The scaffolding protein ZO-1 coordinates actomyosin and epithelial apical specializations in vitro and in vivo

Polarized epithelia assemble into sheets that compartmentalize organs and generate tissue barriers by integrating apical surfaces into a single, unified structure. This tissue organization is shared across organs, species, and developmental stages. The processes that regulate development and maintenance of apical epithelial surfaces are, however, undefined. Here, using an intestinal epithelial-specific knockout (KO) mouse and cultured epithelial cells, we show that the tight junction scaffolding protein zonula occludens-1 (ZO-1) is essential for development of unified apical surfaces in vivo and in vitro. We found that U5 and GuK domains of ZO-1 are necessary for proper apical surface assembly, including organization of microvilli and cortical F-actin; however, direct interactions with F-actin through the ZO-1 actin-binding region (ABR) are not required. ZO-1 lacking the PDZ1 domain, which binds claudins, rescued apical structure in ZO-1–deficient epithelia, but not in cells lacking both ZO-1 and ZO-2, suggesting that heterodimerization with ZO-2 restores PDZ1-dependent ZO-1 interactions that are vital to apical surface organization. Pharmacologic F-actin disruption, myosin II motor inhibition, or dynamin inactivation restored apical epithelial structure in vitro and in vivo, indicating that ZO-1 directs epithelial organization by regulating actomyosin contraction and membrane traffic. We conclude that multiple ZO-1–mediated interactions contribute to coordination of epithelial actomyosin function and genesis of unified apical surfaces.

Epithelial surfaces are composed of sheets of polarized cells. In these, the basolateral surfaces attach to adjacent cells and the underlying basement membrane, whereas apical surfaces interface directly with the extracellular (e.g. luminal) milieu. Apical specializations, including microvilli and cilia, vary by organ and cell type, but a constant feature of apical membranes is their precise integration to form a single, contiguous surface in which it can be difficult to identify boundaries between individual cells. This organization, which presents a united front to pathogens and other external stressors (1), is thermodynamically unfavorable (2,3) and must, therefore, require energy expenditure and be functionally important. Nevertheless, the mechanisms by which contiguous cell populations establish and maintain uniform apical surfaces have not been defined.
Even in the first stages of multicellular organism development, individual epithelial cells integrate to form a uniform apical surface; this structure is maintained during dynamic tissue processes. As early as embryo invagination, the apical surface of polarized epithelia undergoes constriction, in which the apical actomyosin network contracts to decrease cross-sectional apical surface area while maintaining continuous apical surfaces (4 -11). In developing murine intestinal epithelia, uniform apical surfaces are maintained during progression through transient pseudostratified structures (12). At the other end of the spectrum, apical borders are maintained during epithelial cell extrusion until the point where the cell being shed extends beyond adjacent cells (13)(14)(15)(16). In contrast, this tissue organization is lost in early neoplasia and when intercellular junctions are disrupted.
Direct interactions between F-actin and zonula occludens-1 (ZO-1) 2 have been implicated in a number of coordinated cellular processes, including tissue morphogenesis, maintenance of apical structure, and barrier regulation (17)(18)(19). To better define the contributions of these and other ZO-1-mediated interactions to epithelial organization, we generated intestinal epithelial-specific ZO-1 knockout (KO) mice. Apicobasal polarity of these epithelia was maintained, but apical membranes formed convex surfaces that disrupted apical surface continuity. Similar defects were present in ZO-1 knockdown (KD) MDCK monolayers. Here, we show that excessive contraction of subapical, but not perijunctional, actomyosin is responsible for the structural abnormalities induced by ZO-1 deficiency, both in vivo and in vitro. Interactions mediated by the ZO-1 U5 and GuK domains are critical to actomyosin organization. ZO-1 lacking the PDZ1 domain can restore apical structure in ZO-2-sufficient, but not ZO-2deficient, monolayers. This suggests that heterodimerization with ZO-2 can restore PDZ1-dependent ZO-1 functions. We conclude that ZO-1-mediated interactions within the apical junctional complex signal to apical F-actin to maintain uniform apical surfaces and that loss of these signals results in marked deformation of apical membranes and actin-based structures.

In vivo ZO-1 knockout disrupts apical cell structure and apical tissue architecture
The obligate role of ZO-1 during embryogenesis (20) has limited in vivo study of ZO-1 to date. To overcome this obstacle, we generated intestinal epithelial cell-specific ZO-1 knockout (Tjp1 fl/fl ; villin-cre ϩ ) mice (Fig. 1A). These mice grew and reproduced normally with expected Mendelian ratios and were free of spontaneous disease under specific pathogen-free conditions. Western blotting of purified intestinal epithelia and immunostains of intestinal tissue sections confirmed the absence of ZO-1 expression (Fig. 1, A and B). Crypt-villus architecture was maintained (Fig. 1B), but apical surface continuity was markedly disrupted. This included undulation of the apical surface brush border membrane and distinct crevasses at intercellular junctions between ZO-1 KO enterocytes.
To better define the altered apical morphology of ZO-1deficient enterocytes, intact portions of jejunum were fixed and imaged without physical sectioning. Confocal slices of these phalloidin-stained tissues followed by 3D reconstruction showed distinct convex surfaces, with depressed intercellular junctions in ZO-1 KO, but not WT, epithelium ( Fig. 1C and Video S1). These alterations were even more striking when viewed by scanning EM (SEM), where intercellular junctions between ZO-1-deficient enterocytes appeared as deep crevasses (Fig. 1D). In contrast, although the sites of intercellular junctions could be identified in WT enterocytes, they were subtle and were not depressed beneath the apical surface. SEM also revealed that the apical surface of ZO-1 KO epithelia appeared rough in contrast to the velvet-like brush border surfaces of WT epithelia (Fig. 1D). Higher magnification showed that the irregular apical surface of ZO-1 KO epithelia was due to highly variable microvillus length and diameter (Fig. 1D), in stark contrast to the regular length and uniform packing of microvilli in WT enterocytes.
Transmission EM (TEM) confirmed abnormal microvillus organization in ZO-1 KO epithelia (Fig. 1E). Long, nonparallel, thin strands extended beyond the normally uniform microvillus length and frequently included vesicles that were adherent to microvillus tips. Unlike the uniformly distributed and nearly identical-appearing microvillus rootlets seen at regular intervals in WT enterocytes (Fig. 1E), rootlet morphology and spacing were irregular in ZO-1 KO epithelia. Finally, ZO-1 KO intestinal epithelia included an expanded electron-dense zone beneath the apical membrane that obscured microvillus rootlets and excluded organelles (Fig. 1E, brackets). Conversely, tight junction structure was intact (Fig. 1E, right panels, arrowheads). Thus, despite displaying normal growth, reproduction, and overall tissue architecture, intestinal epithelial ZO-1 KO results in striking morphologic abnormalities that include disruption of apical surface architecture and microvilli with normal-appearing tight junctions.

ZO-1 knockdown in vitro phenocopies knockout in vivo
To determine whether in vitro ZO-1 KD recapitulates the striking in vivo effects of ZO-1 KO, WT (T23) and ZO-1 KD Madin-Darby canine kidney (MDCK) II cells were cultured to maturation on semipermeable supports, and apical structure was assessed by SEM and fluorescence microscopy (Fig. 2). As expected, WT monolayers displayed intact apical surfaces that were smooth and uniform. In contrast, apical surfaces of ZO-1 KD monolayers were disrupted by bulbous distensions. As a result, individual cell profiles were emphasized ( Fig. 2A, top panels, arrowheads). Although less well-organized than intestinal epithelial brush borders, microvilli of WT MDCK monolayers were of uniform length. In contrast, ZO-1-deficient MDCK monolayers displayed microvilli with highly variable lengths ( Fig. 2A, bottom panels, arrowheads), similar to ZO-1 KO enterocytes. ZO-2 KD monolayers were indistinguishable from WT, but apical surface and microvillus morphology of ZO-1/ZO-2 double-KD was markedly disrupted (Fig. 2). ZO-1, but not ZO-2, KD in vitro therefore causes disruptions of apical brush border membrane architecture similar to those induced by KO in vivo. In addition to demonstrating that the contributions of ZO-1 to apical architecture are generalizable beyond intestinal epithelium, these data indicate the dominant role of ZO-1, relative to ZO-2, in maintaining apical structure.

Apicobasal polarization is maintained in monolayers with irregular apical surfaces
Biogenesis of apical membranes is precisely controlled and is an essential component of epithelial polarization and brush border organization (18,(21)(22)(23)(24)(25). We therefore hypothesized that the striking morphological disruption of ZO-1-deficient cells might be accompanied by apicobasal polarity defects. Monolayers were stained for the apical marker podocalyxin/ gp135, the apical polarity protein aPKC, and the basolateral adherens junction protein E-cadherin. Despite the abnormalities induced by ZO-1 KD, all lines formed confluent monolayers with correct apical-basal polarization (Fig. 2, B and C). This is consistent with the retained polarized distributions of NHE3 and E-cadherin in vivo (Fig. 1B).
Changes in cell shape are frequently secondary to alterations in actomyosin organization and contractility (4,26). Consistent with this, both x,z reconstructions and 3D surface reconstructions of the dome-like apical surface deformations revealed dense accumulations of F-actin just below the apical membranes of ZO-1 KD epithelia ( Fig. 2 (C and D) and Video S2). This F-actin organization is seen in x,y projections as dense cytoplasmic F-actin rings (Fig. 2B, arrowheads). These were not seen in WT and ZO-2 KD monolayers but were only present in ZO-1 KD and ZO-1/ZO-2 double-KD monolayers. We thus considered the possibility that the F-actin rings were a consequence of increased actomyosin contractility. Increases in actomyosin contraction have been reported in ZO-1/ZO-2 double-KD cells but were associated with perijunctional F-actin and myosin IIB intensification and reorganization into a sarcomere-like banded pattern (26,27). This perijunctional F-actin and myosin IIB reorganization was not present in ZO-1 KD monolayers. However, ZO-1 KD showed clear myosin IIB rings that coincided with the F-actin rings (Fig. 2B, zoomed panels). These were also present in ZO-1/ZO-2 double-KD cells, but were distributed over larger areas, consistent with the more severe phenotype of these cells. These data suggest that ZO-2 can partially compensate for ZO-1 deficiency in terms of perijunctional actomyosin organization. Nevertheless, ZO-1 deficiency alone is sufficient to induce dense, cytoplasmic F-actin rings that are separate from perijunctional F-actin. To avoid confounding results related to the additional changes induced by ZO-1/ZO-2 double-KD, our studies of ZO protein function and regulation of apical structure therefore focused on ZO-1 single-KD monolayers and tissues.
In addition to corroborating the SEM results, orthogonal x,z projections also demonstrated that the apical distensions induced by ZO-1 depletion were enriched in gp135, aPKC, and

ZO-1 controls apical epithelial structure in vitro and in vivo
F-actin (Fig. 2C). Thus, ZO-1 KD has little effect on perijunctional actomyosin but induces striking changes in subapical F-actin and microvillus structure. To better define the alteredmorphologyofZO-1-deficientmonolayers,3Dreconstructions of F-actin distribution within each cell were created, and the height of each cell was measured (Fig. 2D). ZO-1 KD cells were taller and had markedly greater cell height variation than WT cells (p Ͻ 0.0001) (Fig. 2E).
We and others have shown that ZO-1 regulates epithelial proliferation in vitro (18,28,29). We therefore considered the possibility that increased crowding of ZO-1 KD monolayers was responsible for the apical distensions. One could, for example, hypothesize that, to maintain cell volume, cells became thinner and more elongated as proliferation continued in ZO-1 KD monolayers. However, the basolateral cross-sectional area of ZO-1 KD and WT cells in mature monolayers was similar (WT, 157.1 Ϯ 35.2 m 2 (n ϭ 137); KD, 150.4 Ϯ 35.2 m 2 (n ϭ 152); p ϭ 0.12 by two-tailed t test). Further, calcium-switch assays showed that WT and ZO-1 KD monolayers reached steady-state barrier function with similar kinetics (Fig. S1A). Arrowheads, intercellular junctions (low magnification) and abnormal microvilli (high magnification). Bars, 5 m (top panels) and 1 m (bottom panels). B, confluent monolayers were immunostained for podocalyxin/gp135 and E-cadherin, aPKC and F-actin (phalloidin), or myosin IIB. F-actin, aPKC, gp135, and E-cadherin images are maximum projections of the full monolayer thickness. Myosin IIB images are of a single confocal z-slice; zoomed views of the boxed areas are shown below each image. All low magnification images for each antigen are scaled identically; the high-power views of myosin-IIB are scaled individually to demonstrate the rings that colocalize with F-actin. Arrowheads, prominent cytoplasmic (subapical) F-actin and myosin-IIB rings in ZO-1 KD and ZO-1/ZO-2 double-KD monolayers. Bars, 10 m and 30 m for zoomed images. C, projections (x,z) of confluent monolayers stained for gp135 (green), E-cadherin (red), and nuclei (Hoechst; blue) (left panels) or F-actin (phalloidin; red), aPKC (green), and nuclei (Hoechst; blue) (right panels). Bars, 10 m. D, confluent WT and ZO-1 KD monolayers subjected to calcium switch were fixed and stained for F-actin (phalloidin) 24 h after calcium repletion. 3D reconstructions of confocal z-stacks of F-actin images were created, and cell reconstructions are color-coded by height as indicated. Bar, 10 m. E, graphical representation of cell height quantification. Boxes enclose second and third quartiles, the center bar indicates the mean, and bars extend to maximum and minimum values. Data shown are from one experiment representative of three independent studies *, p Ͻ 0.001 by two-tailed t test comparing WT (n ϭ 198) with KD (n ϭ 188).

ZO-1 controls apical epithelial structure in vitro and in vivo
Nevertheless, even during this rapid barrier development, ZO-1 KD cells displayed cytoplasmic F-actin rings when viewed in x,y projections and the corresponding apical distensions seen in x,z reconstructions (Fig. S1, B and C). Together, these data indicate that ZO-1, but not ZO-2, regulates apical membrane morphology and cortical F-actin structure in vivo and in vitro.

ZO-1 structural domains coordinate apical surface development
ZO-1 is a scaffolding protein with multiple structural domains (Fig. 3A) that coordinate binding partners to regulate tight junction and adherens junction protein recruitment and assembly, perijunctional F-actin organization, and epithelial morphogenesis (18, 30 -33). We tested the hypothesis that spe-

ZO-1 controls apical epithelial structure in vitro and in vivo
cific ZO-1 structural domains mediate development of a uniform apical surface by expressing EGFP-tagged ZO-1 mutants in the ZO-1 KD cell line ( Fig. 3A and Fig. S2). Expression of free EGFP had no effect on apical F-actin accumulation or membrane distensions, but EGFP-tagged full-length ZO-1 localized to normal-appearing tight junctions and restored both F-actin organization and a uniform apical surface (Fig. 3, B and C). This was confirmed by SEM, which demonstrated restoration of normal microvillus brush border structure (Fig. 3D). Uniform apical surface restoration was characterized by shorter cells with less variation in cell height (Fig. 4, A and B; p Ͻ 0.0001). In addition to demonstrating that EGFP-tagged ZO-1 is func-tional, these data show that the observed phenotype is specific to ZO-1 loss.
Given striking differences in F-actin organization upon ZO-1 loss, we next tested whether the actin-binding region (ABR) is required for ZO-1-dependent F-actin organization. Expression of EGFP-tagged ZO-1 ⌬ABR in ZO-1 KD cells restored morphology in a manner similar to full-length ZO-1 (Fig. 3, B and C). Specifically, ZO-1 ⌬ABR localized to normal-appearing tight junctions (Fig. 3B), restored both uniform apical surfaces and microvillus architecture (Fig. 3, C and D), and decreased cell height (Fig. 4, A and B; p Ͻ 0.001). Whereas cell height of ZO-1 ⌬ABR -expressing cells was statistically greater than that of

ZO-1 controls apical epithelial structure in vitro and in vivo
cells expressing full-length ZO-1 (Fig. 4B), the absolute difference was small (9.2 Ϯ 1.2 m versus 10.1 Ϯ 1.4 m), and restoration of a uniform apical surface was accompanied by a reduction in cell height variance. This indicates that the ZO-1 ABR is dispensable for both F-actin and apical surface organization.
The U5-GuK region within the N terminus of ZO-1 has been reported to regulate perijunctional actomyosin structure (30,33). In ZO-1 KD cells, EGFP-tagged ZO-1 ⌬U5-GuK localized to tight junctions but was more prominent within cytoplasmic punctae (Fig. 3, B and C). Fluorescence microscopy showed that ZO-1 ⌬U5-GuK failed to eliminate cytoplasmic F-actin rings, as seen in x,y projections, and actually enhanced the density of these rings (Fig. 3B). Consistent with this, SEM showed that apical distensions increased in prominence in monolayers expressing ZO-1 ⌬U5-GuK relative to those expressing free EGFP alone (Fig. 3D). Further, ZO-1 ⌬U5-GuK expression increased both average cell height and cell height variation (Fig. 4, A and B; p Ͻ 0.001). In addition to cell height, the EGFP fluorescence intensity of ZO-1 ⌬U5-GuK -expressing cells varied significantly throughout the monolayers (Fig. 4A). Direct comparisons show a direct correlation between increased EGFP fluorescence intensity and increased cell height (Fig. 4, A and C). These data suggest that, not only did ZO-1 ⌬U5-GuK fail to restore cortical F-actin and apical tissue architecture, but it exacerbated the defect induced by ZO-1 KD.
We considered the hypothesis that ZO-1 ⌬U5-GuK could be acting as a dominant-negative effector of F-actin organization, given that it was likely to bind F-actin but unable to interact with U5-GuK binding partners. This could explain the correlation between increased expression and increased cell height in ZO-1 ⌬U5-GuK -transfected monolayers. We therefore assessed the impact of expressing ZO-1 lacking both the U5-GuK and ABR domains (EGFP-tagged ZO-1 ⌬U5-GuK⌬ABR ). This construct behaved identically to the ZO-1 ⌬U5-GuK construct by increasing cell height in an expression-dependent manner (Figs. 3 and 4). These observations confirm the conclusion above that the ABR is unnecessary for uniform apical surface formation in a ZO-1-deficient background and, more importantly, indicate that the U5-GuK region of ZO-1 is central to organization of apical F-actin and apical cell structure. One possible explanation for this could be defective recruitment of known U5-GuK binding partners to the apical junctional complex. However, stable KD of each of the three proteins known to bind the U5-GuK domain, occludin, shroom2, and ␣-catenin (18, 34 -37), had no effect on F-actin organization or apical surface structure (Fig. 4, D and E).
Based on the striking phenotype induced by deletion of U5 and GuK domains and reported functional differences between these regions (33), we expressed ZO-1 ⌬U5 and ZO-1 ⌬GuK in ZO-1 KD cells. In cells with lower levels of ZO-1 ⌬U5 expression, the protein was recruited to tight junctions and rescued morphology of ZO-1 KD epithelia (Figs. 3 and 4). In cells with higher levels of expression, ZO-1 ⌬U5 was distributed diffusely throughout the cytoplasm and within cytoplasmic punctae (Fig.  3, B and C). These cells stood out due to their dense F-actin rings and large apical distensions. Although we could not correlate morphology with individual cell transgene expression in SEM images, a similar fraction of cells were noted to have dis-tended apical membranes, loss of microvillus architecture, and deep crevasses at intercellular junctions (Fig. 3D). In contrast, ZO-1 ⌬GuK localized to the tight junction and restored cortical F-actin, microvillus architecture, cell height, and apical surface in a manner similar to full-length ZO-1 (Figs. 3 and 4).
Similar to ZO-1 ⌬U5 , ZO-1 ⌬PDZ1 expression resulted in two distinct phenotypes. In cells with low levels of expression, ZO-1 ⌬PDZ1 largely rescued F-actin structure, microvillus architecture, and apical surface structure (Figs. 3 and 4). However, cells with higher levels of ZO-1 ⌬PDZ1 expression displayed dense accumulations of the transgene within apical distensions and apical F-actin rings (Figs. 3 and 4). Similar heterogeneity was apparent in SEM images, where many areas of the monolayer resumed a nearly normal morphology, but clusters of cells with apical distensions and abnormal microvilli persisted (Fig. 3D).

ZO-1 ⌬PDZ1 rescues epithelial architecture in a ZO-2-dependent manner
The data above highlight the unique functions of each of the tested ZO-1 structural domains. The results were, however, difficult to understand in some contexts. For example, why did deletion of U5-GuK domains exacerbate the phenotype of ZO-1 KD monolayers? One potential explanation for these data is that the U5-GuK deletion mutant acted as a dominant-negative protein that inhibited ZO-2 function. A dominant-negative function might also explain why constructs lacking U5 or PDZ1 domains displayed dose-dependent effects, with lower levels of expression restoring function and higher levels having detrimental effects. One could postulate that low-level expression allowed ZO-1/ZO-2 heterodimers to rescue function, whereas higher levels of ZO-1 favored formation of defective ZO-1 dimers that outcompeted ZO-1/ZO-2 heterodimers.
To investigate a potential compensatory role for ZO-2 structural domains in ZO-1 KD cells, EGFP-tagged ZO-1 mutants were expressed in ZO-1/ZO-2 double-KD cells. Monolayers expressing free EGFP displayed the expected cytoplasmic F-actin rings and apical distensions (Fig. 5, A and B). In contrast to ZO-1 single-KD monolayers, ZO-1 ⌬PDZ1 was unable to rescue F-actin organization or apical cell structure in ZO-1/ZO-2 double-KD monolayers, regardless of expression level. These data suggest that ZO-1 ⌬PDZ1 expression rescued ZO-1 single-KD monolayers by dimerizing with ZO-2 to create a fully functional heterodimer (Fig. 5C). The data further indicate that PDZ1mediated interactions are essential to ZO-1-dependent organization of cortical F-actin and assembly of uniform apical surfaces.
As in ZO-1 single-KD cells, expression of either ZO-1 ⌬U5-GuK or ZO-1 ⌬U5-GuK⌬ABR failed to rescue morphology (Fig. 5, A and  B). However, unlike ZO-1 KD cells, expression of these mutants did not appear to exacerbate the phenotype of ZO-1/ZO-2 double-KD cells. As in ZO-1 single-KD cells, low levels of ZO-1 ⌬U5 expression partially restored F-actin structure and reduced the magnitude of apical distensions in ZO-1/ZO-2 double-KD cells. However, high levels of ZO-1 ⌬U5 expression also partially rescued and did not make the phenotype more severe (Fig. 5, A  and B). These data support the hypothesis that the detrimental effect of high-level ZO-1 ⌬U5 expression in ZO-1 single-KD

ZO-1 controls apical epithelial structure in vitro and in vivo
monolayers was due to inhibition of ZO-2 function. These data also suggest that the U5 domain contributes to ZO-1-dependent organization of F-actin and apical membrane structures to only a minor degree.
Finally, expression of ZO-1, ZO-1 ⌬ABR , or ZO-1 ⌬GuK in double-KD cells restored cortical F-actin and apical membrane structure. ZO-2, therefore, did not provide the ABR domain that allowed ZO-1 ⌬ABR to rescue morphology. Thus, despite the profound F-actin disruption, ZO protein ABR domains are not required to restore normal apical structure.

ZO-1 directs uniform apical surface assembly by regulating actomyosin architecture
The dense cytoplasmic actomyosin that characterizes ZO-1 KO and KD epithelia suggests that hypercontractility of this cytoskeletal network may be responsible for the irregular apical surfaces that characterize these epithelia. To test this hypothesis, we screened the effect of pharmacologic agents that target different aspects of cytoskeletal function. In each case, a range of concentrations was tested. To minimize the duration of drug exposure, monolayers were grown to confluence and then cultured in low-calcium medium for 16 h to disrupt cell-cell junctions. Pharmacological agents were added at the time of calcium repletion, and monolayers were harvested after 24 h, at which time uniform apical surfaces are well-established in WT monolayers (Fig. S1).
To determine whether F-actin and apical membrane abnormalities of ZO-1 KD monolayers reflected aberrant microfilament function, monolayers were treated with latrunculin A (latA), which binds actin monomers and prevents their polymerization (38). In WT monolayers, latA disrupted F-actin in a dose-dependent manner. The lowest dose (0.05 M) had no effect, but the intermediate dose (0.5 M) profoundly disrupted cortical F-actin in WT monolayers and also interfered with delivery of gp135 to apical membranes (Fig. 6, A and B). In contrast, 0.5 M latA selectively depleted cytoplasmic F-actin rings within ZO-1 KD monolayers without disrupting other F-actin structures (Fig. 6, A and B). gp135 delivery to uniform apical membranes was also restored by 0.5 M latA. Morphometric analysis confirmed that 0.5 M latA reduced both cell height and the variance of cell height in ZO-1 KD monolayers (Fig. 6, C and D; p Ͻ 0.0001). At the highest dose (1.5 M), latA disrupted perijunctional and cortical F-actin in both WT and ZO-1 KD monolayers. latA reduced, but did not eliminate, the long, thin microvillus extensions induced by ZO-1 KD. latA also induced microvillus clustering in both the presence and absence of ZO-1. Thus, F-actin disruption partially corrected the microvillus phenotype of ZO-1 KD epithelium, suggesting that excessive actin polymerization contributes to the apical abnormalities observed.
Inhibition of F-actin polymerization has multiple effects. These include disruption of myosin II motor activity, which has been linked to epithelial cell shape, morphogenesis, tight junction barrier regulation, and ZO-1 exchange between tight junction and cytoplasmic pools (5, 17, 39 -41). We therefore assessed the impact of myosin motor ATPase inhibitors on apical surface structures (Figs. 7 and 8) (42-47). Blebbistatin did not disrupt apical surfaces in WT cells but reduced the magni-tude of apical distensions and the intensity of cytoplasmic F-actin rings in ZO-1 KD monolayers in a dose-dependent manner (Fig. 7, A-C). Morphometric analysis confirmed that, at the highest dose (100 M), blebbistatin significantly reduced the height of ZO-1 KD epithelia (Fig. 7, C and D; p Ͻ 0.0001). However, blebbistatin also reduced the height of WT epithelia (Fig. 7, C and D; p Ͻ 0.0001) and, paradoxically, induced straightening of perijunctional F-actin profiles at all three doses tested in WT monolayers (Fig. 7, A-C). Blebbistatin fully eliminated the bizarre microvillus extensions that characterized ZO-1 KD epithelia. Thus, whereas blebbistatin-mediated myosin motor inhibition did improve the ZO-1 KD phenotype, rescue was only partial.
A second myosin motor inhibitor, 2,3-butanedione monoxime (BDM), is used infrequently due to numerous reports of off-target effects, primarily in muscle cells (47,48). Nevertheless, it is structurally unrelated to blebbistatin and unlikely to have shared off-target effects despite also inhibiting myosin II activity. In WT monolayers, BDM had no observable effect on F-actin organization or assembly of uniform apical surfaces but did cause a dose-dependent decrease in apical gp135 delivery (Fig. 8, A and B). When ZO-1 KD monolayers were treated with BDM, cytoplasmic F-actin rings were eliminated, and normal apical surface architecture was reestablished in a dose-dependent manner (Fig. 8). As in WT cells, BDM reduced the intensity of gp135 staining at the apical membranes of ZO-1 KD monolayers (Fig. 8, A and B). BDM also reduced the height of both WT and ZO-1 KD monolayers. As a result, ZO-1 KD monolayers treated with the highest dose of BDM (10 mM) had indistinguishable morphology and statistically identical heights to WT monolayers treated with low-dose BDM (1 mM) (Fig. 8, C  and D). Rescue by BDM included normalization of microvillus architecture.
We considered potential off-target effects of BMD. The most likely off-target effect in epithelial cells is inhibition of Ca 2ϩ -dependent myosin light-chain kinase activation. However, a highly specific myosin light chain kinase inhibitor, PIK (49,50), had no effect on apical structure of WT or ZO-1 KD monolayers (data not shown). Notably, relative to blebbistatin, BDM resulted in greater restoration of normal apical morphology, including microvillus structure. This could be interpreted as reflecting off-target effects of BDM or incomplete myosin motor inhibition by blebbistatin. Nevertheless, the overlapping effects of blebbistatin and BDM indicate that myosin motor activity is a significant contributor to the dense cytoplasmic F-actin rings, apical distensions, and abnormal microvilli seen in ZO-1 KD cells.
In contrast to myosin motor inhibitors, pharmacological agents targeting Cdc42 (ML141) and Rho kinase (Y27632) were unable to rescue apical structure in ZO-1 KD monolayers (51,52). Cdc42 inhibition had no effect on WT monolayers, but treatment with an intermediate dose (10 M) increased both the proportion of ZO-1 KD cells with apical distensions and cell height ( Fig. S3; p Ͻ 0.001). Rho kinase inhibition had no effect on cell height but induced aberrant polarization (Fig. S4). Abnormalities included stacked nuclei as well as F-actin-and gp135-positive membrane structures at cell-cell contacts and cell rounding at the highest dose. These changes were present in WT and ZO-1 KD monolayers and were therefore independent of ZO-1 expression.

ZO-1 controls apical epithelial structure in vitro and in vivo Dynamin inhibition corrects apical surface expansion
Myosin motors and actin have been linked to endocytosis and exocytosis (53)(54)(55)(56)(57)(58)(59)(60). We therefore considered the possibility that either excessive delivery of membrane to the apical surface or defective endocytic or exocytic removal contributed to the apical expansion observed. To test this hypothesis, we asked whether inhibition of dynamin II, an effector GTPase that mediates membrane fission, modified the phenotype of ZO-1 KD epithelia. Dynamin inhibition in ZO-1 KD monolayers resulted in flattening of the apical surface into a normal architecture, eliminated F-actin rings, and partially normalized microvillus architecture (Fig. 9). These data suggest that disruption of normal membrane traffic contributes to the expanded apical surface of ZO-1-deficient epithelial cells. These data, however, must be interpreted cautiously, as the inhibitor used, dynasore (61), has been criticized as nonspecific because it also interferes with membrane ruffling and destabilizes F-actin, even in cells lacking all three dynamins (62). Nonetheless, expression of defective dynamin II mutants in epithelial cells has demonstrated a role for dynamin in regulating actomyosin cytoskeletal structure and epithelial polarization (54). This may explain why newer dynamin inhibitors thought to have greater specificity also inhibit membrane ruffling and destabilize F-actin (62).

ZO-1 controls apical epithelial structure in vitro and in vivo Inhibition of F-actin polymerization, myosin motor activity, and dynamin II restores apical epithelial structure in vivo
The ability of F-actin disruption, myosin motor perturbation, and dynamin inhibition to restore cortical F-actin organization and reduce apical membrane expansion in vitro was striking. To determine whether these agents were also able to rescue morphology in vivo, a segment of intestine was functionally isolated while maintaining the neurovascular supply and cannulated. The lumen was perfused with saline containing latA, blebbistatin, BDM, dynasore, or vehicle (Fig. 10A). Examination of tissue sections showed that only latA substantially affected overall intestinal morphology; it induced extensive epithelial shedding in both WT or ZO-1 KO mice. As a result, whole mounts of latA-treated intestines were not amenable to 3D reconstruction.
Without exception, none of the drugs markedly affected apical membrane structure and F-actin organization in WT mice. However, all four drugs corrected apical expansion in vivo and resulted in unified apical surfaces without disruption at depressed cell-cell junctions (Fig. 10B). Confocal microscopy of jejunal whole mounts confirmed that BDM, blebbistatin, and dynasore all restored continuous flat apical surfaces but had no effect on WT epithelium (Fig. 10C). These in vivo data support the in vitro data and, when taken together, indicate that epithelial architectural abnormalities induced by ZO-1 deletion can be normalized by disrupting F-actin polymerization, myosin motor ATPase activity, or dynamin-mediated vesicle trafficking.

Discussion
Epithelial surfaces are critical sites that separate organisms from the external world or define interfaces between distinct internal compartments. These functions require that many epithelial cells work as a single unit. Organization of the basal aspect of these cells is maintained by interactions with the base-

ZO-1 controls apical epithelial structure in vitro and in vivo
ment membrane. However, the factors that coordinate apical surfaces are less well-defined. The widespread presence of uniform apical surfaces, loss of which is well-recognized as a marker of neoplasia, argues that conservation of this single integrated multicellular surface structure is important to epithelial function. Here, we examined the contributions of the apical junctional complex protein ZO-1 to epithelial surface contiguity and apical membrane structure. Our data show that ZO-1 deficiency results in accumulation of dense cytoplasmic F-actin rings that define broad-based apical extensions. We further show that specific ZO-1 domains are required for normal structure, that absence of some domains can be complemented by ZO-2, and that inhibitors of myosin motor ATPase activity and membrane traffic can largely restore apical structure in vitro and in vivo.
Surprisingly, our data indicate that the ZO-1 ABR is not required for ZO-1-dependent organization of apical actomyosin. We considered the possibility that ZO-1/ZO-2 heterodimers could allow the ZO-2 ABR to complement deletion of that domain within ZO-1. However, ZO-1 ⌬ABR was able to restore normal apical structure in ZO-1/ZO-2 double-KD monolayers. Similarly, the GuK domain was not required for rescue of actomyosin organization in ZO-1 KD or ZO-1/ZO-2 double-KD monolayers. Whereas ZO-1 ⌬PDZ1 expression corrected apical defects in ZO-1 KD monolayers when expressed at low levels, it was ineffective when expressed at high levels. Further, ZO-1 ⌬PDZ1 was unable to rescue ZO-1/ZO-2 double-KD monolayers. This suggests that ZO-2 PDZ1 domains complemented ZO-1 ⌬PDZ1 , likely due to heterodimerization between ZO-1 and ZO-2. The requisite role of the PDZ1 domain suggests that ZO-1 interactions with claudins, which bind PDZ1, may be critical to regulation of normal apical structure. Disruption of ZO-1 interactions with other binding partners, including connexins and vinculin (63)(64)(65)(66), which bind to PDZ2 and PDZ3 domains, respectively, could also be considered. Because we did not specifically test ZO-1 mutants lacking those domains, our data do not address this issue, which could be a topic of future studies.

ZO-1 controls apical epithelial structure in vitro and in vivo
Similar data indicate that ZO-2 expression complements ZO-1 ⌬U5 , as low-level expression rescued apical abnormalities in ZO-2-expressing monolayers but not in double-KD monolayers. Strikingly, high-level ZO-1 ⌬U5 expression exacerbated apical abnormalities in ZO-1 KD, but not double-KD, monolayers. This can be best explained as a dominant-negative effect of ZO-1 ⌬U5 on ZO-2 and suggests that, whereas the U5 domain is important to apical structure, the ZO-2 U5 domain can partially compensate for loss of that domain within ZO-1.
ZO-1 deletion induces polymerization of dense F-actin structures beneath the apical membrane. These are distinct from the dense perijunctional F-actin accumulations reported previously in ZO-1/ZO-2 double-KD monolayers (27), which were not seen in ZO-1 KO intestinal epithelia or ZO-1 single-KD monolayers. Notably, latA doses that eliminated cytoplasmic F-actin rings while having milder effects on perijunctional F-actin were able to significantly correct apical abnormalities within ZO-1 KD monolayers. Further, myosin motor ATPase inhibitors (blebbistatin and BDM) also restored apical cell structure and eliminated cytoplasmic F-actin rings without disrupting perijunctional F-actin in ZO-1 KD monolayers. Perfusion of blebbistatin or BDM through the intestinal lumen verified that these drugs had the same restorative effect on intestinal epithelia in vivo.
Given the lack of requirement for the ZO-1 ABR in generating normal apical structure, it is not entirely clear how ZO-1 deficiency so negatively impacts apical F-actin structures. However, a recent report examining the conformation of ZO-1 may provide some insight (67). That study showed that ZO-1 normally assumes an elongated molecular structure and is precisely oriented at apical junctional complexes. Simultaneous KD of ZO-2 and myosin motor ATPase inhibition with blebbistatin allowed ZO-1 to relax into a folded structure in which N and C termini were in proximity to one another (67). Neither blebbistatin nor ZO-2 KD alone were able to cause this change in ZO-1 confirmation. Thus, ZO-2 and cytoskeletal tension were both required for ZO-1 elongation. One could hypothesize that, in turn, ZO-1 regulates the magnitude of cytoskeletal

ZO-1 controls apical epithelial structure in vitro and in vivo
tension. This would predict that ZO-1 deletion could lead to dysregulated and excessive cytoskeletal tension that ultimately caused the apical abnormalities reported here.
Although we did not identify a specific protein, one potential effector within this signaling platform is the small GTPase Cdc42, which regulates tight junction assembly and vesicle trafficking in polarized epithelia (68,69). Consistent with this, a guanine nucleotide exchange factor, Dbl3, positions Cdc42 along the lateral membrane to regulate apical surface size and shape (70). In a more recent report, the same authors speculated that, in addition to promoting apical myosin activation, Cdc42 may reduce perijunctional actomyosin contractility, which results in a contractility gradient that promotes apical expansion (71). Support for that interpretation included the observation that Cdc42 inhibition prevented apical expansion. In contrast, we found that Cdc42 inhibition exacerbated apical surface expansion of ZO-1 KD monolayers (Fig. S3). It is therefore unlikely that the phenotype induced by ZO-1 KO or KD is due to Cdc42 hyperactivation, as occurs after Dbl3 overexpression (71).
Perturbation of vesicular trafficking could provide an alternative mechanism for the apical expansion that occurs in ZO-1-deficient epithelia. Although ZO-1 has not been linked directly to regulation of membrane traffic, it is notable that key components of the exocyst are known to localize to tight junctions (72). Moreover, epithelia with endosomal maturation defects developed expanded apical surfaces similar to those in ZO-1-deficient cells (73). Consistent with this hypothesis, dynamin inhibition completely restored apical surface biogenesis in ZO-1-deficient epithelia, both in vitro and in vivo.
Membrane traffic defects could also contribute to the strikingly abnormal microvillus structure seen in ZO-1-deficient epithelia. One potential mechanism could be defects in vesicle release from microvillus tips (24). This would be consistent with the bizarre elongated microvilli we observed. Although microvillus structure was not the focus of our investigation, the diameter of these extensions, which was approximately half of that of the microvilli from which they arose, would be consistent with a lack of F-actin cores. This contrasts sharply with the relatively normal diameter of the modestly elongated microvilli described in Myo1a KO mice, which display excessive vesicle release (25), as well as the abnormal microvilli induced by in vitro protocadherin 24 KD (74). Notably, microvillus architecture was improved by inhibition of myosin motor activity or dynamin, but not by F-actin disruption. Regardless of the mechanism, it is difficult to rationalize the apical expansion and abnormal microvillus structure of ZO-1-deficient epithelial because ZO-1 is not present in apical membranes or microvilli. One potential explanation could relate to the as yet undefined function of the large, soluble, extrajunctional ZO-1 pool, which has been estimated to represent 60% of total ZO-1 (25,74,75). Overall, our data demonstrate an integral role of ZO-1 in regulating structure of apical membranes, subapical actomyosin, microvilli, and, above all, the unified single surface that epithelial sheets present to the external milieu.

Mice
To generate mice with intestinal epithelial-specific ZO-1 deletion, Tjp1 tm2a (KOMP)Wtsi (MGI:98759) C57BL/6N mice were bred with B6.Cg-Tg(ACTFLPe)9205Dym/J (MGI: 2448985) to delete the En2SA-LacZ-neo cassette between the Frt sites, thereby creating the B6-Tjp1 tm2a line. For experimental use, B6.Cg-Tjp1 tm2a mice were bred with Tg(Vil-cre)20Syr (MGI:3053819) mice to induce intestinal epithelial-specific gene deletion. ZO-1 fl/fl ; vil-cre ϩ and ZO-1 fl/fl ; vil-cre Ϫ mice were maintained on a C57BL/6J background. Knockout was confirmed by Western blotting and immunohistochemistry, as shown in the figures, as well as routine genotyping. Mice of both sexes were used between 8 and 12 weeks of age. Littermates were used for individual experiments. All studies were performed according to protocols approved by institutional animal care and use committees of the University of Chicago, Brigham and Women's Hospital, Boston Children's Hospital, and Harvard Medical School. Animals were maintained in specific pathogen-free environments.

Intestinal perfusion assay
In vivo perfusion was performed as described previously (76). Briefly, adult, 8 -12-week-old mice were fasted for 12-16 h before each experiment, and surgical plane anesthesia was induced with ketamine and xylazine. The abdomen was opened by a midline incision, and a 4 -5-cm loop of jejunum was cannulated at the proximal and distal ends with 0.76-mm internal diameter polyethylene tubing. The lumen was flushed by perfusion with 140 mM NaCl, 10 mM HEPES, pH 7.4, warmed to

ZO-1 controls apical epithelial structure in vitro and in vivo
37°C at 1 ml/min for 10 min using a peristaltic pump. This was replaced by fresh perfusion solution with or without 100 M blebbistatin, 10 mM BDM, 240 M dynasore, or 0.5 M latA, and perfusion continued in a recirculating manner at 1 ml/min for 2 h. The abdominal cavity was covered with moistened gauze, and body temperature was maintained at 37°C using a heat lamp. After perfusion, the animal was sacrificed, and the perfused jejunal segment was excised, fixed in 10% formalin, and embedded in paraffin or used for whole-mount preparation, as described below.
Occludin KD MDCK (clones F36 and F52) were cultured in T23 medium containing 50 g/ml hygromycin B to prevent loss of occludin shRNA (79). Shroom2 KD and ␣-catenin KD MDCK cells were cultured in T23 medium containing 10 g/ml blasticidin (18). All cell lines used in this study tested negative for mycoplasma contamination either at time of cell line acquisition or generation and again at 3-month intervals while being used.

Calcium switch assay
For growth on Transwell filters, cells were plated at a density of 150,000 cells on 0.33-cm 2 surface area, 0.4-m pore Transwell filters (Corning) in 24-well plates. Growth medium was exchanged 48 h after plating. Calcium was depleted after 72 h of culture by washing, followed by incubation in calcium-free suspension-modified Eagle's medium (SMEM, Life Technologies) containing 10% dialyzed fetal bovine serum (Life Technologies) and 15 mM HEPES. After a 16-h incubation, calcium-free medium was replaced with normal LDMEM containing calcium. For pharmacologic inhibitor treatments, monolayers were exposed to the indicated doses of inhibitors upon calcium repletion and cultured with inhibitors for 24 h in the presence of calcium before fixation and immunostaining.

Immunofluorescence
Monolayers were fixed at the indicated time points in either ethanol (to visualize myosin IIB) or paraformaldehyde (to visualize all other assessed proteins). For ethanol fixation, monolayers were rinsed with and then submerged in Ϫ20°C ethanol for 30 min. Subsequent steps were described previously (59). Briefly, cells were incubated in bis(sulfosuccinimidyl)suberate in PBS ϩ (PBS containing 0.1% n-octyl-glutaraldehyde) for 30 min. Monolayers were washed in PBS ϩ , quenched with 100 mM ethylenediamine, and washed again in PBS ϩ . Inserts were then blocked for 1 h at room temperature in blocking buffer containing 1% nonfat milk, 1% fish gelatin, and 1% BSA in PBS ϩ . For PFA fixation, cells were rinsed in PBS and incubated in 4% PFA for 20 min at room temperature. Fixation was quenched by a 15-min incubation with 50 mM NH 4 Cl, and cells were permeabilized with 0.5% Triton X-100 for 10 min. Monolayers were blocked with 10% normal goat serum in PBS for 1 h at room temperature. Primary antibodies were diluted in blocking buffer and incubated overnight at 4°C. Primary antibodies were washed, and monolayers were incubated with a 1:100 dilution of the appropriate Alexa Fluor 488 -, 594 -, or 647-conjugated donkey anti-rabbit, -mouse, or rat F(abЈ) 2 AffiniPure antiserum secondary antibodies (Jackson Immunoresearch) for 1 h at room temperature. F-actin and nuclei were labeled with Alexa Fluor-conjugated phalloidin and Hoechst 33342, respectively. Monolayers were mounted under coverslips in Slowfade Diamond (Life Technologies).
Small intestinal whole mounts were prepared for fluorescent imaging as follows. Animals were sacrificed in accordance with institutional animal care and use committee protocols. A segment of jejunum was excised, cut lengthwise to expose the luminal surface, and then incubated in fixative (0.5% paraformaldehyde, 15% picric acid, and 0.1 M sodium phosphate buffer (pH 7.0)) with gentle shaking for 12 h at 4°C. Specimens were rinsed in ice-cold PBS (three washes, 5 min each), followed by a 3-h incubation in 10% sucrose, 0.1% NaN 3 /PBS (4°C) and an overnight incubation in 20% sucrose, 10% glycerol, 0.1% NaN 3 / PBS (4°C). A 1-cm portion of the fixed intestinal segment was washed in PBS and blocked in 0.5% BSA, 0.1% NaN 3 , 0.3% Tri-

ZO-1 controls apical epithelial structure in vitro and in vivo
ton X-100, 5% normal donkey serum/PBS for 6 h. Tissues were stained with phalloidin (Invitrogen) and Hoechst 33342 overnight at 4°C, followed by five 0.3% Triton X-100/PBS washes (1 h per cycle, 4°C). Sections were washed with cold PBS twice for 10 min each and immersed in 2% PFA/PBS for 2 days at 4°C. The stained whole-mount preparation was placed on a glass slide (villus side up) with imaging spacer (Grace Bio-Labs), and Slowfade Diamond antifade reagent (Life Technologies) was added, followed by a coverslip.
Confocal images of cultured monolayers and intestinal whole mounts were acquired on a Leica DMI6000 microscope equipped with a CSUX Yokogawa spinning disk (Andor Technologies). Laser illumination was supplemented with Borealis illumination technology to ensure uniform illumination across the sample, images were acquired using either a Retiga EXi Fast 1394 (QImaging) or Zyla 4.2 Plus sCMOS (Andor Technologies) camera, and the system was controlled by Metamorph 7. A ϫ20 HC PLAN Apo/0.70 objective (Leica 11506166) was used to capture z-stacks with 0.5-m z-steps, and both ϫ63 HCX PL Apo/1.3 glycerol immersion (Leica 506193) and ϫ100 HC PLAN Apo/1.4 oil immersion (Leica 11506372) objectives were used to capture z-stacks with 0.2-m z-steps. Identical laser intensity, exposure time, and camera gain settings were used to acquire all images of a given fluorescent wavelength or label within individual experiments. After acquisition, select images were deconvoluted where indicated using Autoquant X3. All images were then scaled and cropped, maximal and x,z projections were created in Meta-Morph (Molecular Devices), and rendering and quantification were performed in Imaris version 8.4.1 (Bitplane) as described below. All images are representative of at least three independent experiments.

Confocal z-stack reconstruction and quantification
To generate 3D reconstructions of both intestinal whole mounts and cultured monolayers, confocal z-stacks were viewed and processed with the surface tool, spots tool, or cell module in Imaris.
The surface tool was used to render the apical surface of both intestinal whole mounts and cultured monolayers. First, a region of interest that spanned from immediately basal to the tight junction to just above the apical-most point of the surface was drawn. Next, a surface was built within the region of interest using F-actin as a marker with the following creation parameters: smoothing, grain size 0.125 and 0.5 m (whole mount and cultured monolayer, respectively), without background subtraction. Thresholding was performed manually for each surface rendered.
For cell creation, F-actin (phalloidin) was used as a membrane marker with the following creation parameters: smallest cell diameter 1.28 and 5 m (whole mount and monolayer, respectively), membrane detail 0.128 and 0.227 m (whole mount and monolayer, respectively), and local contrast filter. Intensity and quality measurements were manually thresholded to generate representative renderings. After initial cell generation, cells at the edges of the region of interest (in the x and y directions) and cell renderings that were not within the plane of the epithelium (in the z direction) were eliminated by filtering. The height of each resulting cell rendering was calculated with a bounding box in the z direction and is reported along with cell population statistics. Cells are color-coded by cell height using the same scale for all monolayers.

Scanning EM
Confluent monolayers were rinsed with warm LDMEM without additives and fixed for 1 h at room temperature with 3% glutaraldehyde in 0.1 M sodium cacodylate buffer with 50 mM CaCl 2 (pH 7.4). Fixed cells were rinsed three times for 5 min with 0.1 M sodium cacodylate buffer before a 30-min incubation with 1% osmium tetroxide in 0.1 M sodium cacodylate buffer. Cells were rinsed with water and then incubated for 20 min with 2% tannic acid in water. Monolayers were rinsed with water again before a 20-min incubation with 1% osmium tetroxide in water. Samples were then dehydrated with a series of 10-min ethanol washes (30,50,75, 100, and 100%) and dried overnight in a chemical hood with hexamethyldisilazane. Whole-mount intestinal tissue samples were prepared by the same method, but before fixation, intestines were flushed with 37°C Hanks' balanced salt solution, cut lengthwise to expose the mucosal surface, and fixed overnight at 4°C. A Ted Pella sputter coater was used to sputter monolayer and tissue samples with 8 and 10 nm of platinum-palladium alloy, respectively. All images were acquired in the University of Chicago Materials Research Center using an FEI Nova NanoSEM 230 at 10.0 kV using a TLD detector and a 45-s dwell time.

Transmission EM
For transmission EM, tissue samples were prepared as described previously (80). Briefly, tissues were fixed with 2.5% glutaraldehyde and 4% PFA in 0.1 M sodium cacodylate buffer. Samples were dehydrated and embedded in Spurr resin prior to sectioning. Images were acquired on an FEI Tecnai F30 transmission electron microscope in the Advanced Electron Microscopy Facility at the University of Chicago.

Western blotting
Cell lysates were taken from confluent monolayers in nonreducing sample buffer (0.375 M Tris base, 15% glycerin, 3% SDS, pH 6.8) with 1:200 Protease Inhibitor Mixture Set III (EMD Millipore). Samples were boiled for 5 min followed by sonication on ice. 15 g of lysate was loaded per lane and separated by SDS-PAGE. Protein was then transferred to PVDF membranes as described previously (80). PVDF membranes were blocked with 5% milk in TBST for 1 h, and the indicated proteins were probed with primary antibodies at the dilutions shown in Table  S1. Species-specific horseradish peroxidase-conjugated sec-

ZO-1 controls apical epithelial structure in vitro and in vivo
ondary antibodies were used, and proteins were detected by chemiluminescence.