB+FN (Fig. 2). TGFBI is an extracellular matrix glycoprotein, interacting with FN (Billings et al., 2002), induced by TGF-β and previously reported to be upregulated in glioblastoma vasculature (Mustafa et al., 2012). Strong positive correlations in the expression of TGFBI and FN have been reported in other contexts, such as ovarian cancer (Ahmed et al., 2007). In addition, we observed positive correlations in the expression of FN and LTBPs (Fig. 2), the latter being ECM proteins that bind to FN and that play a central role in the control of TGF-β latency and activation (Robertson et al., 2015). Overall, the positive correlations in the expression of the ECM-localized TGF-β target genes FN1 and TGFBI, as well as in the expression of ECM components involved in TGF-β storage and activation, i.e. FN and LTBPs, suggest a coordinated expression in the main ECM glycoproteins associated with TGF-β activity (Gopal et al., 2017) in glioblastoma vasculature.

The potential role played by oncofetal FN in neo-angiogenesis and vascular morphogenesis is still unclear. EDA and EDB double-null mice die at embryonic stage having several vascular defects (Astrof et al., 2007). By contrast, EDA or EDB single-knockout mice are vital, showing normal vasculogenesis (Fukuda et al., 2002; Muro et al., 2008), and normal physiological and tumor angiogenesis (Astrof et al., 2004). Since EDA and EDB null-mice have phenotypical similarities compared with mice in which different growth factors, including TGF-β1, are knocked out (Kulkarni et al., 1995); a role for EDA and EDB of FN in regulating TGF-β signaling has been postulated (Astrof and Hynes, 2009). In this present work, we demonstrate that gene silencing of EDA+FN and EDB+FN in hCMECs reduces the levels of pSMAD1,5; in addition, pSMAD3 levels were also reduced upon EDA+FN-gene silencing (Fig. 5A). Interestingly, gene silencing of EDB+FN in hCMECs and, to a larger extent, gene silencing of EDA+FN, reduced levels of TGF-β1 in the conditioned medium of these cells (Fig. 5B). This explains the reduction in the levels of pSMAD3, and suggests a role for FN and its splice variants in the control of TGF-β signaling in the ECM by affecting TGF-β latency/activation. Also, treatment of hCMECs with recombinant fragments of FN, including EDA and EDB, reduced SMAD1,5 phosphorylation at both basal levels and upon TGF-β2 treatment, when compared to untreated cells or cells treated with FN fragments lacking EDA and EDB (Fig. 5D,E). The same FN fragments also attenuated induction of pSMAD2,3 upon TGF-β2 treatment. Overall, these data suggest that EDA and EDB of FN modulate TGF-β superfamily signaling in hCMECs. Similarly, a role for EDA+FN in controlling TGF-β activation has been reported in fibroblasts (Muro et al., 2008; Serini et al., 1998; White et al., 2008).

In summary, we show that EDA+FN and EDB+FN are abundantly expressed in glioblastoma vasculature, and that both endothelial cells and pericytes/SVMCs contribute to the expression of oncofetal FN in glioblastoma. In both CD31+ and αSMA+ cells TGF-β expression correlates with EDA+FN and EDB+FN expression. All TGF-β isoforms induce FN, EDA+FN and EDB+FN in hCMECs and CD31+ cells derived from human glioblastoma tissues, and involve SMAD3 and SMAD4. In turn, EDA+FN and EDB+FN modulate TGF-β superfamily signaling mainly affecting the SMAD1,5 signaling branch that also depends on TGF-β superfamily members other than isoforms TGF-β1, TGF-β2 and TGF-β3. Since TGF-β superfamily ligands play a central role in vasculogenesis and angiogenesis, further studies are needed to deepen our knowledge on this new mode of regulation of TGF-β superfamily signaling that involves oncofetal FN in neo-vessels.

Real-time PCR (RT-PCR) was performed using the primers reported in Table S1. mRNA levels were determined by using the ΔC_T method, and ARF1 as housekeeping gene.

Glioblastoma tissues obtained from surgery were immediately dissociated using a papain-based dissociation system (Worthington, Lakewood, NJ). Leukocytes were depleted using anti-human CD45 microbeads (Milteny Biotech, Bergisch Gladbach, Germany). Single-cell quantitative real-time PCR (qRT-PCR) gene expression profiling was performed using C1 Single-Cell Autoprep and BioMark HD instruments (Fluidigm, South San Francisco, CA). Cells were captured on a C1 Single-Cell Preamp IFC (10–17 μm) by using the Fluidigm C1 and capture efficiency was evaluated under an inverted microscope to identify empty sites, sites with debris and multiple cells to exclude them from the final analysis. Preamplified cDNA was generated using the Single Cells-to-CT Kit (Life Technologies, Carlsbad, CA), pooled qPCR primers (Table S1) and Fluidigm STA reagents. Preamplified cDNA was then used for high-throughput qPCR measurement of each amplicon using the BioMark HD system with IFC Controller HX (Fluidigm) and 2× SsoFast EvaGreen Supermix with Low ROX (Bio-Rad, Hercules, CA). Single-cell expression data were collected using the Fluidigm Data Collection software. Quality controls included cDNAs derived from a whole glioblastoma tumor before and after CD45+ depletion and the qPCR Human Reference cDNA, random-primed (Clontech, Mountain View, CA). For all tumors tested, runs were performed successfully. All primers used were tested for amplification efficiency of at least 90% by generating standard curves for each gene. Data were processed by removing cells with low overall expression values, identification of limits of detection for each analyzed gene and cell-to-cell median normalization as described previously (Livak et al., 2013). Informed consent was obtained from all patients and the analysis was performed according to the guidelines of the local ethics committees (Kantonale Ethikkommission Zürich, Switzerland, KEK-ZH-Nr./BASCE-Nr. 2016-00456).

For the detection of soluble FN, EDA+FN and EDB+FN, BRAND Immunograde 96-well plates (Sigma-Aldrich, St. Louis, MO) were coated with 20 µg/ml gelatin in PBS overnight at room temperature, washed with PBS, blocked with 2% bovine serum albumin (BSA) in PBS and then incubated with the conditioned culture medium of cells cultured for 72 h. The conditioned medium was tested at different dilutions in 2% BSA to ensure that the resulting readings were in the linear range. FN, EDA+FN and EDB+FN were detected by using the IST-5 (Klein et al., 2003), IST-9 (Borsi et al., 1987) and C6 (Balza et al., 2009) antibody, respectively, at the final concentration of 2 µg/ml in 2% BSA in PBS. For the detection of insoluble FN, EDA+FN and EDB+FN cells cultured for 72 h in 96-well plates were fixed with methanol, blocked with 2% BSA in PBS and incubated with 0.5 µg/ml IST5 or 2 µg/ml IST9 or C6. The HRP goat anti-mouse IgG (minimal cross-reactivity) antibody (cat. no. 405306, lot no. B167518, BioLegend, San Diego, CA) was used as secondary antibody. 3,3′, 5,5′ tetramethylbenzidine (TMB) (BD Bioscience, San Jose, CA) was used as a reaction substrate. Data for soluble FN, EDA+FN and EDB+FN are expressed as the value of absorbance at 450 nm divided by the protein concentration as determined in undiluted cell conditioned medium using the Bradford Assay (Bio-Rad). In case of insoluble FN, cells seeded in a second 96-well plate and subjected to the same treatments for 72 h, were incubated for 3 h with 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide (MTT). After cell lysis the absorbance at 540 nm was measured and used to normalize the ELISA data.

For immunohistochemical analysis of FN and CD31, methanol-fixed 5 µm cryostat sections of ten human glioblastomas and of four human normal brain tissues derived from patients with epilepsy were used. Analyses were performed according to the guidelines of the local ethics committees (Kantonale Ethikkommission Zürich, Switzerland, KEK-ZH-Nr./BASCE-Nr. 2016-00456). For staining of fibronectins, sections were treated with the above described primary antibodies IST-5, IST-9 and C6 at the final concentration of 2 µg/ml, the peroxidase mouse IgG Vectastain ABC kit (cat. no. PK-4002, Vector Laboratories, Burlingame, CA) and the DAB substrate (Peroxidase blocking, Dako-Agilent, Santa Clara, CA). CD31+ cells were stained using the anti-CD31 monoclonal antibody clone JC70A (cat no. M0823, lot no. 00063340, Dako, Glostrup, Denmark) diluted 1:40, the peroxidase mouse IgG Vectastain ABC kit (Vector Laboratories) and the DAB substrate (Dako-Agilent). For the αSMA IHC analysis, 5 µm formalin-fixed paraffin-embedded sections derived from two out of the six glioblastoma tissues employed also for the single-cell RT-PCR were used. Heat-induced epitope retrieval was performed in 10 mM Tris buffer pH 9.0, 1 mM EDTA. Tissue sections were then treated with the anti-αSMA monoclonal antibody clone 1A4 (cat. no. A5228, batch 056M4828V, Sigma) diluted 1:50, the Histofine simple stain MAX PO (M) universal immuno-peroxidase polymer anti-mouse (cat. no. 414132F, Nichirei Biosciences, Tokyo, Japan) and the DAB substrate (Dako-Agilent). Serial sections were stained for CD31 as above-described after heat-induced epitope retrieval in 0.1 M citrate buffer pH 6.0. Images were acquired using an Axio Scope.A1 microscope equipped with an AxioCam MRc camera and the AxioVision LE64 program (Carl Zeiss, Oberkochen, Germany).

Denatured and reduced (with 5% β-mercaptoethanol) whole-cell lysates or denatured concentrated conditioned media under reducing or non-reducing conditions were separated by polyacrylamide gels and transferred to nitrocellulose membranes (Amersham/Ge Healthcare Life Science, Arlington Heights, IL). Membranes were blocked with 5% skimmed milk or serum bovine albumin (ApplChem, Darmstadt, Germany) and treated with the following primary antibodies: IST-4 (Sekiguchi et al., 1985), IST-9 and C6 (see above) antibodies to fibronectin used at the final concentration of 5 µg/ml; the antibodies specific for SMAD1 (cat. no. 9743, ref: 01/2012), phosphorylated SMAD1/5 [pSmad1/5(Ser463/465)] (cat. no. 9516, ref. 08/2015), SMAD2 (cat. no. 3122, ref: 05/2015), phosphorylated SMAD2 [pSMAD2 (Ser465/467)] (cat. no. 3108, ref: 04/2016), SMAD3 (cat. no. 9513, ref: 07/2014), and SMAD4 (cat. no. 9515, ref: 08/2013) from Cell Signaling Technology (Danvers, MA), used diluted 1:1000 except for anti-pSMAD2 which was diluted 1:500; anti-phosphorylated SMAD3 (cat. no. ab52903, lot no. GR128879-24, Abcam, Cambridge, UK) diluted 1:1000; anti-human LAP/TGF-β1 (cat. no. AF-246-NA, lot no. EF0212101, R&D Systems, Minneapolis, MN) diluted 1:2000, anti-TGF-β1 (cat. no. G122A, lot no. 0000051912, Promega, Madison, WI) diluted 1:1000, anti-GAPDH (cat. no. EB07069, lot no. C2, Everest Biotech, Ramona, CA) at the final concentration of 0.2 µg/ml. Secondary antibodies used were HRP-coupled goat anti-rabbit (cat. no. sc-2004, lot no. B2216) or donkey anti-goat (cat. no. sc-2033, lot no. A2914) (both Santa Cruz Biotechnology, Dallas, TX) or sheep anti-mouse antibodies (cat. no. NA931V, lot no. 9997907, GE Healthcare UK Limited, Amersham, UK), diluted 1:5000. Protein bands were visualized with horseradish peroxidase (HRP)-coupled secondary antibodies followed by enhanced chemiluminescence (Pierce/Thermo Fisher, Madison, WI).

Human cerebral microvascular endothelial cells (hCMEC-D3) were kindly provided by P. O. Couraud (Institut Cochin, Paris, France). The human glioblastoma-derived CD31+ cell lines ZH-459, ZH-464, and ZH-483-2 have been reported previously (Krishnan et al., 2015). The human glioblastoma-derived CD31+ cell lines ZH-613 and ZH-616 were prepared as described (Krishnan et al., 2015). Endothelial cells were cultured in EBM-2 medium (CC-3156, Lonza, Walkersville, MD), containing 0.1 M HEPES (Gibco), 1% v/v CD lipid concentrate (Gibco) and endothelial growth supplements (EGM-2-CC4176, Lonza). HCMEC-D3 were negatively tested for contamination with mycoplasma. Glioblastoma-derived cell lines were not tested for contamination and were not recently authenticated because they were used at low passage number. Informed consent was obtained from all patients.

Transient gene silencing was performed by using the Lipofectamine RNAiMAX reagent (Invitrogen/Life Technologies) and the following small interfering (si)RNAs provided by Dharmacon (Lafayette, CO): ON-TARGETplus, siRNA SMART pools specifically recoginzing SMAD2 (L-003561-00), SMAD3 (L-020067-00), SMAD4 (L-003902-00), TGF-βRII (L-003930–00), ALK1 (L-005302–02), ALK5 (L-003929–00), TGF-β1 (L-012562-00) and non-targeting control (D-001810-10) at a final concentration of 100 nM. For EDA+FN and EDB+FN the following siRNAs provided by Dharmacon were used at the final concentration of 10 nM: (1) EDA+FN: pool of siEDA1 5′-GGTTCTGAGTACACAGTCA-3′ and siEDA2 5′-GGTTCTGAGTACACAGTCA-3′; (2) EDB+FN: a pool of four siRNAs previously described (Khan et al., 2005).

Cells were treated by using the following reagents in EBM-2 medium without supplements: 5 ng/ml TGF-β1, TGF-β2 or TGF-β3 (R&D Systems, Minneapolis, MN), 1 µM of the TGF-βRI-specific kinase activity inhibitor SD-208 (Scios, Fremont, CA) and 10 µM of the JNK-specific inhibitor P600125 (Sigma-Aldrich). The human fibronectin recombinant fragments used in this study have been previously described (Carnemolla et al., 1996).

All experiments were performed at least twice and in triplicates. Statistical analysis for single-cell real-time (scRT)-PCR data was performed in the statistical programming language R using the package ggplot2. All other statistical analysis was performed by using the GraphPad Prism 5 program. Data are shown as the mean±the standard deviation (s.d.). The statistical significance of the RT-PCR and ELISA data was determined by performing one-way ANOVA followed by Tukey's post hoc test at 95% confidence interval (CI).

We thank Prof. Elisabeth Jane Rushing from the Institute of Neuropathology of the University Hospital of Zurich for providing glioblastoma tissues.