The atypical mitogen-activated protein kinase ERK3 is essential for establishment of epithelial architecture

Epithelia contribute to physical barriers that protect internal tissues from the external environment and also support organ structure. Accordingly, establishment and maintenance of epithelial architecture are essential for both embryonic development and adult physiology. Here, using gene knockout and knockdown techniques along with gene profiling, we show that extracellular signal–regulated kinase 3 (ERK3), a poorly characterized atypical mitogen-activated protein kinase (MAPK), regulates the epithelial architecture in vertebrates. We found that in Xenopus embryonic epidermal epithelia, ERK3 knockdown impairs adherens and tight-junction protein distribution, as well as tight-junction barrier function, resulting in epidermal breakdown. Moreover, in human epithelial breast cancer cells, inhibition of ERK3 expression induced thickened epithelia with aberrant adherens and tight junctions. Results from microarray analyses suggested that transcription factor AP-2α (TFAP2A), a transcriptional regulator important for epithelial gene expression, is involved in ERK3-dependent changes in gene expression. Of note, TFAP2A knockdown phenocopied ERK3 knockdown in both Xenopus embryos and human cells, and ERK3 was required for full activation of TFAP2A-dependent transcription. Our findings reveal that ERK3 regulates epithelial architecture, possibly together with TFAP2A.

Extracellular signal-regulated kinase 3 (ERK3), 4 also known as mitogen-activated protein kinase 6 (MAPK6), is an atypical MAPK family member that has been much less characterized than the classical MAPKs, such as ERK1/2, c-Jun N-terminal kinase (JNK), and p38 (1,2). Genetic ablation of the Erk3 gene in mice has revealed that ERK3 plays important roles in fetal growth and lung maturation during embryogenesis (3). Additionally, ERK3 has recently emerged as a potential target for cancer therapy because (a) it promotes cancer cell migration, invasion, and chemoresistance (4 -6), and (b) its overexpression and genomic mutations are observed in multiple human cancers (4,(7)(8)(9). However, molecular and cellular mechanisms by which ERK3 regulates embryogenesis or oncogenesis have not been fully investigated. Moreover, although ERK3 is evolutionarily conserved in vertebrates from fish to humans, its physiological role has not been examined in nonmammalian vertebrate classes.
Epithelia provide physical barriers that separate the internal environment from the external environment and support the structure of organs. Establishment and maintenance of epithelial architecture are essential for embryonic development and adult physiology, whereas impairment of epithelial architecture is a hallmark of cancer progression and metastasis. Transcription factor AP-2␣ (TFAP2A) plays a role in regulating the expression of epithelial genes, such as E-cadherin (encoding a major adherens junction protein) and keratin 14 (encoding a component of epithelial intermediate filaments) (10 -17). Constitutive or conditional knockouts of Tfap2a in mice yield neural crest-related craniofacial defects, as well as malformation of epithelia-containing organs, such as the kidneys, ventral wall, and skin epidermis (18 -22). Moreover, mutations in the human TFAP2A gene are found in patients with branchio-oculo-facial syndrome (23), a congenital developmental disorder characterized by defects in the craniofacial structures, neck skin, eyes, and ears, as well as less frequently occurring kidney malformation. These findings demonstrate the essential role of TFAP2A in diverse developmental processes. However, the regulatory mechanisms of TFAP2A are largely unknown.
In this study, our analyses in X. laevis embryos and human cancer cells demonstrate that ERK3 is crucial for maintaining epithelial cell junction integrity and epithelial tissue architecture in vertebrates. Our transcriptome analyses suggest that TFAP2A is involved in ERK3-dependent gene expression changes. Moreover, we demonstrate that TFAP2A, like ERK3, is required for epithelial cell junction integrity and

ERK3 is required for pronephros and epidermal development in X. laevis embryos
To investigate ERK3 function, we performed knockdown experiments with antisense morpholino oligonucleotides (MOs) against ERK3, ERK3 MO1 for ERK3A alone, ERK3 MO2 for ERK3B alone, and ERK3 MO3 for both ERK3A and ERK3B ( Fig.  2A). Immunoblotting analyses showed that ERK3 MO1 and MO2 specifically blocked the translation of ERK3A and ERK3B mRNAs, respectively (Fig. 2, B and C). ERK3 MO3 blocked the translation of both ERK3A and ERK3B mRNAs (Fig. 2D). To first examine the role of pronephric ERK3 expression, we injected control MO, ERK3 MO1/2 (ERK3 MO1 plus MO2), or ERK3 MO3 into both ventral vegetal blastomeres (also called V2 blastomeres, from which the pronephros arises) (26) of 8-cell stage embryos. The injection of ERK3 MO1/2 or ERK3 MO3, but not that of control MO, caused edema formation (Fig. 3, A and B). As pronephros defects lead to increased water retention and edema in amphibian embryos (27), we next investigated expression of the pan-pronephros marker gene atp1b1 by whole-mount in situ hybridization. ERK3 knockdown led to a reduction of atp1b1 expression, suggesting that pronephros development was inhibited by ERK3 knockdown (Fig. 3, C and E). The reduction of atp1b1 expression in ERK3 morphants was partially rescued by overexpressing N-terminally Myc-tagged ERK3A and ERK3B (Myc-ERK3A and Myc-ERK3B) mRNAs, which were MO-resistant (Figs. 2E and 3 (C and D)). These results suggest that ventral vegetal ERK3 is required for pronephros development. Next, we investigated the role of ventral animal (i.e. presumptive epidermal) ERK3. The injection of ERK3 MO1/2 or ERK3 MO3 into the animal regions of all ventral blastomeres at the 4-cell stage led to epidermal disintegration at the tailbud stage (Fig. 3, F and G). At stage 23, ERK3 morphants began to show local cell detachment from the surface (Fig. 3F, black arrowheads). At stage 39, almost all ERK3 morphants were partly or completely crushed due to epidermal architecture disruption (Fig. 3G). In control MO-injected embryos, epidermal architecture disruption was not observed (Fig. 3, F and G). These results suggest that ERK3 is required for epidermal development. The expression levels of ERK3 were normalized to those of odc in two independent experiments (#1 and #2). The normalized ERK3 expression level at stage 1 was defined as 1.0 in each experiment. B, whole-mount in situ hybridization analysis of ERK3 expression. ec, ectoderm; nc, neural crest; s, somite; nt, neural tube; ep, epidermis; pn, pronephros; b, brain; e, eye. Scale bars, 400 m. Anterior is to the left (stage (St.) 19 -33/34). Dorsal is up for the lateral views (stage 19 -33/34). Shown are representative images of 6 -10 embryos from one experiment using the ERK3A probe. Essentially the same results were obtained for 6 -9 embryos from another experiment using the ERK3B probe.

ERK3 regulates epithelial architecture ERK3 is required for the formation of adherens and tight junctions in X. laevis embryonic epidermal epithelia
We hypothesized that the epidermal disintegration observed in ERK3 knockdown tailbuds might be caused by defective cell junctions in the early stages. To test this hypothesis, embryos after co-injection of CAAX-GFP mRNA with control MO, ERK3 MO1/2, or ERK3 MO3 were subjected to E-cadherin immunostaining at stage 13, when epidermal disintegration had not yet occurred in ERK3 morphants. The GFP signal of the plasma membrane-targeted CAAX-GFP revealed that epidermal epithelial cells in the control embryos were almost polygonal and had nearly straight cell-cell boundaries, but those in ERK3 morphants were less polygonal and had more curved cell-cell boundaries (Fig. 4A). Moreover, junctional E-cadherin staining in epidermal epithelial cells almost completely disappeared in ERK3 morphants (Fig. 4A), suggesting that the adherens junctions were severely defective. Consistent with this, the expression level of E-cadherin mRNA was down-regulated in ERK3 morphants (Fig. 4B). Next, embryos injected with control MO, ERK3 MO1/2, or ERK3 MO3 were sub-jected to immunostaining for ZO-1, a tight-junction protein, at stage 13. In control embryos, the ZO-1 signals in epidermal epithelial cells were detected as continuous straight lines (Fig. 4C). However, in ERK3 morphants, the signals were detected as discontinuous or curved lines (Fig. 4C). The number of gaps in ZO-1 staining in a single confocal section (Fig. 4C, yellow arrowheads), which could be indicative of the absence, apical shift, or basal shift of ZO-1 protein, was significantly higher in ERK3 morphants than in control embryos (Fig.  4D). Also, although ZO-1 mRNA expression was not significantly affected by ERK3 MO1/2 (Fig. 4B), the fluorescence intensity of junctional ZO-1 staining in a single confocal section was slightly but significantly lower in ERK3 morphants than in control embryos (Fig. 4E). These results suggest that the tight junctions were defective in ERK3 morphant. To assess the barrier function of the tight junctions, we used EZ-link Sulfo-NHS-LC-Biotin, a membrane-impermeable reagent that does not pass through the tight-junction barrier (28). The transverse sections of embryos revealed that externally added EZ-link Sulfo-NHS-LC-Biotin was diffused into the internal tissues in ERK3 morphants, suggesting A, The target sequences of three antisense MOs for X. laevis ERK3 homeologs are shown in rectangles. ERK3 MO1 and MO2 were designed to target ERK3A alone and ERK3B alone, respectively. ERK3 MO3 was designed to target both ERK3A and ERK3B. B-E, the indicated sets of MOs (80 ng), ERK3 mRNAs (1.6 ng), and GFP mRNA (800 pg, control) were injected into the animal regions of all blastomeres at the 4-cell stage. C-terminally Myc-tagged ERK3A (ERK3A-Myc) and ERK3B (ERK3B-Myc) mRNAs comprise the 5Ј-UTR and the coding region to contain the entire MO target sequences. N-terminally Myc-tagged ERK3A (Myc-ERK3A) and ERK3B (Myc-ERK3B) mRNAs contain additional AUG and Myc-encoding sequences upstream of the coding region to avoid MO-mediated translational inhibition. To analyze the ERK3A-Myc and ERK3B-Myc proteins, which were unstable due to proteasome-dependent degradation, animal caps (n ϭ 25, each sample) were dissected at stage 9, cultured with 10 M MG132 (a proteasome inhibitor) until stage 12, and then lysed with 50 l of lysis buffer. To analyze the Myc-ERK3A and Myc-ERK3B proteins, injected whole embryos (n ϭ 10, each sample) were harvested at stage 12 and lysed with 200 l of lysis buffer. The protein levels were examined by immunoblotting. The data are representative of two or three independent experiments. B, ERK3 MO1 blocked the translation of ERK3A-Myc mRNA but not that of ERK3B-Myc mRNA. C, ERK3 MO2 blocked the translation of ERK3B-Myc mRNA but not that of ERK3A-Myc mRNA. D, ERK3 MO3 blocked the translation of both ERK3A-Myc and ERK3B-Myc mRNAs. E, Myc-ERK3A and Myc-ERK3B mRNAs were MO-resistant.

ERK3 regulates epithelial architecture
the loss of the barrier function of the tight junctions (Fig. 4F). Taken together, these results suggest that ERK3 knockdown leads to the disruption of adherens and tight junctions in X. laevis embryonic epidermal epithelia.

ERK3 knockdown and ERK3 knockout both disturb epithelial architecture in human epithelial breast cancer cells
So far, our results suggest that ERK3 regulates epithelial cell junction formation during epidermal development in X. laevis.

ERK3 regulates epithelial architecture
To determine whether the role of ERK3 is conserved in mammals, we used the human breast epithelial cancer cell line MCF7, in which ERK3 is highly expressed (according to the public online gene expression database BioGPS (http://www. biogps.org)). 5 ERK3-specific siRNA1 and siRNA2 effectively decreased human ERK3 expression levels in MCF7 cells (Fig. 5A). Bright-field observation indicated that control siRNA-transfected cells, but not ERK3 siRNA-transfected cells, were able to reach confluence (Fig. 5B). ERK3 siRNAs did not affect cell proliferation (Fig. 5C), suggesting that the low confluence in ERK3 siRNA-transfected cells was not due to decreased cell numbers. 3D imaging revealed that control siRNA-transfected cells formed a flat monolayer epithelial sheet, whereas ERK3 siRNA-transfected cells piled up and formed dome-like shaped colonies (Fig. 5, B, E, G, and H). Moreover, junctional E-cadherin and ZO-1 signals in ERK3 knockdown cells were less sharp or less continuous than those in control cells (Fig. 5, D and F), suggesting that ERK3 is required for maintaining cell junction integrity to form a flat monolayer epithelial sheet.
We then generated two stable ERK3-knockout MCF7 cell lines by CRISPR/Cas9-mediated genome editing using ERK3specific sgRNA (Fig. 6A). The two ERK3-knockout clones, which had frameshift mutations caused by different single-nucleotide insertions (Fig. 6B), did not express detectable levels of ERK3 protein (Fig. 6C). The ERK3-knockout cells exhibited lower confluence and a slightly slower proliferation rate than WT cells (Fig. 6, D and E). ERK3-knockout cells, but not WT control cells, piled up to form thickened sheets or colony-like structures (Fig. 6, D, H, I, and J), which were similar to but thinner than those formed by ERK3 knockdown cells (Fig. 5, B, E, G, and H). The milder defects in ERK3-knockout cells compared with ERK3-knockdown cells may be partly due to genetic compensation induced by gene knockout, but not by gene knockdown (29). Additionally, junctional E-cadherin and ZO-1 signals in ERK3-knockout cells were less sharp or less continuous than those in WT cells (Fig. 6, F and G), confirming again that ERK3 is required for cell junction integrity. Collectively, these results indicate that ERK3 is essential for maintaining epithelial cell junction integrity in human epithelial breast cancer cells as well as in the X. laevis embryonic epidermis.

TFAP2A is a candidate factor for contributing to ERK3-dependent gene expression changes
To study how ERK3 exerts its effects, we focused on changes in gene expression profiles during pronephros and epidermal development in X. laevis embryos. The animal caps (ACs) were dissected from embryos injected with control MO, ERK3 MO1/2, or ERK3 MO3 and were cultured alone or in the presence of activin plus retinoic acid (activin/RA) (Fig. 7A). Whereas epidermal cells can be induced from nontreated ACs, pronephric cells can be induced from activin/RA-treated ACs (30). Microarray analysis indicated that 825 genes in nontreated ACs were down-regulated Ͼ2-fold in common by both ERK3 MO1/2 and ERK3 MO3, as compared with control MO (Fig. 7B, orange circle). Additionally, in activin/RA-treated ACs, 294 genes were down-regulated Ͼ2-fold in common by both ERK3 MO1/2 and ERK3 MO3 (Fig. 7B, green circle). We focused on 123 genes that were commonly down-regulated by ERK3 MOs in both activin/RA-treated ACs and nontreated ACs, because they could act downstream of ERK3 in both pronephros and epidermal development. Then we performed a transcription factorbinding motif enrichment analysis by scanning promoter sequences with the online tool Pscan (http://159.149.160.88/ pscan/). 5 Due to the lack of available X. laevis genes in Pscan, we subjected mouse orthologs of the 123 genes to the analysis and found that the binding sequence of TFAP2A was highly enriched in their promoter sequences (Fig. 7C).
In X. laevis, tfap2a is expressed in the prospective epidermis and pronephros (16,31) and is essential for epidermal development (16). Moreover, defects in the skin epidermis and kidneys can be detected in Tfap2a-knockout mice and TFAP2A-mutated human patients with branchio-oculo-facial syndrome (19,(21)(22)(23)32). These reports prompted us to speculate that TFAP2A might cooperate with ERK3 to regulate both pronephros and epidermal development in X. laevis.

TFAP2A knockdown phenocopies ERK3 knockdown in X. laevis embryos and human epithelial breast cancer cells
X. laevis has two tfap2a homeologs, tfap2a.L and tfap2a.S. We examined the spatial expression patterns of tfap2a homeologs by whole-mount in situ hybridization using a probe complementary to the coding region of tfap2a.L, which is expected to cross-hybridize with tfap2a.S due to 95% identity in nucleotide sequences. As reported previously (16,31), tfap2a expres-

ERK3 regulates epithelial architecture
sion was detected in the animal region (presumptive ectoderm) at stage 9 and in the ventral ectoderm (fated to be the epidermis) at stage 12 ( Fig. 8). At the neurula and tailbud stages (stage 17-39), tfap2a was highly expressed in the neural crest, brain, and pronephros and moderately expressed in the epidermis ( Fig. 8). Thus, tfap2a has a partially overlapping expression pattern with ERK3.
These results indicate remarkable overlap between ERK3-regulated genes and TFAP2A-regulated genes. Moreover, the transcriptomic alterations in tfap2a knockdown ACs were highly correlated with those in ERK3 knockdown ACs (Fig. 10C, correlation coefficient 0.651; p Ͻ 0.01). Notably, the correlation coefficient for cell junction-related genes (listed in Table S1) was higher than that for all genes (Fig. 10D, correlation coefficient 0.788; p Ͻ 0.01). The number of cell junction-related genes down-regulated by tfap2a knockdown and ERK3 knockdown (represented by dots in the bottom left area of the graph) was higher than that up-regulated by tfap2a knockdown and ERK3 knockdown (represented by dots in the top right area of the graph) (Fig. 10D), indicating that many cell junctionrelated genes are positively regulated by ERK3 and TFAP2A. Taken together, these results suggest that ERK3 and TFAP2A have an overlapping role in regulating gene expression, especially cell junction-related genes.
We next observed the embryonic phenotypes induced by tfap2a knockdown. The ventral vegetal injection of TFAP2A MO1/2 led to edema as well as reduced atp1b1 expression (Fig.  11, A and B), similar to that of ERK3 MOs (Fig. 3, A-E). Moreover, the ventral animal injection of TFAP2A MO1/2 led to the disruption of adherens and tight junctions in epidermal epithelia at early stages (Figs. 4 (C-E) and 11 (D-F)), resulting in epidermal disintegration at the tailbud stage (Fig. 11C), similar to that of ERK3 MOs (Figs. 3 (F and G) and 4). These results indicate that tfap2a knockdown phenocopies ERK3 knockdown in X. laevis.
We also performed knockdown experiments with TFAP2Aspecific siRNA1 and siRNA2, both of which effectively decreased human TFAP2A expression levels without affecting cell proliferation in MCF7 cells (Fig. 12, A and C). TFAP2A knockdown cells piled up to form thickened sheets or colonylike structures (Fig. 12, B, E, G, and H), which were similar to but thinner than those formed by ERK3 knockdown cells (Fig. 5, B, E, G, and H). The milder defects in TFAP2A knockdown cells compared with ERK3 knockdown cells suggest that the actions of ERK3 are not totally dependent upon TFAP2A. Additionally, junctional E-cadherin and ZO-1 signals in TFAP2A knockdown cells were less sharp or less continuous than those in control cells (Fig. 12, D and F). These results thus indicate that TFAP2A, like ERK3, is required for maintaining cell junction integrity to form a flat monolayer epithelial sheet in human epithelial breast cancer cells.

ERK3 is required for full activation of TFAP2A-dependent transcription
Next, we investigated using X. laevis embryos whether the pronephric and epidermal phenotypes in ERK3 morphants could be rescued by tfap2a overexpression. The injection of tfap2a.L mRNA slightly but significantly rescued the reduction of both pronephric atp1b1 expression and epidermal junctional E-cadherin protein in ERK3 morphants (Fig. 13). These results suggest that ERK3 regulates pronephros and epidermal development at least partly through TFAP2A.
The mRNA expression levels of tfap2a were not significantly affected by ERK3 knockdown (Fig. 14A), raising the possibility that ERK3 regulates TFAP2A in a posttranscriptional or posttranslational manner. We therefore examined whether ERK3 affects the transcription activity of TFAP2A by luciferase assays using the TFAP2-luc reporter (3xAP2-Luc), which contains three tandem repeats of the consensus binding sequence of the TFAP2 family to drive the expression of luciferase (33,34) (Fig.  14B). The TFAP2-luc reporter activity was reduced to an average of 69% in ACs and to an average of 65% in whole embryos by ERK3 knockdown compared with that in the controls (Fig. 14C, green dots), suggesting that the endogenous transcription activity of the TFAP2 family is partly dependent upon ERK3 in X. laevis embryos. Additionally, the high TFAP2-luc reporter activity achieved by injecting tfap2a.L mRNA was reduced to an average of 62% in ACs and to an average of 53% in whole embryos by ERK3 knockdown (Fig. 14C, blue dots), suggesting that TFAP2A requires ERK3 to exert its full transcription activity in X. laevis embryos. Moreover, in the human hepatoma cell line HepG2, in which TFAP2A is not detectably expressed (35), tfap2a.L overexpression increased the TFAP2-luc reporter ERK3 regulates epithelial architecture activity (Fig. 14D). This increase was enhanced by the coexpression of WT ERK3A, but not by that of the catalytically inactive K49R/K50R mutant of ERK3A (KR) (36 -38) (Fig. 14D), suggesting that ERK3 directly or indirectly stimulates the transcription activity of TFAP2A in a kinase activity-dependent manner. Additionally, ERK3 siRNAs, which were effective in HepG2 cells (Fig. 14E), slightly but significantly inhibited the tfap2a.L overexpression-dependent TFAP2-luc reporter activity (Fig. 14F, blue dots), but not the basal one (Fig. 14F, green dots), suggesting again that TFAP2A specifically requires ERK3 to exert its full transcription activity. Collectively, these results suggest that ERK3 cooperates with TFAP2A to regulate gene expression.

Discussion
In this study, we identified the crucial roles of the poorly characterized atypical MAPK ERK3 and the branchiooculo-facial syndrome-related transcription factor TFAP2A in the establishment and maintenance of epithelial architecture. Although TFAP2A directly up-regulates E-cadherin transcription (12)(13)(14)17), its regulatory mechanisms and its role in epithelial cell junction formation were unknown. Our results suggest that ERK3 acts in cooperation with TFAP2A to regulate cell junction-related gene expression, adherens and tight-junction formation, and epithelial architecture establishment in the X. laevis embryonic epidermis and human epithelial breast can-cer cells. Although ERK3 has recently been described as a potential target for cancer chemotherapy (4 -6), our results imply that inhibiting ERK3 may have an adverse effect on epithelial architecture, potentially resulting in the promotion of cancer progression and metastasis. Also, it should be noted that our study identifies for the first time a physiological role of ERK3 in nonmammalian vertebrates.
The mechanism by which ERK3 cooperates with TFAP2A remains unclear. Because ERK3 knockdown did not affect tfap2a mRNA expression levels in X. laevis (Fig. 14A), ERK3 might regulate TFAP2A posttranscriptionally or posttranslationally. There are multiple posttranslational steps for regulating the transcription activity of TFAP2 family proteins: their subcellular localization, dimerization, posttranslational modifications, and physical interactions with other transcription factors or co-activators (39). In our luciferase assays, the kinase activity of ERK3 was required for enhancing the transcription activity of TFAP2A (Fig. 14D). The posttranslational step at which ERK3 activity regulates TFAP2A should be elucidated in future studies.
Our finding that ERK3 and TFAP2A regulate epithelial architecture in X. laevis embryos and human cells might provide a clue to the understanding of cellular mechanisms of kidney malformation in ERK3-or tfap2a-knockdown X. laevis embryos (reported in this study), Tfap2a-knockout mice (19), and human patients with branchio-oculo-facial syndrome (23,32). In X. laevis, the pronephros anlage undergoes mesenchymal-to-epithelial transition (MET) to form the pronephric tubule (which functions as embryonic kidney) during the tailbud stage (40,41). ERK3 and tfap2a were little expressed in the early pronephros anlage at the neurula stage but were highly expressed in pronephric tubules at the tailbud stage (Figs. 1B and 8), suggesting the possibility that ERK3 and TFAP2A play an important role in forming pronephric tubules by regulating the establishment of epithelial architecture during MET. It has been shown that cell junction-related genes are required for pronephric tubule formation in X. laevis (42)(43)(44). ERK3 and TFAP2A may regulate pronephric epithelial architecture through controlling the expression of cell junction-related genes. In mammals, although defects in kidney development have not been reported in Erk3-knockout mice, Erk3 expression in mouse embryonic kidneys has been reported (3). In mouse embryonic kidney development, mesenchymal progenitors undergo MET to form nephric tubules (45,46). We thus speculate that the regulation of epithelial architecture by ERK3 and TFAP2A during and after MET might be involved in both

Whole-mount in situ hybridization
Whole-mount in situ hybridization on albino or WT X. laevis embryos was performed according to a standard protocol (47) and using a robot (InsituPro, Intavis). Briefly, embryos were fixed in MEMFA (100 mM MOPS (pH 7.4), 2 mM EGTA, 1 mM MgSO 4 , and 3.7% formaldehyde) and dehydrated in meth-ERK3 regulates epithelial architecture anol. After rehydration, the embryos were treated with 2.5 g/ml proteinase K in PBS containing 0.1% Tween 20, incubated in 100 mM triethanolamine (pH 8.0), incubated in 100 mM triethanolamine containing 0.25% acetic anhydride, and then refixed in 4% paraformaldehyde in PBS. After washing, the embryos were hybridized with digoxigenin-labeled antisense probes and then subjected to staining with anti-digoxigenin (Roche Applied Science, catalog no. 11093274910, 1:2500). The color reaction was performed using BM purple solution (Roche Applied Science). The digoxigenin-labeled probes were synthesized from cDNAs corresponding to the coding regions of X. laevis ERK3A, ERK3B, tfap2a.L, and atp1b1 (atp1b1.S).

Immunofluorescence staining
X. laevis embryos were fixed in 4% paraformaldehyde in PBS overnight at 4°C for E-cadherin immunostaining or in Dent's solution at Ϫ20°C for ZO-1 immunostaining. Fixed embryos were washed with PBS containing 0.1% Triton X-100 (PBST) and blocked with 10% goat serum in PBST at room temperature for 1 h. The primary antibodies used were anti-E-cadherin (DSHB, 5D3; 2.5 mg/ml) and anti-ZO-1 (Invitrogen, catalog no. 61-7300; 1:200). Bound primary antibodies were detected with Alexa Fluor 568 goat anti-rabbit IgG (Invitrogen, catalog no. A-11036; 1:250) or Alexa Fluor 568 goat anti-mouse IgG (Invitrogen, catalog no. A-11031; 1:250). For immunofluorescence staining of MCF7 cells, the cells were fixed in 4% paraformaldehyde at room temperature for 10 min and then permeabilized with 0.2% Triton X-100 in PBS at room temperature for 10 min. After blocking with 3% BSA in PBS at room temperature for 1 h, the cells were incubated with 3% BSA in PBS containing anti-E-cadherin (Cell Signaling, catalog no. 3195; 1:200) or anti-ZO-1 (1:200) at 4°C overnight, washed with PBS, and then incubated with 3% BSA in PBS containing Alexa Fluor 488 goat anti-rabbit IgG (Invitrogen, catalog no. A-11034; 1:250) for 30 min at room temperature. Nuclei were simultaneously stained with Hoechst. To visualize F-actin, we used Alexa Fluor 568 -conjugated phalloidin (Invitrogen, catalog no. A12380). Cells were finally mounted in Mowiol. The confocal image was obtained by using a confocal laser-scanning microscope (Leica, TCS SP8). For the quantification shown in Figs. 4 (D and E) and 13D, images were acquired at a constant gain and offset for each sample. The fluorescence intensity values of junctional ZO-1 and E-cadherin in the graph were then calculated using TCS SP8 LAS X version 1.1 software (Leica) by subtracting the average fluorescence intensities of ZO-1 or E-cadherin in the midline drawn on the background cytoplasmic area from those in the cell-cell boundary, which was defined by appearance (Fig. 4E) or CAAX-GFP signals (Fig. 13D). For the 3D imaging in Figs. 5 (D-H), 6 (F-J), and 12 (D-H), cells were imaged with the XYZ scanning mode of a Leica TCS SP8 confocal laser-scanning microscope and at a constant gain and offset for each sample. 3D images were then generated by the 3D viewer in the TCS SP8 LAS X version 1.1 software.

Examination of the tight-junction permeability barrier
The indicated MOs were injected into the animal region of both ventral blastomeres at the 4-cell stage. After vitelline membrane removal at stage 23, the injected embryos were incubated with 1 mg/ml EZ-link Sulfo-NHS-LC-Biotin (Thermo Fisher, catalog no. 21335) in 0.1ϫ MBS at 15°C for 10 min, fixed

ERK3 regulates epithelial architecture
in 4% paraformaldehyde in PBS at 4°C overnight, washed with PBS, cryoprotected with 7.5% gelatin and 15% sucrose in PBS, and embedded in OCT compound (Sakura). The embedded embryos were frozen in liquid N 2 and transversely sectioned at 14 m using a CryoStar NX70 (Thermo Fisher). The sections were incubated at 4°C overnight with streptavidin FITC diluted 1:500 in 5% BSA in PBS. All slides were washed with PBS and then mounted with Mowiol.

ERK3 regulates epithelial architecture Cell culture
HepG2 cells and MCF7 cells were cultured in Dulbecco's modified Eagle's medium with 10% fetal bovine serum. For immunostaining, MCF7 cells were plated on noncoated glass coverslips and cultured.

Luciferase assay
To construct 3xAP2-Luc, three tandem repeats of the consensus binding sequence of the TFAP2 family (33,34) were inserted into the pGL4.10 vector (Promega) with the minimal promoter. The Renilla luciferase vector pGL4.74 (Promega) was used for an internal control. In Fig. 14D, the indicated combinations of plasmids were transfected into HepG2 cells using Fugene HD (Promega). In Fig. 14F, the indicated plasmids and siRNAs were simultaneously transfected into HepG2 cells by reverse transfection using FuGENE HD transfection reagent and Lipofectamine RNAiMAX transfection reagent (Invitrogen), respectively. In Fig. 14C, the indicated combinations of plasmids, MOs, and tfap2a.L mRNA were injected into all blastomeres of 2-cell stage X. laevis embryos. Animal caps were dissected from the injected embryos at stage 9. Whole embryos (at stage 13), animal caps (at stage 13), and HepG2 cells (24 h after transfection for Fig. 14D or 48 h after transfection for Fig.  14F) were lysed in 1ϫ Passive Buffer (Promega). Luciferase assays were performed using the Dual-Luciferase Reporter Assay System (Promega). Signals were detected using the Glo-Max Multi ϩ Detection System (Promega). Firefly luciferase activity derived from the 3xAP2-Luc vector was normalized to the internal control Renilla luciferase activity.

Microarray analysis
Sampling schemes are shown in Figs. 7A and 10A. Briefly, control MO (80 ng), ERK3 MO1/2 (40 ng each of MO1 and MO2), ERK3 MO3 (80 ng), or TFAP2A MO1/2 (40 ng each of MO1 and MO2) were injected into the animal regions of all blastomeres at the 4-cell stage. The animal caps were dissected from the injected embryos at stage 9, cultured alone or in the presence of activin plus RA, and harvested at stage 15. For Fig. 7, one experiment was performed. For Fig. 10, two independent replicates were prepared. Total RNA was extracted using TRIzol reagent. The quality of the total RNA was assessed using an Agilent 2100 BioAnalyzer. cDNA synthesis and transcriptional amplification were performed using 250 ng of total RNA with the GeneChip 3Ј IVT PLUS Reagent Kit (Affymetrix, catalog no. 902415). Fragmented and biotin-labeled cDNA targets were hybridized to the GeneChip X. laevis Genome 2.0 Array (Affymetrix) according to the manufacturer's protocol. Hybridized arrays were scanned using an Affymetrix GeneChip scanner. Scanned chip images were analyzed with GeneChip operating software version 1.4 (GCOS). The probe set signal intensities in the raw data (CEL files) were normalized using a robust multiarray average algorithm and Expression Console software. Up-regulated genes and down-regulated genes were identified by statistical and -fold change analyses using Gene-Spring GX (Agilent Technologies). Correlation analyses were performed with R version 3.1.0. Because the genes that had low signals (Ͻ2.0) were considered to not be expressed by Gene-Spring GX, we omitted these genes from further analyses. Gene Ontology term analyses were performed using the DAVID bioinformatics tool from the National Institutes of Health (http://david.abcc.ncifcrf.gov/). 5 For transcription factor-binding motif enrichment analysis in Pscan, we used a mouse transcription factor-binding profile from the JASPAR database.

ERK3 regulates epithelial architecture
to the manufacturer's protocol. PCR products were purified with the QIAquick gel extraction kit (Qiagen) and used as a template for nested PCR using KOD FX Neo and a set of primers (forward, 5Ј-CCGGAATTCCGGGGTTTCTCCGGTTGGTCAGAC-3Ј; reverse, 5Ј-CCGCTCGAGCGGGATACTGACAGAATT-ATGCAAC-3Ј). The nested PCR products were purified with the QIAquick PCR Purification Kit (Qiagen), cloned into the pCS2ϩ vector at the EcoRI/XhoI site, and then sequenced.

Cell proliferation assay
For proliferation experiments, WT cells, CRISPR-edited knockout cells, or siRNA-transfected cells were seeded at 3.5 ϫ 10 5 cells/ The normalized tfap2a expression level in control embryos was defined as 1.0 in each experiment. n.s., not significant by Dunnett's test. B, the TFAP2-Luc reporter (3xAP2-Luc) contains three tandem repeats of the consensus binding sequence (GCCNNNGCC) of the TFAP2 family to drive firefly luciferase expression. C, luciferase assays using X. laevis embryos. The indicated combinations of control MO (40 ng), ERK3 MO1/2 (20 ng each of MO1 and MO2), and tfap2a.L mRNA (1 ng) were co-injected with the 3xAP2-Luc plasmid (20 pg) and the pGL4.74 Renilla luciferase plasmid (20 pg) into the animal regions at the 2-cell stage. Animal caps dissected at stage 9 or whole embryos were lysed at stage 13 for the Dual-Luciferase assay, in which firefly luciferase activity was normalized to the internal control Renilla luciferase activity. Shown are all data points from four independent experiments. The bars represent the average Ϯ S.D. The normalized relative light unit for firefly luciferase in control animal caps or embryos was defined as 1.0 in each experiment. *, p Ͻ 0.05; **, p Ͻ 0.01 by t test. D, 3xAP2-Luc (150 ng) and pGL4.74 (10 ng) plasmids were transfected with the indicated combinations of the expression plasmids for ERK3A WT (300 or 600 ng), ERK3A K49R/K50R (KR) (600 ng), or Myc-tagged tfap2a.L (600 ng) into HepG2 cells. The total amount of plasmids was kept constant by adding the mock vector. After 24 h, the cells were lysed for Dual-Luciferase assays. Shown are all data points from 5-8 independent experiments. The bars represent the average Ϯ S.D. The normalized relative light unit for firefly luciferase in control cells was defined as 1.0 in each experiment. *, p Ͻ 0.05 by Dunnett's test. n.s., not significant. E, real-time quantitative RT-PCR analysis of ERK3 expression. HepG2 cells were transfected with 20 nM control siRNA, ERK3 siRNA1, or ERK3 siRNA2, cultured for 48 h, and then subjected to analyses. The mock plasmid was also transfected to examine the efficiency of siRNAs in the presence of co-delivering plasmid DNAs. Shown are all data points from three or four independent experiments. The expression levels of human ERK3 were normalized to those of human GAPDH. The bars represent the average Ϯ S.D. The normalized ERK3 expression level in control cells was defined as 1.0 in each experiment. **, p Ͻ 0.01 by Dunnett's test. F, the 3xAP2-Luc plasmid (1 g), the pGL4.74 plasmid (40 ng), and the indicated siRNA (20 nM) were simultaneously transfected into HepG2 cells with or without the expression plasmid for Myc-tagged tfap2a.L (0.96 g). The total amount of plasmids was kept constant by adding the mock vector. After 48 h, the cells were harvested for the Dual-Luciferase assay. Shown are all data points from four independent experiments. The bars represent the average Ϯ S.D. The normalized relative light unit for firefly luciferase in control cells was defined as 1.0 in each experiment. **, p Ͻ 0.01 by Tukey's test. n.s., not significant by Tukey's test.

ERK3 regulates epithelial architecture
well in 6-well tissue culture plates (IWAKI). Cell proliferation was determined at 72 h after seeding by the direct counting of harvested cells under a microscope.

Statistics
The t test and z test were performed using Excel (Microsoft). The Tukey, Dunnett, and Mann-Whitney U tests were performed using SPSS (IBM). The results were considered significant when p was Ͻ0.05.
Author contributions-C. T. designed and carried out the experiments with the help of K. M. and M. K.; C. T., K. M., M. K., and E. N. wrote the manuscript; E. N. and M. K. supervised the project; all of the authors discussed the results and commented on the manuscript.