Resolving the heterogeneity of diaphragmatic mesenchyme: a novel mouse model of congenital diaphragmatic hernia

Congenital diaphragmatic hernia (CDH) is a severe developmental defect that affects approximately one in 3000 live births (Greer, 2013) and can have devastating clinical outcomes. Despite advances in surgical repair and neonatal care, the mortality remains high. There is an urgent need to develop new therapies by gaining a better understanding of the pathophysiology of CDH (Reiss et al., 2010). The aetiology of CDH is not well understood owing to the complexity of diaphragm formation (Kardon et al., 2017). The mature diaphragm is composed of several tissues, including connective tissue, muscle progenitors, muscle, tendons, nerves, blood vessels, lymphatics and mesothelium (Ackerman and Greer, 2007). Several embryonic structures are implicated in diaphragm development: the pleuroperitoneal folds (PPFs), the septum transversum (ST), the posthepatic mesenchymal plate (PHMP) and the somites (Carmona et al., 2016; Merrell and Kardon, 2013; Sefton et al., 2018). The ST is a thin layer of mesodermal cells overlying the liver and is formed at approximately embryonic day (E) 8.5 in the mouse. This is followed by the formation of the PPFs, which develop at E10.5-E12.5. A study demonstrated that the fibroblasts of the PPFs play a crucial role in guiding the expansion and movement of the myogenic cells that originate in the somites and are essential for the correct formation of the muscular component of the diaphragm (Merrell et al., 2015). The PHMP first appears at ∼E10.5 and is thought to be derived from the mesenchymal population of the lateral ST (Carmona et al., 2016; Iritani, 1984). The cells of PPFs fuse with the PHMP to form a membranous continuum, separating the thoracic and peritoneal cavities. Subsequently, myoblasts migrate from the somites, leading to the muscularisation of this membrane to form the mature diaphragm.

Several genes have been shown to be involved in normal diaphragm development and are associated with human CDH (Kardon et al., 2017). One of these is WT1 transcription factor (WT1). Mutations in WT1 have been described in patients with CDH (Antonius et al., 2008; Holder et al., 2007; Scott et al., 2005; Suri et al., 2007). In the mouse, Wilms tumor 1 homolog (WT1) was first described as a transcriptional regulator with a large array of target genes, but it also has RNA-editing functions (Bharathavikru et al., 2017; Hastie, 2017; Toska and Roberts, 2014). Wt1 plays an essential role in the development of multiple organs (Kreidberg et al., 1993; Moore et al., 1999) and is also crucial for maintaining adult tissue homeostasis (Chau et al., 2011). Homozygous Wt1 null mouse embryos die at ∼E13.5 and have diaphragmatic hernias (Clugston and Greer, 2007; Kreidberg et al., 1993). During diaphragm development in the mouse, Wt1 is expressed in the PPFs, PHMP, ST, mesothelium and lateral wall body mesenchyme (Carmona et al., 2016; Paris et al., 2016). Given the wide expression of Wt1 in structures that are involved in diaphragm development, a tissue-specific approach is essential for delineating the role of Wt1 and the role of the cells that express Wt1 in the underlying pathophysiology of CDH.

Mesenchymal cells are present throughout the diaphragm, but their origins and cell types are not well defined or understood. One mesenchymal cell population, the connective tissue fibroblasts, for which GATA binding protein 4 (GATA4) and transcription factor 4 (TCF4) are the best markers (Merrell et al., 2015; Paris et al., 2016), is crucial for guiding the migration of myoblasts during diaphragm development, as shown by the conditional deletion of Gata4 using the Prx1-Cre mouse model (Merrell et al., 2015). The TCF4/GATA4-expressing connective tissue fibroblast population does not overlap substantially with the WT1-expressing non-muscle mesenchyme in the diaphragm (Paris et al., 2016), suggesting that they are distinct cell populations. Furthermore, it has been shown previously that WT1⁺ mesenchymal cells largely corresponded to TWIST1, but that still explains only at most a 70% co-expression in an E13.5 diaphragm, providing further evidence of the complexity and heterogeneity of the mesenchymal cells that make up diaphragm (Paris et al., 2016).

To delineate the heterogeneity of the ill-defined mesenchymal cells in the diaphragm, we generated a mouse model in which Wt1 was deleted conditionally in the Prx1-Cre lineage. In this model, mutant embryos can survive in utero but die shortly after birth, which we believe is attributable to the formation of diaphragmatic hernias. In addition to the CDH phenotype, we show that the developmental origin(s) of the non-muscle mesenchymal cells in the PPF is different from those in the PHMP. Moreover, we show data providing cellular insights into the roles of PPF mesenchymal cells during the formation of diaphragm.

In our model, male Prx1^Cre/+;Wt1^loxp/+ mice were crossed with female Wt1^loxp/loxp mice to inactivate Wt1 conditionally using Prx1-Cre. No live mutants (Prx1^Cre/+;Wt1^loxp/loxp) were present when the litters were genotyped at ∼2-3 weeks of age (n>50). Given the important role of Wt1 in regulating key developmental processes (Chau and Hastie, 2012), we suspected that the phenotypes of the mutants probably resulted in embryonic lethality. However, mutant embryos appeared to be grossly normal (externally) at all stages analysed (E11.5, E12.5, E14.5, E16.5, E18.5 and E19.5). The number of mutant embryos obtained at each stage is summarised in Table S1. When the pregnant dams were left to give birth, it was apparent that mutant pups were born alive but died within a few hours. Obtaining mutant mice that survived until birth led us to hypothesise that their death might have been caused by an inability to breathe. Diaphragmatic defects typically result in disrupted breathing (Greer, 2013). As mentioned previously, Wt1 null mouse embryos also develop diaphragmatic hernias (Kreidberg et al., 1993). Therefore, we hypothesised that the Prx1^Cre/+;Wt1^loxp/loxp embryos might have diaphragmatic hernias.

We analysed deceased [postnatal day (P) 0] and E19.5 mutant embryos and found large holes in their diaphragms (Fig. 1A-G). Younger mutant embryos (E14.5 and E16.5) were also found to have diaphragmatic holes (Fig. 1H-J and K-R, respectively), often accompanied by liver herniation into the thoracic cavity (Fig. 1I,J,L,M). Between 80% and 90% of CDH is Bochdalek type, characterised by hernias in the posterolateral region of the diaphragm. In more than 85% of cases, Bochdalek hernias are left sided (Longoni et al., 2006). A description of the phenotypes of the Prx1^Cre/+;Wt1^loxp/loxp embryos at different stages is summarised in Table S2, with the left-sided hernia being the most common defect (45%). Twenty percent of the mutants had bilateral hernias, and 5% had holes on the right side only. Finally, 30% of the mutants either did not exhibit any obvious diaphragmatic phenotype or did not have fully formed holes, but a thinning of the diaphragm was observed.

The initial separation of the thoracic and peritoneal cavities is established by the PPFs, which are typically formed between E10.5 and E12.5 (Carmona et al., 2016; Merrell et al., 2015). By E12.5, a continuous band of cells has developed, completely separating the two cavities. Haematoxylin and Eosin (H&E)-stained sections of E11.5 and E12.5 mutant embryos (Fig. 2A-D′, sectioned transversely) showed that this continuous band of cells failed to form correctly (Fig. 2B,B′,D,D′). Large gaps between the PPFs and the PMPH were observed in the mutants (Fig. 2B,B′,D,D′), leading to the failure of separation of the two cavities.

To understand the cause of this defect, we first needed to know the location of Prx1-Cre expression (i.e. in which cells Wt1 was being deleted). Immunofluorescence was performed on sections from Prx1^Cre/+;R26R^mTmG/+ lineage-tracing embryos using anti-green fluorescent protein (anti-GFP) and anti-WT1 antibodies. In this model, cells in which Prx1-Cre is or has been expressed are GFP⁺. As shown in Fig. 3, WT1 expression was detected in the PHMP and PPFs (indicated in red). Although the cells of the PPFs expressed GFP and were thus Prx1-Cre lineage positive (indicated in green), no expression of GFP was detected in the cells of the PHMP (Fig. 3A,A′). This was particularly obvious in the E11.5 embryos. At E11.5, the PPFs and PMHPs have begun to fuse, with boundaries still clearly visible between the two structures. This boundary is distinctly marked by the clear domains of GFP-expressing cells. To demonstrate this better, consecutive sections in this region were obtained and stained (Fig. 3A,B, where B is anterior to A). In Fig. 3A′,A″, the PPFs (GFP⁺) and PHMPs (GFP⁻) can be seen abutting one another, whereas in Fig. 3B′,B″, it is clear that the PPFs (GFP⁺) are merging with the PHMPs (GFP⁻). At E12.5, once the continuous band has been formed, the population of GFP⁺ cells (PPF derived) infiltrating the PHMP is still distinctive (Fig. 3C).

Like other visceral organs, the diaphragm is lined by a mesothelial layer. The mesothelial cells of the diaphragm express endogenous WT1 (as indicated by the yellow arrows in Fig. 3C′); however, this mesothelial layer does not express GFP and therefore is not derived from the Prx1-Cre lineage and thus not from the PPFs. Using this model, we have demonstrated that only the PPFs arise from Prx1-Cre-expressing cells and not the PHMP. Therefore, despite WT1 being expressed in both the PPFs and PHMP, it will be deleted only in the cells of the PPFs in our model.

To illustrate this better, a second mouse model was generated. Prx1^Cre/+;Wt1^GFP/+ males were crossed with Wt1^loxP/loxP females. In Prx1^Cre/+;Wt1^loxP/GFP offspring, one copy of the Wt1 allele is flanked by loxP sites, whereas the other copy has a GFP knock-in at exon 1, which disrupts function. Therefore, WT1-expressing cells will be GFP⁺. Moreover, upon Cre-mediated loxP recombination (driven by Prx1-Cre), the second copy of Wt1 is conditionally deleted. The mutant cells remain GFP⁺ despite no functional WT1 being present. Staining with anti-GFP and anti-WT1 antibodies reveals cells that have once expressed WT1 but no longer do so. Such cells will be positive for GFP but negative for endogenous WT1. This is illustrated in Fig. 4 (E11.5, E12.5 and E13.5), where endogenous WT1 expression (red) was detected in PPFs and PHMP in the control embryos (Fig. 4A,A′,C), but its expression was completely lacking in the PPFs and unaffected in the PHMP of the mutant embryos (Fig. 4B,B′,D,F). GFP expression was observed in the disrupted PPFs, thus confirming that WT1 would normally be expressed in these cells (Fig. 4B,B′,D-F). The presence of these GFP⁺ cells also suggests that Wt1 deletion in these cells did not lead to cell loss. Intriguingly, the GFP signal in the PPFs of the mutant embryos was much stronger than that in the control littermates (Fig. 4). It is plausible that this was a consequence of an accumulation of cells in the PPFs owing to defective cell movement. Identifying PPF and PHMP structures is not trivial, because sections have to be obtained at precisely the same level from control and mutant embryos. This model provides a reassuring way of identifying the regions in which Wt1 has been deleted, which is convenient for subsequent studies of pathways that are potentially disrupted.

Previously, we have shown that Wt1 plays a major role in regulating the epithelial and mesenchymal cell states in several mesodermal tissues (Martínez-Estrada et al., 2010). To test the hypothesis that the formation of the continuous band of cells is disrupted owing to defects in PPF cell migration, we analysed the expression levels of markers for epithelial-to-mesenchymal transition (EMT). We stained sections from control (Prx1^+/+;Wt1^GFP/loxP) and mutant (Prx1^Cre/+;Wt1^GFP/loxP) embryos with an anti-E-cadherin (CDH1; an epithelial marker) antibody and an anti-vimentin (VIM; a mesenchymal marker) antibody. Sections were co-stained with an anti-WT1 antibody or an anti-GFP antibody to demonstrate deletion of Wt1 and regions of PPFs and PHMPs. Despite strong expression of CDH1 in the lung bud (a positive control), no expression of CDH1 was detected in the PPFs of the mutant or control embryos (Fig. S1C-D′; CDH1 is indicated in green and GFP in red). Therefore, we did not observe any change in CDH1 expression in the PPFs between the control and mutant embryos. We also compared expression of VIM between control and mutant embryos (VIM is indicated in green and WT1 in red), but no changes were observed (Fig. S1A-B′). Next, we investigated whether cell proliferation within the mutant PPFs was reduced. We co-stained sections with anti-Ki67 (also known as Mki67) and anti-GFP antibodies (Fig. S2; Ki67 is indicated in red and GFP in green). Quantification of the percentage of cells that are positive for Ki67 expression suggests no significant differences between control and mutant embryos (n=3 animals for each genotype, Mann–Whitney U-test, two tailed, P-value>0.5). Ki67 marks nuclei in all active phases of the cell cycle. Inclusion of a more specific mitotic marker, such as phosphohistone H3, can provide a more reliable indication about whether the lack of a continuous band of cells is caused by a disruption in PPF cell proliferation.

The retinoic acid (RA) signalling pathway plays a key role in development of the diaphragm. A common mouse model of CDH uses nitrofen to inhibit RA synthesis (Greer et al., 2003; Mey et al., 2003). We analysed the expression of RALDH2 (also known as ALDH1A2), which catalyses the formation of RA from retinaldehyde. Importantly, WT1 has been shown to activate Raldh2 transcriptionally in epicardial cells (Guadix et al., 2011). Surprisingly, we observed an increase in RALDH2 expression in the PPFs of the mutant embryos (Fig. S3B,B′). The increase in RALDH2 expression in the mutant rudiments is likely to be attributable to individual cells expressing higher levels of RALDH2. Mutant cells in the affected area expressed higher levels of RALDH2 per cell (n=4 animals) compared with the control cells (n=3 animals, Mann–Whitney U-test, two tailed P-value=0.05; Fig. S3E), quantified by measurement of the RALDH2 immunofluorescence intensity.

To understand plausible molecular pathways that might be underlying the defects in our model, we generated a mouse model to allow the precise isolation of cells in the diaphragmatic region in which Wt1 has been deleted (Prx1^Cre/+;R26R^tdRFP/+;Wt1^loxP/GFP). In this model, cells that are derived from the Prx1-Cre lineage and have had Wt1 deleted will express both GFP and red fluorescent protein (RFP). We used fluorescence-activated cell sorting (FACS) to sort these cells from the diaphragmatic region of E11.5 embryos. Control cells were sorted from the same region isolated from a comparable model, differing only by the absence of loxP sites flanking one of the Wt1 alleles (Prx1^Cre/+;R26R^tdRFP/+;Wt1^+/GFP), therefore avoiding the deletion of Wt1 in the Prx1-Cre lineage. We measured expression levels of known EMT markers in these cells, including snail family zinc finger 1 (Snai1) and 2 (Snai2). Fig. S4 shows that the levels of these EMT markers are not altered, which is consistent with the CDH1 and VIM results shown in Fig. S1.

Defects in the PPFs have been suggested to be the cause of CDH in several mouse models (Merrell et al., 2015). The PPFs have been shown to give rise to the central tendon and the muscle connective tissue fibroblasts, characterised by TCF4 and GATA4 expression (Merrell et al., 2015). In addition, work performed by Paris et al. (2016) shows that WT1-expressing cells in the PPFs are non-muscle mesenchymal cells, and the majority of them do not express TCF4. This is supported by our data showing that the majority of WT1⁺ cells (indicated by GFP staining, green) do not express TCF4 (indicated in red; Fig. 5A′). Quantification shows that 6.7% of WT1⁺ cells express TCF4 (s.d.=0.9%, n=3 animals). Together, these results suggest that the TCF4-expressing cells (muscle connective tissue fibroblasts) and the WT1-expressing cells in the PPFs are likely to be two distinct cell types. In addition to the central tendon and the muscle connective tissue fibroblasts, our data suggest that the PPFs also give rise to the non-muscle mesenchyme (marked by WT1).

As PPFs expand, connective tissue fibroblasts (TCF4⁺ and GATA4⁺) guide migration of myoblasts. We examined the expression of TCF4 in our model, in which Wt1 is deleted in the non-muscle mesenchymal cells by Prx1-Cre. Unexpectedly, TCF4 expression (indicated in red) was almost completely absent in the mutant PPFs at both E12.5 and E13.5 (Fig. 5A-E). Single-channel images representing GFP expression are included in Fig. S5. Despite dissimilar PPF/PHMP appearance, the sections were taken at comparable levels between control and mutant embryos using their positions relative to the lung, liver and oesophagus for guidance. The positioning of PPF/PHMP relative to other organs is often used where 3D imaging of embryos is not readily accessible (Carmona et al., 2016; Coles and Ackerman, 2013; Domyan et al., 2013; Paris et al., 2016; Wat et al., 2012). In mutant embryos, defective PPFs appear as residual rudiments, whereas PPF and PHMP have already formed a continual band in control embryos (E12.5 and E13.5). This absence of TCF4⁺ cells in mutant PPF rudiments was not expected, because WT1 is not expressed in the muscle connective tissue fibroblasts in PPFs (Paris et al., 2016), suggesting plausible interactions between these two populations of cells.

In addition to labelling muscle connective tissue fibroblasts, TCF4 is a transcription factor that binds to B-catenin (CTNNB1) (Korinek et al., 1997) in developing diaphragm (Paris et al., 2016). The Wnt/CTNNB1 signalling pathway is also known to act downstream of WT1 (Kim et al., 2009). The reduction in TCF4 expression in the mutant PPFs in our model could therefore be attributable to the disruption of the Wnt/CTNNB1 pathway as a result of reduced Wt1 expression. Alternatively, the disappearance/absence of muscle connective tissue fibroblasts, which are marked by TCF4, could also explain the reduction. To test the second possibility, sections were stained with an anti-GATA4 antibody, which also marks muscle connective tissue fibroblasts (Merrell et al., 2015). At E12.5, GATA4 expression was found at high levels in cells of the PHMP and in the continuous band of cells separating the cavities (Fig. 6A,A′). The expression was particularly strong in the mesothelium of the PPFs but much weaker (if not absent) in the mesenchymal/fibroblast cells within the PPFs (Fig. 6A,A′). As with the TCF4 staining, no GATA4-expressing cells were found in the PPFs in which Wt1 was deleted (Fig. 6B-C′; E12.5 embryos from the Prx1^Cre/+;Wt1^loxP/GFP model, where GATA4 is indicated in red and WT1 and/or GFP in green). We also checked GATA4 expression at E13.5. Likewise, no GATA4 expression was detected in the rudiment of the PPFs in the mutant embryos (Fig. 6E-F′) compared with clear expression in the controls (Fig. 6D,D′), suggesting an absence of muscle connective tissue fibroblasts in the mutant embryos. Single-channel images representing WT1 and GFP expression are shown in Fig. S6.

The GATA4- and TCF4-expressing muscle connective tissue fibroblasts have been shown to play a crucial role in guiding the migration of myoblasts (Merrell et al., 2015). Myogenic progenitors begin to migrate from the somites to the diaphragm at around E12.5 (Merrell and Kardon, 2013). To test whether the myoblasts were affected in our model, we stained sections with an anti-MYOD (also known as MYOD1) antibody. In the control E12.5 diaphragm, MYOD⁺ cells were clearly visible in the PPF/PHMP continuum (indicated in red; Fig. 7A), whereas there was a complete absence of MYOD⁺ cells in the mutant PPF rudiments (Fig. 7B-C′). MYOD⁺ cells were also absent in the mutant PHMPs, because PPFs failed to reach PHMPs to form the PPF/PHMP continuum. The myoblasts were present in the lateral body wall, suggesting that their formation was not affected (Fig. 7B,C; myoblasts in lateral body wall are circled). In the mutant diaphragm at E16.5, there was muscle formation. Additionally, a thickening of the diaphragm was observed at the edges of the hernia, consistent with previous reports of disrupted migration in other models of CDH (Clugston et al., 2006). Sections were stained with an anti-MF20 antibody, which detects the heavy chain of myosin II, indicated in green, and WT1 is indicated in red (Fig. 7D-E′). Single-channel images are included in Fig. S7. WT1 expression was lost in the mesenchymal cells of the diaphragm (white asterisk) but not in the mesothelial cells (white arrows). These data suggest that deletion of Wt1 in the non-muscle mesenchyme results in an absence of muscle connective tissue fibroblasts owing to unknown mechanisms and a failure of myoblast migration to the PPFs, ultimately leading to disrupted diaphragm formation.

CDH is a complex condition, and many aspects of this disease are poorly defined, particularly the underlying causal mechanisms. The formation of diaphragm is a delicate process that involves multiple cell types arising from several regions (somites, lateral plate mesoderm, neurons and ST). Different cell types interact with one another (for example, muscle connective tissue fibroblasts guide the migration of the myogenic progenitors; Merrell et al., 2015), and crosstalk between signalling pathways is common. The complexity of diaphragm development no doubt contributes to the complex aetiology of CDH. Here, we use transgenic mouse models to dissect the role of one crucial gene, Wt1, in the mesenchymal cells of the PPFs.

Initially, we demonstrate the heterogeneous nature of the mesenchymal cell populations within the PPFs. As described previously, Prx1-Cre (expressed in the lateral plate mesoderm during development) gives rise to muscle connective tissue fibroblasts, marked by TCF4 and GATA4 (Merrell et al., 2015). In agreement with a previous study (Paris et al., 2016), we show that most of the TCF4⁺ (and GATA4⁺) cells in the PPFs do not express WT1. This suggests that in addition to the muscle connective tissue fibroblasts (Merrell et al., 2015), Prx1-Cre-expressing cells give rise to another cell type in the developing diaphragm, with mesenchymal properties (marked by WT1). We next show that these WT1-expressing PPF cells (from the Prx1-Cre lineage) have the ability to expand and form a continuous band of cells with PHMP, ultimately sealing off the thoracic and peritoneal cavities. It is interesting to note that in the CDH model in which Gata4 is deleted in the muscle connective tissue fibroblasts using Prx1-Cre (Prx1Cre^Tg/+;Gata4^∇/flx;Rosa26^LacZ/+), B-gal⁺ sacs are present covering herniated regions, and not as holes in this tissue (Merrell et al., 2015). Given that our data demonstrate that cells in the Prx1-Cre-expressing lineage can give rise to mesenchymal cells that are able to expand, it is likely that these mesenchymal cells might contribute to the thin membranous sac formed in the model described by Merrell et al. (2015). However, in our model, in which Wt1 is deleted in mesenchymal cells of the PPFs using Prx1-Cre, we observe a failure of closure of the body cavities at early stages and holes/hernia at later stages. Together, these results suggest that the WT1-expressing mesenchymal cells of the PPFs might be one driving force behind expansion of PPFs towards the PHMP.

Moreover, our data suggest that cells of the PHMP might also move towards the PPFs. In the diaphragms of E12.5 embryos (at which point the continuous PPF/PHMP membrane is almost complete), GATA4 expression is high in the cells of the PHMP, and most of these GATA4⁺ cells co-express WT1. Our staining suggests that the GATA4⁺ cells in the PHMP migrate towards the PPFs, because most of the cells in the PPFs do not express GATA4. This forms a distinct boundary at the point at which the PPFs (GATA4⁻) and PHMP (GATA4⁺) meet. A similar staining pattern has been described in a recent study in which G2-Gata4-Cre was used to delete Wt1 (Carmona et al., 2016).

Connective tissue fibroblasts (indicated in yellow in the cartoon illustration; Fig. 8) are absent from the PPFs in our model, in which Wt1 is deleted in the non-muscle mesenchymal cells (indicated in green in the cartoon illustration; Fig. 8) using Prx1-Cre. Interactions between two cell types are common. For example, in skeletal muscle, ablation of muscle satellite cells (PAX7⁺) severely reduces the expansion of muscle connective tissue fibroblasts (TCF4⁺) during regeneration (Mathew et al., 2011). The molecular mechanisms that might govern the interactions between these cell types during development of the diaphragm remain incompletely understood (cartoon illustration; Fig. 8).

In addition to connective tissue fibroblasts (yellow; Fig. 8) and non-muscle mesenchymal cells (green; Fig. 8), there is an additional player in diaphragm development, the myoblasts (brown; Fig. 8). Absence of connective fibroblasts has previously been shown to cause defects in guiding the myogenic progenitors towards PPFs (Merrell et al., 2015). Staining with a muscle progenitor marker, MYOD, confirmed that this is also the case in our model. It should be noted that muscle fibres are present in the mutant diaphragm at later stages (e.g. E16.5), with possible disrupted orientation. In addition, shorter and thicker segments of diaphragm are present adjacent to the hernia in our mutant embryos at later stages. It is possible that muscle progenitors still retain some ability to move into the PPFs (or ‘ride on’ other moving cells as the embryo grows and expands in size), albeit less effectively, in the absence of the guiding connective tissue fibroblasts.

These data led us to investigate the underlying molecular mechanisms of the paused PPFs. Defects in EMT would be a plausible cause; however, we do not see changes in EMT markers at the protein (e.g. CDH1 and VIM by immunostaining) or mRNA level (e.g. Snai1 and Snai2 by quantitative PCR). An elegant study showed that conditional deletion of Wt1 using a G2-Gata4-Cre results in the mice developing diaphragmatic hernias. Defects in EMT (as indicated in their study by an increase in CDH1, an epithelial cell marker) were suggested to be the underlying cause (Carmona et al., 2016). The Cre activity of the G2-Gata4-Cre is widely detected in lateral plate mesoderm at E7.75, but is then restricted in the ST and in visceral mesoderm at E9.5 (Rojas et al., 2005). At E10.5 and E11.5, the Cre activity of G2-Gata4-Cre is detected strongly in PHMP, with hardly any in the PPFs (Carmona et al., 2016; Rojas et al., 2005). In our model, we show that cells of the Prx1-Cre lineage are retained in PPFs and start to move from PPFs towards the PHMP. The difference between these results in EMT might be attributable to the different mouse models used; here, we delete Wt1 in PPFs using Prx1-Cre, whereas the model described by Carmona et al. (2016) resulted in the deletion of Wt1 predominantly in the PHMP. It is plausible that the cells that express Wt1 in the PPFs differ from those that reside in the PHMP, because the two structures are derived from different regions during development.

Interestingly, we observed an increase in RALDH2 expression in the mutant PPFs deleted for Wt1 in our model. We have shown previously that Wt1 regulates the RA pathway by direct transcriptional activation of Raldh2 in the epicardium (Guadix et al., 2011). This unexpected increase in RALDH2 in diaphragms with Wt1 deleted has also been described in work performed by Carmona et al. (2016). In their model, the authors reasoned that the increase in RALDH2 was attributable to the intermediate mesoderm of the renal ridges persisting in the PPFs. In situ hybridization of Pax2 (a marker of the intermediate mesoderm) provided evidence for their argument. However, no PAX2 expression was observed in the PPFs in our model (using an anti-PAX2 antibody; Fig. S3C). Clear PAX2 staining was observed in the neural tube, suggesting that the antibody worked well (Fig. S3D). Again, the difference between these results might be attributable to Wt1 deletion using different Cre lines, hence affecting different regions or structures. Furthermore, RA is well known for its role in regulating the differentiation status of cells (Niederreither and Dollé, 2008), and RA levels have been shown to regulate the migration of cells during heart development (Wang et al., 2018). The elevated expression of RALDH2 seen in the mutant PPF rudiments where Wt1 has been deleted could lead to changes in the differentiation status of cells in PPFs, and thereby affect their ability to expand.

Wt1 plays crucial roles in diaphragm development, and its importance has been demonstrated in previous studies using mouse models. Wt1 is expressed in several structures that are involved in diaphragm formation, including ST, PHMP, mesothelium, PPF and lateral body wall mesenchyme. Our model complements other studies by delineating the role of Wt1 and the WT1⁺ mesenchymal cells by deleting Wt1 in the PPFs and adjacent lateral body wall mesenchyme using Prx1-Cre. An early study, in 1993, in which Wt1 was inactivated globally, resulted in embryonic lethality and multiple developmental defects, including defects in diaphragm formation (Kreidberg et al., 1993). This global Wt1 knockout model was subsequently reanalysed and reported to have left-sided posterolateral defects (Clugston et al., 2006). A study performed by Paris et al. (2016) focused on the role of mesothelium in diaphragm development. Their Wt1 loss-of-function experiments were carried out using homozygous Wt1^{GFPCre/GFPCre} embryos (i.e. a global Wt1 knockout). In our model, despite the strong expression of Wt1 in mesothelium, we have shown clearly that there is no Cre activity detected in mesothelium using Prx1-Cre. In other words, mesothelium is not targeted in our model. As mentioned above, a conditional deletion of Wt1 using G2-Gata4-Cre leads to Wt1 deletion in ST and PHMP (Carmona et al., 2016; Rojas et al., 2005). Wt1 deletion by G2-Gata4-Cre also leads to severe impairments of coronary vascular development and causes embryonic lethality around E15.5 (Cano et al., 2016). Interestingly, all the Wt1 CDH models described above result in embryonic lethality, whereas our mice can survive until birth, and die shortly after owing to breathing difficulties. A left-sided dorsal defect is most frequent in our model (45%), followed by bilateral defects (20%) and thinning or no obvious phenotypes (15% each). The thinning of the diaphragm observed in some of our mutant embryos resembles diaphragmatic eventration, which is a rare condition that usually occurs as a result of defects/delays in the migration of myoblasts, leading to a structural deficiency of diaphragmatic muscle (Groth and Andrade, 2009).

About 25% of individuals with CDH also have abnormalities of one or more major body systems, and 50-60% of CDH cases are isolated, which means that affected individuals have no other major malformations. Gross examination of our mutants with Wt1 deleted did not reveal any major malformations, but it is possible that some abnormalities might not be obvious until the pups are older. Mortality rates in infants with CDH remain high despite advances in surgical repair. Pulmonary complication is a major cause of CDH mortality. Herniated viscera compress the lung and lead to pulmonary hypoplasia. However, abnormal lung development can precede the herniation of abdominal content into the thoracic cavity, suggesting that it can be a primary rather than a secondary defect (Donahoe et al., 2016). Lung hypoplasia has been observed in the mouse model where Wt1 is inactivated globally (Kreidberg et al., 1993). Relating to our model, the Prx1-Cre transgene has been shown to label some cells in the lungs (Merrell et al., 2015); however, the exact cell type is not clear. By contrast, Wt1-expressing cells have been shown to contribute to pulmonary endothelial and smooth muscle cells, bronchial musculature, tracheal and bronchial cartilage, and CD34⁺ fibroblast-like interstitial cells (Cano et al., 2013). Despite the compelling evidence suggesting that our model is also likely to have a lung phenotype, we cannot be sure without a detailed analysis.

Other components that are also crucial in diaphragm development are blood vessels and the nervous system (Merrell and Kardon, 2013), but these have not been analysed in our model. Furthermore, with advances in single-cell RNA-seq technique, analysis of diaphragm development using this approach will provide further insight into the cellular heterogeneity of the diaphragm, in addition to the pathways, regulatory mechanisms and possible interactions between different cell types during diaphragm development.

In summary, we illustrate the heterogeneous nature of the cell population that forms the PPFs, using several new mouse models. We show that the conditional deletion of Wt1 in the non-muscle mesenchyme using Prx1-Cre results in diaphragmatic hernias, and we illustrate that the mesenchymal cells of the PPFs are capable of expanding towards the PHMP. Connective tissue fibroblasts are absent in the PPFs in which the non-muscle mesenchyme is deleted for Wt1, suggesting possible interactions between these two cell types. This absence of connective tissue fibroblasts could be a plausible explanation for the evident disruption to the migration; however, more work is required to delineate the exact underlying molecular mechanism.

Animals used in this study were housed at the animal facilities at the University of Edinburgh, with procedures performed under Personal and Project Home Office Licences. Targeted deletion of Wt1 (Prx1^Cre/+;Wt1^loxp/loxp) was achieved by crossing homozygous mice (Martínez-Estrada et al., 2010) carrying a loxP-flanked Wt1 allele into a mouse strain of Cre recombinase expression driven by the Prx1 transgene (Logan et al., 2002). The Wt1-GFP mouse line (Wt1^GFP/+) used in this study was made by Hosen et al. (2007). GFP is a knock-in at the first exon of Wt1 and is expressed under the endogenous transcriptional regulatory elements of Wt1. The Prx1-Cre lineage-tracing model (Prx1^Cre/+;R26R^mTmG/+) was made by crossing male Prx1-Cre mice (Logan et al., 2002) with female R26R^mTmG/mTmG double fluorescence reporter mice (Muzumdar et al., 2007). Cre expression, following Prx1-derived enhancer element, mediates a switch of Tomato to eGFP expression. The model to obtain mutant diaphragmatic cells that are expressing WT1 and have been conditionally deleted for Wt1 by the Prx1-Cre (Prx1^Cre/+;W1^GFP/loxp;R26R^tdRFP/+) was generated by crossing Prx1^Cre/+;Wt1^GFP/+;R26R^tdRFP/tdRFP male with female Wt1^loxp/loxp mice. To obtain counterpart control cells in the above model, male Prx1^Cre/+;Wt1^GFP/+;R26R^tdRFP/tdRFP mice were crossed with wild-type females. The R26R^tdRFP/+ line was generated as described by Luche et al. (2007).

Samples were fixed in 4% paraformaldehyde (dissolved in PBS) at 4°C overnight, unless otherwise stated. The next day, samples were washed three times in PBS (10 min each) and stored in 70% ethanol before preparation for paraffin embedding. A Tissue-Tek VIP Jr. Vacuum Infiltration Processor was used for paraffin wax embedding. Paraffin sections were cut at a thickness of 5-6 μm using a microtome, mounted on SuperFrost Plus Microscope slides and dried at 50°C overnight.

Immunofluorescence staining was performed using a similar protocol to the one described previously (Chau et al., 2014). The primary antibodies used in this study were anti-WT1 (Abcam, ab89901, 1:1000), anti-Ki67 (Abcam, ab15580, 1:1000), anti-GFP (Abcam, ab5450, 1:1000), anti-PAX2 (Biolegend, 901001, 1:100), anti-MF20 (DSHB, Ab_2147781, 1:20), anti-VIM (Santa Cruz Biotechnology, sc-7557, 1:100), anti-CDH1 (BD, 610181, 1:100), anti-RALDH2 (Santa Cruz Biotechnology, sc-22591, 1:200), anti-GATA4 (Santa Cruz Biotechnology, sc-25310, 1:100), anti-TCF4 (Cell Signaling Technology, C48H11, 1:100) and anti-MYOD (Santa Cruz Biotechnology, sc-32758, 1:100). Alexa-Fluor 488- or 594-conjugated antibodies were used as secondary antibodies. Sections were stained with 4′,6-diamidino-2-phenylindole (DAPI) and mounted using VectorShield. A Zeiss Axioplan II microscope was used to view immunofluorescence and H&E-stained sections. Image capture was performed using the open source microscopy software: uManager.

Slides were dewaxed in xylene and rehydrated in a series of ethanol washes, followed by washing in tap water and staining with Mayer's Haematoxylin. After washing in tap water, sections were differentiated in 1% HCl in 70% ethanol for a few seconds and washed with tap water. Slides were stained with saturated lithium chloride solution for a few seconds before washing in tap water. Slides were then stained with Eosin, rinsed in tap water, rehydrated and mounted using DPX mounting medium.

Tissues of the thoracic region/upper abdominal region (with lung, heart and liver removed) were digested into a single-cell suspension using collagenase (1 mg/ml collagenase and 4 mg/ml bovine serum albumin, dissolved in PBS) for 30-60 min at 37°C with shaking. Collagenase activity was stopped by washing the cells in PBS containing 5% fetal calf serum. Cells were pelleted by centrifugation at 300 g for 5 min. Cells were filtered using a 40 µm cell strainer and subjected to FACS (BD FACSAria II System). RNA from the sorted cells was extracted using TRIzol reagent.

Complementary DNA was synthesised from RNA using a QuantiTect Whole Transcriptome kit (Qiagen), following the manufacturer's protocol. Quantitative PCR was performed using the Universal Probe Library system (Roche) using a LightCycler 480 II machine. The cycling conditions were pre-incubation (10 min, 95°C), amplification (95°C for 10 s, 60°C for 30 s and 72°C for 1 s) and cooling (40°C for 30 s). The amplification was repeated 50 times. Primer and probe information for each gene used is listed in Table S3.

We thank Anna Thornburn for technical assistance, Allyson Ross for her technical support and expertise in histology, Craig Nicol and Connor Warnock for their support in the graphic design and cartoon preparation, flow cytometry facilities at the Institute of Genetics and Molecular Medicine (IGMM) and Institute of Immunology and Infection Research, IGMM Advanced Imaging Resource, and staff at the animal facility.