Establishment of a neuroepithelial barrier byClaudin5a is essential for zebrafish brain ventricularlumen expansionJingjing Zhanga, Jörg Piontekb, Hartwig Wolburgc, Christian Piehlb, Martin Lissa,b, Cécile Ottena, Annabel Christa,Thomas E. Willnowa, Ingolf E. Blasigb,1, and Salim Abdelilah-Seyfrieda,1

aMax Delbrück Center for Molecular Medicine, D-13125 Berlin, Germany; bLeibniz Institute for Molecular Pharmacology, D-13125 Berlin, Germany; andcInstitute of Pathology, University of Tübingen, D-72076 Tübingen, Germany

Edited by Igor B. Dawid, National Institute of Child Health and Human, Bethesda, MD, and approved November 23, 2009 (received for review October19, 2009)

Lumen expansion driven by hydrostatic pressure occurs duringmany morphogenetic processes. Although it is well established thatmembers of the Claudin family of transmembrane tight junctionproteins determine paracellular tightness within epithelial/endothe-lial barrier systems, functional evidence for their role in the morpho-genesis of lumenized organs has been scarce. Here, we identifyClaudin5a asa core componentof an early cerebral-ventricular barriersystem that is required for ventricular lumen expansion in thezebrafish embryonic brain before the establishment of the embry-onic blood–brain barrier. Loss of Claudin5a or expression of a tightjunction-opening Claudin5amutant reduces brain ventricular volumeexpansion without disrupting the polarized organization of the neu-roepithelium. Perfusion experiments with the electron-dense smallmolecule lanthanum nitrate reveal that paracellular tightness of thecerebral-ventricular barrier decreases upon loss of Claudin5a. Geneticanalyses show that the apical neuroepithelial localization of Clau-din5a depends on epithelial cell polarity and provide evidence forconcerted activities between Claudin5a and Na+,K+-ATPase duringluminal expansion of brain ventricles. These data establish an essen-tial role of a barrier-forming Claudin in ventricular lumen expansion,thereby contributing to brain morphogenesis.

brain morphogenesis | lumen formation | Na+,K+-ATPase | tight junctions |cell polarity

Brain morphogenesis in the zebrafish embryo involves a well-studied ventricular lumen expansion process which occurs at

early stages of development between 17 and 21 h after fertilization(hpf) (1). Genetic analyses showed that the osmoregulatory ionpumpATPase, Na+/K+ transporting, alpha 1 polypeptide (Atp1a1)is critically important for lumen expansion which suggests a role inthe generation of hydrostatic pressure (1). However, the embryoniccerebral barrier expected to maintain luminal fluids and ions withinthe brain ventricles remains unknown. It is known that the blood–brain barrier forms 2 days after brain ventricle expansion (2). Also,the choroid plexus, which develops from the ependymal layer liningthe ventral floor of the cerebral ventricle, is not functional at thesestages and forms the blood-cerebrospinal fluid barrier only at laterstages and in the adult fish (3, 4). To elucidate the nature of thecerebral-ventricular barrier system required for initial lumen ex-pansion, we focused our attention on the neuroepithelium liningthe cerebral cavities. We hypothesized that Claudins (Cldn) maycontribute to the cerebral-ventricular barrier based on their es-tablished roles in the regulation of tight junction (TJ) barriers (5, 6).Claudins are characterized as either barrier- or pore-forming. In arecent study, Bagnat and colleagues demonstrated an involvementof the pore-forming Cldn15 in gut lumen expansion in the zebrafishembryo (7). Moreover, the C-terminal half of Clostridium per-fringens enterotoxin (C-CPE), a polypeptide with inhibitory activityto several barrier-forming Claudins including Cldn3, Cldn4, andCldn6, affected murine blastocoel cavity expansion, which isanother lumen expansion process that requires hydrostatic pressure

(8). Together, these studies implied an important function ofClaudins in brain ventricular lumen expansion.Here, we report that Cldn5a is essential for the establishment of

an embryonic cerebral-ventricular barrier system within the neu-roepithelium lining the brain ventricles. Tightening of this para-cellular barrier requires Claudin-Claudin trans interactions via thesecond extracellular loop of Cldn5a. We propose that during ven-tricular lumen expansion, Cldn5a seals the neuroepithelial layerTJs and maintains the fluid pressure that depends on Atp1a1activity. Therefore, concerted activities of these two proteins con-tribute to brain morphogenesis.

Results and Discussionclaudin5a Is Expressed Within the Brain During Ventricular LumenExpansion. There are at least 20 Claudin family members in tele-osts (9, 10). Database searches (http://zfin.org/) combined withwhole-mount in situ hybridization analyses revealed that twozebrafish clnd5 genes encoding the predicted orthologs of the hu-man four-transmembrane pass TJ protein Cldn5 are expressed inthe zebrafish embryo (cldn5a: gi|47086108|; cldn5b: gi|53733860|)(Fig. S1) and that zebrafish cldn5a has a strong neuroepithelialexpression (Fig. 1A; http://zfin.org/) whereas cldn5b is endothelial-specific (zgc:103419; http://zfin.org/). At 14 hpf, which is just beforebrain ventricle formation and initial expansion, cldn5a is expressedwithin the developing central nervous system with a particularlystrong expression in hindbrain and spinal cord regions and someweaker expression in somedorsal portionsof themidbrain (Fig. 1A).By 30 hpf, expression of cldn5a is within the entire central nervoussystem and particularly strong within ventral neuroepithelial cellslining brain ventricles within the hindbrain and some regions of themidbrain and forebrain (Fig. 1A′). This expression pattern suggestsa role of Cldn5a within developing ependymal cells lining the brainventricles.Ananti-mammalianCldn5 antibody that recognizes bothCldn5a and Cldn5b isoforms in the zebrafish revealed strong ex-pression of Cldn5 isoforms within the spinal cord and dorsal aorta(Fig. 1B). Overall, the zebrafish Cldn5a protein sequence is 56%identical and 73% similar to its human protein ortholog and iden-tities are higher for the two extracellular loops (ECLs). These data,combined with the established role of murine Cldn5 in maintainingthe integrity of the endothelial blood–brain barrier (11), suggested a

Author contributions: J.Z., J.P., C.P., T.E.W., I.E.B., and S.A.-S. designed research;J.Z., J.P., H.W., C.P., M.L., C.O., A.C., and S.A.-S. performed research; J.P. and C.P. contrib-uted new reagents/analytic tools; J.Z., J.P., H.W., A.C., T.E.W., I.E.B., and S.A.-S. analyzeddata; and I.E.B. and S.A.-S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence may be addressed. E-mail: iblasig@fmp-berlin.de or salim@mdc-berlin.de.

This article contains supporting information online at www.pnas.org/cgi/content/full/0911996107/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.0911996107 PNAS | January 26, 2010 | vol. 107 | no. 4 | 1425–1430

DEV

ELOPM

ENTA

LBIOLO

GY

potential involvement of Cldn5a in the formation of a cerebral-ventricular barrier in zebrafish.

Loss of Claudin5a Disrupts the Neuroepithelial Paracellular BarrierFunction. To test whether zebrafish Cldn5a has a similar functionfor the integrity of the brain neuroepithelial layer lining the brainventricles, we knocked down its expression by using antisensemorpholino oligonucleotides (MOs) directed against the 5′UTRof the gene (12). The efficiency of the cldn5a knock-down wasverified by using the anti-mammalian Cldn5 antibody (Fig. 1C).Tissue sections through the hindbrain ventricular zone revealedthat in wild-type (WT) embryos Cldn5a colocalized together withthe apical marker atypical protein kinase C (aPKC) (13) at apicalsurfaces of neuroepithelial cells lining the brain ventricles (Fig.1C). Injection of MOcldn5a completely abrogated the neuro-epithelial protein expression between 20 and 30 hpf withoutaffecting the apical localization of aPKC (Fig. 1C and Fig. 2D).As expected, the endothelial Cldn5b expression was not affectedin cldn5a morphants (Fig. 1C, arrows).

Next, we assessed the possibility that the paracellular barrier ofthe neuroepithelial layer lining the brain ventricles could beaffected by the loss of Cldn5a. To this end, we performed ven-tricular injections of lanthanumnitrate (molecularmass of 0.5 kDa)at 20 hpf and analyzed electron micrographs of hindbrain sectionsto detect the localization of this electron-dense material, which isroutinely used to assess the integrity of the TJ (14, 15). Whereas inWT embryos, lanthanum nitrate was enriched within intercellularspaces apical from the TJ of neuroepithelial cells and was notdetectable basally of the TJ (Fig. 1D), loss of Cldn5a resulted in thepermeation of the marker throughout the intercellular space (Fig.1E). The specificity of this effect could be verified by coinjection ofMOcldn5a together with cldn5a mRNA lacking the MO targetsequence (rescue), which completely restored the integrity of theTJ and prevented lateral diffusion of lanthanum nitrate (Fig. 1F).The finding that, upon loss of Cldn5a, lanthanumnitrate can diffusethrough intercellular spaces is in fair agreement with earlierobservations in Cldn5-deficient mice that revealed a leaky blood–brain barrier formed by endothelial cells for substances with amolecular mass of <0.8 kDa (11). Together, our analyses demon-strated that the early neuroepithelial layer lining brain ventriclesrepresents a cerebral-ventricular barrier and that Cldn5a deter-mines the permeability of hindbrain neuroepithelial cell TJs forsmall molecules.

A Tight Junction-Opening Mutant of Claudin5a Impairs Tight JunctionIntegrity and Affects Brain Ventricular Lumen Expansion. To test therelevance of Cldn5a-mediated tightening of the TJ for ven-tricular lumen expansion and brain morphogenesis, we attemp-ted to identify a specific Cldn5a mutant with a reducedtightening function. The ECL2 region has been shown to beinvolved in Claudin-Claudin trans interactions and formation ofTJ strands between two opposing cells (16), and the essentialresidues are conserved between classic Claudins and betweenspecies as shown for Cldn5 and Cldn3 (Fig. 2A and Fig. S2) (17).All Cldn5 proteins contain the aromatic amino acid Tyr148, whichhas been shown to be essential for TJ strands formation oftransfected human embryonic kidney (HEK) cells via an inter-cellular aromatic core (16). However, until now it has not beenclarified whether the ECL2 contributes to TJ tightness. To testthis possibility and, in particular, to assess the functional con-sequences of replacing Tyr148 for paracellular permeability, weused Madin-Darby canine kidney (MDCK-II) epithelial cells thatstably express either murine Cldn5WT or mutant Cldn5Y148A

(Fig. 2C). The tightness of the respective epithelial sheets wastested by measuring the transepithelial permeation of the tracerfluorescein (molecular mass of 376 Da). We found that themutant protein was strongly expressed and correctly localized toapical junctions, just like the WT protein (Fig. 2C). As a result,the permeability of the MDCK-II cell layer to fluorescein sig-nificantly increased >5-fold for cells expressing Cldn5Y148A

compared to cells expressing the WT form and was 2.5-foldhigher for cells expressing Cldn5Y148A when compared to vectorcontrol transfected cells (Fig. 2B). Our data demonstrates thatthe ECL2 domain of a Claudin contributes to TJ tightness. Untilnow, evidence has only been provided for a tightening role ofECL1 (18). Together, these results suggest that the two ECLscontribute independently or cooperatively to the TJ. Moreover,these findings pointed at a dominant barrier-disrupting effect ofmurine Cldn5Y148A on TJ permeability and opened the possi-bility to use this mutant as a tool to specifically inhibit formationof a paracellular barrier.Next, we investigated the effects of the zebrafish Cldn5aY148A

mutant on the initial ventricular lumen expansion, which occurs≈20 hpf (1). To assess the expression and localization of themutantprotein within the zebrafish embryo, we coinjectedMOcldn5a, whichis directed against the 5′UTR, together with mRNAs that lack thistarget sequence and encode zebrafish Cldn5aWT or mutant

nitca5ndl

C

CKPa5nd l

C

sc

dacv

B

30 hpf

MOcldn5a

WTC

V

V

20 hpf

V VV

WT MOcldn5a Rescue20 hpf

D E F

hbhb

hbA A’ cldn5a

14 hpf 30 hpf

sc

mb

mb

fbfb

mb

schb

mb

Fig. 1. Loss of Claudin5a disrupts the neuroepithelial paracellular barrierfunction. (A and A′) Whole-mount in situ hybridization of claudin5a (cldn5a)expression at two developmental time points. Dorsal views onto the brainare shown magnified within the insets. (A) Before brain ventricle formation(14 hpf), cldn5a is strongly expressed within the developing central nervoussystem including the hindbrain (hb) and spinal cord (sc). Weaker expressionof cldn5a is found within the dorsal midbrain region (mb). (A′) Duringventricle expansion (30 hpf), cldn5a is strongly expressed within the spinalcord and the neuroepithelial ventricular zones of ventral hindbrain andmidbrain. There is also strong expression in the forebrain (fb) ventricularzone. (B) Confocal microscopic images of cross-sections through the trunkregion. Strong expression of Cldn5 proteins is present within the spinal cord(sc) and the dorsal aorta (da) but not the cardinal vein (cv) at 30 hpf. (C)Injection of MOcldn5a efficiently blocks expression of Cldn5a protein withinthe apical neuroepithelium lining the brain ventricles (V). Endothelialexpression of Cldn5b is not affected in cldn5a morphants (arrows). (D–F)Electron micrographs of neuroepithelial cells covering the cerebral ventricles(V) after intraventricular injection of the electron-dense molecule lantha-num nitrate. In the WT, paracellular clefts are tight for the tracer, whichaccumulates in a dot-like pattern at the TJ (arrows). Knock-down of cldn5aresults in diffusion of lanthanum nitrate into the paracellular space betweencells (arrows). In rescue embryos, in which the cldn5a knock-down was res-cued by concomitant cldn5a mRNA injection, electron dense material isconfined to apico-lateral membranes of neuroepithelial cells similar to thedistribution in WT embryos. (Scale bars: B and C, 50 μm; D, 2 μm.)

1426 | www.pnas.org/cgi/doi/10.1073/pnas.0911996107 Zhang et al.

Cldn5aY148A. We found that both Cldn5a proteins localized cor-rectly to the apical side of neuroepithelial cells (Fig. 2D). The brainventricular lumen was visualized and quantified by in vivo labelingof embryos using sodium green indicator in combination with re-cording of confocal microscopic Z-scan projection stacks (Fig. 3).Subsequently, 3D reconstructions were generated for a total of 4–5embryos per sample group and volume measurements were gen-erated by using Volocity software (Improvision). These measure-ments revealed a strong reduction of the mid- and hindbrainventricular volume in cldn5a morphants (Fig. 3 B and F) and inembryos coinjected with MOcldn5a and cldn5aY148A mRNA (Fig.3 E and F) when compared with WT (Fig. 3 A and F) or withembryos rescued by coinjection ofMOcldn5a together with cldn5aWT

mRNA (Fig. 3 D and F). We also coinjected MOcldn5a+p53 toprevent potential MO-induced p53-mediated apoptosis (19) andfound that the ventricular lumen expansion was comparable to theknock-down of Cldn5a alone (Fig. 3 B, C, and F). Consistent withthis finding, we also observed that loss of Cldn5a does not increaseapoptotic cell death nor impair proliferation rates within the

hindbrain region at 20 hpf. Whole-mount immunohistochemistryusing antibodies against phosphorylated histone H3 to detectmitotic cells or against caspase 3 to detect apoptotic cells werecomparable for both WT and cldn5a morphants [mitotic indexdetermined for four embryos each (mean ± SD): WT (117/2,732)= 4.28 ± 0.15; MOcldn5a, (128/3,005) = 4.25 ± 0.28; MOcldn5a+p53

(123/2,787) = 4.41 ± 1.2; P = 0.886 (WT versus MOcldn5a), P =0.888 (WT versus MOcldn5a+p53). Apoptotic events counted for fiveembryos each (mean ± SD): WT (5/1,386) = 0.36 ± 0.25; MOcldn5a

(7/1,257)= 0.56± 0.11, P=0.309]. Measurements taken at 30 hpfof live embryos injected with fluorescein isothiocyanate (FITC)70-kDa dextran into the hindbrain ventricle revealed a sustainedreduction of ventricle size in cldn5a morphants also at laterstages of development (Fig. S3). Taken together, our datarevealed that the early neuroepithelial layer lining the brainventricles is part of an embryonic cerebral-ventricular barriersystem at 20 hpf, which is required for the initial ventricularexpansion process and that Claudin-Claudin trans interaction byCldn5a provide an essential tightening role within this tissue.

Claudin5a Localization Depends on Neuroepithelial Tissue Integrity.Loss of the cell polarity regulators Crumbs homolog 2 (Crb2)[in zebrafish known as Oko meduzy (Ome)] or Membraneprotein, palmitoylated 5a (Mpp5a) [in zebrafish known as Nagieoko (Nok)] causes a severe disruption of the brain ventricularzone (20–23) and a failure of brain ventricles to expand between20 and 30 hpf (Fig. S4) (1). Similarly, heart and mindla1 (hadla1)embryos that are deficient for the osmoregulatory ion pumpAtp1a1 display brain ventricle expansion defects by 30 hpf (Fig.S4) (1). The phenotypic similarities between these mutants andcldn5a morphants suggested a common molecular pathwayinvolved in the establishment of the cerebral-ventricular barriersystem. We therefore assessed whether loss of cell polarity or ofAtp1a1, which is strongly expressed within the neuroepithelium(Fig. 4B), affected the apical localization of Cldn5a. Indeed, lossof Mpp5a or Crb2 affected the integrity of the neuroepithelium,disrupted the contiguous ventricular apical zone, and abolishedthe apical accumulation of Cldn5a, aPKC, and the focal local-ization of the junctional marker protein Zonula occludens-1(ZO-1) (Fig. 4A and Fig. S4) (24). In contrast, loss of Atp1a1 didnot disrupt the contiguous apical ventricular localization of

A B C D E

F

* * *Volume brain ventricles

[x106µm3]1.41.21.00.80.6

0.00.20.4

MBV HBVMBV HBV MBV

MBV MBV

HBV

HBV HBV

A

20 hpf

B CWT MOcldn5a

MOcldn5a + cldn5aWT MO cldn5a + cldn5aY148AD E

MO cldn5a+p53

Fig. 3. Loss of Claudin5a affects expansion of the brain ventricular lumen at20 hpf. (A–E) Shown are confocal microscopic Z-scans of in vivo sodium greenindicator-labeled embryos. The ventricular lumen is indicated by false-coloring(orange). Indicated are ventricles of the midbrain (MBV) and hindbrain (HBV).(F) Quantifications of ventricular volume were generated from detailed 3D-reconstructions of confocal Z-scan sections for a total of 4–5 embryos persample group by using Volocity software (unit size is [×106 μm3]). Data rep-resent mean ± SEM, n ≥ 4, *, P = 0.016 in all three cases. (Scale bar: 100 μm.)

A

NC

ECL2

Y148*

outer

inner

2 3

ECL1

1 4

0.0

0.5

1.0

1.5

2.0

2.5

3.0

wt Y148A

ytilibaemr ep

evitaler

**B

*

Fluorescein (10 min)

vector Cldn5 Cldn5

C vector control Cldn5 Cldn5 wt Y148A

AN

D5ndl

CC

KPa

5ndlC

5ndlC

WT MOcldn5a MO cldn5a MO cldn5acldn5a wt cldn5a Y148A+ +D

Fig. 2. The TJ opening Claudin5aY148A mutant increases TJ permeability. (A)Schematic representation of the four-transmembrane pass TJ protein Clau-din5a (Cldn5a). The two extracellular loops (ECL), facing the paracellularspace, are involved in claudin-claudin trans interactions. The position of anaromatic residue within ECL2, which is conserved between all classic claudins(17) and which is essential for trans interactions, is highlighted in red. ECL2,which is conserved between all classic claudins (17) and which is essential fortrans interactions. (B) Stable transfection of murine Cldn5wt reduces theparacellular permeability offluorescein (filter culture) compared to the vectorcontrol, whereas transfection of the mutant Cldn5Y148A causes an increase offluorescein permeability. Data represent mean± SEM, n≥ 10; *, P< 0.05, **, P< 0.01. (C) Stable transfection of murine Cldn5wt and Cldn5Y148A intoMDCK-IIcells. The Cldn5 variants are strongly enriched at cell-cell contacts as detectedby immunocytochemistry against the N-terminal FLAG-tag. (Scale bars: 10μm.) (D) Injection of mRNA encoding zebrafish Cldn5awt or Cldn5aY148A intocldn5a morphants results in the correct localization of the proteins at theventricular surface of the 30 hpf neural tube as detected by confocal micro-scopy of sectioned immunohistochemical stainings. (Scale bar: 50 μm.)

Zhang et al. PNAS | January 26, 2010 | vol. 107 | no. 4 | 1427

DEV

ELOPM

ENTA

LBIOLO

GY

Cldn5a, aPKC, or ZO-1 (Fig. 4A and Fig. S4). To further assessthe role of the ion pump function for ventricular expansion,embryos were treated with ouabain, a potent inhibitor of Na+,K+-ATPase activity, between 18 and 30 hpf. By this treatment,ventricular lumen expansion was severely impaired (Fig. S5),although neuroepithelial tissue integrity was not affected (Fig.4A). Together, these findings confirmed that neuroepithelialtissue polarity and the osmoregulatory function of Atp1a1 are aprerequisite for the expansion of brain ventricular lumen.Moreover, the correct localization of Cldn5a depends on neu-roepithelial tissue polarity.Embryos lacking Cldn5a or expressing the TJ-opening

Cldn5aY148A mutant display significantly reduced brain ventricles,which is phenotypically reminiscent of cell polarity mutants, em-bryos lacking Atp1a1 or ouabain-treated embryos (Fig. S4 and S5).We therefore tested whether, conversely, loss of Cldn5a affects thepolarized organization or integrity of theneuroepithelial ventricularzone.However, the focal localizationof the junctionalmarkerZO-1was not changed in cldn5a morphants compared with WT at 30hpf (Fig. 4A). Moreover, loss of Cldn5a did not affect the neuro-epithelial localization of Atp1a1 as detected by using the a6F

antibody that recognizes different isoforms of the Na+,K+-ATPase(Fig. 4B) (25). These findings demonstrated that cell polarity andtissue integrity of the neuroepithelium are not disrupted incldn5a morphants.Our work sheds light on a poorly understood aspect of brain

morphogenesis during early zebrafish development. We haveshown that the embryonic neuroepithelial layer lining the devel-oping brain ventricles is organized as a cerebral barrier and thatCldn5a-mediated paracellular tightness of this tissue is requiredfor ventricular lumen expansion. Unlike in crb2 or mpp5a cellpolarity mutants, tissue integrity of the neuroepithelial layer is notaffected in cldn5a morphants. Therefore, Cldn5a tightens thecerebral-ventricular barrier by modifying the TJ physiology ratherthan by disrupting organization of the neuroepithelium. Based onour results, we favor a model according to which cell polarityregulators of the Crumbs complex are essential for ventricularlumen formation via regulation of neuroepithelial integrity, whichis required for TJ formation and Cldn5a apical localization. Inaddition, Crumbs proteins have been shown to be required for theestablishment of an apical membrane compartment, which isessential for lumen generation in cyst-formation assays inMDCK-II cells (26). A potential direct involvement of Crumbs complexproteins in brain ventricular lumen formation independent ofapical Cldn5a localization remains to be tested. Cldn5a actsdownstream of cell polarity regulation in sealing neuroepithelialTJ and maintaining ventricular fluid pressure. The generation offluid pressure may depend on Atp1a1 activity during the brainventricle expansion process between 17 and 30 hpf (Fig. 4C). Thiswork on the barrier-forming Cldn5a and a published study (7) onthe pore-forming Cldn15 highlight the diversity of mechanisms bywhichClaudin-mediatedmodulation of TJ physiology controls thematuration of different organ lumens during development.It is a well established fact that the ventricular wall of the adult

mammalian brain does not represent a cerebral barrier system andthat neuroepithelial ependymal junctions are leaky (27). There-fore, the embryonic cerebral-ventricular barrier system describedin our study should be considered transient and may only beessential for the initial ventricular expansion rather than formaintenance of ventricular shape. A similar maturational role hasbeen described in the neonatal mouse kidney for Cldn6 and Cldn9,which are not expressed in the adult proximal tubuli (28). BothCldn6 and Cldn9 are the closest relatives of Cldn5 in the phylo-genetic tree of all claudins (17). Because Cldn5 is not obviouslyexpressed in the ventricular zone of early mouse embryos betweenembryonic day (E) 9.5–E11.5 (Fig. S6; http://developingmouse.brain-map.org/), it is possible that other Claudins may serve acomparable function within the murine embryonic brain.Based on our study, we envision that substances interfering

with Cldn5-mediated paracellular barrier tightening could assistan improved drug delivery into the brain because Cldn5 is es-sential for blood–brain barrier integrity (11). Such substancescould be identified using zebrafish brain ventricle expansion asan in vivo assay system.

Materials and MethodsFish Stocks and Maintenance. General zebrafish maintenance and embryocollection was carried out according to standard conditions. Embryos werestaged at 28.5 °C. The following fish strains were used: AB, hadla1, nokm520,and omem289 (21, 23, 29).

DNA Constructs, Site-Directed Mutagenesis, and mRNA Synthesis. Full-lengthcldn5a cDNA was produced by RT-PCR from 24 hpf embryonic cDNA by usingthe following primers and subcloned into pCS2+ expression vector:

cldn5a_fw: 5′-CCGCTCGAGATGGCCTCCGCGGCTTTG-3′,cldn5a_re: 5′-GCTCTAGATCACACGTAATTCCTCTTG-TC-3′.

WT MOcldn5a

V

V

DNA Na+,K+-ATPaseB

Cell Polarity

Crb2Mpp5a

Fluid pressureAtp1a1

Cerebral-ventricularbarrier sealing

Cldn5a

Ventricle lumen expansion

C

aPKC ZO-1AWT

MOcldn5a

ouabain

nok / mpp5a

ome / crb2had / atp1a1

Fig. 4. Loss of Claudin5a does not affect neuroepithelial tissue integrity. (A)Shown are apical views of immunohistochemical stainings onto the hind-brain ventricular zone of cell polarity mutants, Cldn5a- or Atp1a1-deficientembryos, or ouabain-treated embryos at 30 hpf. The neuroepithelial local-ization of Zonula occludens-1 (ZO-1) or aPKC is only affected in the cellpolarity mutants nokm520/mpp5a and omem289/crb2. (Scale bar: 20 μm.) (B)Loss of Cldn5a does not affect the neuroepithelial localization of Na+,K+-ATPase which is labeled with the a6F antibody. Shown are confocal micro-scopic images of sections of immunohistochemical stainings of the hindbrainand ventricle. (Scale bar: 20 μm.) (C) Schematic diagram of developmental/cellular processes contributing to brain ventricle expansion. Our study sug-gests that the cell polarity regulators Crb2 and Mpp5a are essential forneuroepithelial integrity and maintenance of the TJ. Moreover, Crumbscomplex proteins may be directly required for lumen formation (Results andDiscussion). Tightness of the TJ is, at least in part, regulated by Cldn5a, whichseals the neuroepithelial layer to maintain the fluid pressure, which maydepend on the ion pump activity of Atp1a1. Ventricular fluid accumulationdrives expansion of brain ventricles and tissue morphogenesis. crumbs2(crb2); heart and mind (had); membrane protein, palmitoylated 5a (mpp5a);nagie oko (nok); oko meduzy (ome); Ventricle (V).

1428 | www.pnas.org/cgi/doi/10.1073/pnas.0911996107 Zhang et al.

Site-directed mutagenesis of cldn5aY148A was performed by using theQuikChange kit (Stratagene) using the following primers:

cldn5aY148A_fw: 5′-TATCATCTCCGACTTCGCTAACCCG-CAGGTGCTGC-3′,cldn5aY148A_re: 5′-GCAGCACCTGCGGGTTAGCGAAGT-CGGAGATGATA-3′.

The resultant clones were cut with XhoI/XbaI and subcloned into the pCS2+expression vector. For mRNA synthesis, the SP6 or T7 mMessage mMachine kits(Ambion) were used. Digoxigenin-UTP-labeled riboprobes for whole-mount insitu hybridizations were synthesized according to the manufacturer’s instruc-tions (Roche) by using the pCS2+ vectors with full-length cldn5a.

Mammalian expression vectors for murine Cldn5 were based on pEYFP-N1(Clontech). FLAG-cldn5-YFP was generated by amplification of murine cldn5(accession number NM_013805) using pGTCl-5 as template and primers, oneencoding the N-terminal FLAG epitope. To generate FLAG-cldn5, site-directedmutagenesis was performed as described in ref. 16 introducing a stop codon3′ behind the cldn5 ORF. The amino acid substitution Y148A was againintroduced by site-directed mutagenesis.

Zebrafish Microinjections and Morpholinos.MOs (Gene Tools) were injected ata concentration of 150 μmol/L (for MOcldn5a) or 100 μmol/L (for MOp53). MOsequences were as follows:

MOp53: 5′-GCGCCATTGCTTTGCAAGAATTG-3′ (19),MOcldn5a: 5′-AGGCCATCGCTTTCTTTTCCCACTC-3′.

For mRNA rescue experiments, 75–100 pg of mRNA and 150 μM ofMOcldn5a were coinjected. All types of injections were performed by usingthe MPPI-2 pressure injector (Applied Scientific Instrumentation) and MM33micromanipulator (Maerzhaeuser).

Whole-Mount in Situ Hybridization, Tissue Sectioning, and Immunohistochemistry.Whole-mount in situ hybridizations were basically performed as described inref. 30. For documentation, stained embryos were cleared in benzyl:benzoate(2:1) and embedded in Permount. Images were recorded on a Zeiss Axioplanmicroscope with ×10 objective by using a SPOT digital camera (DiagnosticInstruments) andMeta Morph software (Visitron). Images were processed withPhotoshop software (Adobe). Immunohistochemistry was performed asdescribed in ref. 13. For tissue sectioning of Cldn5a immunostainings, embryoswere fixed in trichloroacetic acid for 1 h on ice and subsequently postfixed in2% paraformaldehyde overnight for hardening of the tissue. Transverse sec-tioning was performed according to Trinh and Stainier (31). The followingantibodies were used: mouse anti-mammalian Cldn5 (1:100; Invitrogen), rabbitanti-aPKCζ that recognizes both isoforms of aPKCs (1:100; Santa Cruz Bio-technology), rabbit anti-phospho-Histone H3 (1:800; Millipore), rabbit anti-Caspase 3 (1:200, BD Biosciences), mouse anti-ZO-1 (1:200; Zymed Labo-ratories), mouse anti-Na+,K+-ATPase (1:100; α6F, Hybridoma Bank), goat anti-rabbit Cy5 (1:100; Jackson ImmunoResearch Laboratories), goat anti-mouseFITC (1:100; Jackson ImmunoResearch Laboratories), rhodamine phalloidin(1:100, Invitrogen), 4′,6-Diamidino-2-phenylindole dihydrochloride (1:1000,Sigma). Cell proliferation and apoptosis quantifications were performed asdescribed in ref. 1. To prevent bleaching during confocal microscopic imaging,samples were embedded in SlowFade Gold Antifade reagent (Invitrogen).Confocal images were obtained with the Zeiss LSM510 Meta confocal micro-scope by using ×40 and ×1 zoom. LSM 510 software was used to capture theimages. Images were processed by using Photoshop software (Adobe).

Sequence Analysis and Phylogenetic Analysis. We identified zebrafish cldn5aand cldn5b based on homology searches with Mus musculus cldn5 by usingthe NCBI BLAST search tool (http://blast.ncbi.nlm.nih.gov/Blast.cgi). The geneaccession numbers of the claudin genes are as follows: cldn5a, GI47086108;cldn5b, GI53733860. Protein alignments and phylogenetic analysis wereperformed with MacVector software by using the ClustalW slow/accuratemethod. The accession numbers of protein sequences that were used for thephylogenetic analysis shown in Fig. S1 and the alignment shown in Fig. S2are as follows: Cldn3/h (Danio rerio) NP_571842; Cldn3 (Mus musculus)NP_034032; Cldn3 (Homo sapiens) NP_001297; Cldn3b (Takifugu rubripes)AAT64048; Cldn5 (Musmusculus) NP_038833; Cldn5 (Homosapiens) NP_003268;Cldn5a (Danio rerio) NP_998439; Cldn5b (Danio rerio) NP_001006044; Cldn5a(Takifugu rubripes) AAT64067; Cldn6 (Homo sapiens) NP_067018; Cldn6 (Mus

musculus) NP_061247; Cldn9 (Homo sapiens) NP_066192; Cldn9 (Mus mus-culus) NP_064689.

Electron Microscopy. Brain ventricles of zebrafish embryos were injected with≈10 nL of 1% lanthanum nitrate together with 2.5% glutaraldehyde in0.1 M cacodylate buffer for immediate preservation of ventricular tissues.Subsequently, embryos were fixed with 2.5% glutaraldehyde (Paesel-Lorei)buffered in 0.1 M cacodylate buffer (pH 7.4). Thereafter, the whole fish waspostfixed in 1% OsO4 in 0.1 M cacodylate buffer and then dehydrated in anethanol series (50, 70, 96, and 100%). The 70% ethanol was saturated withuranyl acetate for contrast enhancement. Dehydration was completed inpropylene oxide. The specimens were embedded in Araldite (Serva). Ultra-thin sections were made on a FCR Reichert Ultracut ultramicrotome (Leica),mounted on pioloform-coated copper grids, contrasted with lead citrate,and analyzed and documented with an EM10A electron microscope (Zeiss).

Cell Culture. MDCK-II cells were grown at 10% CO2 in DMEM supplementedwith 10% FBS, 2 mM L-glutamine, 100 U penicillin, and 100 mg/mL strep-tomycin. Transfections of MDCK-II cells were performed with Lipofectamine2000 (Invitrogen) or Cell Line Nucleofection Kit L (Amaxa). Stable lines wereselected by adding 1 mg/mL G418 (Calbiochem). To exclude clonal variations,at least three independent cell clones were isolated of Cldn5wt, Cldn5Y148A,or vector control and were used for permeation studies. For immunocy-tochemistry, cells were cultured on glas coverslips for 5 days, fixed andstained as described in ref. 32 by using goat anti-rabbit Cy3 antibody (1:250)and DAPI (4′,6-diamidino-2-phenylindole dihydrochloride, 1:300, MolecularProbes). For confocal imaging, a Zeiss LSM 510 Meta confocal microscopeand a ×100 objective was used.

Paracellular permeation studies were done with confluent cells on transwellfilter inserts (Millipore). Cells were washed twice with PBS and incubated withHanks’ balanced salt solution (Gibco); 100 μM Na-fluorescein was added to theapical side. One hundred-μL aliquots were removed from the basolateral res-ervoir 10min after incubation. Thefluorescein concentrationwasmeasured in afluorescence microplate reader (Tecan). The permeability coefficient was cal-culated as Pcoeff ¼ Va

ðA×C0Þ×dCdt ;where Va is volume in apical chamber, A is filter

surface area, C0 is initial apical fluorescein concentration, and dC/dt is initialslope of the concentration versus time curve (33).

Brain Ventricle Volume Measurement and Pharmacological Treatment. Sodiumgreen indicator (Molecular Probes) was added directly to the fishmedium at afinal concentration of 5 μM. At 20 hpf, embryos were incubated in themedium for 1 h before mounting and imaging them under the microscope.Z-stack images of the whole ventricle zones were captured by using the ZeissLSM Meta software. Three-dimensional projections and brain ventricle vol-ume measurements were generated from the stack of images by usingVolocity (Improvision) software, and the values of several experiments wereanalyzed and plotted in Microsoft Excel by using Mann–Whitney statisticaltesting. Injection of FITC 70-kDa dextran (Invitrogen) into 30 hpf brainventricles was performed as described in ref. 1. Ouabain (Sigma) treatmentwas performed as described in ref. 29 from 18 to 30 hpf at 1 mM concen-tration. Differential interference contrast images were then taken on a ZeissAxioplan microscope (Zeiss) by using a SPOT digital camera (DiagnosticInstruments) and Metamorph software (Zeiss). Images were processed withAdobe Photoshop 6.0. False-coloring of the ventricular lumen was per-formed by using Adobe Photoshop 6.0.

Whole-Mount In Situ Hybridization in Mice. Embryos were dissected frompregnant mice at the designated time point and fixed overnight in 4%paraformaldehyde in PBS at 4°C. Whole-mount in situ hybridizations werecarried out essentially as described in ref. 34. The localization of signals wasstudied in whole embryos or in plastic sections thereof. First-strand mousecldn5 cDNA was used as in situ probe.

ACKNOWLEDGMENTS. We are indebted to M. Furuse (Kyoto University,Japan), J. Herz, N.D. Lawson, J. Malicki, and B. Weinstein for sharingreagents and tools and to Robby Fechner for expert technical assistancewith the fish facility. We thank Gabi Frommer-Kästle for expert help withelectron microscopy. Nicole Cornitius and Jana Richter helped with themolecular laboratory work. We thank Alistair Garratt, Matthias Selbach,Michael Gotthardt, Erez Raz, and members of the S.A.-S. and I.E.B. labora-tories for technical help, discussions, and comments on the manuscript, andin particular Justus Veerkamp for help with ventricular volume measure-ments. We thank M. Affolter and H. Belting for sharing unpublished results.This work was supported by the German Research Council Deutsche For-schungsgemeinschaft Grant BL308/7-3 (to I.E.B. and S.A.-S.).

Zhang et al. PNAS | January 26, 2010 | vol. 107 | no. 4 | 1429

DEV

ELOPM

ENTA

LBIOLO

GY

1. Lowery LA, Sive H (2005) Initial formation of zebrafish brain ventricles occursindependently of circulation and requires the nagie oko and snakehead/atp1a1a.1gene products. Development 132:2057–2067.

2. Jeong JY, et al. (2008) Functional and developmental analysis of the blood-brainbarrier in zebrafish. Brain Res Bull 75:619–628.

3. Bill BR, et al. (2008) Development and Notch signaling requirements of the zebrafishchoroid plexus. PLoS One 3:e3114.

4. García-Lecea M, Kondrychyn I, Fong SH, Ye ZR, Korzh V (2008) In vivo analysis ofchoroid plexus morphogenesis in zebrafish. PLoS One 3:e3090.

5. Furuse M, Tsukita S (2006) Claudins in occluding junctions of humans and flies. TrendsCell Biol 16:181–188.

6. Furuse M, Moriwaki K (2009) The role of claudin-based tight junctions inmorphogenesis. Ann N Y Acad Sci 1165:58–61.

7. Bagnat M, Cheung ID, Mostov KE, Stainier DY (2007) Genetic control of single lumenformation in the zebrafish gut. Nat Cell Biol 9:954–960.

8. Moriwaki K, Tsukita S, Furuse M (2007) Tight junctions containing claudin 4 and 6 areessential for blastocyst formation in preimplantation mouse embryos. Dev Biol 312:509–522.

9. Kollmar R, Nakamura SK, Kappler JA, Hudspeth AJ (2001) Expression and phylogenyof claudins in vertebrate primordia. Proc Natl Acad Sci USA 98:10196–10201.

10. Loh YH, Christoffels A, Brenner S, Hunziker W, Venkatesh B (2004) Extensive expansionof the claudin gene family in the teleost fish, Fugu rubripes. Genome Res 14:1248–1257.

11. Nitta T, et al. (2003) Size-selective loosening of the blood-brain barrier in claudin-5-deficient mice. J Cell Biol 161:653–660.

12. Nasevicius A, Ekker SC (2000) Effective targeted gene ‘knockdown’ in zebrafish. NatGenet 26:216–220.

13. Horne-Badovinac S, et al. (2001) Positional cloning of heart and soul reveals multipleroles for PKC lambda in zebrafish organogenesis. Curr Biol 11:1492–1502.

14. Whittembury G, Rawlins FA (1971) Evidence of a paracellular pathway for ion flow inthe kidney proximal tubule. Electromicroscopic demonstration of lanthanumprecipitate in the tight junction. Pflugers Arch 330:302–309.

15. Wolburg H, et al. (2008) Epithelial and endothelial barriers in the olfactory region ofthe nasal cavity of the rat. Histochem Cell Biol 130:127–140.

16. Piontek J, et al. (2008) Formation of tight junction: determinants of homophilicinteraction between classic claudins. FASEB J 22:146–158.

17. Krause G, et al. (2008) Structure and function of claudins. Biochim Biophys Acta 1778:631–645.

18. Wen H, Watry DD, Marcondes MC, Fox HS (2004) Selective decrease in paracellularconductance of tight junctions: Role of the first extracellular domain of claudin-5.MolCell Biol 24:8408–8417.

19. Robu ME, et al. (2007) p53 activation by knockdown technologies. PLoS Genet 3:e78.20. Bit-Avragim N, et al. (2008) Divergent polarization mechanisms during vertebrate

epithelial development mediated by the Crumbs complex protein Nagie oko. J Cell Sci121:2503–2510.

21. Malicki J, Driever W (1999) oko meduzy mutations affect neuronal patterning in thezebrafish retina and reveal cell-cell interactions of the retinal neuroepithelial sheet.Development 126:1235–1246.

22. Omori Y, Malicki J (2006) oko meduzy and related crumbs genes are determinantsof apical cell features in the vertebrate embryo. Curr Biol 16:945–957.

23. Wei X, Malicki J (2002) nagie oko, encoding a MAGUK-family protein, is essential forcellular patterning of the retina. Nat Genet 31:150–157.

24. Itoh M, et al. (1993) The 220-kD protein colocalizing with cadherins in non-epithelialcells is identical to ZO-1, a tight junction-associated protein in epithelial cells: cDNAcloning and immunoelectron microscopy. J Cell Biol 121:491–502.

25. Drummond IA, et al. (1998) Early development of the zebrafish pronephros andanalysis of mutations affecting pronephric function. Development 125:4655–4667.

26. Schlüter MA, et al. (2009) Trafficking of Crumbs3 during cytokinesis is crucial forlumen formation. Mol Biol Cell 20:4652–4663.

27. Brightman MW, Reese TS (1969) Junctions between intimately apposed cellmembranes in the vertebrate brain. J Cell Biol 40:648–677.

28. Abuazza G, et al. (2006) Claudins 6, 9, and 13 are developmentally expressed renaltight junction proteins. Am J Physiol Renal Physiol 291:F1132–F1141.

29. Shu X, et al. (2003) Na,K-ATPase is essential for embryonic heart development in thezebrafish. Development 130:6165–6173.

30. Thisse C, Thisse B, Schilling TF, Postlethwait JH (1993) Structure of the zebrafish snail1gene and its expression in wild-type, spadetail and no tail mutant embryos.Development119:1203–1215.

31. Trinh LA, Stainier DY (2004) Fibronectin regulates epithelial organization duringmyocardial migration in zebrafish. Dev Cell 6:371–382.

32. Blasig IE, et al. (2006) On the self-association potential of transmembrane tightjunction proteins. Cell Mol Life Sci 63:505–514.

33. Blasig IE, Mertsch K, Haseloff RF (2002) Nitronyl nitroxides, a novel group ofprotective agents against oxidative stress in endothelial cells forming the blood-brainbarrier. Neuropharmacology 43:1006–1014.

34. Hammes A, et al. (2001) Two splice variants of the Wilms’ tumor 1 gene have distinctfunctions during sex determination and nephron formation. Cell 106:319–329.

1430 | www.pnas.org/cgi/doi/10.1073/pnas.0911996107 Zhang et al.