The occludin and ZO-1 complex, defined by smallangle X-ray scattering and NMR, has implicationsfor modulating tight junction permeabilityBrian R. Tasha,1,2, Maria C. Bewleya,1, Mariano Russob, Jason M. Keilc, Kathleen A. Griffina, Jeffrey M. Sundstromd,David A. Antonettic, Fang Tiana, and John M. Flanagana,3

aDepartments of Biochemistry and Molecular Biology, bCellular and Molecular Physiology, and dOphthalmology, Pennsylvania State University College ofMedicine, H171, 500 University Drive, Hershey, PA 17033; and cDepartment of Ophthalmology and Visual Sciences, Kellogg Eye Center, 1000 Wall Street,Ann Arbor, MI 48105

Edited by Axel T. Brunger, Stanford University, Stanford, CA, and approved May 21, 2012 (received for review January 2, 2012)

Tight junctions (TJs) are dynamic cellular structures that are criticalfor compartmentalizing environments within tissues and regulat-ing transport of small molecules, ions, and fluids. Phosphoryla-tion-dependent binding of the transmembrane protein occludinto the structural organizing protein ZO-1 contributes to the regula-tion of barrier properties; however, the details of their interactionare controversial. Using small angle X-ray scattering (SAXS), NMRchemical shift perturbation, cross-saturation, in vitro binding, andsite-directed mutagenesis experiments. we define the interfacebetween the ZO-1 PDZ3-SH3-U5-GuK (PSG) and occludin coiled-coil(CC) domains. The interface is comprised of basic residues in PSGand an acidic region in CC. Complex formation is blocked by a pep-tide (REESEEYM) that corresponds to CC residues 468–475 andincludes a previously uncharacterized phosphosite, with the phos-phorylated version having a larger effect. Furthermore, mutationof E470 and E472 reduces cell border localization of occludin. To-gether, these results localize the interaction to an acidic regionin CC and a predominantly basic helix V within the ZO-1 GuKdomain. This model has important implications for the phosphor-ylation-dependent regulation of the occludin∶ZO-1 complex.

membrane-associated guanylate kinase ∣ tricellulin ∣ calmodulin

Tight junctions (TJs) are highly polarized gates that control theflux of fluids, proteins, and even ions across sheets of endothe-

lial or epithelial cells (1, 2). These barriers function in a range oftissues, including the vasculature of the central nervous system,kidney, and gut epithelium. Dysregulation of barrier properties isassociated with a host of disease states including cancer, stroke,diabetic retinopathy, and inflammatory bowel syndrome (3–5).Furthermore, viral and bacterial pathogens exploit specific TJproteins to gain access to host cells. Thus, understanding the con-tribution of TJ components in barrier formation and regulationmay aid the development of therapies to restore barrier prop-erties, control barrier properties to promote drug delivery to re-gions of the CNS, and for the development of novel antibacterialand antiviral compounds.

Occludin is an integral membrane, vesicle-trafficking, MALand related proteins for vesicle trafficking and membrane link(MARVEL) protein (6), that has a role in regulating TJ proper-ties (7). For example, in endothelial cells, it regulates TJ barriersin response to cytokines, such as IFN-γ (8, 9), and growth factors,such as VEGF (10). Removal of the cytoplasmic C-terminalcoiled coil domain (CC; residues 413–522) of occludin leads tocytoplasmic localization with increased tracer flux through thejunctions and an inability to maintain the apical localization ofmarker proteins (11, 12), identifying the CC as a key elementin regulation (13). The CC directly associates with the SH3-U5-GuK domains of the membrane-associated guanylate kinasehomolog (MAGUK) protein, ZO-1 (14). ZO proteins arecharacterized by their core PDZ3-SH3-GuK (PSG) domains,

preceded by two additional PDZ domains and separated byunique regions (U1–6), each with distinct functions (15).

The interaction of ZO-1 and occludin is modulated by phos-phorylation of residues in CC that affects the properties of TJs(7). Biochemical and biophysical experiments have led to two dis-tinct models for the complex (Fig. 1). In the first, (Model 1) acluster of basic residues on CC (K433, K444, K485, K488, K504,and K511) form the interface with ZO-1 (16). In the second(Model 2), the interface was formed between the acidic surfaceof CC and regions of the ZO-1 U5 and GuK domain (17–19). Thecrystal structures of ZO-1 PSG and SH3-GuK domains (20–22)seemed to support the Model 2, since the surface of the GuKdomain is largely basic (Fig. 1B). The potential contributions ofthe ZO-1 U6 (not part of the minimal CC binding domain) or U5motifs to the interface are unknown as they are absent from thestructures. Thus, no clear consensus exists on how occludin andZO-1 interact or how phosphorylation modulates this interaction.

In this study, we present a structural model for the minimalcomplex between occludin and ZO-1 (CC∶PSG) where residues468–475 (REESEEYM) in the CC acidic head directly interactwith helix V of ZO-1 GuK. This model is consistent with ourdata from SAXS, NMR, in vitro, and ex vivo binding studiesand in transiently transfected Madin–Darby canine kidney(MDCK) cells. Furthermore, we found that a phosphopeptide(YREEpS471EEYM) bound approximately 20-fold tighter thanits nonphosphorylated equivalent to PSG, implicating S471 phos-phorylation in regulating binding. In light of these results, wediscuss roles for CC phosphorylation in complex stability andTJ regulation.

ResultsCC and PSG Form an Extended Complex in Solution. To discriminatebetween the opposing models for the ZO-1∶occludin complex(16, 19), we measured the SAXS curves of the minimal bindingdomains (CC and PSG) alone and in their binary complex(Fig. 2A). From the Guinier plot, all samples were monodisperse(Fig. 2A and SI Appendix, Fig. S1A) and monomeric, based onthe relative scattering intensities at zero angle and Porod volumes

Author contributions: B.R.T., M.C.B., J.M.S., D.A.A., F.T., and J.M.F. designed research; B.R.T.,M.C.B., M.R., J.M.K., K.A.G., F.T., and J.M.F. performed research; B.R.T., M.C.B., K.A.G.,J.M.S., D.A.A., F.T., and J.M.F. contributed new reagents/analytic tools; B.R.T., M.C.B.,M.R., J.M.K., K.A.G., J.M.S., D.A.A., F.T., and J.M.F. analyzed data; and B.R.T., M.C.B.,D.A.A., F.T., and J.M.F. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1B.R.T. and M.C.B. contributed equally to this work.2Present address: Renal Electrolyte and Hypertension Division, University of Pennsylvania,Philadelphia, PA 19104.

3To whom correspondence should be addressed. E-mail: jmf27@psu.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1121390109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1121390109 PNAS ∣ July 3, 2012 ∣ vol. 109 ∣ no. 27 ∣ 10855–10860

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

of CC and PSG. The molecular weight of the CC∶PSG complexwas consistent with a 1∶1 stoichiometry, and this was confirmedin a titration experiment (SI Appendix, Fig. S1B). The radius ofgyration (Rg) extrapolated to zero concentration of CC andPSG were 23.2� 0.4 Å (approximately 24 Å, 1WPA.PDB,3G7C.PDB) (7, 16) and 28.9� 0.3 Å (approximately 29 Å,3SHW.PDB) (22), respectively, and 35.6� 0.6 Å for the complex(Fig. 2A). These values correspond to a centers-of-mass separa-tion of 51� 2 Å from the parallel axis theorem (23), suggestingan extended complex. Further, the distance distribution func-tions, PðrÞ, and maximum chord length, Dmax, for the CC andPSG alone and in complex support this conclusion (Fig. 2B).The shape of the PðrÞ curve for CC is indicative of a rod

(Dmax ¼ 80 Å), while PSG is more compact (Dmax ¼ 95 Å),consistent with their crystal structures (CC Dmax approximately75 Å, 1XAW.PDB; PSG Dmax approximately 93 Å, 3SHW.PDB).The complex is significantly longer (Dmax ¼ 140 Å).

The SAXS-derived molecular envelope of CC was rod-like(Fig. 2C), while PSG was boot-shaped in high or low NaCl(Fig. 2D and SI Appendix, Fig. S1 C and D) with the SH3 atthe heel, GuK forming the ankle, and PDZ3 the toes. The envel-ope of the complex was extended at the ankle, presumably due tobinding of the CC (Fig. 2E). Crystal structures of CC and PSGwere fit as rigid molecules to the SAXS data (SI Appendix,Fig. S2) (24), either unconstrained or constraining the relativeplacement of CC and PSG (SI Appendix, Fig. S2). Constrainingeither E470 or 472 (CC, acidic head) to be <7 Å from K760(PSG, GuK, helix V) resulted in the best fit to the data (SIAppendix, Fig. S2) and gave marginally better χ 2 values than with-out constraints. This arrangement is consistent Model 2 (17–19).Constraints involving K433 (CC), to test Model 1, gave muchpoorer fits to the scattering data (SI Appendix, Fig. S2). More-over, constraining the model to be a homodimer of PSG, ratherthan a CC∶PSG complex, fit the data poorly (χ 2 ¼ 3.1), consis-tent with the observation that dimerization occurs through PDZ2,which was not in our construct (25).

The binding of PSG to CC was characterized in a quantitativecapture assay (SI Appendix, Fig. S3). PSG (2 μM) was soluble anddid not bind to GST beads alone or in the presence of GST, evenin 50 mM NaCl. It was efficiently captured on these beads byGST-CC (SI Appendix, Fig. S3A, 2 μM). The amount capturedwas saturable and the maximum relative intensity for the cap-tured PSG and CC was consistent with a 1∶1 stoichiometry at50 mM NaCl (SI Appendix, Fig. S3 B and C). Under these con-ditions, the apparent dissociation constant, Kd, was 1.7� 0.2 μMand increased with NaCl concentrations (SI Appendix, Fig. S3C,6� 1 μM at 100 mM NaCl; 12� 2 μM at 150 mM NaCl,). Thesedata indicate a net displacement,Δn, of 1.7 Naþ plus Cl− in form-ing the GST-CC∶PSG complex (SI Appendix, Fig. S3C), consis-tent with an ionic interface.

CC Binds to Basic Residues in Helix V of the ZO-1 GuK Domain. Todefine the CC binding surface on PSG, we focused on the regionin and around residues 749–768 (helix V, GuK), a conserved, pre-dominantly basic region that contributes to calmodulin (CaM)

Fig. 1. Ribbon diagrams showing location of key residues in CC and PSG.Residues characterized in detail are labeled and shown in stick representa-tion. (A, Top) Ribbon diagram of CC. K433, K444, K485, K488, K504, andK511 (basic face) (16) are blue. (Middle) Surface representation with arrowsand a circle highlighting the basic (blue) and acidic (red) surfaces, respec-tively. (Bottom) Residues in the loop between the first two helices are drawnas sticks (acidic, red; hydrophobic, yellow; phosphorylatable Ser, purple). Allother residues are shown as a cyan ribbon. (B, Top) Ribbon diagram of PSG(PDZ3, orange; SH3, mid-blue; GuK, green) showing basic residues in helix V(blue). The asterisk (*) denotes the ends of the U5 motif. (Middle) Surfacerepresentation colored as in A. A circle highlights the region around helixV. (Bottom) Residues within helix V involved in CaM binding (20).

Fig. 2. Analysis of the SAXS data for CC, PSG, and their binary complex.(A) Plot of log of scattered intensity versus scattering angle, q, for CC (black),PSG (red), and their binary complex (blue); (Inset) Guinier plot for each. (B) PðrÞcurves. (C) Fit of CC (cyan ribbon), (D) PSG (PDZ3, red; SH3, blue; GuK, green),and (E) the binary complex, into their respective ab initio envelopes (beige net).

Fig. 3. CaM competes with CC for binding to the ZO-1 GuK domain. (A) Ad-dition of increasing amounts of purified CaM inhibits the capture of PSG byGST-CC. A representative Coomassie-stained gel of the GST-CC and PSGbound to glutathione resin as a function of added CaM with a summaryof the amount of captured PSG relative to GST-CC relative (n ¼ 3). For statis-tical significance: * p < 0.05, ** p < 0.01, and *** p < 0.001. (B) The chargeswaps in helix V of PSG 2D/E and 3D/E cannot bind CaM (20) and are notcaptured by GST-CC on glutathione beads.

10856 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1121390109 Tash et al.

binding (20). The CaM∶PSG complex by SAXS was elongated(Rg 35.4� 0.4 Å) with a 1∶1 stoichiometry. The SAXS-derivedenvelope suggested that CaM bound to PSG in a similar positionto CC (SI Appendix, Fig. S4). Further, CaM inhibited, in a dose-dependent manner, the capture of PSG by GST-CC (Fig. 3A),suggesting that they share a common binding site on ZO-1.The basic residues in helix V GuK are a primary determinantof CaM binding (20). Therefore, we tested the effect on CC bind-ing of the variants used in the CaM studies [(20); K749D/R752D/K753E (PSG 3D/E) and K760E/K763E (PSG 2D/E) in helix VofZO-1 GuK)]. Both PSG variants were captured in significantlyreduced amounts by GST-CC (Fig. 3B), demonstrating thatCC and CaM are direct competitors for PSG and share a bindingsite that includes basic residues within helix V GuK.

The Acidic Head of CC Forms the Interface. In the optimal SAXS-based model, the acidic head of CC forms part of the ZO-1 bind-ing site; however, the data does not conclusively exclude bindingvia the opposite end of CC near its N and C termini (residues414–439 and 508–522). To resolve this question, we performedNMR chemical shift perturbation and cross-saturation experi-ments. Backbone resonance assignment of 109 of 112 residuesin CC was accomplished with the transverse relaxation optimizedspectroscopy (TROSY) form of standard triple resonance experi-ments (SI Appendix, Fig. S5) (26, 27) and provided the basisfor identification of the CC-binding surface. Chemical shift per-turbation experiments comparing CC and CC∶PSG identified

two clusters of residues that showed small but detectable changesin the TROSY spectrum. The first cluster was located within helix1 (N454, E456, and R459) and the second in the acidic head(Fig. 4A and SI Appendix, Fig. S6; L464, D465, Y467, E470,S471, E473, M475, and A478). No clusters in chemical shift per-turbations were observed for residues at the opposite end of theCC. The observed chemical shifts are relatively small because theCC∶PSG interaction is most likely dominated by side chains withlittle perturbation of the backbone environment. There wasfurther evidence of complex formation when comparing theTROSY spectra of 15N, 2H labeled (random fractionally deuter-ated) PSG with and without unlabeled CC (SI Appendix, Fig. S7).Upon addition of CC numerous resonances appeared between8.1–8.5 ppm 1H and 120–126 ppm 15N and a large peak asso-ciated with an unfolded resonance signal disappeared, suggestingsome structural rearrangement in PSG upon complex formation.

To determine whether the chemical shift perturbations resultfrom direct binding or allosteric effects, NMR cross-saturationexperiments (28) were performed. Perdeuterated ½15N�-CC wasbound to protonated, unlabeled PSG at 2∶1, 1∶1, and 1∶2 molarratios. Due to the tendency of free PSG to aggregate at high pro-tein concentrations and prolonged incubation at 27 °C, total PSGwas kept at 100 μMand short collection times were used (<24 h).To overcome the sensitivity limitations, buffers containing 40%and 70% D2O were used instead of the more usual 90% D2O.To avoid potential false positives arising from the spin diffusioneffect, we used a conservative criterion for identifying perturbed

Fig. 4. The acidic head of CC mediates interactionwith ZO-1. (A) Close-up of TROSY spectrum of per-deuterated ½15N�-apo (red) CC overlaid with PSG-bound perdeuterated ½15N�-CC (black) showing theshifts for residues affected by the addition of CC(Y467, A478) and those unaffected (K433). (B) 1Dtraces of the residues in A from cross-saturation re-laxation experiments for saturation at 15 ppm (red)and 1 ppm (black). (C) The in vitro capture efficiencyof PSG by GST-CC is reduced by charge reversals atresidues 465∕469, 470, 472, and 473 in the acidichead of CC. (D) The variant GST-CC, except K433D,had a similar effect upon the capture of full-lengthZO-1 from MDCK lysates. Relative binding (belowthe gel) was the intensity ratio of the ZO-1 bandin each lane and the wt condition.

Tash et al. PNAS ∣ July 3, 2012 ∣ vol. 109 ∣ no. 27 ∣ 10857

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

resonances. L464, D466, Y467, and A478 were at the interface inthese experiments (Fig. 4B and SI Appendix, Fig. S8). No residuesaround the N and C termini showed significant cross-saturationrelaxation effects. Taken together, the SAXS and NMR experi-ments demonstrate that CC binds with its acidic head in proximityto PSG, while the opposite end, containing its N and C termini, isfurthest from this interface.

To probe the contribution of specific residues in the acidichead to complex stability, the effect of single (E470K, E472K, andE473K) and double (D465K/E469K) substituted variants of GST–CC were evaluated in PSG capture assays (Fig. 4C). The E473Kvariant marginally reduced capture (82% of wt, p < 0.001) whilemore dramatic reductions were observed for the E470K, E472K,and D465K/E469K variants [43%, 47%, and 23% of wt (allp < 0.001), respectively]. The E470K, E472K, and D465K/E469KGST-CC variants also showed reduced capture of full-length ZO-1from MDCK cell lysates (Fig. 4D; 18%, 49%, and 20% of wt, re-spectively, p < 0.001) confirming the importance of the negativecharges on CC in mediating its interaction with ZO-1. By contrast,the K433D substitution, located on the basic face of CC, displayedopposite effects in vitro (Fig. 4C) and in lysate (Fig. 4D). In vitro,K433D enhanced capture of PSG (136% of wt, p < 0.001) but, asreported previously (16), it abolished the interaction with full-length ZO-1 (14% of wt, p < 0.001), suggesting that K433 maybe involved in a contact present in full-length ZO-1 (29) but notin PSG. A possible candidate is the acidic U6 motif (residues 803–888), that regulates complex stability but is not required for thebinding of theminimal interacting domains (29). However, we can-not exclude other possibilities, such as post-translational modifica-tions of ZO-1 or additional factors present in the extract.

E470 and E472 in CC Contribute to TJ Localization in Cells. SinceE470K and E472K substitutions had profound effects on ZO-1binding, full-length V5-tagged occludin or variants containingthese substitutions were transiently transfected into MDCK cells.Expression of wt occludin resulted in immunostaining at the cell

membrane, with the majority occurring at cell∶cell contacts,colocalizing with ZO-1 and endogenous occludin (Fig. 5A). The3D rendering of these confocal images revealed good colocaliza-tion of wt occludin with ZO-1 at the apical membrane border(Fig. 5B). By contrast, the E470K and E470K/E472K occludinvariants showed increased localization in the cytoplasm and at thelateral border (Fig. 5B) and reduced staining at cell contacts, withloss of ZO-1 colocalization. To quantify the change in localiza-tion, the ratio of Triton X-100-soluble and insoluble expressedoccludin was compared to ZO-1. As expected, the majority of wtoccludin was found in the Triton X-100-insoluble fraction, alongwith ZO-1. However, the E470K and E470K/E472K variants werelocated in the soluble fraction, with a 66% (p < 0.01) decreasein the ratio of Triton X-100-insoluble∶soluble occludin for the dou-ble variant (SI Appendix, Fig. S9). These results suggest that theinteraction of ZO-1 with the acidic head of CC is necessary foreither the transport to, or maintenance of, occludin at TJs.

Phosphorylation of S471 May Enhance Occludin∶ZO-1 Complex Stabi-lity. The two acidic residues, E470 and E472, with a key role inlocalizing occludin at TJs, flank S471, a previously identifiedphosphosite (7). To investigate the effect of S471 phosphoryla-tion, occludin containing a S471D substitution (S471D) was tran-siently transfected into MDCK cells. Using confocal microscopy,S471D demonstrated good apical border colocalization withZO-1 (Fig. 5).

To further define the effect of S471 phosphorylation, peptideswith and without a phosphogroup on S471 (residues 468–475,REESEEYM, and REEpSEEYM) were tested for their abilityto inhibit PSG capture by GST-CC (Fig. 6A). Both peptidessignificantly inhibited the capture of PSG [19% and 33% of wt,(p < 0.001), respectively], with the phosphorylated peptide beingmore efficient than the nonphosphorylated peptide (p < 0.01).This observation was similar in experiments to capture endogen-ous ZO-1 from MDCK cell lysates (Fig. 6B; 32% and 41%,respectively, p < 0.05). By contrast, peptides containing S490

Fig. 5. Charge reversal mutations in the acidichead disrupt occludin localization. Confocal imagesof immunocytochemistry for ZO-1, occludin, andV5-tagged occludin exogenously expressed inMDCK cells transiently transfected with V5 taggedwt, E470K, E470K/E472K, or S471D variants of occlu-din. (A) Max projected confocal image. Arrows indi-cate failure of E470K/E472K mutant to colocalizewith ZO-1. There is good localization for S471D.Bar represents 10 μm. (B) Topical surface of V5 stain-ing and V5 staining plus ZO-1 rendered in 3Dvolume. Bar represents 2 μm.

10858 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1121390109 Tash et al.

(residues 486–494, QVKGSADYK and QVKGpSADYK), a sec-ond validated phosphosite in CC (7, 10) and a control peptideencompassing K433 (residues 430–438, QLYKRNFDT) did notsignificantly alter capture efficiency in vitro or ex vivo.

To quantitate binding of the S471-containing peptides, a fluor-escently labeled peptide corresponding to residues 467–475 CC[FITC—(miniPEG)—YREESEEYM] was synthesized. Basedupon the changes in fluorescent anisotropy of this peptide as afunction of PSG concentration, the Kd for labeled peptide bind-ing was 1.8� 0.4 μM (Fig. 6C and SI Appendix), although thefluorophore and the linker appeared to contribute slightly to theinteraction. In competition assays, the equivalent unlabeled pep-tide (Ac—YREESEEYM) bound with a Kd of 15� 3 μM, whilethe serine phosphorylated peptide (Ac—YREEpSEEYM) in-creased the affinity by approximately 20-fold (Kd ¼ 0.70�0.05 μM). By contrast, a scrambled peptide (Ac-EEYRSYMEE)did not affect binding of the labeled peptide (Fig. 6C). Taken as awhole, these results confirm that the acidic head of CC directlymediates its interaction with the GuK domain of PSG with highaffinity and phosphorylation of S471 enhances occludin∶ZO-1complex stability.

DiscussionPrevious studies of TJs have established a close associationbetween the scaffolding protein, ZO-1, and the transmembraneprotein, occludin, and identified multiple phosphorylation sites aspotential regulators of their interaction. However, the moleculardetails of this interaction were unclear. The results described heresuggest a model for complex formation utilizing two distinct in-terfaces that contain phosphorylatable residues within or aroundtheir periphery (Fig. 7). The core of the interface is electrostatic,involving D465, E469, E470, E472, and E473 in CC and residuesK749, R752, K753, K760, and K763 in PSG. Consistent with thismodel, the stability of the complex is salt-dependent. In addition,helix V (residues 751–765) (20) is used by two other MAGUKfamily members for binding their phosphosites (30, 31). Finally,the coiled-coil domain of tricellulin, a homolog of occludin pre-sent at tricellular junctions, lacks the acidic residues in its head

domain and does not interact strongly with PSG in SAXS andcapture assays (SI Appendix, Fig. S10) (32).

The argument for a second binding interaction between occlu-din and ZO-1 is based upon differences in the effects of someamino acid substitutions in CC on their capture efficiency ofPSG and longer ZO-1 constructs. Charge reversal substitutionsof lysine residues in the basic face of CC did not affect captureof PSG in vitro but, consistent with earlier data, showed reducedcapture of longer ZO-1 constructs (1–888 and full-length ZO-1)that contain U6 (16). Therefore, we speculate there is a secondionic interface that involves interactions between the basic face ofCC, including K433, and acidic residues on the surface of the full-length ZO-1, perhaps within the U5 or U6 motifs (16, 29, 33) orinvolving a bridge with another cellular factor. The presence ofthis secondary interface may enhance the stability of the complexin the cellular environment and/or contribute to the regulationof the complex. One very attractive possibility is that binding ofZO-1 binding to the secondary interface controls access to thelysine-rich face of CC, which contains seven of the 12 conservedlysine residues in occludin, and thus its ubiquitin-dependent en-docytosis observed in endothelial cells (10).

In vivo, the occludin∶ZO-1 interaction is modulated by phos-phorylation events (7, 34–38). In occludin, four of these sites seg-regate structurally and possibly functionally—two adjacent to theprimary interaction site (S471 and Y474) in the acidic head andtwo in the secondary interface (S490 and S508) on the basic face.Y474 phosphorylation is associated with binding p85 of phospha-tidyl inositol 3-kinase at the leading edge of migrating cells (39)while S490 and S508 are associated with vascular permeabilityafter VEGF treatment (7) or HIV encephalitis (40). Our resultssuggest a role for S471 phosphorylation in ZO-1 binding. The CCpeptide 468–475 containing S471 and Y474, is more effective atblocking PSG and ZO-1 capture when phosphorylated at S471(pS471), suggesting that phosphorylation here may enhance com-plex formation in vivo. Indeed, in MDCK cells, the S471D-occlu-din variant exhibited at least as good colocalization with ZO-1 aswt (Fig. 5). The results are consistent with the recent findingthat the GuK domain has evolved as a phosphopeptide-bindingmodule (31). The Kd for pS471 peptide binding to PSG (0.7 μM)is very similar to that of p-LGN binding to SAP97 SH3-GuK(0.22 μM). However, the Kd values of the equivalent nonpho-sphorylated peptides are different (S471 to PSG, approximately15 μM; LGN to SPA97, >100 μM). Interestingly, occludin pos-sesses the requisite arginine at −3 relative to the phosphositeof GuK binding peptides (31); however, ZO-1 does not containthe sequence required for high affinity binding. These results areconsistent with S471 phosphorylation strengthening or modulat-ing binding to ZO-1 rather than providing an all or none bindingmotif. By contrast, phosphorylation of S490 reduces its inter-action with endogenous ZO-1 (7). While the effect of S508 phos-phorylation on ZO-1 binding is not known, its location adjacentto K504 and K511, which are required for complex formationwith endogenous ZO-1 (16), suggests that S508 phosphorylationwould reduce ZO-1 binding. Further, phosphorylation of either

Fig. 6. A phosphopeptide containing S471 competes with CC binding toPSG. (A) In vitro, only inclusion of peptides (200 μM) spanning residuesthe acidic head, 468–475 (�pS471) inhibited capture of PSG by GST-CC onglutathione beads. (B) In MDCK lysates, only the peptides in A significantlyreduced capture of full-length ZO-1 by GST-CC (n ¼ 3). (C) Equilibrium bind-ing of the fluorescent-labeled peptide, FITC-(miniPEG)-YREESEEYM, to PSGin the absence and presence of unlabeled peptides. In all experiments, 3 nM-labeled peptide was titrated with purified PSG. A representative titrationand fit to a single binding site model is shown (▪, solid line). Inclusion ofa nonfluorescently labeled version of this peptide (○, long dash, weak com-petition) or this peptide with S471 phosphorylated (▾, short dash, strongcompetition) but not a scrambled version of this sequence (Δ, no competi-tion) reduced PSG binding compared to the unlabeled peptide alone.

Fig. 7. Working model of the CC∶PSG interaction and the role of the variousphosphosites on CC. Domains shown as rectangles, colored as Fig. 1. Phospho-sites are ovals that are shaded when phosphorylated. U5 and U6 are repre-sented as lines.

Tash et al. PNAS ∣ July 3, 2012 ∣ vol. 109 ∣ no. 27 ∣ 10859

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

S490 or S508 correlates with opening of the barrier in endothelialcells (7, 40), suggesting that the basic face of CC is involved inregulating TJ barrier properties in vivo. Five additional phospho-sites that lie in a conserved region just upstream of CC (Y398,Y402, T403, T404, and T408) also regulate phosphorylation-de-pendent complex formation. Based on geometric considerations,they cannot exert their effect directly at the primary interactionsite and may modulate binding through a different site, perhapsclose to the proposed secondary interaction site.

In summary, our model for the complex between CC and PSG,provides a framework for understanding the functional role ofthe occludin∶ZO-1 complex in TJ permeability via occludin phos-phorylation and offers new insight into the regulation of barrierproperties. These studies highlight the direct interaction of theCC, through its acidic head, with the basic region surroundinghelix V in PSG. Undoubtedly, additional contacts exist in thelarger TJ complex that contribute to complex organization;however, the relative contribution and mechanism of action ofoccludin phosphorylation on its interaction with ZO-1 is emer-ging. A complete understanding of the interface may uncovernew therapeutic strategies to selectively alter TJ permeabilityin the treatment of disease or delivery of drugs to the centralnervous system.

MethodsE. coli expressed human CC and human PSG were expressed as tobacco etchvirus (TEV) protease cleavable His6-tagged fusion proteins; the TEV-cleavedform of each was used for SAXS and NMR experiments. A GST-tagged versionof the CC (413–522) was expressed and purified using standard protocols (SIAppendix, Methods) and used in vitro and in cell lysate capture assays. Thefinal concentrations of proteins used in individual experiments are given inthe text and the figure legends. SAXS experiments were conducted at X9,National Synchrotron Light Source, Upton, NY, using a standard configura-tion and software. NMR experiments were conducted on Bruker Avance600 MHz and 850 MHz spectrometers, equipped with cold probes using stan-dard protocols. In-cell based assays were conducted with MDCK cells usingstandard protocols. Additional details are available in the SI Appendix.

ACKNOWLEDGMENTS.We thank Drs. Marc Allaire and Lin Yang for discussionsand technical assistance with small angle X-ray scattering. We thank thePennsylvania Lions Sight Conservation and Eye Research Foundation(J.M.F.), the American Diabetes Association Grant 7-07-RA-34 (to J.M.F.),National Eye Institute Grant EY012021 (to D.A.), and the Pennsylvania Tobac-co Funds (J.M.F., F.T., and D.A.) for financial support. The Kellog Eye Centerimagine core is supported by National Institutes of Health Grants EY07003and P60DK020572. The National Synchrotron Light Source and beamlineX9 is supported by the US Department of Energy, Office of Science, Officeof Basic Energy Sciences, under Contract No. DE-AC02-98CH10886.

1. Van Itallie CM, Anderson JM (2004) The molecular physiology of tight junction pores.Physiology (Bethesda) 19:331–338.

2. Forster C (2008) Tight junctions and the modulation of barrier function in disease.Histochem Cell Biol 130:55–70.

3. Edelblum KL, Turner JR (2009) The tight junction in inflammatory disease: Communi-cation breakdown. Curr Opin Pharmacol 9:715–720.

4. Fanning AS, Mitic LL, Anderson JM (1999) Transmembrane proteins in the tightjunction barrier. J Am Soc Nephrol 10:1337–1345.

5. Mitic LL, Van Itallie CM, Anderson JM (2000) Molecular physiology and pathophysiol-ogy of tight junctions I. Tight junction structure and function: Lessons from mutantanimals and proteins. Am J Physiol Gastrointest Liver Physiol 279:G250–G254.

6. Sanchez-Pulido L, Martin-Belmonte F, Valencia A, Alonso MA (2002) MARVEL: Aconserved domain involved in membrane apposition events. Trends Biochem Sci27:599–601.

7. Sundstrom JM, et al. (2009) Identification and analysis of occludin phosphosites: Acombinedmass spectrometry and bioinformatics approach. J Proteome Res 8:808–817.

8. Marchiando AM, et al. (2010) Caveolin-1-dependent occludin endocytosis is requiredfor TNF-induced tight junction regulation in vivo. J Cell Biol 189:111–126.

9. Van Itallie CM, Fanning AS, Holmes J, Anderson JM (2010) Occludin is required forcytokine-induced regulation of tight junction barriers. J Cell Sci 123:2844–2852.

10. Murakami T, Felinski EA, Antonetti DA (2009) Occludin phosphorylation and ubiqui-tination regulate tight junction trafficking and vascular endothelial growth factor-induced permeability. J Biol Chem 284:21036–21046.

11. Balda MS, et al. (1996) Functional dissociation of paracellular permeability and trans-epithelial electrical resistance and disruption of the apical-basolateral intramembranediffusion barrier by expression of a mutant tight junction membrane protein. J CellBiol 134:1031–1049.

12. FuruseM, et al. (1994) Direct association of occludin with ZO-1 and its possible involve-ment in the localization of occludin at tight junctions. J Cell Biol 127:1617–1626.

13. Nusrat A, et al. (2000) The coiled-coil domain of occludin can act to organize structuraland functional elements of the epithelial tight junction. J Biol Chem 275:29816–29822.

14. Fanning AS, Anderson JM (2009) Zonula occludens-1 and -2 are cytosolic scaffolds thatregulate the assembly of cellular junctions. Ann NY Acad Sci 1165:113–120.

15. Bauer H, Zweimueller-Mayer J, Steinbacher P, Lametschwandtner A, Bauer HC (2010)The dual role of zonula occludens (ZO) proteins. J Biomed Biotechnol 2010:402593, 10.1155/2010/402593.

16. Li Y, Fanning AS, Anderson JM, Lavie A (2005) Structure of the conserved cytoplasmicC-terminal domain of occludin: Identification of the ZO-1 binding surface. J Mol Biol352:151–164.

17. Muller SL, et al. (2005) The tight junction protein occludin and the adherens junctionprotein alpha-catenin share a common interaction mechanism with ZO-1. J Biol Chem280:3747–3756.

18. Schmidt A, Utepbergenov DI, Krause G, Blasig IE (2001) Use of surface plasmon reso-nance for real-time analysis of the interaction of ZO-1 and occludin. Biochem BiophysRes Commun 288:1194–1199.

19. Schmidt A, et al. (2004) Occludin binds to the SH3-hinge-GuK unit of zonula occludensprotein 1: Potential mechanism of tight junction regulation. Cell Mol Life Sci61:1354–1365.

20. Lye MF, Fanning AS, Su Y, Anderson JM, Lavie A (2010) Insights into regulated ligandbinding sites from the structure of ZO-1 Src homology 3-guanylate kinase module.J Biol Chem 285:13907–13917.

21. Nomme J, et al. (2011) The Src homology 3 domain is required for junctional adhesionmolecule binding to the third PDZ domain of the scaffolding protein ZO-1. J Biol Chem286:43352–43360.

22. Pan L, Chen J, Yu J, Yu H, Zhang M (2011) The structure of the PDZ3-SH3-GuK tandemof ZO-1 suggests a supramodular organization of the MAGUK family scaffold proteincore. J Biol Chem 286:40069–40074.

23. Engelman DM,Moore PB (1975) Determination of quaternary structure by small angleneutron scattering. Annu Rev Biophys Bioeng 4:219–241.

24. Petoukhov MV, Svergun DI (2005) Global rigid body modeling of macromolecularcomplexes against small-angle scattering data. Biophys J 89:1237–1250.

25. Fanning AS, Lye MF, Anderson JM, Lavie A (2007) Domain swapping within PDZ2 isresponsible for dimerization of ZO proteins. J Biol Chem 282:37710–37716.

26. Salzmann M, Wider G, Pervushin K, Senn H, Wuthrich K (1999) TROSY-type triple-resonance experiments for sequential NMR assignments of large proteins. J Am ChemSoc 121:844–848.

27. Salzmann M, Pervushin K, Wider G, Senn H, Wuthrich K (1998) TROSY in triple-resonance experiments: New perspectives for sequential NMR assignment of largeproteins. Proc Natl Acad Sci USA 95:13585–13590.

28. Takahashi H, Nakanishi T, Kami K, Arata Y, Shimada I (2000) A novel NMR method fordetermining the interfaces of large protein-protein complexes. Nat Struct Biol7:220–223.

29. Fanning AS, et al. (2007) The unique-5 and -6 motifs of ZO-1 regulate tight junctionstrand localization and scaffolding properties. Mol Biol Cell 18:721–731.

30. Reese ML, Dakoji S, Bredt DS, Dotsch V (2007) The guanylate kinase domain of theMAGUK PSD-95 binds dynamically to a conserved motif in MAP1a. Nat Struct Mol Biol14:155–163.

31. Zhu J, et al. (2011) Guanylate kinase domains of theMAGUK family scaffold proteins asspecific phospho-protein-binding modules. EMBO J 30:4986–4997.

32. Raleigh DR, et al. (2010) Tight junction-associated MARVEL proteins marveld3, tricel-lulin, and occludin have distinct but overlapping functions.Mol Biol Cell 21:1200–1213.

33. Riazuddin S, et al. (2006) Tricellulin is a tight-junction protein necessary for hearing.Am J Hum Genet 79:1040–1051.

34. Balda MS, Anderson JM, Matter K (1996) The SH3 domain of the tight junction proteinZO-1 binds to a serine protein kinase that phosphorylates a region C-terminal to thisdomain. FEBS Lett 399:326–332.

35. Elias BC, et al. (2009) Phosphorylation of Tyr-398 and Tyr-402 in occludin prevents itsinteraction with ZO-1 and destabilizes its assembly at the tight junctions. J Biol Chem284:1559–1569.

36. Kale G, Naren AP, Sheth P, Rao RK (2003) Tyrosine phosphorylation of occludin attenu-ates its interactions with ZO-1, ZO-2, and ZO-3. Biochem Biophys Res Commun302:324–329.

37. Raleigh DR, et al. (2011) Occludin S408 phosphorylation regulates tight junctionprotein interactions and barrier function. J Cell Biol 193:565–582.

38. Rao RK, Basuroy S, Rao VU, Karnaky KJ, Jr, Gupta A (2002) Tyrosine phosphorylationand dissociation of occludin-ZO-1 and E-cadherin-beta-catenin complexes from thecytoskeleton by oxidative stress. Biochem J 368:471–481.

39. Du D, et al. (2010) The tight junction protein, occludin, regulates the directionalmigration of epithelial cells. Dev Cell 18:52–63.

40. Yamamoto M, et al. (2008) Phosphorylation of claudin-5 and occludin by rho kinase inbrain endothelial cells. Am J Pathol 172:521–533.

10860 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1121390109 Tash et al.