The tight junction protein Claudin-5 limits endothelial cell motility

The formation of a vascular network is an essential process during early embryogenesis. This involves the migration of arterial and venous angioblasts (endothelial progenitor cells) to the embryonic midline, where they assemble into vessel beds, establish cell junctions, rearrange into tubular structures and generate a lumenized network of blood vessels (Jin et al., 2005; Strilić et al., 2010; Yu et al., 2015). Subsequently, endothelial cells (ECs) that form the inner linings of arteries and veins adapt to their different physiological roles within the cardiovascular network by acquiring morphological and molecular differences. For instance, arterial ECs establish tight junction (TJ)-based barrier systems, which seal blood vessels against the paracellular movement of water, whereas venous ECs form less-dense TJs that are more leaky (Schneeberger, 1982).

The barrier properties of many tissues are dependent on claudins, a family of four-transmembrane TJ proteins (Kollmar et al., 2001; Krause et al., 2008; Lal-Nag and Morin, 2009; Loh et al., 2004). These proteins have also been shown to possess adhesive properties (Kubota et al., 1999). Functional studies in mouse revealed an important role of claudin-5 (Cldn5) in controlling the permeability of the cerebrovascular blood–brain barrier (Nitta et al., 2003). In zebrafish, Cldn5a has an essential function in the tightening of a neuroepithelial barrier during hydrostatic pressure-driven brain ventricle expansion (Zhang et al., 2010), but its potential developmental roles within the vasculature have not yet been characterized (Fleming et al., 2013; Jin et al., 2005; Xie et al., 2010).

Besides the canonical roles of claudins on the regulation of the physiological barrier functions, recent studies have also revealed an involvement of claudins in tumor progression which included affecting the migration, invasion and metastasis of cancer cells (Ma et al., 2017). For instance, during carcinogenesis of colorectal cancer, the changes of cell interactions and tissue morphology are accompanied by expression changes of various claudin genes (Bujko et al., 2015). In breast cancer, rather than tightening the cell barrier, Cldn5 affects the cell motility through the N-WASP and ROCK signaling pathways (Escudero-Esparza et al., 2012). One way claudins may change endothelial or epithelial cell motility is by impacting cell adhesion.

In zebrafish, two cldn5 paralogs (cldn5a and cldn5b) are expressed during embryonic developmental stages (van Leeuwen et al., 2018; Zhang et al., 2010). Cldn5a has a neuroepithelial and dorsal aorta expression, while Cldn5b is restricted to the arterial vessel system including the dorsal aorta (Zhang et al., 2010). We previously showed that Cldn5a has an essential sealing function in a neuroepithelial barrier during hydrostatic pressure-driven brain ventricle expansion (Zhang et al., 2010). In contrast, whether Cldn5 proteins have developmental roles in the zebrafish vasculature has not yet been determined. During zebrafish embryonic vasculogenesis, Cldn5 proteins are first detectable at 18 h post fertilization (hpf) in arterial endothelial cells marking endothelial cell–cell junctions of the trunk region (Jin et al., 2005). In the zebrafish brain, cldn5a is expressed in the cerebrovasculature after 48 hpf, while cldn5b is expressed within cerebral vascular endothelial cells as early as 30 hpf (Zhang, 2010). Functional studies revealed that the zebrafish blood–brain barrier matures after 2 days post fertilization (dpf) (Fleming et al., 2013; Jeong et al., 2008; van Leeuwen et al., 2018; Xie et al., 2010). However, previous studies did not address whether endothelial Cldn5 proteins may have developmental roles during vasculogenesis before 48 hpf.

In this study, we quantify the important contribution of Cldn5 to EC adhesive forces and motility. We also address the role of Cldn5a in the nascent zebrafish dorsal aorta and find that it limits EC motility during vasculogenesis and is required for the lumenization of the dorsal aorta.

The TJ protein Cldn5 has a well-characterized role in controlling the permeability properties of endothelial tissues and of the blood–brain barrier (Chen et al., 2017b; Nitta et al., 2003). To investigate to what extent Cldn5 impacts EC motility, we applied in vitro analyses based on the knockdown of Cldn5 in murine brain microvascular ECs (bEnd.3 cells). We reasoned that a reduction of Cld5 may result in a quantifiable increase in cell motility. When transfecting these ECs with mouse-specific Cldn5 short hairpin RNA (shCldn5-bEnd.3), Cldn5 protein expression was efficiently blocked in comparison with control vector-transfected cells (shCtrl-bEnd.3) (Fig. 1A). The expression of the TJ-associated protein ZO-1 (also known as TJP1) was not affected by this knockdown approach (Fig. 1A). Subsequently, we used these cells for a well-established trans-well membrane migration assay (Justus et al., 2014). When control bEnd.3 or shCldn5-bEnd.3 cells were incubated in trans-well insets for 24 h, shCldn5-bEnd.3 cells exhibited a significantly higher degree of motility and migrated through the trans-well membrane towards the bottom more efficiently than control bEnd.3 cells. This migration behavior was assayed using a Crystal Violet staining protocol that clearly marks ECs. It allowed us to quantify the proportion from total of ECs that migrated to the outside membrane of the well (Fig. 1B,C). We also observed an enhanced motility of shCldn5-bEnd.3 cells compared with control bEnd.3 cells when EC migration was further stimulated by adding endothelial cell growth factor (ECGF) to the bottom well (Fig. 1B,C). To further support this finding, a classic wound healing assay was performed with shCtrl-bEnd.3 and shCldn5-bEnd.3 cells. After 10 or 22 h of culturing, shCldn5-bEnd.3 cells had recovered more strongly compared with control bEnd.3 cells (Fig. 1D,E). This finding indicates that bEnd.3 cells have an increased migration ability upon loss of Cldn5. To confirm these results, we also generated stabilized bEnd.3 Cldn5^−/− lines by CRISPR/Cas9-mediated knockout of Cldn5 in bEnd.3 cells (Fig. S1A,B). The wound healing experiment revealed that within 22 h of culturing, bEnd.3 Cldn5^−/− cells had more strongly recovered and showed higher rate of motility compared with the wild-type bEnd.3 cells (Fig. S1C). Meanwhile, we performed a CRISPR/Cas9-mediated knockout of Cldn5 in another endothelial line, human brain microvascular ECs (HBMECs), which under wild-type conditions express high levels of Cldn5 (Fig. S2A,B). Within 12 h of culturing, HBMEC Cldn5^−/− cells showed a higher rate of motility and covered the wound area faster than the control HBMEC cells (Fig. S2C). These results suggest that the expression of Cldn5 has a limiting effect on EC motility.

Claudins are transmembrane proteins, and can interact with other claudins through homo- or hetero-philic trans-interactions between neighboring cells. Cldn5 seals the paracellular space by interacting, via face-to-face homophilic trans-interactions, between neighboring cells during the formation of TJ strands (Piorntek et al., 2008). Our findings showing that Cldn5 limits the motility of murine ECs in a transwell membrane migration assay, suggest that it might play a role in cell adhesion that goes beyond its well-established function in tight junctional sealing. To measure the strength of Cldn5-dependent adhesive forces between ECs, we performed a bead-rolling assay (Strilić et al., 2010). This assay measures the ability of EC-covered beads to roll down a slightly tilted EC monolayer-coated plate in a gravity-driven manner. Hence, the rolling distance covered by the beads inversely correlates with the adhesive strength of cell–cell contacts (Strilić et al., 2010). After incubation of ECs on collagen-coated glass beads, we found that Cldn5 was depleted from shCldn5-bEnd.3 ECs (Fig. 2A,B). In the bead-rolling-assay (Fig. 2C), beads coated with shCldn5-bEnd.3 ECs covered a significantly longer rolling distance on either shCtrl-bEnd.3 EC- or shCldn5-bEnd.3 EC-coated tilting plates, compared to a setup in which both beads and tilting plate were coated with the control-silenced shCtrl-bEnd.3 ECs (Fig. 2D,E). Similarly, beads coated with control-silenced ECs rolled a significantly longer distance when placed on tilting plates coated with Cldn5-depleted ECs compared with the distance covered over tilting plates coated with control-silenced ECs or over basal control tilting plates (Fig. 2D,E; Movies 1–4). Similar results were observed from the bead-rolling analyses with stabilized Cldn5-knockout bEnd.3 cells, where beads coated with bEnd.3 Cldn5^−/− cells covered a significantly longer rolling distance on both wild-type bEnd.3 EC- and bEnd.3 Cldn5^−/− EC-coated tilting plates, compared to a setup in which both beads and tilting plate were coated with the wild-type bEnd.3 ECs (Fig. S1D). These results demonstrate that Cldn5 generates cell–cell adhesion forces between adjacent ECs.

To quantify the strengths of adhesive forces between ECs, we also used atomic force microscopy (AFM) (Puech et al., 2005). We cultured control bEnd.3 ECs for at least 3 days on glass beads until they expressed Cldn5 (Fig. 2A,B) prior to fixing these beads onto the AFM cantilever (Fig. 2F). Next, the AFM cantilever holding bEnd.3 cell-coated beads was allowed to attach to the basal substratum with cultured ECs for at least 5 min to ensure a full face-to-face contact. Using AFM, we recorded the force needed to separate EC-coated beads from contacting with basal ECs on a disc. These measurements revealed adhesive forces of 1245.04±280.18 pN (means±s.d.) between Cldn5-expressing control ECs, while adhesion forces were only 423.36±212.41 pN between Cldn5-deficient shCldn5-bEnd.3 ECs, 453.78±83.14 pN between Cldn5-deficient ECs and control ECs, and 381.38±97.29 pN between control ECs and Cldn5-deficient ECs (Fig. 2G). Taken together, these measurements demonstrate that Cldn5 strongly affects EC adhesion.

The above findings suggest that Cldn5 may play some developmental role by affecting EC adhesion and motility. The zebrafish genome contains two genes encoding paralogs of mammalian Cldn5, Cldn5a [zgc:85723] and Cldn5b [zgc:103419], of which Cldn5a is more similar to mammalian Cldn5 (van Leeuwen et al., 2018; Zhang et al., 2012; Zhang et al., 2010). To elucidate the precise pattern of gene expressions of zebrafish cldn5a and cldn5b during early embryonic vascular development, we performed whole-mount in situ hybridizations (WISH) on 30 hpf zebrafish embryos. Consistent with previous reports, cldn5a was expressed within the zebrafish dorsal aorta and in neuroepithelial cells of the brain and spinal cord (Fig. 3A) (van Leeuwen et al., 2018; Zhang et al., 2010). In comparison, cldn5b was expressed in a partially overlapping pattern in arterial ECs, including the dorsal aorta but not the caudal vein (Fig. 3B).

To characterize developmental roles of cldn5a and cldn5b within the vasculature, we generated loss-of-function mutants using CRISPR/Cas9 genome-editing strategies. This approach yielded two cldn5a loss-of-function alleles (cldn5a^Δ8 with an 8 bp deletion and cldn5a^Δ14 with a 14 bp deletion) encoding predicted truncations of Cldn5a at the second transmembrane domain (Fig. 3C). This approach also resulted in a cldn5b loss of function allele (cldn5b^Δ130 with a 130 bp deletion) causing a similar predicted truncation at the second transmembrane domain (Fig. 3D).

Since both cldn5 paralogous genes were co-expressed within the dorsal aorta, we first characterized the morphology of this blood vessel in cldn5b and cldn5a single mutants and also in cldn5a^Δ14/Δ14;cldn5b^Δ130/Δ130 double mutants. We found that cldn5b^Δ130/Δ130 single mutants did not show any obvious phenotype compared with wild-type (Fig. 3E,F,K). Strikingly, in cldn5a^Δ14/Δ14 single mutants and cldn5a^Δ14/Δ14;cldn5b^Δ130/Δ130 double mutants, the dorsal aorta failed to lumenize by 30 hpf (Fig. 3G–I,K). Consequently, the blood vessel diameter was massively reduced from 35.00±3.79 μm (means±s.d.) in wild type (Fig. 3E; n=5 embryos) to 11.40±1.02 μm in cldn5a^Δ14/Δ14 mutants (Fig. 3G; n=5 embryos), 9.00±1.26 μm in cldn5a^Δ8/Δ8 mutants (Fig. 3J; n=5 embryos) and 10.80±4.17 μm in cldn5a^Δ14/Δ14;cldn5b^Δ130/Δ130 double mutants (Fig. 3H; n=5 embryos). Immunohistochemical stainings using an anti-pan-Cldn5 antibody on tissue cross sections of embryos revealed that Cldn5 immunoreactivity, which was present in the dorsal aorta of wild-type embryos (Fig. 3E′), was entirely absent only in cldn5a^Δ14/Δ14;cldn5b^Δ130/Δ130 double mutants rather than in the individual mutants (Fig. 3F′–H′). Hence, both Cldn5 proteins have a partially overlapping expression within the dorsal aorta.

A similar failure of dorsal aorta lumen expansion was observed after antisense morpholino oligonucleotide (MO)-mediated knockdown of cldn5a (Zhang et al., 2010; Fig. S3A,C; Fig. 3L) and of cldn5a together with cldn5b (Fig. S3A,D; Fig. 3L), but not of cldn5b alone (Zhang, 2010; Fig. S3A,B; Fig. 3L). The diameter of the dorsal aorta lumen diminished from 39.80±4.40 μm in wild-type (n=5 embryos) to 5.20±2.98 μm in cldn5a morphants (n=5 embryos) and 9.40±4.72 μm in cldn5a/b double morphants (n=5 embryos). To verify the specificity of cldn5a MO, we co-injected exogenous cldn5a mRNA together with cldn5a MO, which significantly reduced the proportion of embryos with a defective dorsal aorta lumen expansion (Fig. 3L). When performing immunohistochemical stainings, the MO-based gene knockdown of both cldn5a and cldn5b entirely depleted Cldn5 protein from the dorsa aorta (Fig. S3D′,D″).

Morphological inspections also showed that the loss of Cldn5a or of Cldn5a/b caused vascular changes that went beyond mere lumen expansion defects. Vibratome cross sections of the trunk region revealed that the dorsal aorta and caudal vein were in tight contact in cldn5a^Δ14/Δ14 single mutants and cldn5a^Δ14/Δ14;cldn5b^Δ130/Δ130 double mutants (Fig. 3G″,H″), whereas these two vessels were clearly separated in wild-type (Fig. 3E″) and in cldn5b^Δ130/Δ130 single mutants (Fig. 3F″). Similarly, in cldn5a single morphants or cldn5a/b double morphants, some regions along the dorsal aorta were in close proximity with the posterior caudal vein, which is suggestive of a failure of these two vessel types to separate (Fig. S3C″,D″).

To complement the other knockout and knockdown approaches, we also used a novel CRISPR/Cas13a-mediated gene knockdown strategy (Abudayyeh et al., 2017; Gootenberg et al., 2017; Liu et al., 2017) for cldn5a. In cldn5a crispants, the diameter of the embryonic dorsal aorta was reduced to 14.82±5.00 μm at 30 hpf (n=4 embryos), which was significantly smaller than that of the only cldn5a crRNA-injected control embryos (32.85±3.69 μm, n=4 embryos) (Fig. S4A,B). To verify the efficacy of this method, we performed quantitative reverse transcription-PCR (qRT-PCR) of cldn5a mRNA levels, which were significantly reduced in cldn5a crispants (Fig. S4C). These findings suggest that Cldn5a plays a role in the separation of the dorsal aorta from venous ECs and in the lumen expansion of the dorsal aorta.

Because the expression of Cldn5 had a limiting effect on murine EC motility in in vitro assays, we wondered whether a loss of Cldn5a had an equal impact on the motility of dorsal aorta ECs during vasculogenesis. To address this question, we traced the migration of arterial ECs during early vasculogenesis between 17 and 24 hpf. Time-lapse imaging in the endothelial-specific transgenic reporter line Tg(fli1a:nEGFP)^y7 combined with single-cell track analyses revealed that the motility of arterial ECs significantly increased with 5.35±0.33 μm/30 min (means±s.d.) in cldn5a morphants compared with 4.29±0.25 μm/30 min in wild-type embryos (Fig. 4A,B; Movies 5,6). Notably, the direction of ECs was primarily along the antero-posterior body axis in cldn5a morphants, whereas wild-type ECs had prominent trajectories in the dorsal-to-ventral direction (Fig. 4A,B). These data indicate that loss of Cldn5a causes a failure of morphogenesis due to reduced adhesion and increased motility of dorsal aorta ECs.

Adhesive mechanisms play important roles during various morphogenetic processes (Kim et al., 2017; Yi et al., 2008). Molecular studies have shown that expression of cadherin is important for the control of cell adhesion and motility during development (De Pascalis and Etienne-Manneville, 2017). A loss of claudins in Madin–Darby canine kidney (MDCK) cells revealed alterations in the actomyosin cytoskeleton. This may impact cadherin-mediated cell adhesion (Otani et al., 2019). To assess whether the loss of Cldn5 affected the expression of ve-cadherin, we first performed WISH in cldn5a morphant zebrafish. We found that ve-cadherin mRNA expression was not affected upon loss of Cldn5a at 20–22 hpf (Fig. 5A). We also performed immunofluorescence imaging on vibratome cross sections through the zebrafish trunk region and found that loss of Cldn5a had no effect on the expression and localization of VE-Cadherin in aortic ECs (Fig. 5B). Similarly, VE-Cadherin was not reduced in both shCtrl-bEnd.3- or shCldn5-bEnd.3-ECs (Fig. 5C). Hence, the increase in EC motility in cldn5 morphants is not related to a reduced expression of VE-Cadherin.

PECAM1 and integrin-β1 are typical endothelial adhesive molecules that may affect EC motility. Next, we investigated whether their expression was altered upon loss of Cldn5 in ECs. Based on immunohistochemical stainings, we found that the expression of PECAM-1 and integrin-β1 was reduced upon loss of Cldn5 (Fig. 5D,E), while levels of filamentous actin (F-actin), an essential cytoskeletal protein, were not altered in shCldn5-bEnd.3 cells when compared with shCtrl-bEnd.3 cells (Fig. 5F).

Our findings based on AFM measurements demonstrate that Cldn5 limits motility of murine ECs and that this may be important also within the nascent dorsal aorta of zebrafish. Unlike VE-cadherin, which is expressed ubiquitously within the vasculature, Cldn5 proteins are restricted to arterial vessel beds and play an important role in controlling the selective adhesive forces required for the segregation of arterial versus venous endothelial progenitor cells. This effect may be mediated directly, via homophilic trans-interactions between opposing Cldn5 proteins, or by regulating the expression of other cell adhesion factors such as PECAM-1 and integrin-β1 (Fig. 6). In tune with such a model, loss of zebrafish Cldn5a resulted in increased motility and reduced cohesion of dorsal aorta progenitor ECs. Its loss also resulted in a failure of dorsal aorta to expand its lumen and to separate from caudal vein ECs. These findings suggest that Cldn5 may have dual functions. First, we suggest that it may provide the adhesive forces needed for limiting EC motility and for the developmental sorting of vascular progenitor cells. However, whether motility defects are directly caused by defects in cell adhesion remains to be clarified and will require additional experiments. Later, Cldn5 may contribute to the physiological adaptation of arterial ECs to their characteristic vessel permeability properties. Our findings are consistent with increasing evidence from in vitro cell studies reporting that claudins regulate adhesion, migration and proliferation of tumor cells (Escudero-Esparza et al., 2012; Kwon, 2013). The knockdown or overexpression of Cldn5 in hCMEC/D3 cells affects EC proliferation and migration, which causes a disruption of the blood–brain barrier permeability and an increased infiltration of tumor cells (Shmakov et al., 2017). Increased Cldn5 expression results in decreased cell motility and in reduced adhesion to the extracellular matrix (Escudero-Esparza et al., 2012). In sum, our study points at an adhesive role of Cldn5 during vasculogenesis of the zebrafish dorsal aorta. Intriguingly, this may be an evolutionarily conserved role related to non-canonical adhesive roles ascribed to claudin-related proteins that are essential for the assembly of the Drosophila heart tube (Yi et al., 2008).

Zebrafish transgenic lines Tg(gata1:DsRed) (Traver et al., 2003), Tg(kdrl:eGFP)^s843 (Jin et al., 2005) and Tg(fli1:nEGFP)^y7 (Siekmann and Lawson, 2007) were maintained according to standard laboratory procedures (Kimmel et al., 1995; Westerfield, 2007). Fertilized embryos were raised in 0.003% phenylthiourea (P7629, Sigma), which was added to the embryo medium at 24 hpf to inhibit pigment formation. Handling of zebrafish was performed in accordance with Guangdong State Regulations on Laboratory Animal Management, and German and Brandenburg state law, carefully monitored by the local authority for animal protection (LAVG, Brandenburg, Germany).

CRISPR/Cas9 technology was utilized to generate cldn5a and cldn5b mutants. Cas9 mRNA was synthesized using T7 mMESSAGE mMACHINE^® Kit (AM1344, Ambion, USA) and guide RNA was in vitro transcribed following the T7 MAXIscript^® Kit manual (AM1312, Ambion, USA). To generate stable cldn5a or cldn5b mutants, guide RNA (150 ng/μl) and Cas9 mRNA (250 ng/μl) were co-injected into one-cell stage wild-type embryos. Founders and F1 generation animals were genotyped by DNA sequencing. Two cldn5a mutant alleles (cldn5a^Δ14 and cldn5a^Δ8) and one cldn5b mutant (cldn5b^Δ130) were identified via sequencing. Adult zebrafish were genotyped using PCR amplification products generated using specific primers (Table S1).

In brief, cas13a from Leptotrichia shahii was cloned in pCS2+ plasmid and the plasmid was linearized with XbaI. Then, cas13a mRNA was amplified using mMESSAGE mMACHINE^®SP6 Transcription Kit (Roche) with the promoter of SP6. Zebrafish cldn5a crRNA (5′-GGTGGGGTTATAGCTTCCCCTGATTTTGGAGCGCGACGACGCTCAAGACCGAGGAG-3′) was amplified using the mMESSAGE mMACHINE^®T7 in vitro transcription kit. For injection into the one-cell stage embryos, the following concentration was used: cas13a, 1500 ng/µl and cldn5a crRNA, 500 ng/µl. Only cldn5a crRNA of 500 ng/µl was injected as control. For the analysis of the knockdown efficiency, qRT-PCR of cldn5a was performed with RNA extracted from different groups of embryos. Primers used for cldn5a crRNA amplification and qRT-PCR are listed in Table S1.

Zebrafish cldn5a or cldn5b cDNAs were amplified from a standard cDNA library and were cloned into pGEM-T vector. To synthesize Dig-labeled antisense mRNA probes, recombinant plasmids were linearized with NcoI (R0193V, New England BioLabs, Ipswich, MA) and transcribed with SP6 RNA polymerase using Dig RNA Labeling kit (11175025910, Roche, Switzerland). Whole-mount in situ hybridization (WISH) was performed as previously described (Hauptmann and Gerster, 1994). Following WISH of ve-cadherin, whole embryos were incubated in 30% sucrose for 24 h at 4°C and then embedded within an optimum cutting temperature compound (4583, SAKURA), flash-frozen in liquid nitrogen, and cryosectioned into 10-μm-thickness sections, followed by 0.05% Neutral Red (N3246, Thermo Fisher, USA) staining for 12 s.

Embryos were fixed with 4% PFA overnight at 4°C. Serial trans-sections of embryonic trunk vessel parts with a thickness of 150 μm were made using a VT1000S vibratome (Leica, Germany). Immunofluorescence stainings were performed as follows. Embryonic tissue sections were blocked for 60–100 min in PBS with 0.1% Tween 20 and 1% DMSO (PBDT) plus 5% (v/v) normal goat serum (NGS) at room temperature (RT), then incubated with goat anti-mouse Cldn5 antibody (35-2500, Thermo Fisher Scientific; diluted 1:200 in PBDT, 5% NGS and 1% BSA) or anti-VE-Cadherin antibody (diluted 1:50 in PBDT, 5% NGS and 1% BSA) (Blum et al., 2008) overnight at 4°C. During the second day, sections were washed eight times for 10 min each time in PBDT, plus 1% NGS and 0.1 M NaCl (last wash in PBDT plus 1% NGS). Secondary antibodies conjugated with Alexa Fluor 647 (115-605-003, Jackson; diluted 1:200 in PBDT, 5% NGS and 1% BSA) and Rhodamine–phalloidin fluorescent dye (R415, Invitrogen, diluted 1:400 in PBDT+5% NGS+1% BSA) against F-actin were used to incubate embryonic tissue sections overnight at 4°C. After staining, sections were washed eight times for 10 min each time in PBDT plus 1% NGS and 0.1 M NaCl. Finally, slices were kept in the dark in PBS with 0.1% Tween 20 (PBST) (pH 7.4) at 4°C and mounted in anti-Slow Fade (S36936, Life Technologies) prior to imaging.

Morpholino antisense oligonucleotides (MOs) were purchased from GeneTools, and dissolved to 1 mM stock solutions in RNase-free water. One-cell stage embryos were injected with 4 nl of 150 μM cldn5a-MO (Zhang et al., 2010) or 300 μM cldn5b-MO (Zhang, 2010) for single MO-mediated knockdowns, and 4 nl of 100 μM cldn5a-MO together with 200 μM cldn5b-MO for double MO-mediated knockdowns. MO sequences are listed in Table S1. For rescue experiments, full-length cDNA encoding Cldn5a was generated by PCR and fused to the 3′ terminus of an egfp cDNA fragment. The egfp-cldn5a was then cloned into pCS2+ vector. The mMESSAGE mMACHINE Kit (AM1344, Ambion, USA) was used to synthesize capped mRNA. One-cell stage embryos were injected with 200 μM of cldn5a-MO or 200 μM cldn5a-MO together with 200 ng/μl egfp-cldn5a mRNA in a total volume of 4 nl.

Embryos with immunohistochemical staining were mounted in anti-Slow Fade (S36936, Life Technologies, USA). WISH stainings of embryos were imaged on an Olympus BX51/DP71 microscope (TOKYO, JAPAN) and images of frozen sections of WISH-stained embryos were captured using an Olympus BX/DP80 microscope (TOKYO, JAPAN). All images of cells were acquired on a Leica M205FA stereo microscope (Leica, Germany).

For time-lapse confocal microscopy imaging (as shown in Fig. 4) Tg(fli1:nEGFP)^y7 zebrafish embryos were dechorionated at the 17-somite stage and mounted with a lateral orientation on a glass bottom dish in 0.8% low-melting agarose (16520-100, Thermo Fisher Scientific, USA), with E3 medium containing 0.4 mg/ml Tricaine (3-amino benzoic acidethylester, Sigma-Aldrich, A-5040), positioned within a chamber and maintained at 28°C. Images for each embryo were recorded every 30 min during a period of 7 h on a LSM 710 confocal microscope (Zeiss) with a 20× objective. For image processing, Imaris (Bitplane, Version 7.7) was used to perform semi-automatic cell tracking of dorsal aorta ECs. For statistical analysis, the track length of each EC cell within a 5-somite wide region was divided by its total track duration (with a threshold of 3 time points). The average trajectories for each ECs’ displacement within 30 min were analyzed by an unpaired two-tailed Student's t-test with Microsoft Excel 2010.

Murine brain microvascular endothelial cells (bEnd.3) obtained from Shanghai Bioleaf Biotech were maintained in Dulbecco's modified Eagle's medium (C11995500BT, Gibco, Thermo Fisher Scientific) supplemented with 10% FBS at 37°C in a 5% CO₂ incubator. A Cldn5 knockdown cell line (shcldn5-bEnd.3 cell) was generated as described in our previous study (Liao et al., 2016). Human brain microvascular endothelial cells (HBMECs) were purchased from MingzhouBio (https://www.mingzhoubio.com/, cat. no. C-12287). The HBMEC Cldn5^−/−-mutated line was generated by CRISPR/Cas9 strategy using sgRNA (5′-CGACAGACCCGCGGGGCAAA-3′). The bEnd.3 Cldn5^−/− mutated line was generated by CRISPR/Cas9 strategy using sgRNA (5′-CCACAACATCGTGACGGCGCAGA-3′). The expression of endothelial Cldn5 was then analyzed by western blotting and immunolabeling with anti-Cldn5 antibody. For EC culturing, the culture medium was changed every 2 days and the cells were cultured for ∼5 days until formation of a monolayer was completed. For immunostainings, monolayers of bEnd.3 or HBMECs were cultured on beads or coverslips and were fixed with acetone for 10 min on ice. After blocking (1% BSA and 0.2% Tween-20 in PBS) for 1 h, fixed cells were then incubated with primary antibodies diluted 1:200 in blocking solution, overnight at 4°C. The following primary antibodies were used: pan-Cldn5 monoclonal antibody (Invitrogen, 35-2500), ZO-1 polyclonal antibody (Invitrogen, 40-2200), VE-cadherin polyclonal antibody (Santa Cruz Biotechnology, sc-28644), PECAM-1 (CD31) monoclonal antibody (eBioscience, 14-0311-81), integrin-β1 monoclonal antibody (Abcam, ab24693). For secondary antibody stainings, cells were subsequently incubated with Alexa Fluor 488-conjugated goat anti-mouse IgG (1:200; Jackson ImmunoResearch), Alexa Fluor 647-conjugated goat anti-rabbit IgG (1:200; Jackson ImmunoResearch), Rhodamine–phalloidin (1:500; Invitrogen) and DAPI (1:1000; Sigma-Aldrich) for 1 h at 37°C. Finally, stained cells were imaged using a Leica TCS SPII 5 confocal microscope (Leica, Solms, Germany).

The AFM assay was performed as previously reported (Chen et al., 2017a; Flach et al., 2011). In brief, bEnd.3 or shcldn5 bEnd.3 cells were cultured on collagen-coated glass plates/slides or glass beads until a cell monolayer formed. The plates/slides were placed onto an AFM-compatible chamber. A clean cantilever coated with Cell-Tak (Corning CoStar Corporation, Cambridge, MA) was made to adhere to the cell-covered bead. During each cycle of measurements, the AFM cantilever carrying the cell-covered bead was incrementally lowered by 0.5–2 µm steps to make contact with the cells covering the basal plate/slide until the first force curve was generated. The cell on the cantilever interacted with the cell on the disk for 5 min before the cantilever was moved upwards to separate completely. This process was carried out in an incubated chamber with stable 37°C and 5% CO₂ conditions. Data from five independent sample tests was collected and normalized by using the JPK image processing software (Bruker/JPK NanoWizard, Berlin, Germany).

Bead rolling measurements were conducted as previously described (Strilić et al., 2010). Briefly, bEnd.3, bEnd.3 Cldn5^−/− or shcldn5-bEnd.3 cells were cultured on six-well cell culture plates or collagen-coated glass beads until a cell monolayer formed. The cell culture plate was placed on a fixed trestle with an inclination angle of 21° under a stereo-microscope (M205FA, Leica, Germany). The cell-covered bead then rolled upon the plate/slide containing the cell monolayer culture. Images of rolling beads were captured continuously every 30 s for a total duration of 5 min. The bead rolling distance was measured directly according to the start and end positions.

Cell migration assays were performed using modified 24-well TransWell inserts (pore size, 8 μm; diameter, 6.5 mm; Costar, Acton, MA) according to the manufacturer's instruction. Briefly, bEnd.3, bEnd.3 Cldn5^−/− or shcldn5-bEnd.3 cells, with a density of 10⁵ cells/well cultured in 200 μl 10% FBS medium, were seeded within the upper chamber, and 1.1 ml 10% FBS medium with or without endothelial cell growth factor (92590, Millipore) was added within the lower chamber. After a 24 h incubation, cells on the upper or bottom surface were fixed with methanol and stained with 0.1% Crystal Violet (C0775, Sigma) for quantifications of the migration rate using a stereo microscope (M205FA, Leica, Germany).

BEnd.3, HBMEC, shcldn5-bEnd.3 bEnd.3 Cldn5^−/− and HBMEC-Cldn5^−/− cells were seeded into six-well dishes independently and cultured until ∼90% confluency. A sterilized pipette tip was used to scratch wounding across the cell monolayer. Migration of cells was observed and imaged under the microscope at serials of time point. Three wells of each group were quantified for each experimental condition. Experiments were carried out at least three times per group.

All statistical analyses were performed using Student's t-test (for two groups) or one-way ANOVA (for three or more groups) with Prism 5 software (GraphPad Software). In detail, a two-tailed method was applied for unpaired Student's test for both in vivo and in vitro experiments. The diameter of vessels and all in vitro cellular data are shown as means±s.d. Means±s.e.m. were used for trunk region vessel phenotypic ratios. 95% confidence intervals of data analyses were acquired for in vivo and in vitro experiments.

We thank Dr Dong Liu (Nantong University, China) for providing the pCS2-Cas13a plasmid. Thanks also to P. Yu, B. Wuntke, A. Hubig for technical assistance and to Y. Zheng, M. Kneisler for fish husbandry. For discussions of the project and critical reading of the manuscript we are indebted to all members of the Zhang's and Seyfried's groups and Russ Hodge.