The Robo4–TRAF7 complex suppresses endothelial hyperpermeability in inflammation

Blood vessels function as a barrier to separate the blood from organ tissues. Endothelial cells (ECs) on the inner surface of the vasculature play essential roles in the maintenance and regulation of this barrier function (Mehta and Malik, 2006; Weber et al., 2007). Endothelial permeability is regulated by various molecules such as vascular endothelial growth factor (VEGF), lipopolysaccharide (LPS) and tumor necrosis factor α (TNFα). These factors alter the cytoskeleton and cell adhesion molecules in ECs and regulate angiogenesis and immune cell trafficking to maintain vascular homeostasis. However, excessive inflammatory cytokines in inflammatory diseases induce extreme vascular permeability and lethal symptoms such as septic shock and pulmonary edema (Madge and Pober, 2001; Zhu et al., 2012).

Roundabout guidance receptor 4 (also known as Roundabout homolog 4, Robo4) is an endothelial-specific receptor that is localized in both the plasma membrane and the cytoplasm (Huminiecki et al., 2002; Okada et al., 2014, 2008, 2007; Park et al., 2003; Sheldon et al., 2009; Zhang et al., 2016). Previous studies using Robo4-knockout mice have shown that Robo4 is not essential for developmental vasculogenesis and angiogenesis, but suppresses pathological angiogenesis (Jones et al., 2008; Koch et al., 2011; Zhang et al., 2016). In pathological angiogenesis, Robo4 and its binding proteins suppress VEGF-induced vascular permeability by attenuating VEGF signaling. The N-terminal domain of Robo4 interacts with Unc-5 netrin receptor B (Unc5B) and Slit guidance ligand 2 (Slit2). The binding of Robo4 to the transmembrane receptor Unc5B induces Unc5B downstream signaling, which results in suppression of VEGF receptor 2 (VEGFR2, also known as KDR) phosphorylation, and of vascular hyperpermeability (Koch et al., 2011; Suchting et al., 2005; Zhang et al., 2016). The binding of Robo4 to Slit2 also suppresses VEGF-induced cell migration by modulating the functions of Robo1 and GTPases, including Rac1 and Cdc42 (Enomoto et al., 2016; Jones et al., 2008; Kaur et al., 2006; Sheldon et al., 2009). Further, the C-terminal domain of Robo4 regulates cell migration by interacting with the focal adhesion-associated protein paxillin and the cytoskeleton-regulating protein enabled homolog (Enah) and Wiskott-Aldrich syndrome protein (Jones et al., 2009; Park et al., 2003; Sheldon et al., 2009).

Although the relationship between Robo4 and VEGF signaling has been well studied, Robo4 functions in inflammation are not well understood. A previous study suggested a potential role of Robo4 in inflammation by showing that Slit2 suppresses vascular permeability in mouse sepsis and influenza infection models (London et al., 2010). This study provided an important insight that Robo4 could be a potential therapeutic target in infectious and inflammatory diseases related to vascular hyperpermeability. However, the study was performed following administration of exogenous Slit2, which has been shown to regulate endothelial cell (EC) function in a Robo4-independent pathway (Rama et al., 2015). Thus, to clearly elucidate the physiological roles of Robo4 in inflammation, Robo4 needs to be evaluated in inflammatory models without exogenous Slit2.

In this study, we analyzed Robo4 functions in inflammation using in vivo and in vitro inflammatory models. We successfully demonstrated that Robo4 decreased vascular leakage and improved survival in endotoxemic mice. Moreover, Robo4 suppressed endothelial hyperpermeability in inflammation through cooperation with a newly identified Robo4-binding protein, TNF receptor-associated factor 7 (TRAF7). Our findings indicate a new regulatory mechanism for inflammatory hyperpermeability via the Robo4–TRAF7 complex and suggest that this complex could be used as a therapeutic target for the treatment of inflammatory diseases.

To analyze the function of Robo4, Robo4^−/− mice were generated by replacing exons 2 and 3 with a neomycin resistance cassette (Fig. 1A). Generation of Robo4^−/− mice was confirmed using PCR genotyping (Fig. 1B). Robo4^−/− mice showed no obvious pathological phenotype and were able to breed. In endotoxemia model mice, Robo4 deficiency was associated with significantly lower survival compared with Robo4^+/+ mice (Fig. 1C). To investigate the mechanism modulating this lower survival rate accompanying Robo4 deficiency, the vascular permeability of Robo4^−/− and Robo4^+/+ mice was analyzed with Evans blue dye (Fig. 1D). Compared with Robo4^+/+ mice, Robo4^−/− mice injected with PBS showed increased extravasation of the dye in organs. A significant increase was observed in the lungs, in which Robo4 is highly expressed (Okada et al., 2007; Park et al., 2003). Similarly, Robo4^−/− mice injected with LPS showed increased extravasation of the dye compared to Robo4^+/+ mice injected with LPS. In particular, significant increases in extravasation were observed in the heart, lungs and small intestine. Taken together, these results indicate that Robo4 depletion decreases the survival rates of mice with endotoxemia, possibly by increasing vascular permeability in the heart, lungs and small intestine. This suggests that Robo4 suppresses inflammatory mediator-induced vascular hyperpermeability.

To investigate whether Robo4 regulated endothelial hyperpermeability induced by inflammatory mediators, we employed an in vitro model using human umbilical vein endothelial cells (HUVECs) and TNFα, which is known to induce vascular hyperpermeability in sepsis (McKenzie and Ridley, 2007; Mong et al., 2007; Qiu et al., 2011). We first examined the effects of Robo4 knockdown on TNFα-induced hyperpermeability. HUVEC monolayers transfected with control siRNA (siCont) or siRNA against Robo4 (siRobo4) were treated with TNFα, and endothelial permeability was analyzed using measuring transendothelial electrical resistance (TEER; Fig. 2A). TEER was observed to increase immediately after TNFα treatment in HUVECs transfected with both siCont and siRobo4. This increase could be caused by TNFα-induced upregulation of cAMP and S1P, as suggested in previous reports (Pober et al., 1993; Xia et al., 1998). Transfection with siRobo4 induced significantly lower TEER than siCont. In addition, transfection with siRobo4 did not affect EC viability in the WST-8 assay (Fig. S1A). These results indicate that Robo4 knockdown increases endothelial permeability without affecting cell viability.

We next investigated the effects of Robo4 overexpression on endothelial permeability. HUVEC monolayers infected with adenoviral vectors encoding AcGFP as a control (Ad-Cont) or Robo4 (Ad-Robo4) were stimulated with TNFα, and TEER was measured (Fig. 2B; Fig. S2A). Ad-Robo4 significantly suppressed the decrease in TEER induced by TNFα compared with Ad-Cont, indicating that Robo4 overexpression suppressed endothelial permeability. In addition, Ad-Robo4 completely cancelled the enhanced hyperpermeability mediated by siRobo4 (Fig. 2C). Taken together, these results indicate that Robo4 suppresses TNFα-induced endothelial hyperpermeability.

To investigate the mechanisms through which Robo4 suppressed TNFα-induced hyperpermeability, the subcellular localization and expression levels of the adherens junction protein VE-cadherin (also known as CDH5) were analyzed. VE-cadherin is normally localized at junctions between ECs, but redistributed upon TNFα stimulation to increase permeability (Nwariaku et al., 2002). On observation of immunofluorescence staining using HUVECs transfected with siCont, VE-cadherin was shown to be located at junctions, and TNFα treatment led to redistribution of VE-cadherin from junctions (Fig. 2D). Transfection with siRobo4 decreased VE-cadherin localization at junctions between ECs, both in HUVECs treated with or without TNFα. In addition, treatment with siRobo4 did not alter total and cell surface levels of VE-cadherin in HUVECs, but slightly decreased the total VE-cadherin level in HUVECs treated with TNFα (Fig. 2F; Fig. S3). In contrast, transfection with Ad-Robo4 increased the amount of VE-cadherin localized at junctions and suppressed TNFα-induced redistribution of VE-cadherin in HUVECs compared with cells treated with Ad-Cont (Fig. 2E). These results indicate that Robo4 suppresses endothelial hyperpermeability by stabilizing VE-cadherin at junctions.

Since destabilization of VE-cadherin has been reported to induce extravasation of immune cells (Corada et al., 1999; Gotsch et al., 1997), we investigated whether Robo4 regulates immune cell trafficking. Transfection with siRobo4 significantly increased transmigration of monocytic U937 cells through a HUVEC layer treated with TNFα (Fig. 2G). In contrast, siRobo4 transfection led to decreased expression of intercellular adhesion molecule-1 (ICAM-1) and E-selectin by decreasing the TNFα-induced nuclear localization of the NF-κB p65 (also known as RELA) and p50 (also known as NFKB1) subunits (Fig. S4). These results suggest that Robo4 suppresses extravasation of immune cells by stabilizing VE-cadherin at junctions without inhibiting the NF-κB pathway.

To investigate the essential domain of Robo4 for the inhibition of hyperpermeability and redistribution of VE-cadherin, we prepared Robo4 mutants lacking N- or C-terminal domains (Fig. 5A; Fig. S2B,C). Adenoviral expression of the Robo4 mutant lacking the N-terminal domain (ΔN28–447), as well as wild-type Robo4, inhibited TNFα-induced hyperpermeability and redistribution of VE-cadherin (Fig. 3A,C). In contrast, both of the Robo4 mutants lacking the C-terminal domain (ΔC785–1007 and ΔC570–1007) failed to inhibit hyperpermeability and redistribution of VE-cadherin (Fig. 3B,D). Taken together, these results indicate that the C-terminal, but not the N-terminal, domain of Robo4 is essential for inhibition of TNFα-induced endothelial hyperpermeability and redistribution of VE-cadherin.

To investigate how Robo4 regulates endothelial permeability via its C-terminal domain, Robo4-interacting proteins were purified from HUVECs infected with Ad-Cont or Ad-Robo4-FLAG by means of immunoprecipitation using anti-FLAG antibodies. The precipitated proteins were enzymatically digested, and the resulting peptides were analyzed using mass spectrometry. These data identified TRAF7 as a novel Robo4-binding protein (Fig. 4A). The interaction between Robo4 and TRAF7 was confirmed through co-immunoprecipitation using COS-7 cells (Fig. 4B). Taken together, these results indicated that Robo4 interacts with TRAF7 in ECs.

To identify the Robo4 domain that interacts with TRAF7, co-immunoprecipitation assays were performed with Robo4 mutants (Fig. 5A,B). Robo4 mutants lacking the N-terminal domain (ΔN28–228 and ΔN28–447) showed similar or strong binding to TRAF7 compared with wild-type Robo4. In contrast, Robo4 mutants lacking the C-terminal domain (ΔC785–1007, ΔC711–1007, and ΔC570–1007) showed weaker binding to TRAF7 than wild-type Robo4; ΔC570–1007 completely lost its binding affinity for TRAF7. These results indicate that Robo4 interacts with TRAF7 via its C-terminal domain.

We next investigated the effects of the interaction between Robo4 and TRAF7 on the subcellular localization of each protein in HUVECs using immunofluorescence staining (Fig. 5C). Robo4–FLAG was localized in the cytoplasm, particularly around the nucleus. Additional expression of TRAF7–myc did not alter this localization. In contrast, the additional expression of Robo4–FLAG shifted localization of TRAF7–myc from cytoplasm to the perinuclear region (Fig. 5D) and caused TRAF7–myc to colocalize with Robo4–FLAG (Fig. 5C). This Robo4-mediated alteration of TRAF7 localization was not observed with Robo4 mutants lacking the C-terminal domain. Similar results were obtained in HUVECs treated with TNFα (Fig. S5). Taken together, these findings indicate that Robo4 interacts with TRAF7 via the C-terminal domain and regulates TRAF7 localization, further suggesting that Robo4 modulates TRAF7 function by regulating its localization.

To investigate TRAF7 functions in ECs, we analyzed the effects of TRAF7 knockdown on endothelial permeability using siRNA against TRAF7 (siTRAF7). siTRAF7 transfection resulted in a significantly lower TEER in HUVECs than siCont transfection, both before and after TNFα stimulation (Fig. 6A). In contrast, infection with an adenoviral vector encoding TRAF–myc (Ad-TRAF7-myc) suppressed the TNFα-induced decrease in TEER (Fig. 6B). In addition, treatment with siTRAF7, unlike with siRobo4, reduced cell viability in the WST-8 assay (Fig. S1B). These results indicate that TRAF7 suppresses TNFα-induced endothelial hyperpermeability and that it is important for EC viability.

We next analyzed whether TRAF7 affects the localization and the protein expression of VE-cadherin. Through immunofluorescence staining, we observed that siTRAF7 decreased VE-cadherin localization at cell junctions and induced intercellular gaps between HUVECs before and after TNFα stimulation (Fig. 6C). In contrast, Ad-TRAF7-myc treatment increased VE-cadherin localization at junctions, particularly in HUVECs treated with TNFα (Fig. 6D). In addition, siTRAF7 transfection reduced protein expression of VE-cadherin in HUVECs treated with or without TNFα (Fig. 6E). These results indicate that TRAF7 stabilizes VE-cadherin expression and its localization at junctions, and suppresses endothelial hyperpermeability. Furthermore, siTRAF7 transfection decreased expression of ICAM-1 and E-selectin by decreasing TNFα-induced nuclear localization of the NF-κB p65 and p50 subunits (Fig. S4). This suggested that TRAF7 suppresses endothelial hyperpermeability without inhibiting NF-κB pathway.

The functional analyses of Robo4 and TRAF7 indicated that both proteins similarly suppress endothelial permeability, suggesting that this function stems from the Robo4–TRAF7 complex. To determine which of these factors is a major regulator of endothelial permeability, we investigated the effects of Robo4 overexpression on TNFα-induced hyperpermeability with or without TRAF7 knockdown in HUVECs (Fig. 6F). The Ad-Robo4-mediated suppression of the TNFα-induced decrease in TEER was observed after transfection with siCont but not with siTRAF7. This indicates that TRAF7 is essential for Robo4 function. We next investigated the effects of TRAF7 overexpression on hyperpermeability with or without Robo4 knockdown. Interestingly, the Ad-TRAF7-myc-mediated suppression of the TNFα-induced decrease in TEER was observed after transfection with both siCont and siRobo4 (Fig. 6G). This suggests that Robo4 is not essential for TRAF7 function. Taken together, these results suggest that TRAF7 is a major regulator of endothelial permeability and that Robo4 functions as a modulator for enhancing TRAF7 function.

In the present study, we demonstrated that Robo4 suppresses vascular hyperpermeability and improves the survival of endotoxemic mice without administration of exogenous Slit2. We further identified TRAF7 as a novel binding protein of the Robo4 C-terminal domain and demonstrated that Robo4 suppresses TNFα-induced hyperpermeability by modulating TRAF7 function, thereby regulating VE-cadherin localization and protein expression. Thus, our findings support a novel Robo4-mediated regulatory mechanism of endothelial hyperpermeability in inflammation (Fig. 7).

In a Miles assay, we employed the endotoxemia model using Robo4-knockout mice and observed a significant increase in vascular leakage in the heart, lung and small intestine. Since we and others have previously reported that Robo4 is highly expressed in lung and heart vasculature (Okada et al., 2007, 2008; Park et al., 2003), Robo4 depletion in these organs may strongly affect vascular permeability. This also supports an organ-specific function for Robo4 in inflammation, and vascular heterogeneity dependent on Robo4. Based on our Miles assay results, we speculated that Robo4 suppresses some inflammatory signaling that induces vascular permeability. Since TNFα, one of the major inflammatory mediators, is known to be expressed at the early stage of endotoxemia and to induce vascular leakage (Ashkenazi et al., 1991; Norman et al., 1996; Tracey et al., 1987; Walsh et al., 1992), we focused on TNFα and successfully demonstrated suppression of TNFα signaling by the Robo4–TRAF7 complex. However, it is still possible that Robo4 regulates hyperpermeability and inflammatory responses induced by other inflammatory mediators, as we have shown previously (Shirakura et al., 2018).

Previous reports have demonstrated that Robo4 regulates VEGF-induced vascular permeability through the N-terminal domain via two mechanisms. In the Robo4-Slit2 pathway, Slit2 binds to the N-terminal domain of Robo4 and induces downstream signaling, which suppresses VEGF-induced hyperpermeability in pathological angiogenesis (Jones et al., 2008). In the Robo4-Unc5B pathway, the N-terminal domain of Robo4 binds to the netrin receptor Unc5B and suppresses VEGF receptor 2 activation and vascular permeability (Koch et al., 2011; Zhang et al., 2016). In contrast, the Robo4–TRAF7 complex, identified in this study, represents a new model of inflammatory endothelial permeability regulation. In this model, the C-terminal domain of Robo4, but not the N-terminal domain, is necessary for the suppression of endothelial hyperpermeability in inflammation.

The C-terminal domain of Robo4 was found to bind TRAF7. TRAF7 belongs to the TRAF family of proteins (TRAF1–TRAF7), which regulate inflammation and apoptosis (Bouwmeester et al., 2004; Xu et al., 2004; Yoshida et al., 2005; Zotti et al., 2011). Unlike other TRAF family members, TRAF7 does not contain a TRAF-C domain, which interacts with TNF receptors, but instead contains a unique WD40 domain (Xie, 2013). Although TRAF7 has been shown to interact with mitogen-activated protein kinase kinase kinase 3 (MAP3K3), modulate its activity, and regulate TNFα signaling in cell line-based assays (Bouwmeester et al., 2004; Xu et al., 2004), TRAF7 function in ECs is unclear. In this study, for the first time, we demonstrate that TRAF7 regulates endothelial permeability in inflammation. Surprisingly, our results indicate that TRAF7 is essential for Robo4-mediated suppression of hyperpermeability, but not vice versa, suggesting that Robo4 functions as an upstream modulator of TRAF7 to regulate endothelial permeability in ECs.

One possible mechanism through which Robo4 modulates TRAF7 is via regulation of TRAF7 localization. Robo4 has been shown to be localized in the cell membrane and cytoplasm, and to shuttle between these compartments (Sheldon et al., 2009; Zhang et al., 2016). Consistent with previous reports, we observed Robo4 in the perinuclear region, whereas TRAF7 was observed in cytoplasm. Co-expression of Robo4 and TRAF7 altered the localization of TRAF7 but not of Robo4, and Robo4 and TRAF7 almost completely colocalized. Interestingly, Robo4 mutants lacking the C-terminal domain, which did not suppress permeability, also did not alter TRAF7 localization. These results suggest that Robo4 modulates TRAF7 localization and function by interacting with TRAF7 via the Robo4 C-terminal domain. Interestingly, we have previously reported that TNFα induces Robo4 expression via NF-κB signaling. This suggests that Robo4 induced by TNFα increases the formation of Robo–TRAF7 complex molecules in ECs and suppresses hyperpermeability, thus serving as a negative feedback mechanism. To test this hypothesis, further detailed study to elucidate the localization and interaction of endogenous Robo4 and TRAF7 would be needed.

In conclusion, our study successfully demonstrated that Robo4 suppresses inflammatory endothelial hyperpermeability via the Robo4 C-terminal domain and its binding protein TRAF7. The Robo4–TRAF7 complex could thus serve as a novel therapeutic target in inflammatory diseases related to vascular hyperpermeability.

The targeting vector was prepared through BAC recombination-mediated genetic engineering using the Escherichia coli strain EL350 (Lee et al., 2001). Briefly, the mouse Robo4 gene (extending from base pair −2560 to exon 11) was inserted into the vector. A neomycin resistance cassette flanked by two loxP sites was inserted between exons 1 and 2 of the resulting plasmid, and the cassette was then removed using Cre recombinase in EL350. Another neomycin resistance cassette flanked by a loxP site and two FRT sites was inserted between exons 3 and 4 of the resulting plasmid, and exons 2 and 3, flanked by loxP sites, were removed using Cre recombinase. The resulting targeting vector was transfected into embryonic stem cells derived from 129/Sv mice using electroporation, and successfully transfected cells were selected through resistance to G418 selection antibiotic. Surviving embryonic stem colonies were picked and processed for genomic DNA extraction and Southern blot analysis. Correctly targeted embryonic stem clones were used to generate Robo4^−/− mice. The established mice were backcrossed with C57BL/6N mice for more than 10 generations. Genotyping polymerase chain reaction (PCR) was performed using mouse tail DNA and specific primers (Table S1).

For analysis of survival, Robo4^−/−, Robo4^+/−, and Robo4^+/+ male littermates (8–10 weeks old) were used to establish an endotoxemia model, as previously described (Yano et al., 2006). Survival was assessed over 96 h in Robo4^−/− and Robo4^+/+ mice that received an intraperitoneal injection of LPS (16.5 mg/kg body weight; Sigma-Aldrich). For permeability assays, Robo4^−/− and Robo4^+/+ littermates (8–12 weeks old) were intraperitoneally injected with LPS (16.5 mg/kg body weight). Six hours later, the mice were intravenously administered 100 µl of 1% Evans blue dye in phosphate-buffered saline (PBS). One hour later, the mice were perfused with PBS containing 2 mM ethylenediaminetetraacetic acid (EDTA) during anesthesia with isoflurane, and organs were harvested. Evans blue dye was eluted by mincing and incubating the organs in formamide for 2 days. The eluted dye was quantified by measuring the optical density at 620 nm. All the animal studies were approved by the ethics committee of Osaka University.

Human umbilical vein endothelial cells (HUVECs; Lonza) were cultured in EGM-2-MV medium (Lonza). Human embryonic kidney (HEK293) cells and African green monkey SV40-transfected kidney fibroblasts (COS-7 cells; ATCC, Manassas, VA, USA) were cultured in Dulbecco's modified Eagle's medium (Nacalai Tesque) supplemented with 10% fetal bovine serum (FBS), 100 IU/ml penicillin, and 100 µg/ml streptomycin. Monocytic U937 cells were cultured in RPMI 1640 (Sigma-Aldrich) containing 10% FBS and 2% penicillin-streptomycin. All cells were cultured at 37°C in an atmosphere containing 5% CO₂. All cells were free of mycoplasma contamination.

siRNA against Robo4 (SI03066896) and its control (AllStars Negative Control) were purchased from Qiagen. siRNA against TRAF7 (10620319) and its control (Stealth RNAi siRNA Negative Control, Med GC) were purchased from Invitrogen. Each siRNA was transfected using Lipofectamine RNAiMAX (Invitrogen).

The DNA fragments encoding Robo4 and TRAF7 with or without peptide tags (FLAG and Myc) were amplified by PCR using HUVEC cDNA and specific primers (Table S1) and inserted into pcDNA3 (Invitrogen) or the adenoviral shuttle vector pHMEF5 (Kawabata et al., 2005). The DNA fragments for Robo4 deletion mutants were prepared by PCR from pHMEF5-Robo4-FLAG using specific primers (Table S1). Each shuttle vector was digested, and the expression cassette was purified and inserted into the parental adenoviral vector pAdHM4. The resulting plasmids were linearized and transfected into HEK293 cells using Lipofectamine 2000 (Invitrogen). Amplified adenoviral vectors were purified by centrifugation on a CsCl₂ gradient and quantified as described previously (Maizel et al., 1968).

HUVECs (4×10⁴ cells) were seeded onto cell culture inserts with pore size of 0.4 µm (BD Falcon), treated with siRNAs (2.5 pmol) and/or adenoviral vectors (3000–10,000 virus particles/cell), and incubated for 48 h. TNFα (400 ng; Wako Pure Chemicals) was added to the upper chamber, and TEER was measured using a Millicell ERS-2 Voltohmmeter (Merck Millipore). The TEER value was calculated by the following formula: (resistance of experimental wells−resistance of blank wells)×0.32 (the membrane area of the cell culture insert), according to the manufacturer's instructions.

HUVECs (8×10⁵ cells) were transfected with siRNA (50 pmol) or adenoviral vectors (3000–10,000 virus particles/cell) and incubated for 48 h. The resulting cells were treated with TNFα for 24 h, and total cell lysates were extracted. For the detection of NF-κB, HUVECs were treated with TNFα for 30 or 60 min, and nuclear extract was prepared using the Qproteome Cell Compartment Kit (Qiagen). Western blotting was performed with the lysates and antibodies against Robo4 (N-17, Santa Cruz Biotechnology, 1:100; or AF2366, R&D systems, 1:500), TRAF7 (H-300, Santa Cruz Biotechnology, 1:2000), FLAG (F1804, Sigma-Aldrich, 1:1000), VE-Cadherin (F-8, Santa Cruz Biotechnology, 1:2000), Myc (9B11, Cell Signaling Technology, 1:1000), NF-κBp65 (C-20: Santa Cruz Biotechnology, 1:2000), NF-κBp50 (H-119, Santa Cruz Biotechnology, 1:2000), LaminB2 (LN43, Abcam, 1:2000) and GAPDH (MAB374, Merck Millipore, 1:10,000) and secondary antibodies conjugated to horseradish peroxidase (Jackson ImmunoResearch Laboratories Inc., PA). The resulting data were quantified by measuring intensities of the bands using ImageJ/Fiji software (Schindelin et al., 2012).

HUVECs (8×10⁵ cells) were transfected with siRNA (50 pmol) and incubated for 48 h. The resulting cells were treated with TNFα for 24 h, and harvested using Hanks' Balanced Salt Solution containing 5 mM EDTA. The cells were incubated with R-Phycoerythrin (PE)-conjugated mouse anti-human VE-cadherin (16B1, Thermo Fisher Scientific, 1:40) or PE-conjugated mouse IgG1 (12-4714-81, Thermo Fisher Scientific, 1:40) for 30 min on ice and then washed with PBS containing 2% FBS. Fluorescence intensity of the cells was analyzed using a flow cytometer (FACSCalibur, Becton-Dickinson).

HUVECs (4×10⁴ cells) were seeded onto the FluoroBlok insert with pore size of 8.0 µm (Corning Life Sciences), treated with siRNAs (2.5 pmol), incubated for 48 h, and treated with TNFα (200 ng) for 24 h. U937 cells (1×10⁶ cells) labeled with Cellstain Calcein AM solution (Dojindo) were added onto the HUVECs and incubated for 15 min. The transmigrated U937 cells were counted using a BZ-X700 fluorescence microscope and BZ-X analyzer software (KEYENCE).

HUVECs (2×10⁵ cells) were transfected with siRNA (25 pmol), incubated for 48 h and treated with TNFα (80 ng) for 30 min. Total RNA from cells was prepared using the RNeasy Mini Kit (Qiagen) and reverse-transcribed with Superscript VILO Master Mix (Invitrogen). Real-time PCR was performed using the cDNA, specific primers (Table S1), and QuantiTect SYBR Green PCR Kit (Qiagen). Copy numbers were calculated from the standard curve prepared using known amounts of plasmids including target sequences. The expression levels of Robo4, ICAM-1, and E-selectin (Sele) were normalized against the GAPDH level.

HUVECs were infected with adenoviral vectors to express Robo4–FLAG or Ac green fluorescent protein (AcGFP) and incubated for 36 h. The resulting cells were suspended in 1 ml lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 1 mM EDTA, and Roche protease inhibitor cocktail). The resulting cell lysates were immunoprecipitated and eluted with the FLAG Immunoprecipitation Kit (Sigma-Aldrich). The precipitated proteins were digested with trypsin and analyzed by means of liquid chromatography tandem mass spectrometry (LC-MS/MS) using the Ultimate 3000 HPLC/UPLC system and Q Exactive spectrometer (Thermo Fisher Scientific). The resulting data were analyzed using Mascot Software (Matrix Science) to identify the peptides specifically included in the Robo4–FLAG sample but not in the AcGFP sample.

COS-7 cells (3×10⁶ cells) were transfected with 3 μg of each expression vector using Lipofectamine 2000 and cultured for 24 h. The resulting cells were lysed in 1 ml lysis buffer and used for immunoprecipitation with the FLAG Immunoprecipitation Kit. The precipitated proteins were analyzed using western blotting. The data were quantified by measuring the intensity of the bands using ImageJ/Fiji software.

HUVECs (2×10⁵ cells) were seeded on gelatin-coated coverslips, treated with adenoviral vectors (3000–10,000 virus particles/cell) or siRNAs (12.5 pmol), and incubated for 48 h. The resulting cells were treated with or without TNFα for 24 h and fixed with 4% paraformaldehyde. For staining of Robo4–FLAG and TRAF7–myc, the cells were permeabilized with PBS containing 0.5% sodium dodecyl sulfate and 4 mM dithiothreitol, and blocked with PBS containing 1% bovine serum albumin. For the staining of VE-cadherin, the cells were permeabilized with PBS containing 0.3% Triton X-100 and blocked with PBS containing 1% bovine serum albumin. The resulting blocked cells were incubated with antibodies against FLAG (F1804, Sigma-Aldrich, 1:1000), TRAF7 (H-300, Santa Cruz Biotechnology, 1:1000), or VE-cadherin (F-8, Santa Cruz Biotechnology, 1:50), followed by incubation with secondary antibodies conjugated with Alexa Fluor 488 or Alexa Fluor 555 (Life Technology). The coverslips were mounted with VECTASHIELD Mounting Medium with DAPI (Vector Laboratories) and analyzed using a BZ-X700 fluorescence microscope (KEYENCE). Fluorescence intensity in the whole cell and the non-junctional region was measured using ImageJ/Fiji software, and the ratio of non-junctional VE-cadherin was calculated. TRAF7 localization was also analyzed using ImageJ/Fiji by measuring fluorescence intensity of TRAF7 within various regions that were marked by progressively increasing by 1 pixel. The relative intensity of TRAF7 was calculated by dividing the intensity in each region with that of the whole cell.

HUVECs (4×10⁴ cells) were transfected with siRNAs (2.5 pmol) and incubated for 48 h. Cell viability was analyzed using the Cell Counting Kit-8 (Dojindo). Briefly, the cells transfected with siRNAs were incubated in medium containing 10% WST-8 reagent for 1 h, after which their optical density was measured at 450 nm.

Data are expressed as the mean±standard error (s.e.m.). No statistical methods were used to determine the sample size before experiments. Animals were selected for experiments based on certain criteria established: their genotypes, proper age, and sex. No randomization and blinding were used. Normality and variance were tested by the Shapiro–Wilk test and the Brown–Forsythe test, respectively. For samples with a normal distribution and equal variance, P-values were calculated by Student's t-test and ANOVA followed by Tukey–Kramer test or Dunnett's test. For samples with non-normal distribution and equal variance, P-values were calculated by the Kruskal–Wallis test. The statistical significance of differences in the means was determined by the tests stated in the figure legends. P-values of <0.05 were considered to be statistically significant.

We thank Dr Hiroyuki Mizuguchi and Yurie Nakano for excellent assistance.