HNF1B controls epithelial organization and cell polarity during ureteric bud branching and collecting duct morphogenesis

Branching morphogenesis is the process by which simple epithelium arborizes into an elaborate network. The molecular and cellular events underlying branching morphogenesis have only begun to be characterized. Several genetic studies have shown that the GDNF and FGF pathways are necessary to promote the formation of new branches in different organs (Gjorevski and Nelson, 2010). This process also requires a number of cellular events, such as cell migration, adhesion and proliferation. These are guided by intracellular signaling triggered by stimuli at the cell surface, which lead to active actin cytoskeleton reorganization and extracellular matrix remodeling (Kim and Nelson, 2012).

During definitive mouse kidney development, the ureteric bud (UB) is induced by signals from adjacent metanephric mesenchyme (MM) and emerges from the caudal part of the Wolffian duct (WD) at embryonic day 10.5 (E10.5). This initial outgrowth is regulated by the GDNF/Ret signaling cascade and is mainly due to localized cell proliferation and cell movements into a pseudostratified epithelium (Michael and Davies, 2004; Costantini, 2012). The UB remains pseudostratified during the first branching event (E11.5) and is subsequently reshaped into a simple cuboidal epithelial tube (Chi et al., 2009a). From E11.5 to E15.5, the UB expands into the MM forming new branches and inducing nephrogenesis at each branch tip. After E15.5, UB growth slows and the trunks of the branches elongate, differentiating into principal and intercalated cells (Costantini, 2012). This stereotyped process of branching gives rise to the complex network of collecting ducts (CDs) that will transport the renal filtrate from the nephrons to the ureter and, subsequently, the bladder.

The POU-homeodomain transcription factor HNF1B plays a crucial role in the early differentiation of various cell lineages and organs in vertebrates (Barbacci et al., 1999; Wild et al., 2000; Bohn et al., 2003; Coffinier et al., 2002; Gresh et al., 2004; Haumaitre et al., 2005; Lokmane et al., 2008). In humans, HNF1B heterozygous mutations cause the complex syndrome renal cysts and diabetes (RCAD) that is characterized by kidney, genital tract and pancreas abnormalities, as well as early onset of diabetes (Barbacci et al., 2004; Bellanné-Chantelot et al., 2004; Haumaitre et al., 2006; Heidet et al., 2010). Constitutive inactivation in the epiblast by tetraploid aggregation has shown that HNF1B is involved in WD maintenance, the timing of UB outgrowth and its early branching, as well as in the induction of nephrogenesis (Lokmane et al., 2010). However, the premature arrest of UB growth after the first branching event in these tetraploid rescued embryos precluded the examination of later aspects of branching and CD morphogenesis (Lokmane et al., 2008, 2010).

Using conditional inactivation in nephron progenitors, we uncovered an additional and later role of HNF1B in the acquisition of a proximal-intermediate nephron segment fate and subsequent differentiation of all nephron segments (Heliot et al., 2013). This function appears to be conserved in vertebrates (Heliot et al., 2013; Massa et al., 2013; Naylor et al., 2012). It has also been shown that Hnf1b conditional inactivation in renal medullar tubules leads to cystic kidneys after birth (Gresh et al., 2004; Hiesberger et al., 2005). Importantly, human fetuses carrying heterozygous mutations in HNF1B exhibited bilateral dysplastic kidneys with cysts derived from all nephron segments and the collecting system (Haumaitre et al., 2006), thus indicating an early developmental role in human kidney morphogenesis. However, and despite previous studies, the role of this factor during CD morphogenesis has not yet been addressed.

Here, we examine the function of HNF1B during renal branching using conditional inactivation in the UB and its derivatives. By using a reporter strategy that allows high-resolution analysis of cell behaviors, we show that deletion of Hnf1b affects cell movements during mitosis. We found that HNF1B is required both at the tip domain to maintain cell-cell contacts and within the UB trunks for proper epithelial cell organization and further differentiation of the collecting system.

To examine the specific function of HNF1B during UB branching, we used the HoxB7-Cre line that mediates recombination in the WD and derivatives from E9.5 (Yu et al., 2002). We assessed deletion efficiency by generating two mouse lines: one carrying the ROSA26^{loxP-mtomato-loxP-mEGFP} reporter and Hnf1^fl/f, which expresses the EGFP or tomato marker at the cell membrane, depending on the expression of Cre-recombinase, and the other with the Rosa^EYFP/+ reporter. HNF1B-GFP co-immunostaining at E13.5 showed the specificity of the deletion in the UB compartment because as developing nephrons, MM and stroma did not express GFP (Fig. 1A,C). As expected, in controls, HNF1B⁺ nuclei were detected in both UB branches and developing nephrons (Fig. 1A). In mutants, HNF1B was present in early nephron structures, whereas no specific HNF1B staining was observed in GFP⁺ cells (Fig. 1B). Accordingly, we found a 30% decrease in wild-type Hnf1b transcripts at E13.5 (Fig. 1E). However, some cells were GFP negative and HNF1B positive, apparently owing to a partial mosaicism of the HoxB7-Cre. This mosaic deletion of Hnf1b then allowed us to compare the fate of HNF1B-postitive and HNF1B-negative cells. Interestingly, at E13.5 we found more HNF1B⁺ cells within the mutant tips than in the trunks (59.7% and 24.2%, respectively; Fig. 1F). By contrast, the same proportion of cells from tips or trunks in heterozygous kidneys lacked EYFP staining (12.6% and 11%, respectively). At later stages, HNF1B⁺ cells were almost undetectable in mutant embryos in the future CDs and developing ureter, whereas the percentage at the UP tips increased up to 90%. Thus, mutant cells appear to be progressively excluded from the UB tips to populate the trunks (Fig. 1F). Interestingly, we have previously observed a similar behavior by Hnf1b mutant cells in Hnf1b^−/−: WT diploid chimeras (Lokmane et al., 2010).

The extensive cell rearrangements that occur during UB branching likely contribute to the differential mosaicism observed within the tip and trunk domains, which leads to a strong reduction in Hnf1b mutant cells within the UB tips while the future CDs contain mostly mutant cells. This differential mosaicism allowed the examination of HNF1B function at later stages during collecting duct morphogenesis.

Gross morphology analyses were performed at different developmental stages and in newborn pups, in controls (HoxB7^Cre/+;Hnf1b^lacZ/+) and mutants (HoxB7^Cre/+;Hnf1b^fl/lacZ), from the same litter either carrying or lacking the allele ROSA26^{loxP-mtomato-loxP-mEGFP} (see mouse genotype details in the Materials and Methods). Mutant kidneys exhibited delayed UB branching and irregular WDs, often exhibiting ectopic buds, at E13.5 (96%) (Fig. 1D and Fig. 2A′).

Mutant newborns were obtained at the expected Mendelian ratio; however, the majority of them died between P2 and P15 (Table 1). At P0, the mutants displayed multiple urogenital tract abnormalities, including severe bilateral hypoplasia, medullar cysts and hydroureters, and in some cases kidney/ureter duplications or unilateral agenesis (Table 1; Fig. 2B,B′). Further analysis of either E16.5-17.5 mutant embryos or P0 newborns showed normal insertion of the ureters into the bladder in both females and males, with no evidence of ureter obstruction that could explain hydroureters (data not shown). Additionally, mutants exhibited cysts both in the female (oviducts) and male (epidydimus) genital tracts (Fig. 2B′ and data not shown).

Histological examination of mutant kidneys showed strongly reduced branching and nephron number, as well as the development of medullar cysts (Fig. 2C-D′). In mutants, nascent nephrons appeared to develop normally as manifested by WT1 expression in future podocytes of well-formed S-shaped bodies (Fig. 2E,E′). However, further differentiation of nephron segments was perturbed. Notably, the clusters of proximal tubules, stained for AQP1, and the primitive loop of Henle, stained for NKCC2, were underdeveloped and partially disorganized (Fig. 2F-G′). This could be, at least in part, a secondary consequence of a mispatterned collecting duct system.

Thus, inactivation of Hnf1b in the UB and collecting system leads to severe urogenital phenotypes that resemble congenital anomalies of the kidney and urinary tract (CAKUT), one of the predominant developmental disorders that represents about 30% of the anomalies found during the prenatal period.

The ureteric tree develops from stereotypical branching events of the UB. The analysis of these events has been advanced by the application of optical projection tomography (OPT) and the development of software to quantify kidney morphometrics (Short et al., 2014). We examined the branching of the UB epithelia at E13.5 and E16 in 3D by OPT using Troma-1 staining the UB trunks and Ezrin enriched at the tips (Fig. 3A-D″). Following OPT, reconstructed mutant kidneys showed a significant decrease in branching morphogenesis at E13.5 (Fig. 3A-B″). At E16, CDs developed cysts mainly around the pelvic area (Fig. 3C-D″). Skeletal and segmental metrics were assessed using the Tree Surveyor program (Short et al., 2013). We observed that the number of segments and tips were decreased at E13.5 (68% and 66%, respectively) (Fig. 3E). Consistently, mutant kidneys exhibited a decrease in kidney ureteric tree volume of 68% at E13.5 (Fig. 3F) and 60% at E16 (not shown). At E11.5 we observed mis-shaped UB tips (157° in control versus 106° in mutants, data not shown) as observed previously (Lokmane et al., 2010). Then, at E13.5, after remodeling, the angle of the first branching event was reduced by 26° in mutants (Fig. 3G). When assessing all branch segments per generation, a significant dysmorphology was observed in the mutant kidneys, as indicated by the decreased volume and the more curved segments (Fig. 3H,I). All together, this analysis shows that deletion of Hnf1b in WD and UB causes significant changes in the spatial morphology of the ureteric tree, highlighting the requirement of HNF1B for the UB branching pattern program. These results also suggest that HNF1B is necessary for ureteric tree epithelial organization. These early defects likely affect the later kidney growth and function observed in our mutants.

To further understand UB branching defects, we used the ROSA26^{loxP-mtomato-loxP-mEGFP} reporter system to investigate the behavior of mutant cells versus wild type during UB branching by time-lapse imaging of ex vivo mouse kidney cultures (Movies 1 and 2) using the novel culture method FiZD (Fixed Z-dimension) (Saarela et al., 2017). First, we confirmed the decreased branching and differential cell recruitment phenotypes that were observed in mutant kidneys in vivo; indeed, after 56 h or 5 days of culture, mutant kidneys were significantly less branched than controls (compare Movies 1 and 2; Fig. 4A-D′). In addition, while GFP⁺ and GFP⁻ cells were randomly distributed within the UB trunks and tips in the controls. By contrast, mutant cells (GFP⁺) were increasingly excluded from the UB tip domain (Fig. 4C-D′). Moreover, the epithelial organization of UB branches was also defective, and the lumen was barely detectable in some regions of UB trunks (Fig. 4C,D′). These results confirm that Hnf1b-deficient cells have a cell-autonomous deficiency in their ability to contribute to the UB tips.

Active cell proliferation occurs at the UB tips, which are major sites for the generation of new branches formed through bifurcation events (Michael and Davies, 2004; Watanabe and Costantini, 2004). After quantifying phosphorylated-histone H3 staining, we did not detect any difference between control and mutant UBs at E12.5 (data not shown). As proliferation is a very dynamic process, we also developed computational parameters to detect mitotic GFP⁺ cells in time-lapse movies. Automated detection (Fig. 5A) and quantification of the proliferation rate showed no significant difference in the first hours of kidney culture (Fig. 5B). As recently reported, UB tip cells exhibit mitosis-associated cell dispersal (MACD) (Packard et al., 2013), a process that has also been observed in other branching organs, such as the pancreas (Shih et al., 2016). During MACD, premitotic cells delaminate from the epithelium into the lumen where they divide, after which one daughter cell reinserts into the same site, while the other daughter cell reinserts at position one to three cells away (Fig. 5C) (Packard et al., 2013). Occasionally, we observed tips composed of mainly mutant cells. In this context, we found that cells can divide but did not delaminate from the epithelium, and daughter cells did not dissociate from each other (Fig. 5D). This was confirmed by automatic detection and segmentation of mitotic and daughter cells. We found a reduced distance between the position of the former parent and the two daughter cells in mutants compared with controls 25 min after division (6.6 µm and 9.3 µm, respectively), as well as a reduced distance between mutant daughter cells after mitosis compared with control cells (17 µm in controls versus 9.6 µm in mutants) (Fig. 5G,H). Moreover, in a mutant context, the majority of mutant daughter cells remained in contact 25 min after mitosis (17% of shared border in mutants compared with 1% in controls) (Fig. 5H).

Interestingly, when the neighboring cells were mainly wild type, mutant daughter cells were able to separate from each other (Fig. 5E). Quantifications in this context showed that the distances between daughter and parent cells as well as between the two daughter cells were not statistically different from that of control cells. These distances were significantly higher at 15 and 20 min after mitosis when compared with mutant cell behavior in a mutant environment (Fig. 5F,G). Our observations also suggest that proper MACD depends on cell-cell interactions with the neighboring cells. Altogether, these results indicate an early requirement for HNF1B at the tip domain for cell rearrangements and within the trunk to maintain the organization of the epithelium.

Improper epithelial organization and lack of a lumen in mutant UB trunks, led us to analyze apico-basal polarity both in vivo and ex vivo. PCK and laminin bind to the apical and basal side of the UB, respectively (Fig. 6A-A″). In some regions of the mutant UB, PCK staining was extended along the basolateral sides and there was a strong decrease in basal laminin staining (Fig. 6B-B″). Staining of adjacent sections showed that these regions were GFP⁺ (Fig. 6C). Moreover, staining with the apical polarity protein aPKC revealed that the organization of the tip domain epithelium was impaired in mutant kidneys (Fig. 6D,D′). In addition, qRT-PCR analyses showed that markers of polarized epithelium, Cdh16 (Thomson and Aronson, 1999), Pkhd1 (Wang et al., 2004) and Cys1 (Tao et al., 2009), were strongly downregulated in our mutants at E13.5, before cyst initiation (Fig. 6E). Interestingly, these genes are direct targets of HNF1B (Bai et al., 1995; Gresh et al., 2004; De Vas et al., 2015; C.H. and S.C., unpublished ChIP-seq data on E14.5 kidneys, Table S1; full details of the ChIP-seq data will be published elsewhere).

Transmission electron microscopy (TEM) analysis was performed at E15. We selected mis-shapen ureteric bud tips, which are indicative of being composed primarily of mutant cells, and found that epithelial organization was impaired in these UBs. Indeed, the basement membrane was more diffuse or even undetectable in mutants (Fig. 6F,G, insets 1 and 4). Moreover, lateral cell-cell junctions were impaired as the space between cells was greatly increased (Fig. 6F,G, inset 3 and 6). However, apical tight junctions appeared to form correctly (Fig. 6F,G, insets 2 and 5). Together, these data suggest that HNF1B is necessary to establish UB apico-basal polarity.

We have previously shown that HNF1B, although not involved in its initial induction in the intermediate mesoderm, directly controls the transcription of Pax2 in the nephric duct and UB, but not in the adjacent MM (Lokmane et al., 2010). Remarkably, in our mutants, we clearly observed that the E13.5 UB trunks, which were deficient for Hnf1b, lacked the expression of Pax2, whereas its expression was maintained in the tips that were HNF1B⁺ (Fig. 7A,B), thus confirming that HNF1B positively regulates Pax2.

qRT-PCR and in situ hybridization analyses at E13.5 showed that Ret expression was decreased in UB branches (Fig. 7C,D,G). qRT-PCR also indicated that levels of Wnt11 were partially reduced in mutant kidneys (Fig. 7G). Besides, in mutants UB tips enriched in GFP-negative cells, co-immunostaining of WNT11 and GFP showed WNT11 at the tip domain, similar to controls (Fig. 7E,F). However, few cells expressing GFP (thus, lacking HNF1B) lacked WNT11 staining (Fig. 7F′). The analysis of HNF1B staining in adjacent sections confirmed expression of HNF1B in those regions WNT11⁺/GFP⁻ (data not show). Altogether, these results indicate a strict correlation between the presence of HNF1B and WNT11, further suggesting that HNF1B is involved in the maintenance of the GDNF/Ret/Wnt11 regulatory feedback loop (Majumdar et al., 2003).

qRT-PCR analysis indicated that other components of the pathway were similarly downregulated, notably Gfrα1. There was, however, a bias in these results, given the observed recruitment of wild-type cells into the mutant tips and also the enrichment of Ret and other downstream effectors to the UB tips. We therefore performed similar analysis in Hnf1b-Sox2-Cre mutant line (Hayashi et al., 2002), in which Hnf1b is inactivated in the epiblast and consequently less mosaic, and found a stronger downregulation of Ret and other components of the pathway, notably Gfra1 and Etv5 (Fig. 7G). Our HNF1B-ChIP-seq analysis on E14.5 kidneys (C.H. and S.C., unpublished) indicated that, in addition to the already known targets Pax2 and Wnt9b (Lokmane et al., 2010), HNF1B was bound to the intronic sequences of Gfra1 and Etv5 (Gfra1: Chr19: 58470549-58472071: Etv5: chr16: 22428681-22429881) (data not shown). Interestingly, the Gfra1 and Etv5 intron sequences are conserved in mammals and contained, respectively, three and one HNF1 consensus binding sites with high motif specificity scores. We confirmed HNF1B binding to these sequences by performing ChIP-qPCR on chromatin isolated from E14.5 mouse kidneys and found that, relative to the IgG control, HNF1B was strongly enriched at both Gfra1 and Etv5 peak sequences (Fig. 7H).

We confirmed HNF1B binding to these sequences by performing ChIP-qPCR on chromatin isolated from E14.5 mouse kidneys and found that relative to the IgG control, HNF1B was strongly enriched at both Gfra1 and Etv5 peak sequences (by 178.83- and 49.95-fold, respectively) (Fig. 7H).

Altogether, these results show that HNF1B is required for Pax2 expression in the UB epithelium, and for the maintenance of key Ret/GDNF pathway components. Our results further suggest that HNF1B transcriptionally regulates Gfrα1 and Etv5 in vivo. Together these molecular misregulations explain, at least at the earlier UB branching steps, the behavior of Hnf1b-mutant cells during branching and their inability to contribute to the UB tips.

As shown by histological and OPT analyses, Hnf1b mutant kidneys exhibited a severely mispatterned ureteric tree network and developed medullar cysts by E16 (Fig. 2D′ and Fig. 3D-D″). Interestingly dilatations also occurred in the trunk regions of the UB after 3 days of culture (Fig. 4D′ and Movie 2). Immunostaining for E-cadherin (all kidney tubules) and pan-cytokeratin (CDs) at E17.5 showed that the dilated tubules were CDs (Fig. 8A-D′).

Cystogenesis in embryonic kidneys has been reported to be associated with several cellular defects, including increased proliferation, apoptosis, changes in cell-cell and cell-matrix interactions as well as cell polarity (Wilson, 2011). Investigating apicobasal polarity, we found that pan-cytokeratin staining exhibited expanded expression along the basal and lateral regions of mutant CDs, instead of the normal apical localization (Fig. 8D′). Staining with α-acetylated tubulin, a component of cilia, indicated that the percentage of cells with cilia in mutant CDs was significantly reduced compared with controls (−25%) (Fig. 8E-G). In mutant CDs, the apical marker Muc1 was interrupted in several regions of dilated ducts (Fig. 8E,F) and aPKC staining appeared also disorganized and abnormally localized (Fig. 8H,I). Interestingly, similar results were observed ex vivo. Troma 1 and laminin, which bind, respectively, to the apical and basal side of the UB trunk, showed a strong decrease in mutant tubules. These regions were specifically GFP⁺ and thus, HNF1B⁻ (Fig. S2). This suggests that the early defects of cell polarity observed in Hnf1b mutants affect cilia maintenance. Moreover, neither the water channel AQP2 nor the lectin DBA could be detected in mutants, and the transcript levels of specific markers of principal (Atp6v1) and intercalated cells (Foxi1 and pendrin) were significantly decreased (Fig. 8J-M, see also mRNA-seq data, Table S1), further indicating a global differentiation defect of the CD system. Further analysis of proliferation using the mitosis marker phosphorylated histone H3, showed an increase in the numbers of proliferating cells in the epithelium of mutant CDs when compared with control embryos. At this stage, no change was observed in the number of apoptotic cells assessed by TUNEL (Fig. 8O). These results together suggest that HNF1B acts as a key regulator of UB morphogenesis by exerting direct control over several genes involved in epithelial polarity and ensuring maintenance of tubule size during embryogenesis.

To further elucidate the cellular and molecular components regulated by HNF1B during CD morphogenesis, RNA sequencing was performed on E15.5 control and mutant kidneys, a stage at which the UB trunks were essentially devoid of HNF1B-positive cells and the CDs began to exhibit dilatations. We found that 2511 genes were differentially expressed, of which 328 were upregulated and 499 were downregulated, with a P-value (FDR or adjusted P-value) of 0.05 and an absolute fold-change cutoff value 1 in a log₂ scale. In contrast to our E13.5 qRT-PCR results, the expression of Ret and downstream effectors (Etv4/5), as well as Gfra1, were essentially not affected at this later stage. This is likely due to the fact that UB tip cells represent a minor number relative to that of the whole kidneys combined with the enrichment of wild-type cells to the tips.

Given the indirect effects upon Hnf1b deletion in the collecting system on nephron numbers and their further segment differentiation (Fig. 2), we retrieved from the total list of 499 genes downregulated in whole kidneys, those genes expressed in the UB, cortical and medullar collecting ducts using GUDMAP (www.gudmap.org) (256 genes, Table S1A), and then clustered them using ToppGene informatics (Chen et al., 2009) (Table S2). Remarkably, our RNA-seq data identified several genes differentially expressed with greater fold-changes either in the CDs, including Tmem27, Foxi1, Slc26a4 (pendrin), Slc7a8, Slc7a12, Slc44a4, Scnn1a, Kcnj16 and Aqp2, or in the medullar CDs and renal pelvis, such as Foxa1, Fxyd3, Upk3a, Upk2 and Paqr5. Unexpectedly, the previously identified targets of HNF1B, such as Cdh16, Pkhd1 and Crb3 (Verdeguer et al., 2010), Cys1 and Glis3 (De Vas et al., 2015), and Wnt9b (Lokmane et al., 2010), although downregulated, were all under the cutoff value and therefore not included in the GO analysis (Table S1B).

GO terms included transporter activity, trans-membrane transport, metabolism and lipid metabolic process (Table S2). As expected, several of the downregulated genes had HNF1B consensus binding sites and most of them were found bound by HNF1B by E14.5 kidney-ChIP-seq (C.H. and S.C., unpublished). Notably, the downregulated collecting duct genes were mainly associated with abnormal renal/urinary system physiology, kidney collecting duct epithelium and ureter morphology, as well as hydronephrosis. Of interest among the novel putative HNF1B targets identified (Table S1C) are: the transcription factor Foxa1, which is expressed in the renal and pelvic urothelium and prospective medullar CDs (Yu et al., 2012); the gene encoding the nuclear steroid hormone receptor ESRRG, which is involved in early UB branching and renal papilla development (Berry et al., 2011); Lgals3, which is implicated in ureteric bud branching and cyst growth (Bullock et al., 2001; Chiu et al., 2006) as well as the potassium channels Kcnj1 and Kcnj16, Pdzk1 and several SLC transmembrane transporters, including choline, bicarbonate anion exchanger and metal ion transporters (Table S1C) (Yan et al., 2008; Paulais et al., 2011).

Consistent with the histological and TEM results at E13.5, cellular component analyses indicated that the most significantly downregulated genes were membrane components, such as the cell-junction markers Gjb1, Cldn19 and Itgb6, which are additionally putative HNF1B targets (Table S1C). Interestingly, qRT-PCR at E13.5 showed that these genes were also downregulated at this earlier stage (data not shown). Conversely, the most significantly upregulated genes were cytoskeleton components such as Mybph or Krt5 (Table S1D).

Conditional inactivation of Hnf1b in the WD and UB derivatives, using the HoxB7cre line, uncovered a crucial role of this transcription factor in cellular organization and epithelial polarity during UB branching and CD morphogenesis. The early mosaic activity exhibited by the Hoxb7cre combined with the progressive exclusion of mutant cells from the UB tips allowed both high-resolution imaging of cell behaviors during the early steps of UB branching and the analysis of later functions of HNF1B during collecting duct maturation. We found that mutant embryos displayed severe urogenital phenotypes associated with UB branching defects and abnormal epithelium organization. Together, these data indicate an initial crucial requirement for HNF1B in the acquisition of a UB tip domain and branching, followed by a role in tubular epithelial organization and cortico-medullar organization of the ureteric tree.

In addition to proliferation, active cell movements (requiring constant turnover of cellular adhesions and cytoskeleton) have been implicated in UB morphogenesis (Chi et al., 2009a; Kuure et al., 2010). It has been shown that GDNF/Ret signaling promotes cell movements in the nephric duct that lead to the primary UB (Chi et al., 2009b). Moreover, in chimeric embryos, Ret mutant cells are excluded from the nascent tips despite the ureter being populated with them. These cell-sorting behaviors depend in turn on RET-signaling levels, as cells with higher signaling activity migrate more efficiently towards the budding sites (Riccio et al., 2016).

Using a dual fluorescent reporter system, we have been able to describe temporal and dynamic behaviors of mutant and wild-type cells during the branching process. As previously reported using diploid chimeras (Lokmane et al., 2010), we observe that cells lacking Hnf1b are progressively excluded from the UB tips, behaving as Ret, Etv4 or Fgfr2-null UB cells (Leclerc and Costantini, 2016). Interestingly, the deficiency in the contribution of Hnf1b mutant cells to the UB tips appears as severe as that reported for Ret^−/− cells and higher than that of Fgfr2^−/− (Chi et al., 2009b). Altogether, these observations suggest that HNF1B acts cell-autonomously to establish a specialized tip domain.

We found decreased expression of some GDNF/Ret signaling components in E13.5 mutants, which together may contribute to the abnormal UB cell rearrangements and defective branching phenotype observed. Notably, we showed that HNF1B is a transcriptional regulator of Gfra1 and Etv5. It is interesting to note that HNF1B acts at several levels of the GDNF/Ret cascade as Gfra1 is a Ret co-receptor and Etv5 a downstream effector. An early role of Gfra1 in UB induction has been established (Keefe Davis et al., 2013). Although these two genes have individually limited impact on branching morphogenesis after UB induction (Keefe Davis et al., 2013; Lu et al., 2009), we can hypothesize that combined downregulation of Gfra1 and Etv5 contribute to the defective maintenance of the GDNF/Ret signaling. In addition, Hnf1b-mutant cells exhibit abnormal cell-cell contacts and basal membrane in UB tips. Interestingly abnormal cell-cell contacts have also been reported in Mek1/2 mutants (Ihermann-Hella et al., 2014), further supporting the requirement of HNF1B for the maintenance of the GDNF/Ret pathway. In addition, Hnf1b mutant cells show defective apicobasal polarity and cell-cell adhesion in nascent trunks. This may suggest that HNF1B has a more-general role in epithelial cell polarity than merely regulating RET signaling. Moreover, in contrast to Ret^−/− cells, but similar to Etv4/5 mutant cells, Hnf1b mutant cells (Kuure et al., 2010) were not excluded from the common nephric duct (data not shown). We can therefore ask whether HNF1B acts either sequentially or in parallel to maintain epithelial organization and GDNF/Ret signaling.

Interestingly, in our mutants, Pax2 is virtually absent in the mutant UB branches and CDs, and previous studies have shown that decreased levels of Pax2 in the Pax2^Neu/+ heterozygous mutants is associated with reduced branching (Porteous et al., 2000). Although the precise role of Pax2 in branching has not been directly addressed in the ureteric cell lineage, it has been shown that in addition to its role in Gdnf induction in the MM, Pax2 has an intrinsic function in UB branching (Brophy et al., 2001). It has also been shown that Ret in the developing kidney is responsive to Pax2 gene dose (Clarke et al., 2006). Moreover, the analysis of a recently generated hypomorphic allele of Pax2 indicates the crucial role of this transcription factor in maintaining epithelial polarity of the nephric duct. Interestingly, the Pax2 hypomorphic phenotype exhibited, in addition to hypoplasia, duplicated ureters, cystic kidneys and fewer nephrons (Soofi et al., 2012), thus sharing several of the phenotypes observed in our mutants.

UB-tip cells display a specific proliferation mode, morphology and gene expression pattern (Schmidt-Ott et al., 2005; Yu et al., 2012). Tip cells are also progenitor cells: while one part of the pool remains at the tip to create new branches, the other part is left in the trunk to differentiate into CDs (Shakya et al., 2005). MACD may be one of the processes underlying this asymmetric cell division of progenitor cells. Analysis of our data suggested that the defective dispersal of mutant daughter cells during MACD could contribute to the UB branching defect. However, MACD contributes to relatively short-range cell movements and clonal analysis has recently shown that many cell rearrangements occur between cell divisions (Chi et al., 2009a). Moreover, although Ret signaling does promote directed cell movements in the UB tips, it apparently has no effect on MACD (Riccio et al., 2016). Although the developmental consequences of altered MACD remain to be elucidated, we can speculate that the loss of apicobasal polarity could be contributing to the cell dispersal defect observed in our mutants. Furthermore, most cells do not move as isolated entities in vivo but interact with their neighbors during migration (Ghabrial and Krasnow, 2006; Ewald et al., 2008). Our results also show the importance of the neighboring cells during MACD for cell movements, suggesting that this process relies on cell-cell interactions. It is possible that interactions among the mutant cells are impaired in the UB, thereby leading to reduced branching.

We found that the degree of mosaicism is directly linked to the severity of the phenotype both in vivo and in vitro as cultured UBs with few GFP⁻ cells did not develop any further branching after the ‘T’ shape stage. It is interesting to note that in Hnf1b-Sox2-Cre mutants and the previously described tetraploid chimeras, branching stops after the ‘T’-stage (Lokmane et al., 2010).

Owing to the mosaicism of the HoxB7-Cre line, a sufficient amount of wild-type cells was present to enable continued branching after the ‘T’-stage. OPT analysis showed that specific branching parameters were also severely affected. This defective UB architecture disrupts the cortico-medullary axis of the kidneys and affects the number and organization of developing nephrons and is very likely to be responsible for the observed perinatal lethality.

Late-stage embryos or P0 mutants exhibit severe urogenital abnormalities reminiscent of CAKUT. Interestingly, heterozygous mutations in the human HNF1B gene are the most common cause of monogenic CAKUT and account for 5 to 31% of cases, depending on the cohort examined (Raaijmakers et al., 2015). Mutant CDS also exhibit abnormal localization of apical markers and disruption of tissue architecture, leading to cystic dilatations. Our results demonstrate that HNF1B controls the expression of several genes that exhibit apico-basal polarity, such as Atp6v1b1, Slc26a4, Pdzk1, Aqp5, Hpn and Tmem27 (Table S2).

Previous studies in postnatal cystic kidneys, upon Hnf1b inactivation in medullar renal tubules, reported reduced expression of genes involved in cystogenesis, including Pkd2, Pkhd1, Nphp1, Tg737/polaris (Ift88) and Umod (Gresh et al., 2004), thus potentially linking the HNF1B transcriptional regulatory cascade to ciliogenesis. However, this expression analysis was performed in severe postnatal cystic kidneys. In our mRNA-seq analysis Pkd2, Nphp1 and Tg737/polaris were not downregulated in mutants. Moreover, other known HNF1B targets, such as Pkhd1 (Gresh et al., 2004; Hiesberger et al., 2005) and Cys1 (De Vas et al., 2015) were downregulated in our mutants, but under the cutoff value. More importantly, the cystic phenotypes of either Pkdh1 or Cys^cpk mutant mice do not phenocopy that observed in our mutants. Pkhd1 mutants had normal kidney morphology at birth and renal cysts were first detected postnatally (45 days of age) (Williams et al., 2008). Similarly, Cys^cpk mouse developed cysts only postnatally, initially in the proximal tubules and subsequently by P15 in the collecting ducts (Ricker et al., 2000). Notably, we found that CDs are either dilated or develop cysts from around E16. Similar developmental phenotypes have been observed in the context of the RCAD pathology in human fetuses (Chen et al., 2010), and increasing amounts of data report renal developmental abnormalities during gestation (Clissold et al., 2015). Thus, although reduced expression of Pkhd1 and Cys1 may contribute to some aspects of renal phenotype, the mechanisms of cyst formation are not simply explained by downregulation of these genes.

In our mutants, defects in epithelial organization and polarity precede cyst initiation. Moreover, non-dilated tubules contained apparently structurally normal cilia, thus suggesting that HNF1B does not regulate the expression of an essential structural component of cilia. However, cystic tubules contained fewer cells with primary cilia, suggesting that loss of cilia may be secondary to the loss of apicobasal polarity or due to increased proliferation.

The RNA-seq analysis shows upregulation of genes encoding cytoskeleton proteins and a downregulation of membranous genes, encoding protein components of the cell membrane and cell-cell junctions. Interestingly, similar defects have been observed in cell-cell junction mutants in other branching organs (Plosa et al., 2014; Shih et al., 2016). Remarkably, several of the most downregulated genes in mutants are associated with abnormal collecting duct epithelium, ureter morphology and pelvic urothelium differentiation, as well as hydronephrosis. Our study also shows that these deregulations can be both direct and indirect, as in our E14.5 ChIP-seq data HNF1B was found to be recruited on consensus binding sites in the promoter/enhancer regions of several genes (C.H. and S.C., unpublished; Table S1). Of interest is the identification, among the most downregulated genes, of novel HNF1B putative target genes, notably Foxa1, Esrrg, Lgals3, Kcnj1, Kcnj16, Atp6v0a4 and Cldn19, which can be associated with several of the pathogenic characteristics of the RCAD disease (Bullock et al., 2001; Chiu et al., 2006; Lee et al., 2006; Yan et al., 2008; Berry et al., 2011; Paulais et al., 2011; Norgett et al., 2012). In particular, our data uncover an unappreciated role of HNF1B in renal and pelvic urothelium differentiation, likely via the transcriptional control of Foxa1, which in turn regulates uroplakin genes (Varley et al., 2009; Yu et al., 2012). These findings are therefore expected to provide a deeper understanding of the complex syndrome RCAD and be a valuable resource to orientate further investigations.

In conclusion, our results show that HNF1B is an essential transcriptional regulator required for UB branching, morphogenesis and patterning. It seems to be required cell-autonomously for the acquisition of a specialized UB-tip domain, at least in part, through the establishment and maintenance of cell-cell adhesion, as well as Pax2 expression and the maintenance of Ret signaling. HNF1B is subsequently required in the trunk for the establishment of normal apicobasal polarity and tubular organization.

Hnf1b-null (Hnf1b^lacZ/+) mice in which the lacZ gene replaced the first exon (Barbacci et al., 1999) and HoxB7-Cre (Yu et al., 2002) or Sox2-Cre (Hayashi et al., 2002) transgenic mice were kept as heterozygotes. The Hnf1b conditional knockout (Hnf1b^Flox/Flox) carrying LoxP sites flanking exon 4 was as previously described (Heliot et al., 2013). The R26RYFP line [B6.129X1-Gt (ROSA)26Sor^tm1(EYFP)Cos/J] was from Jackson Laboratory. (ROSA)26Sor^{tm4(ACTB-tdTomato,-EGFP)Luo}/J (R26-mT/mG) reporter mice and Hnf1b^Flox/Flox mice were kept as double homozygous.

HoxB7-Cre;Hnf1b^lacZ/+ compound heterozygous males and Hnf1b^fl/fl females were crossed to generate HoxB7-Cre;Hnf1b^lacZ/fl embryos and were referred to as mutants. Hnf1b^lacZ/fl and Hnf1b^+/fl embryos that lacked the HoxB7-Cre transgene were referred to as controls, due to the lack of a phenotype in heterozygous embryos for the Hnf1b-lacZ-null allele (Barbacci et al., 1999; Kornfeld et al., 2013). HoxB7-Cre;Hnf1b^+/fl embryos were referred to as heterozygous and showed no phenotype (Fig. S1).

ROSA26^{loxP-mtomato-loxP-mEGFP};Hnf1^fl/fl compound homozygous females were crossed with HoxB7-Cre;Hnf1b^lacZ/+ males to generate either mutant embryos (HoxB7^Cre/+; Hnf1b^fl/lacZ; ROSA26^{loxP-mtomato-loxP-mEGFP}) or control embryos, lacking the Hnf1b^lacz null allele (HoxB7^Cre/+; Hnf1b^fl/+; ROSA26^{loxP-mtomato-loxP-mEGFP}). In both control and mutant embryos, the expression of the Cre recombinase deleted the mtomato gene, encoding the membrane-bound tomato protein (mT) and induced the membrane-bound EGFP.

All animal experiments described herein were approved by the Ethic commitees of the University Pierre et Marie Curie and the Ministère de l'Education Nationale, de l'Enseignement Superieur et de la Recherche (N° 04811.01).

E11.5 kidneys were micro-dissected from HoxB7Cre^+/−;Hnf1b^flox/+;R26^mT/mG or HoxB7Cre^+/−;Hnf1b^flox/lacZ;R26^mT/mG embryos and placed on Transwell plates (Costar 3450, Corning). The tissues were cultured in DMEM GlutaMAX (GibcoBRL) medium supplemented with 10% FBS, 0.5 mM HEPES and 1% penicillin/streptomycin. For imaging of the kidney cultures on the top of the filter (Costantini et al., 2011), standard six-well glass-bottomed plates were used (BD Falcon). The novel organ culture set-up (FiZD) used for time-lapse imaging is described in detail elsewhere (Saarela et al., 2017). The culture plates were inserted into an on-stage incubator on a Zeiss LSM780 confocal microscope at 37°C and 5% CO₂ (OkoLab Bold Line Temperature Controller, Gas Controller, cage enclosure and top stage incubator). Time-lapse images of the kidney explants were captured at 5 to 10 min intervals and processed with the Zen Blue program (2012, Zeiss). Huygens Professional (Scientific Volume Imaging) was used for z-stack deconvolution and pictures were prepared using Imaris (Bitplane) or Fiji (Schindelin et al., 2012).

Mitotic GFP⁺ cells are of circular shape with bright boundaries and dark centers; they were detected using multi-scale template matching. Templates were learnt from a few manually marked mitotic cells. Detections persisting at the same location in multiple consecutive frames were filtered and only the last among them was retained. The number of cells in a frame was estimated using the kidney volume and average cell size in that frame. The volume of kidney was computed by segmenting it using thresholding followed by morphological filtering and the average cell size in a frame was estimated from the mean size of detected mitotic cells. Manually annotated cell (only mitotic and daughter) markers were used to guide the segmentation. Hessian ridge enhancement (Hodneland et al., 2013) was used to remove small bright regions inside cell bodies and enhance the signal at cell membrane to fill small gaps. Cell centroids from cell segmentation were used to measure the distance of daughter cells from the site of mitosis and from each other. The proportion of shared border between a daughter pair was computed by normalizing the number of pixels where both daughter cells were in contact with the perimeter of the smallest daughter cell. If both daughters touched each other at a time point, they were counted as an in-contact pair.

Kidneys from embryos/newborns were fixed, embedded in paraffin wax, sectioned and then stained with Hematoxylin and Eosin or antibodies as described previously (Lokmane et al., 2008). We used rabbit anti-HNF1B (Santa Cruz Biotechnology, H-85, sc-22840; 1:100), chicken anti-GFP (Aves Labs, GFP-1020; 1:300) (GFP antibody was used to detect YFP as they share GFP epitope antibody recognition), biotinylated LTA (Vector Laboratory, B-1325; 1:100), rabbit anti-NKCC2 (Biosciences, SPC-401D; 1:200), rabbit anti-AQP2 (kindly provided by M. Knepper, NIH, Bethesda, USA), rabbit anti-PAX2 (Covance, PRB-276P; 1:200), rabbit anti-AQP1 (Interchim, Alpha Diagnostic, AQP11-A; 1:200), rabbit anti-aPKCz (Santa Cruz, C-20, sc-216; 1:50), biotinylated DBA (Vector Laboratory, B-1035; 1:100), mouse anti-β-catenin (BD Transduction Laboratories, 610153; 1:100), rabbit anti-phosphoHistone H3 (Upstate Cell Signaling, 06-570; 1:200), rabbit anti-WT1 (Santa Cruz Biotechnology, C-19, sc-192; 1:100), rabbit anti-Laminin (Sigma-Aldrich, L-9393; 1:100), mouse anti-E-cadherin (BD Transduction Laboratories, 610182; 1:50), mouse anti-pan-cytokeratin (Sigma-Aldrich, C1801; 1:200), Armenian hamster anti-MUC1 (Neomarkers, HM-1630-P1; 1:100), mouse anti-acetylated alpha-tubulin (Sigma-Aldrich, T6557; 1:100) and rabbit anti-Wnt11 (GeneTex, GTX105971; 1:100). Staining of kidney culture was performed after 5 days of culture. After fixation in 4% paraformaldehyde (PFA), permeabilization with PBS+0.5% Triton X-100 and blocking with 5% fetal bovine serum (FBS) in DMSO, we used rat anti-Troma-1 (developed by R. Kemler and obtained from DSHB; 1:500) and rabbit anti-Laminin (Sigma-Aldrich, L9393; 1:500) as primary antibodies. Terminal deoxynucleotidyltransferase-mediated dUTPbiotin nick end-labeling (TUNEL) was performed as described previously (Paces-Fessy et al., 2012).

In situ hybridization on paraffin sections were performed as described previously (Lokmane et al., 2008).

Whole-mount E13.5 (n=10 controls and 6 mutants) and E16 (n=6 controls and 6 mutants) kidneys were fixed in 4% paraformaldehyde (PFA) and processed for optical projection tomography (OPT) as described previously with some modifications (Chi et al., 2011). Fixed kidneys were stained with rat anti-Troma-1 (DSHB; 1:500) and anti-Ezrin antibodies (a kind gift from Monique Arpin and Sylvie Robine, Institut Curie, Paris, France) at 4°C for 2 days and then extensively washed in TBST (Tris-buffered saline with 0.01% Triton X-100 and 10% FBS). Secondary antibodies were applied at 4°C for 2 days followed by several washes in TBST. Samples were embedded in 1% low melting point agarose. Agarose blocks were placed in absolute methanol and cleared with benzyl alcohol/benzyl benzoate solution (1:2); images were captured using a 3001 OPT scanner (Bioptonics Microscopy). Pictures were prepared using Imaris (Bitplane) and morphometric parameters were calculated using Tree Surveyor Software (Short et al., 2013).

Total RNA extraction from embryonic kidneys, reverse transcription and qPCR were performed as described previously (Paces-Fessy et al., 2012). Number of kidney samples is indicated in figure legends. The following primers were used (forward/reverse): cyclophilin A, CAGGTCCTGGCATCTTGTCC/TTGCTGGTCTTGCCATTCCT; Hnf1b wild-type, GGCCTACGACCGGCAAAAGA/GGGAGACCCCTCGTTGCAAA; Ret, ACACCTTCGGACTCACTGCT/CAGGGGTGTCTCCTCTTCTG; Gfra1, CTTCGTGCTGCCACTCCTGG/CTCAGTGTGCGGTACTTGGTG; Etv5, CGTTTGATCTTGGTTGGAGG/GCTGAAGCACAAGTTCCTGA; Wnt11, GCAGTGCAACAAGACTTCCA/TCCACCACTCTGTCCGTGTA; Wnt9b, CGAGGAGATGCGAGAGTGC/GGAAGGGTGTCAGGACCTC; Pax2, GGGCATCTGCGATAATGACA/GGTCCGGATGATCCTGTTGAT; Cdh16, GTCCCCACCATCCCTTACAA/CTCCCTCCTTCAAACCGAGT; Pkhd1, TGCTCCTCAGGCAGGCAATCG/ACCTGTACCCTGGGGTGGCTT; Cys1, AGAGGAGCTCATGGCGAGCATT/GCCTGTGGCACAGATGCCAAGA; Foxi1, TCCCATGGCTACTGAGGTTG/CCTCTCCACCATGACAGCAT; Atp6v1, CCCTGAAGTCCCTTCAAACA/TGGACCAGGTCAAGTTTGC; and pendrin, GGAACTGCAGCTAGTAGGGC/CCCAAATACCGAGTCAAGGA.

Chromatin immunoprecipitation (ChIP) was performed on E14.5 kidneys using the rabbit HNF1B antibody (H85, Santa Cruz) as described by Heliot and Cereghini (2012). Each ChIP was validated by assaying the two well-known HNF1B targets genes Ksp-cadherin and Wnt9 (first intron) and a negative control (Heliot and Cereghini, 2012; Heliot et al., 2013; Lokmane et al., 2010) (data not shown). The following primers were used (forward/reverse): Gfra1, GAACCCCTGTCTTGGGCCT/AACCCGCATGGCTCTGTG; and Etv5, TTTTTCCACAGGCTGCACAAT/GGGCCCACTGGCAGACTT.

E15 control (n=2) and mutant (n=2) kidneys were dissected cut into quarters and fixed in fresh 1% glutaraldehyde/4% PFA in 0.1 M PBS. Sample processing was carried out using the Tecnai GS Spirit Bio Twin microscope (FEI Europe) and the images were captured with a Quemesa CCD camera controlled by the iTEM software (Olympus Soft Imaging Solutions).

RNA was extracted from E15.5 kidneys, two controls and two mutant embryos (four kidneys of each genotype in total) from the same litter. The raw bam files, one for each RNA-seq sample, were summarized to a gene read counts table, using the Bioconductor open source software package GenomicRanges. In the final read counts table, each row represents one gene; each column represents one RNA-seq sample; and each cell represents the corresponding read counts associated with each row and column. The gene counts table was normalized for inherent systematic or experimental biases (e.g. sequencing depth, gene length, GC content bias) using the Bioconductor package edgeR after removing genes that had zero counts over all RNA-seq samples (Moulos and Hatzis, 2015). Functional enrichment gene lists were established with ToppGene suite (Chen et al., 2009).

For comparison of several means, normality of the variables was checked and we performed Student's t-tests with a confidence interval of 95%. *P<0.05; **P<0.01 ***P<0.001; error bars indicate s.e.m.

We thank Marie Breau and Daniel Darby for critical reading of the manuscript; Paula Haipus for technical assistance in kidney organ culture; and Eduard Manzoni and Edwige Declercq for animal care. We thank Pantelis Hatzis and Martin Reczko (Alexander Fleming Institute, Genomics Facility, Greece) for mRNA sequencing. We also thank Sylvie Schneider-Maunoury (UMR7622 CNRS-UPMC) for support to complete this study. The anti-Troma-1 antibody was obtained from the Developmental Studies Hybridoma Bank (DSHB), which was created by the NICHD of the NIH at The University of Iowa, Department of Biology, Iowa, USA.