mab21-l3 regulates cell fate specification of multiciliate cells and ionocytes

Cell fate specifications of multiciliate cells (MCCs) and ionocytes are commonly suppressed by the Notch pathway in developing epithelia, but are governed by different master regulators, suggesting the existence of a common regulator linking the Notch pathway to both MCC and ionocyte specifications. Here we show that a mab21 family gene, mab21-l3, represents the missing link. In Xenopus embryonic epidermis, mab21-l3 expression is specifically found in MCCs and ionocytes and is downregulated by the Notch pathway. Knockdown of mab21-l3 in Xenopus downregulates both MCC-specific and ionocyte-specific master genes, resulting in drastic loss of MCCs and ionocytes. In mouse tracheal epithelial cells, mab21-l3 expression is also downregulated by the Notch pathway and is required for MCC differentiation. Moreover, conditional gain of function of mab21-l3 rescues Notch-induced loss of MCCs and ionocytes in Xenopus. These results indicate that mab21-l3 acts downstream of the Notch pathway in cell fate specifications of MCCs and ionocytes. The Notch signalling pathway has important roles in embryonic development. Here the authors show that an evolutionarily conserved gene, mab21-l3, is inhibited by Notch signalling and regulates specification of developing multiciliate cells and ion-transporting ionocytes in frog embryos.

F low-generating multiciliate cells (MCCs) and ion-transporting ionocytes are found in epithelia of various organs in vertebrates. MCCs have hundreds of hair-like motile protrusions, multicilia, to generate extracellular fluid flow along epithelial surfaces 1 , while ionocytes (also known as 'mitochondria-rich cells', which include intercalated cells in mammalian kidney and clear cells in mammalian epididymis) selectively secrete or absorb ions and other small molecules to regulate body fluid homeostasis [2][3][4] . Both cells are essential for human physiological processes, and malformation of them is associated with human diseases, including bronchiectasis, hydrocephalus and distal tubule acidosis 1,2 . Cell fate specifications of MCCs and ionocytes are known to be suppressed by the Notch pathway in various developing epithelia [5][6][7][8][9][10][11][12][13][14] . However, MCCspecific and ionocyte-specific transcriptional programmes are governed by different master regulators: a coiled-coil protein multicilin 15 and a forkhead transcription factor foxj1 (refs [16][17][18] for MCCs and another forkhead transcription factor foxi1 (refs 7-9,19-21) for ionocytes. This raises the possibility that an unidentified common regulator(s) links the Notch pathway to both MCC and ionocyte specifications.
In this study, we identify an evolutionarily conserved mab21 family gene, mab21-like 3 (mab21-l3), as a key common regulator representing the missing link. Our results suggest that mab21-l3 acts downstream of the Notch pathway in early phases of both MCC and ionocyte specifications, by regulating expression of master regulators for MCCs (multicilin and foxj1) and ionocytes (foxi1).

Results
Xenopus mab21-l3 is expressed in both MCCs and a-ionocytes. The Xenopus embryonic epithelium has been used as a model system to understand epithelial cell specification as it contains MCCs, ionocytes and mucus-secreting goblet cells 7,21,22 . During our screen identifying genes that show a characteristic expression pattern in the embryonic epithelium, we found that an evolutionarily conserved gene, mab21-l3, showed a salt-andpepper pattern of expression (see below). Mab21 was first identified in Caenorhabditis elegans, and three orthologous genes, mab21-like 1 (mab21-l1), mab21-like 2 (mab21-l2) and mab21-l3, have been identified in vertebrates 23,24 . Among these three genes, mab21-l3 has been poorly studied. Xenopus mab21-l3 was recently shown to regulate dorsal development 24 , but its detailed expression pattern and role in epidermal development have not been examined. We first analysed the spatial expression pattern of Xenopus mab21-l3 by whole-mount in situ hybridization ( Supplementary Fig. 1a). The expression was first detected in the presumptive ectoderm in the animal region at the early gastrula stage (stage 10.5), and became restricted to the ventral ectoderm with a spotted pattern at the late gastrula stage (stage 12/13). At neurula and tailbud stages (stage [14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30], mab21-l3 is expressed in the epidermis with a salt-and-pepper pattern. In the Xenopus embryonic epidermis, two subtypes of ionocytes, aand b-ionocytes, exist, both of which, like MCCs, are known to be distributed in a salt-and-pepper pattern 7,21 . We therefore examined whether mab21-l3 is expressed in MCCs and ionocytes. Embryos hybridized with a mab21-l3 antisense probe were subjected to double immunostaining with antibodies against acetylated a-tubulin (an MCC marker) and ae1 (an a-ionocyte marker; Fig. 1a). Cells with strong mab21-l3 expression (black arrowheads) were co-stained with the anti-ae1 antibody, while cells with weak mab21-l3 expression (yellow arrowheads) were co-stained with the anti-acetylated a-tubulin antibody. mab21-l3expressing cells were positive for either ae1 or acetylated atubulin staining, suggesting that the mab21-l3 expression level in b-ionocytes is very low, if expressed. In situ hybridization combined with single immunostaining produced essentially the same results ( Supplementary Fig. 1b,c). Thus, mab21-l3 is strongly expressed in a-ionocytes, weakly expressed in MCCs and little or very weakly expressed in b-ionocytes.
We then followed the behaviour of mab21-l3-expressing cells during early development, in which MCC progenitors (red) and ionocyte progenitors (yellow) commonly arise from uncommitted cells (green) in the inner cell layer of ventral ectoderm during gastrula stages (from stage 10 to 13) and then radially intercalate into the outer layer in a regular spacing pattern from mid to late  indicates that mab21-l3 is strongly expressed in a-ionocytes and weakly expressed in MCCs at stage 31. Left panels are lateral views. Anterior is to the left and dorsal is upwards. Right panels are magnified images of the epidermis. mab21-l3 mRNA-expressing cells detected with in situ hybridization are dark blue. MCCs and a-ionocytes were identified by staining with anti-acetylated a-tubulin (red) and anti-ae1 (green) antibodies, respectively. Yellow arrowheads show MCCs weakly expressing mab21-l3. Black arrowheads show a-ionocytes strongly expressing mab21-l3. Essentially the same results were obtained in 19 embryos from five independent experiments. Scale bars, 500 mm (left panels) and 50 mm (right panels). (b) mab21-l3 is initially expressed in the inner cell layer, but later expressed in the outer cell layer, of Xenopus embryonic epidermis. The left panels show vertical sections of embryos processed for whole-mount in situ hybridization against mab21-l3. The right diagram shows the current model for MCC and ionocyte specification. In all panels, apical is at the top. Scale bars, 40 mm. Essentially the same results were obtained in two to five embryos of each stage from one to three independent experiments. neurula stages (from stage 17 to 21; see Fig. 1b, right). After intercalation is completed, these progenitors mature into MCCs and a/b-ionocytes, respectively, in the outer layer 5-7,21 (see Fig. 1b, right). At stages 11-13, mab21-l3 is expressed in the inner cell layer of the epidermis (Fig. 1b, left). At stage 17, mab21-l3expressing cells are located around the border between two layers. At stages 25-30, mab21-l3 expression is observed in the outer cell layer with a spotted pattern. These results demonstrate that the behaviour of mab21-l3-expressing cells closely resembles that of MCC and/or ionocyte progenitors, and thus raise the possibility that mab21-l3 is involved in MCC/ionocyte specification. mab21-l3 is required for the formation of MCCs and ionocytes. We next performed knockdown of mab21-l3 with an antisense morpholino oligonucleotide (MO). Mab21-l3 MO, but not a fivebase mismatch control (mab21-l3 5mis MO), markedly blocked translation from mab21-l3 mRNA ( Supplementary Fig. 2). We co-injected control MO, mab21-l3 MO or mab21-l3 5mis MO with a lineage tracer fluorescein dextran into the left ventral blastomere of 4-cell-stage embryos. Injected embryos, in which uninjected right sides serve as an internal control, were analysed by whole-mount in situ hybridization for marker gene expression. Injection of mab21-l3 MO, but not that of control MO or mab21-l3 5mis MO, led to a drastic decrease in the number of cells expressing the a-ionocyte marker ae1 at stage 22 ( Fig. 2a,b). Similarly, cells showing the normal expression level of pendrin, a b-ionocyte marker, were drastically reduced by mab21-l3 MO at stage 22 (Fig. 2c,d). Next we performed immunofluorescence staining of MO-injected embryos with an antibody against acetylated a-tubulin, an MCC marker. The number of MCCs was also drastically decreased in mab21-l3 morphants (Fig. 3a-c).
These results indicate that mab21-l3 is required for the formation of both ionocytes and MCCs.
mab21-l3 regulates early specification of MCCs and ionocytes. Although mab21-l3 seems to be scarcely expressed in b-ionocytes, mab21-l3 MO markedly decreased the number of b-ionocytes at tailbud stages (Fig. 2c,d), leading to the hypothesis that mab21-l3 functions in ionocyte specification at earlier stages. It has been known that the master regulator for ionocyte specification, foxi1, starts to be expressed in progenitor cells at gastrula stages 7 . Whole-mount in situ hybridization showed that injection of mab21-l3 MO, but not that of control MO or mab21-l3 5mis MO, led to a dramatic decrease in foxi1 expression at the late gastrula stages (stages 12 and 13; Fig. 4a; Supplementary Fig. 3a). The results of real-time quantitative PCR (qPCR) also indicated that mab21-l3 MO significantly reduces the expression level of foxi1 at stages 12 and 13 ( Fig. 4d; Supplementary Fig. 3d). These results suggest that mab21-l3 functions in ionocyte specification at the upstream of foxi1. Next we examined the effect of mab21-l3 knockdown on the expression of the established master regulator for MCC specification, foxj1. Both whole-mount in situ hybridization and real-time qPCR analyses revealed that mab21-l3 MO led to a decrease in foxj1 expression at the late gastrula stage (Fig. 4b,  A previous study reported that depletion of the ionocyte master regulator foxi1 leads to the loss of MCCs at later stages (apparent stages 26-27, although the authors did not mention the stage number) 25 , raising the possibility that the loss of MCCs observed in mab21-l3 knockdown embryos is secondarily caused by the loss of ionocytes. To address this issue, we examined whether foxi1 knockdown leads to the loss of MCCs at earlier stages. At stage 22, ae1 or pendrin expression was drastically decreased, but a-tubulin expression was not or only slightly affected, by the translation-blocking foxi1 MOs (MO1 plus MO2) 26 or the spliceblocking foxi1 MO (SBMO) 25 (Supplementary Fig. 4). Also, another group reported that a-tubulin expression (at apparent stage 25, although the authors did not mention the stage number) is not affected by foxi1 knockdown in the related species X. tropicalis 21 . Taken together, the loss of ionocytes does not largely affect early specification of MCCs before stage 25, but it affects maturation of MCCs at later stages. Our results also demonstrate that mab21-l3 knockdown decreases expression of MCC markers at earlier stages such as stages 12, 13 and 17 (Figs 4b-d and 5; Supplementary Fig. 3b-d). Thus, the loss of MCCs in mab21-l3 knockdown embryos occurs at very early stages and is not secondarily caused by the loss of ionocytes. Also, we examined whether co-expression of foxi1 rescues the loss of ionocytes and MCCs in mab21-l3 knockdown embryos. Injection of mab21-l3 MO alone resulted in the loss of ae1, pendrin and a-tubulin expressions ( Supplementary Fig. 5). Co-injection of mab21-l3 MO with foxi1 mRNA was able to induce weak ae1 expression as well as strong pendrin expression, but was unable to restore a-tubulin expression ( Supplementary Fig. 5), indicating that the loss of MCCs in mab21-l3 knockdown embryos can be observed even in the condition activating the ionocyte differentiation programme. Taken together, requirement of mab21-l3 in MCC specification seems to be independent of the presence or absence of ionocytes, thus decreasing the possibility that the loss of MCCs observed in mab21-l3 knockdown embryos is secondarily caused by the loss of ionocytes.
Next we addressed whether mab21-l3 is sufficient for MCC/ ionocyte specification. Mab21 family proteins have been suggested to function as a transcriptional regulator 27,28 , and exogenously expressed mab21-l3 protein was localized to the nucleus in Xenopus ectoderm ( Supplementary Fig. 1d). Because injection of synthesized mRNA encoding myc-tagged wild-type mab21-l3 induced cell death at blastula stages in a preliminary experiment (18/18 (100%) death, 1 ng mab21-l3-myc mRNA per embryo; 12/20 (60%) death, 400 pg mab21-l3-myc mRNA per embryo), we used a glucocorticoid-inducible form of mab21-l3 (mab21-l3-hGR), whose nuclear translocation can be conditionally induced by dexamethasone (DEX). Embryos unilaterally injected with mab21-l3-hGR mRNA were treated with DEX at gastrula stages, and then cultured until tailbud stages, at which the number of MCCs, a-ionocytes or b-ionocytes in the injected side was compared with that in the uninjected side. DEX treatment weakly but significantly increased the number of MCCs, but not that of a-ionocytes or b-ionocytes ( Fig. 6a; Supplementary Fig. 6). Moreover, overexpression of mab21-l3-hGR increased the expression level of multicilin in both ectodermal explants and whole embryos at stage 12 ( Fig. 6b), although DEX dependency was low possibly because of a short treatment period (from stage 10 to 12). These results taken together suggest that mab21-l3 is sufficient to promote  (a-c) Control MO, mab21-l3 MO or mab21-l3 5mis MO (a,c, 30 ng each; b, 40 ng each) was co-injected with fluorescein dextran (100 ng) into the left ventral blastomere at the four-cell stage, and embryos at stage 13 were subjected to whole-mount in situ hybridization. Ventral views of embryos are shown in left six panels, and fluorescence views reveal distribution of injected materials. foxi1 (a), foxj1 (b) and multicilin (c) expressions were decreased in mab21-l3 MOinjected embryos (foxi1, 100%, n ¼ 17; foxj1, 81%, n ¼ 32; multicilin, 95%, n ¼ 21), but not affected in those injected with control MO (foxi1, 100%, n ¼ 21; foxj1, 91%, n ¼ 35; multicilin, 94%, n ¼ 31) or mab21-l3 5mis MO (foxi1, 100%, n ¼ 14; foxj1, 100%, n ¼ 11; multicilin, 80%, n ¼ 30). The middle 12 panels show higher magnification of boxed areas in left 6 panels. The right graphs show the ratio of the foxi1-, foxj1-or multicilin-expressing cell number per field in the MO-injected side to that in the uninjected side. Shown is the average from embryos in two or three independent experiments. n, the number of embryos scored. (d) The indicated MO (60 ng each) was injected into two ventral blastomeres at the four-cell stage, and embryos at stage 13 were subjected to real-time qPCR analysis. Shown is the average of three or seven independent experiments. (e) Mab21-l3 MO enhances cell proliferation. Left panels, the indicated MO (80 ng each) was injected into two left blastomeres at the four-cell stage, and embryos were fixed at stage 13 for phospho-histone H3 staining. Dorsal is upwards. A, anterior; P, posterior. The right graph shows the ratio of the number of phospho-histone H3-positive cells in the injected side to that in the uninjected side. Shown is the average from embryos (control MO, n ¼ 15; mab21-l3 MO, n ¼ 19) in two independent experiments. (a-e) Error bars represent s.d. *Po0.05, **Po0.01 by Tukey's test (a-d) or unpaired Welch's t-test (e). Scale bars, 500 mm (a-c, left 6 panels; e) and 50 mm (a-c, middle 12 panels). NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7017 ARTICLE specification of MCCs, but not ionocyte specification, which may require an additional factor(s).
In the Xenopus embryonic epidermis, MCCs and ionocytes arise from the inner cell layer where mab21-l3 is expressed (see Fig. 1b), whereas mucus-secreting goblet cells arise from the outer cell layer. We investigated whether mab21-l3 affects differentiation of goblet cells. The expression of Xeel, the goblet cell marker 22 , was reduced by mab21-l3 knockdown at stages 17 and 22 ( Supplementary Fig. 7a,b). However, at stage 11, expression of outer-layer-specific markers, WRPW-bHLH protein enhancer of split-6e (esr6e) and grainyhead-like 3 (grhl3) 29,30 , was not significantly affected by mab21-l3 knockdown ( Supplementary  Fig. 7c). This result suggests that at stage 11 mab21-l3 is dispensable for specification of outer-layer cells including goblet cell progenitors. At stage 12, expression of grhl3 was unaffected, but expression of esr6e or epidermal keratin (xk81a1, another outer-layer marker) was downregulated by mab21-l3 knockdown ( Supplementary Fig. 3e). Thus, outer layer-specific gene expression was initially unaffected, but partly affected later in mab21-l3 knockdown embryos. We also found that the Notch pathway, the established negative regulator for MCC and ionocyte specification, was able to inhibit differentiation of goblet cells at later stages ( Supplementary Fig. 8a). We therefore speculate that differentiation of outer-layer progenitors into goblet cells is disturbed at later stages in mab21-l3 MO-injected embryos, possibly due to secondary effects caused by the loss of MCCs and ionocytes at early stages.
Because both ionocytes and MCCs arise from the inner layer of early ectoderm (see Fig. 1b, right), it is possible that the loss of ionocytes and MCCs in mab21-l3 knockdown embryos is secondarily caused by global defects of inner-layer patterning at early stages. To address this point, we examined expression of inner-layer markers, p63, vestigial-like 4 (vgll4) and hyaluronan synthase 1 (has1) 30,31 . At stage 12, mab21-l3 knockdown significantly decreased expression of p63, but not that of vgll4 or has1 ( Supplementary Fig. 3e), suggesting that inner-layer-specific gene expression is at least partly normal in the absence of mab21-l3. Thus, mab21-l3 does not play a crucial role in controlling all aspects of inner-layer patterning.  mab21-l3 acts downstream of the Notch pathway. Next we investigated the relationship between mab21-l3 and Notch signalling. The intracellular domain of Notch (NICD) has been widely used to induce Notch pathway activation. Overexpression of Xenopus NICD (xNICD) led to marked reduction in mab21-13 expression at stages 13 and 31/32 (Fig. 7a,b; Supplementary  Fig. 8a). We also co-injected xNICD mRNA with fluorescein dextran into one ventral blastomere at the 32-cell stage, and the injected embryos, which were cultured until stage 27, were hybridized with a mab21-l3 antisense probe, followed by immunostaining with an antibody against E-cadherin to visualize the plasma membrane (Fig. 7c). mab21-l3 expression could be seen in xNICD-negative epithelial cells, but not in xNICD-positive epithelial cells (Fig. 7c), suggesting that the inhibitory action of Notch on mab21-l3 expression may be cell autonomous. We then searched for a mediator that acts downstream of Notch to repress mab21-l3 transcription. Certain members of the transcriptional repressor Hes family, including hes1 and hes5, are known as major targets of the Notch pathway 32 . Thus, a Xenopus close homologue of mammalian hes5, esr6e, could be a candidate, because esr6e is known to be induced by the Notch pathway to repress MCC specification 5 . Injection of esr6e mRNA decreased mab21-l3 expression at stages 13 and 31/32 (Fig. 7a,b). Moreover, expression of the a-ionocyte marker ae1 and the b-ionocyte marker pendrin was downregulated by injection of esr6e mRNA, but was upregulated by that of esr6e Db (a dominant negative form of esr6e) 5 mRNA (Supplementary Fig. 8b). These results indicate that esr6e mediates the action of Notch signalling to repress mab21-l3 transcription and MCC/ionocyte specification. We reasoned that ectopic expression of mab21-l3 should rescue Notch-dependent inhibition of MCC/ionocyte specification, if the inhibition is mediated through Notch-dependent downregulation of mab21-l3. We first confirmed Notch inhibition of MCC/ionocyte specification. Injection of xNICD mRNA alone led to the decrease in foxi1 and multicilin expressions at stage 13 ( Fig. 8a,b, second left). At later stages, xNICD suppressed ae1 expression, pendrin expression and acetylated a-tubulin abundance (Fig. 9a,b, second top panels), as also indicated by the bar graph (Fig. 9c, white and light grey bars), which represents the ratio of the number of ae1-, pendrin-or acetylated a-tubulinpositive cells in the injected side to that in the uninjected side. These results are consistent with the previous finding that the Notch pathway suppresses both MCC and ionocyte specification. For rescue experiments, we used mab21-l3-hGR. When embryos co-injected with xNICD and mab21-l3-hGR mRNA were cultured with DEX from the late blastula stage (stage 9), the xNICDinduced decrease in foxi1 or multicilin expression was cancelled by mab21-l3-hGR in a DEX-dependent manner (Fig. 8a,b, rightmost and second right). Moreover, mab21-l3-hGR with DEX restored ae1 expression, pendrin expression and acetylated a-tubulin abundance at later stages (Fig. 9a,b, first and second bottom panels), as indicated also by the bar graph (Fig. 9c, dark grey and black bars). Thus, exogenous expression of mab21-l3 is sufficient to overcome Notch-dependent inhibition of MCC/ ionocyte specification. It should be noted that DEX treatment from the late gastrula stage (stage 13) could not restore a-tubulin expression ( Supplementary Fig. 9), suggesting that mab21-l3 begins to function at late blastula or early gastrula stages, the timing before MCC/ionocyte progenitors are formed. These results taken together demonstrate that mab21-l3 regulates early phases of MCC/ionocyte specification by acting downstream of the Notch pathway and upstream of master regulators (Fig. 9d).
In response to overexpression of xNICD, the numbers of b-ionocytes, MCCs and a-ionocytes showed a 36, 26 and 19% decrease, respectively (Fig. 9c), suggesting that the sensitivity to Notch activation of b-ionocyte, MCC and a-ionocyte ARTICLE specification is high, medium and low, respectively. Interestingly, expression of mab21-l3 was almost undetectable, weak and strong in b-ionocytes, MCCs and a-ionocytes, respectively (Fig. 1a). Thus, the responsiveness to Notch activation was inversely correlated with the expression level of mab21-l3. These results suggest that the expression level of mab21-l3 is one of the factors influencing Notch responsiveness and cell fate decision in MCC/ ionocyte development.
Mouse Mab21-l3 is required for MCC differentiation. We finally examined whether the role of mab21-l3 is conserved in mammals. Overexpression of mouse Notch intracellular domain (mNICD) in mouse tracheal epithelial cells (mTECs) markedly upregulated mouse Hes5, an established Notch-inducible gene that shows close homology to Xenopus esr6e, and significantly downregulated mouse Mab21-l3 (Fig. 10a). These results indicate that the Notch pathway is able to inhibit mab21-l3 expression in mice as well as in Xenopus. Next we assessed a role of mouse Mab21-l3 in MCC differentiation. It has been previously established that MCCs start to be induced at about 3-4 days after exposure of confluent mTECs to an air-liquid interface (ALI) 33 . First we examined changes in gene expression before and after ALI exposure. Multicilin exhibited peak expression around 2 days after ALI exposure (ALI þ 2), while Mab21-l3 exhibited peak expression around the starting time of ALI exposure (ALI þ 0) (Fig. 10b), consistent with our hypothesis that mab21-l3 acts upstream of multicilin to regulate early phases of MCC specification. In agreement with this result, the BioGPS database (http://biogps.org/#goto=genereport&id= 126868) indicates that human MAB21-L3 gene exhibits peak expression at ALI þ 0 in the previous microarray data of ALI-treated bronchial epithelial cells (the GEO accession number GSE5264) 34 . We next designed two short hairpin RNAs (shRNAs) targeting mouse Mab21-l3, mMab21-l3 shRNA #1 and mMab21-l3 shRNA #2, both of which inhibited Mab21-l3 expression in NIH3T3 cells (Fig. 10c). Each shRNA and GFP was co-expressed in mTECs by lentiviral delivery, and at 6 days after ALI exposure (ALI þ 6), the multicilia were visualized by immunostaining with the antiacetylated a-tubulin antibody. While multicilia were formed in 17% of GFP-positive cells expressing control shRNA, they were formed in only 8 and 7% of GFP-positive cells expressing mMab21-l3 shRNA #1 and #2, respectively (Fig. 10d,e). We also examined tracheal progenitor cells called basal cells, which are known to be abundantly present in the mTEC culture 33 . The percentage of cells positive for the basal cell marker p63 was not significantly changed by mMab21-l3 shRNAs ( Supplementary  Fig. 10), ruling out the possibility that Mab21-l3 knockdown has nonspecific toxic effects on mTECs. These results indicate that Mab21-l3 is required for MCC differentiation in mouse.

Discussion
In this study, we uncover a missing molecular link between the Notch pathway and MCC/ionocyte specification. We have identified an evolutionarily conserved gene mab21-l3 as a bifunctional regulator of both MCC and ionocyte specification. Moreover, our rescue experiments suggest that Notch/esr6edriven repression of mab21-l3 is a key step to restrict the number of MCCs and ionocytes to an appropriate level. We have also shown that mab21-l3 functions upstream of the established master regulators for MCCs (multicilin and foxj1) and ionocytes (foxi1). As mab21-l3 expression is positively regulated by foxi1 (ref. 24), there is a feedback relationship between foxi1 and mab21-l3 (see Fig. 9d). This positive feedback may ensure robust progression of ionocyte differentiation. Also, miR-449 micro-RNAs, whose genomic locus is close to the multicilin gene 15 , could be functionally linked with mab21-l3, because they have been shown to regulate MCC differentiation through inhibition of the Notch signalling pathway in both Xenopus embryonic epidermis and mammalian airway epithelial cells 35 .
Although our results were mainly obtained in the Xenopus embryonic epidermis, two main points were also shown to be conserved in mammals; that is, the Notch pathway-dependent repression of mab21-l3 expression and the essential role of mab21-l3 in MCC specification. Because the Notch pathway is known to restrict MCC and ionocyte specification in various organs, including mouse lung 12,13 , mouse kidney 14 , zebrafish pronephros 10,11 and zebrafish skin 8,9 , the Notch/mab21-l3 signalling axis could have a conserved role in MCC/ionocyte specification in vertebrates.
In our rescue experiment in Xenopus, the sensitivity to Notch activation was low in a-ionocytes (which strongly express mab21-l3), medium in MCCs (which weakly express mab21-l3) and high in b-ionocytes (which show no or little mab21-l3 expression). This implies that the expression level of mab21-l3 is one of the factors that influence the responsiveness to Notch. Because a search of NCBI GEO and BioGPS databases reveals that mammalian mab21-l3 is expressed in various tissues including oesophagus, rectum and skin, mab21-l3 may fine-tune the responsiveness to Notch in many other biological contexts.
Previous studies have suggested that the mab21 family member could act as a transcriptional regulator 27  and mab21-l3-hGR (100 pg) mRNAs were co-injected with the tracer fluorescein dextran (100 ng) into the left ventral blastomere at the four-cell stage, and injected embryos were cultured with or without DEX (10 mM) from stage 9. At stage 13, embryos were harvested for whole-mount in situ hybridization against foxi1 (a) or multicilin (b). Ventral views of embryos are shown. Fluorescence views reveal distribution of injected materials. Boxed areas in left eight panels are shown at higher magnification in right panels. Scale bars, 500 mm (left panels) and 50 mm (right panels). foxi1 and multicilin expressions were markedly decreased in xNICD mRNA-injected embryos (foxi1, 92%, n ¼ 12; multicilin, 86%, n ¼ 28) or DEX-untreated embryos injected with xNICD and mab21-l3-hGR mRNAs (foxi1, 90%, n ¼ 20; multicilin, 80%, n ¼ 30), but not largely affected in tracer-injected embryos (foxi1, 100%, n ¼ 16; multicilin, 94%, n ¼ 17) or DEX-treated embryos injected with xNICD and mab21-l3-hGR mRNAs (foxi1, 58%, n ¼ 19; multicilin, 66%, n ¼ 29), in two independent experiments. mab21-l3-hGR also suggest that mab21-l3 primarily functions in the nucleus. We thus speculate that mab21-l3 protein may interact with additional nuclear proteins such as transcription factors, coactivators, corepressors or chromatin-remodelling factors to activate transcription of genes required for cell fate specification. Because mab21-l3 plays a dual role in MCC and ionocyte specification, the constituents of the mab21-l3 complex may vary in a cell-type or context-dependent manner. Detailed molecular mechanisms of the action of mab21-l3 should be elucidated in future studies.
Whole-mount in situ hybridization. Whole-mount in situ hybridization 36 on albino or wild-type Xenopus embryos was performed using a robot (InsituPro, Intavis). Embryos were fixed in MEMFA (100 mM MOPS, pH 7.4, 2 mM EGTA, 1 mM MgSO 4 and 3.7% formaldehyde) for 2 h, dehydrated and stored in methanol. After rehydration, embryos were treated with 5 mg ml À 1 proteinase K in PBS containing 0.1% Tween20, incubated in 100 mM triethanolamine (pH 8) for 12 min, incubated in 100 mM triethanolamine (pH 8) containing 0.25% acetic anhydride for 12 min and then refixed with 4% paraformaldehyde in PBS for 20 min. After washing, embryos were hybridized with digoxigenin-labelled probes for 18 h at 58°C, and then subjected to anti-digoxigenin antibody staining. The colour reaction was performed using BM purple. The digoxigenin-labelled antisense and sense probes were synthesized from full-length cDNA sequences of mab21-l3, a-tubulin, foxj1, foxi1, ae1, pendrin and Xeel. For whole-mount in situ hybridization combined with immunostaining, we omitted the proteinase K step.
Culture of mTECs. Primary mTEC culture was performed according to the published protocol 33 . mTECs were isolated from tracheas of 7 weeks old, wild-type C57BL/6 J male mice, seeded onto Transwell Clear permeable membranes (Corning, #3460) coated with rat-tail collagen, and then cultured with mTEC plus medium 33 . For mNICD overexpression experiments, cells were spin infected with the lentivirus in the presence of 5 mg ml À 1 polybrene (Sigma) at 2 days after seeding 38 , and then harvested at confluency for real-time qPCR analysis. For MCC differentiation experiments, cells reaching confluency were exposed to an ALI and cultured with mTEC basic medium 33 containing 2% Nu-Serum (Becton-Dickinson). For shRNA experiments, cells were spin infected with the shRNAexpressing lentivirus in the presence of 5 mg ml À 1 polybrene at 2 days after seeding 38 , and then exposed to ALI after reaching confluency. All mouse experiments were conducted in accordance with the Regulation on Animal Experimentation at Kyoto University and approved by the Animal Experimentation Committee of Kyoto University.
Statistics. Cell counting was done using Image J Software. The t-test was performed using the EXCEL Software (Microsoft). The Tukey's test and the Dunnett's test were performed using the SPSS Software (IBM). Results are considered significant when Po0.05.