Tissue-specific roles for sonic hedgehog signaling in establishing thymus and parathyroid organ fate

In mice, thymus and parathyroid organogenesis begins at E9.5, when the third pharyngeal pouch (3rd pp) is formed. The current model for 3rd pp patterning suggests that opposing signaling pathways induce the dorsal parathyroid [sonic hedgehog (SHH)] and ventral thymus (BMP2/4, FGF8/10) domains (Gordon and Manley, 2011). The parathyroid marker Gcm2 is expressed at E9.5, whereas the earliest defined thymus-specific marker, Foxn1, is not detected until E11 (40 somites) in the most ventral regions of the pouch (Gordon et al., 2001). By E11.5 (48 somites) Foxn1 is strongly expressed throughout the thymus domain of the developing organ primordium (Gordon et al., 2001), and most cells have acquired either a parathyroid or thymus fate.

SHH signaling is active in both the dorsal pouch endoderm and adjacent neural crest (NC) mesenchyme by at least E9.5, during pouch formation (Moore-Scott and Manley, 2005; Grevellec et al., 2011). In the absence of SHH there is no parathyroid domain and Foxn1 expression expands throughout the pouch and into the pharynx (Moore-Scott and Manley, 2005), although the exact role of SHH signaling in parathyroid fate establishment is not clear (Grevellec et al., 2011). Gcm2 is required for survival of the parathyroid domain but it does not appear to be responsible for fate commitment (Liu et al., 2007). In Gcm2 null mice, a Gcm2-negative domain that expresses Tbx1 and Casr at E10.5 is present and survives until E12, when it undergoes coordinated apoptosis (Liu et al., 2007). Tbx1 is a good candidate for mediating the effects of SHH on 3rd pp patterning because it is a known target of SHH signaling in pharyngeal arch mesoderm (Garg et al., 2001; Yamagishi et al., 2003). These data led us to propose that an Shh-Tbx1-Gcm2 regulatory pathway is responsible for establishing initial parathyroid fate (Liu et al., 2007). Consistent with this model, we showed that TBX1 inhibits Foxn1 expression when misexpressed in thymic epithelial cells (Reeh et al., 2014).

Both initial patterning and later thymus and parathyroid organogenesis depend on epithelial-mesenchymal interactions between pouch endoderm and NC cells (Gordon and Manley, 2011). Splotch mice have a null mutation in Pax3, resulting in a loss of NC cells (Conway et al., 2000; Epstein et al., 2000; Pani et al., 2002). E12.5 thymus lobes in Splotch embryos are hyperplastic and ectopic (Griffith et al., 2009). Increased thymus size correlates with decreased parathyroid size, and is due to a shift in the location of the organ domain boundaries in the developing 3rd pp (Griffith et al., 2009). This result showed that signals from NC cells to the developing endodermal primordia determine the location of the border between thymus and parathyroid domains, and thus affect pouch patterning.

In the current study we investigated the respective contributions of SHH signaling within NC mesenchyme and pouch endoderm during 3rd pp patterning and organ development. We used tissue-specific Cre driver strains to selectively delete or ectopically activate SHH signaling in NC mesenchyme or pharyngeal endoderm by expression or activity of the SHH signaling transducer smoothened (Smo). Loss of SHH signaling to NC mesenchyme did not shift domain borders (as in Splotch mice), but did affect patterning and proliferation of the endodermal primordia. Furthermore, in contrast to Shh null embryos, loss of SHH signaling within pouch endoderm did not prevent Gcm2 expression. These results show that SHH signaling within NC cells or endoderm is sufficient for parathyroid fate specification. We further show that, during normal development, establishment of the border between thymus and parathyroid fate involves a transient stage of cell mixing between the two organ fates. The resolution of this cell mixing to non-overlapping organ domains may be mediated by differential cell adhesion, and is dependent on SHH signaling to the NC mesenchyme. This result indicates that epithelial-mesenchymal signals that mediate the establishment of organ borders are SHH dependent. We also show that SHH signaling within the endoderm is sufficient to induce Tbx1 expression and suppress Foxn1 expression, but that this effect is blocked in the most ventral pouch, and is not sufficient to induce Gcm2. Our data suggest that high levels of Bmp4 expression in the most ventral pouch might ‘protect’ those cells from the effects of ectopic SHH signaling, preventing Tbx1 expression and thus allowing ventral pouch cells to differentiate as thymus despite enforced Smo activation.

We conditionally deleted Smo or induced expression of the constitutively active form, SmoM2, from NC mesenchyme using Wnt1Cre (Danielian et al., 1998) or endoderm using Foxa2^CreERT2 (Park et al., 2008). Efficiency of deletion or activation was assessed indirectly using Ptch1 expression as an indicator of SHH signaling (Fig. S1). Deletion of Smo by Wnt1Cre efficiently removed SHH signaling from NC cells surrounding the 3rd pp (Fig. S1D,E), while the tamoxifen-inducible Foxa2^CreERT2 efficiently removed SHH signaling throughout the 3rd pp, but not from the main pharyngeal endoderm (Fig. S1D,F). Similarly, induction of the R26SmoM2 allele using Wnt1Cre caused strong upregulation of SHH signaling throughout the pharyngeal arch mesenchyme (Fig. S1G,H), while Foxa2^CreERT2 induced SHH signaling strongly in the pouch endoderm (Fig. S1G,I). Although endoderm-specific induction was particularly efficient in the dorsal pouch, we consistently found a few cells in the ventral domain that did not strongly upregulate Ptch1 (Fig. S1I′). This is consistent with our previously published data showing that in some cases the Foxa2^CreERT2 strain can display low-level inefficient deletion in the 3rd pp (Chojnowski et al., 2014). However, as the number and the location of undeleted cells vary between individual pouches, and the phenotypes observed should have cell-autonomous effects, any results that were consistently observed between embryos were unlikely to be significantly affected by the presence of a few cells in which SHH signaling was either not deleted or not upregulated.

We hypothesized that SHH signaling to NC mesenchyme is responsible for boundary placement between the thymus and parathyroid domains in the 3rd pp, while SHH signaling to 3rd pp endoderm is necessary to establish parathyroid fate. To test this model we deleted Smo from either NC mesenchyme (Wnt1Cre) (Fig. 1B,E) or pharyngeal endoderm (Foxa2Cre^ERT2) (Fig. 1C,F) and examined expression of thymus and parathyroid-specific markers. Gcm2 expression is absent in Shh null embryos (Moore-Scott and Manley, 2005). We were therefore surprised to find that Gcm2 expression was largely unaffected by tissue-specific loss of SHH signaling. Although there were minor differences in overall primordium size, Gcm2 was expressed in ∼25% of the E10.5 3rd pp after deletion of Smo from either NC mesenchyme (Fig. 1B,E) or endoderm (Fig. 1C,F), similar to controls (Fig. 1A,D).

We also assessed the expression of Fgf8 and Tbx1, which are known targets of SHH signaling (Garg et al., 2001; Yamagishi et al., 2003). Both Fgf8 and Tbx1 have restricted expression patterns in the 3rd pp at E10.5, and have been implicated in thymus and/or parathyroid fate specification and organogenesis (Jerome and Papaioannou, 2001; Frank et al., 2002; Manley et al., 2012). No significant differences were seen in the expression of these markers at E10.5 in embryos when Smo was deleted from either the endoderm (Fig. S2C,D,K,L) or NC mesenchyme (Fig. S2A,B,I,J). These data indicate that, in contrast to Shh null mutants, initial 3rd pp patterning is normal when SHH signaling is selectively deleted in either NC cells or 3rd pp endoderm.

Our previous data showed that in mice, organ fates within the pouch are initiated in the most dorsal (parathyroid) and ventral (thymus) domains, then ‘spread’ to all cells of the pouch (Gordon et al., 2001; Griffith et al., 2009). To examine this process of ‘fate spreading’ with greater temporal resolution, we assessed the patterns of FOXN1 and GCM2 expression in wild-type embryos at 1-2 somite intervals from the earliest stage when both FOXN1 and GCM2 are present within the 3rd pp, which is at E11 (40 somites), to E12.5 (60 somites) when the developing thymus and parathyroid organs separate from each other.

At E11 (40 somites; Fig. 2A), GCM2⁺ cells are confined to the most dorsal-anterior region and FOXN1 is present in the most ventral-posterior region, with scattered marker-negative cells in both domains. There is also a central region composed of cells that express neither marker (Gordon et al., 2001; Griffith et al., 2009) (Fig. 2A, between the blue dashed lines). By 44 somites, the primordium is larger and most epithelial cells have acquired one or other organ-specific marker, with FOXN1⁺ or GCM2⁺ cells mixing in the central region (Fig. 2B, blue dashed lines). This mixing is maintained through the 47-somite stage (Fig. 2C-E), then resolved such that thymus and parathyroid cells assort into distinct domains with a well-defined border. By 56 somites, separation of the two organ domains begins (Fig. 2F, Table 1).

These data support a model in which parathyroid and thymus cell fates are initially established at the dorsal and ventral ends of the pouch, respectively. Then, over about 1 day of embryonic development, the marker-negative cells in the central domain undergo a cell fate decision to assume either organ fate, and assort into distinct domains to resolve a clear organ boundary.

To determine whether SHH signaling to the NC mesenchyme affects organ domain specification, we examined the number and organization of FOXN1⁺ and GCM2⁺ cells in Wnt1Cre;Smo^fx/fx embryos at E11 (40 somites). To quantify patterning phenotypes, we divided the 40-somite stage primordium into five regions from ventral-most to dorsal-most (Fig. 3A). In control embryos, all FOXN1⁺ cells were located in the three most ventral regions of the primordium (regions 1-3) (Fig. 3B,B′). GCM2⁺ cells were present in the three most dorsal regions of the primordium (regions 3-5). At this stage, the central region 3 is the region of ‘intermingling’ where FOXN1⁺, GCM2⁺ and marker-negative cells were all present.

Following deletion of Smo from the NC mesenchyme, FOXN1⁺ cells were primarily located in regions 1 and 2, with a few cells in region 3 and none in region 5 (Fig. 3C-C″), similar to controls (Fig. 3B-B″). By contrast, GCM2⁺ cells were found within region 2 at a significantly higher frequency than in controls (Fig. 3B′,C′,D). Because of this altered distribution of GCM2⁺ cells, we determined whether the organ boundary resolution at E11.5 was affected (Fig. 3E,F). Ectopic GCM2⁺ cells were located within the most ventral domain at E11.5 (Fig. 3F″,G); presumably, these are related to the cells that were observed in region 2 at 40 somites. The total numbers of FOXN1⁺ and GCM2⁺ cells were similar between Wnt1Cre;Smo^fx/fx embryos and control littermates at E11.5 (48 somites) (Fig. 3H), suggesting that the ectopic GCM2⁺ cells were the result of a cell assortment defect rather than mis-specification.

The transition from a mixed population of cells to a well-defined border could occur by selective apoptosis, or by cells assorting into uniform populations based on differential cell adhesion. There is little or no apoptosis in the primordia at these stages, making this mechanism unlikely (Gordon and Manley, 2011). To test whether a defect in differential cell adhesion could underlie this phenotype, we examined E-cadherin expression in the normal primordium and after deletion of Smo from the NC mesenchyme. In control embryos E-cadherin was differentially expressed in the thymus and parathyroid domains, with higher levels in FOXN1⁺ cells (Fig. 4A-A″). By contrast, in the Wnt1Cre;Smo^fx/fx primordia E-cadherin levels did not correspond to the organ domain boundary defined by FOXN1; both high and low levels of E-cadherin were found within FOXN1⁺ cells (Fig. 4B-B″).

These data are consistent with the idea that cells in the thymus-parathyroid primordium assort into the organ domains according to their cell adhesive properties. The differential cell adhesion model predicts that when different tissues are dissociated, mixed, and reaggregated in vitro, they will preferentially assort based on their differential adhesive properties (Steinberg, 1970; Nose et al., 1988; Friedlander et al., 1989). Thymic epithelial cells have previously been shown to use E-cadherin for cell adhesion (Lee et al., 1994; Müller et al., 1997). Since parathyroid cells express a lower level of E-cadherin than cells in the thymus domain at E11.5 (Fig. 4C-C″), differential cell adhesion could be the mechanism driving cellular organization during normal development. To test whether the differential expression levels observed were sufficiently different to drive cell assortment based on organ identity, E12.0 primordia from Gcm2-EGFP reporter embryos were isolated and dissociated to a single-cell suspension, then allowed to reaggregate as a mixed population (Fig. 4D). After 4 days in culture, cells were organized into distinct GFP⁺ and GFP⁻ clusters that correlated with low and high E-cadherin levels (Fig. 4E-G). This result suggests that cells from the thymus-parathyroid primordium can reorganize in vitro, consistent with differential cell adhesion.

Taken together, these results suggest that cell assortment in the central domain during normal development may be dependent on differential E-cadherin levels, and that this is influenced by SHH signaling to NC mesenchyme. Differential cell adhesion might mediate both the initial location of parathyroid cell specification and the ability to assort into separate thymus and parathyroid organ domains. It is also possible that inefficient cell segregation is a secondary consequence of the broader domain of cell mixing, resulting in cells being ‘trapped’ outside their normal organ domain.

In contrast to the complete absence of parathyroid domain in the Shh null mutants (Moore-Scott and Manley, 2005), deletion of Smo from the pharyngeal endoderm resulted in comparatively mild effects on 3rd pp patterning and organ development. Only the most dorsal region 5 was abnormal in embryos following loss of SHH signaling to the endoderm, where a small patch of FOXN1⁺ cells was present in five out of six primordia examined at 40-41 somites (Fig. 5A,C″). These FOXN1⁺ cells were completely separated from the primary thymus domain in regions 1 and 2 (Fig. 5C,C′). By the 48-somite stage, this dorsal FOXN1⁺ region was continuous with the main FOXN1 domain (Fig. 5E), and the total cell number within the primordium was comparable between mutant and control embryos (Fig. 5F). Thus, deletion of SHH signaling in pouch endoderm results in a mostly normal primordium, except for an abnormal anterior dorsal distribution of FOXN1⁺ cells.

As neither of the tissue-specific deletions of Shh phenocopied the Shh null or Splotch mutant loss-of-function (LOF) phenotypes, we performed gain-of-function (GOF) experiments to test whether ectopic SHH signaling was sufficient to affect the patterning and development of the 3rd pp. We again looked at pouch patterning and found that early Tbx1 and Fgf8 expression was normal at E10.5 when the SHH pathway was activated in the NC mesenchyme (Fig. S2E,F,M,N) and 3rd pp endoderm (Fig. S2G,H,O,P).

Ectopic expression of activated Smo (R26SmoM2) in NC mesenchyme by Wnt1Cre resulted in reduced FOXN1⁺ and GCM2⁺ cell numbers and levels of marker expression at 40 somites. Although the anterior-posterior distribution of FOXN1⁺ and GCM2⁺ cells within the five regions was comparable to wild type (Fig. 6A), marker-positive cells were found primarily within the dorsal side of the pouch, with ventral cells largely marker negative at this stage (Fig. 6A,C-C″). By the 50-somite stage, all cells within the primordium expressed either FOXN1 or GCM2, indicating delayed marker expression on the ventral side (Fig. 6E,E′). The overall primordium size was similar between mutants and controls at 40 somites (Table 2), but by the 50-somite stage it was reduced in the mutants (Fig. 6F), consistent with reduced proliferation (Fig. S3). Relative parathyroid size was also reduced from 26.7% to 15.7% of the total primordium (Table 3). These results showed that ectopic activation of SHH signaling within NC mesenchyme results in delayed initial cell fate specification and reduced proliferation within the developing primordium, followed by increased thymus and reduced parathyroid fate specification.

We next assessed the effects of ectopic expression of activated Smo (R26SmoM2) throughout the pharyngeal endoderm using Foxa2Cre^ERT2. Our initial hypothesis predicted an expanded GCM2 domain within the primordium at the expense of the thymus domain in these mutants. While the FOXN1 domain was restricted to the most ventral region at 40 somites (Fig. 7A,C), the number of GCM2⁺ cells was similar to controls (Fig. 7F). Overall 3rd pp size was normal following activation of the SHH pathway in the pouch endoderm (Table 2). However, the distribution of GCM2⁺ cells was more ventral, extending into region 2 at 40 somites (Fig. 7C′). The restriction in FOXN1⁺ cells and normal GCM2⁺ cell numbers resulted in a substantial proportion of the central pouch expressing neither marker (Fig. 7A,C). This delayed patterning was superficially similar to that seen at this stage when the SHH pathway was activated in NC mesenchyme (Fig. 6). However, these phenotypes persisted at 48 somites following SHH pathway activation in the pouch endoderm (Fig. 7E), resulting in a domain of FOXN1⁻ and GCM2⁻ cells in the central region (Fig. 7E′, asterisk).

The number of marker-negative cells plus FOXN1⁺ cells was comparable to the total number of FOXN1⁺ cells in a wild-type pouch at this stage (Fig. 7F). This suggests that this marker-negative region is likely to represent cells located within the region of the primordium that would normally have been part of the FOXN1⁺ thymus domain, but had failed to turn on Foxn1. Thus, ectopic SHH signaling did not lead to an increase in the number of cells expressing Gcm2, nor did it completely block Foxn1 expression. Instead, activation of the pathway prevented Foxn1 expression from expanding within the central pouch domain.

In the Gcm2 null mutant, a presumptive parathyroid domain that expresses Tbx1 is present at E11.5 (Liu et al., 2007). Therefore, we tested whether ectopic SHH signaling within the pouch endoderm in Foxa2Cre^ERT2;R26SmoM2 embryos induces ectopic Tbx1 expression. Tbx1 is normally restricted to the anterior-dorsal domain of the E11.5 pouch (Fig. 8B,F,N) (Manley et al., 2004; Liu et al., 2007), co-expressed with Gcm2 (Fig. 8E,M) and complementary to Foxn1 (Fig. 8A,I,J). In embryos in which SHH signaling was activated within the pouch endoderm, Tbx1 was expressed both in its normal domain with Gcm2, and in an ectopic domain in the Foxn1- and Gcm2-negative central pouch endoderm at 48 somites, as assessed by both RNA expression and protein analysis (Fig. 8D,H). In addition, Tbx1 expression extended partially into the Foxn1 domain in the mutants (Fig. 8C,D), where Tbx1 and Foxn1 levels were inversely correlated (Fig. 8L,P). However, Tbx1 expression did not extend into the most ventral domain (Fig. 8D,H). As we previously showed that ectopic expression of Tbx1 in the thymus domain represses Foxn1 expression (Reeh et al., 2014), these data suggest that ectopic TBX1 downstream of SHH signaling represses Foxn1 in this central domain. Furthermore, cells in the most ventral primordium did not express Tbx1 and retained Foxn1 expression, despite ectopic activation of the SHH pathway as shown by Ptch1 activation (Fig. S1I,I′).

Shh is not expressed in NC mesenchyme adjacent to the 3rd pp. Therefore, in order for SHH signaling in NC mesenchyme to affect the 3rd pp, a second signaling molecule downstream of SHH would need to act directly on the pouch endoderm. Previous studies demonstrated that FGF10 and BMP2/4 act downstream of SHH in palatal mesenchyme (Lan and Jiang, 2009). Both Fgf10 and Bmp4 are expressed in the mesenchyme surrounding the pouch, and have been shown to influence parathyroid and thymus organogenesis, although their precise roles are unclear (Gordon and Manley, 2011). Most relevant to this study, Bmp4 expression is expanded in pouch endoderm of Shh null mutants, consistent with the expansion of thymus fate and Foxn1 expression (Moore-Scott and Manley, 2005).

Fgf10 expression was similar in control and mutant embryos from all four genetic manipulations at E11.5 (Fig. S4), suggesting that FGF10 does not mediate the effects of SHH signaling. Similarly, Bmp4 expression was not affected by activation of Smo in the NC mesenchyme (Fig. 9A,B). More significantly, Bmp4 expression was unaffected by activation of the SHH signaling pathway in the pouch endoderm (Fig. 9C,D), even though Foxn1 expression was restricted to the most ventral primordium. This result is particularly important, as BMP4 is implicated in promoting thymus fate. These data suggest that BMP4 expression within and near the central domain could be involved in the failure of Gcm2 expression to expand after ectopic activation of Smo and Tbx1 in the endoderm.

We noted in the process of analyzing these mutants that it was often difficult to generate comparable planes of section for control and mutant embryos when SHH signaling had been altered in the NC mesenchyme. We have previously shown that the shape of the pharynx is altered in Shh null mice (Moore-Scott and Manley, 2005). In wild-type E11.5 embryos, the pharynx has an arch-like shape and the primordium is located adjacent and ventral to the pharynx (Fig. S5A,D,G,J). When SHH signaling was manipulated in 3rd pp endoderm, pharynx shape remained normal (Fig. S5F,L). However, pharynx shape was altered when SHH signaling was manipulated in NC mesenchyme. In the absence of SHH signaling, (Wnt1Cre;Smo^fx/fx), the dorsal-ventral width of the pharynx was increased and the primordium was more dorsally located within the embryo (Fig. S5C). When SHH signaling was activated (Wnt1Cre;R26SmoM2), the pharynx was flattened, placing the primordium more laterally (Fig. S5I). These data indicate that SHH signaling to NC mesenchyme helps to shape the pharynx. These shape changes could place the different regions of the pouch into new signaling environments, indirectly influencing 3rd pp patterning.

The mechanisms by which the 3rd pp is patterned into thymus and parathyroid fates remain poorly understood. We previously showed that SHH is required to specify the parathyroid domain in the 3rd pp, and that although Shh itself is not expressed in the pouch, cells in the dorsal pouch endoderm and the neighboring NC mesenchyme are subject to SHH signaling (Moore-Scott and Manley, 2005; Gordon and Manley, 2011). Furthermore, inhibition of SHH signaling in the chick results in a failure to initiate Gcm2 expression and enhances Bmp4 expression (Grevellec et al., 2011), consistent with the Shh null mouse phenotype (Moore-Scott and Manley, 2005). In the current study we used tissue-specific GOF and LOF mouse mutants to dissect the tissue-specific roles of SHH signaling during 3rd pp patterning. Interestingly, neither NC-specific nor endoderm-specific deletion of SHH signaling recapitulated the Shh null phenotype, suggesting that SHH signaling within either of these cell types is sufficient for parathyroid fate specification. We also showed that ectopic SHH signaling in pouch endoderm induced Tbx1 and suppressed Foxn1 expression in the medial pouch, but was not sufficient to induce Gcm2 expression in these cells. Finally, the most ventral pouch cells specified thymus fate and turned on Foxn1 despite enforced SHH signaling. Taken together, these results indicate that cells in the dorsal and ventral 3rd pp are differentially sensitive to the levels of SHH signaling, and that this is part of a complex and robust signaling network that controls cell fate establishment during thymus and parathyroid development.

We propose a model (Fig. 10) in which SHH is necessary but not sufficient for parathyroid fate. In the absence of SHH, Gcm2 is not expressed and parathyroids are absent (Moore-Scott and Manley, 2005; Grevellec et al., 2011). Interestingly, SHH signaling in either the endoderm or NC mesenchyme alone is sufficient but not necessary for Gcm2 expression and parathyroid fate establishment. It is unlikely that these phenotypes are caused by SHH signaling occurring prior to Cre-mediated deletion because Foxa2Cre^ERT2 was activated at E5.5, prior to 3rd pp formation, and Wnt1Cre acts prior to NC cell migration. It is formally possible, but unlikely, that Gcm2 expression is dependent on a Smo-independent role for SHH, in which case deleting Smo would not recapitulate the Shh null phenotype (Jenkins, 2009). It is also possible that another cell type is involved, although it is unclear at this point what that cell type would be. We believe that the most plausible explanation is that SHH signaling to either the endoderm or mesenchyme alone is sufficient to promote Gcm2 expression in the anterior-dorsal 3rd pp. In the endoderm this could be by direct signaling, but in the mesenchyme there would have to be a second, as yet unidentified, signal acting downstream of Shh. Taken together, these data demonstrate that both parathyroid fate and Gcm2 expression are controlled by multiple factors acting directly within the endoderm downstream of SHH signaling, as well as indirectly from the adjacent NC-derived mesenchyme.

When SHH signaling was deleted in the NC mesenchyme, we observed a phenotype wherein a subset of GCM2⁺ cells was mixed with FOXN1⁺ cells in a more ventral location. This is reminiscent of an intermingling phenotype seen in the spinal cord of mutants for Gli3, a negative regulator of SHH signaling. In Gli3 mutants, V2 neurons intermingle with EN1⁺ V1 neurons (Persson et al., 2002). However, although these two phenotypes appear similar they are likely to arise via different mechanisms. The intermingling in the Gli3 mutant spinal cord is likely to be due to mis-specification, whereas that seen in the Wnt1Cre;Smo^fx/fx mutants is likely to be a cell assortment defect that is due to misregulation of E-cadherin. During normal development, we showed that E-cadherin levels differ between the thymus and parathyroid domains and that these cells are capable of reorganizing by homotypic adhesion. In the Wnt1Cre;Smo^fx/fx mutants, loss of SHH signaling to the NC led to misregulation of E-cadherin levels in the primordium, resulting in cells that were unable to sort correctly. The fact that cell numbers were unchanged in these mutant primordia also supports a cell sorting defect, rather than a cell fate switch.

Ectopic activation of the SHH pathway in the thymus domain caused a ventral expansion of Tbx1 expression and suppression of Foxn1 expression within the central pouch. This result is consistent with our previous study showing that ectopic activation of Tbx1 in the ventral thymus domain inhibits Foxn1 expression but fails to induce Gcm2 (Reeh et al., 2014). There are two possible reasons why Gcm2 expression did not expand. First, that an SHH-independent signal acts together with the SHH signaling pathway. One candidate for this pro-parathyroid signal is FGF10, which is expressed in the proximal NC mesenchyme adjacent to the parathyroid domain in the 3rd pp (Gardiner et al., 2012), and is unaffected in all of the SHH pathway mutants examined in the current study. Also, Fgf10 null mutants have reduced parathyroid size (J.G. and N.R.M., unpublished data), further supporting a role for FGF10 in parathyroid development. Second, it is possible that an inhibitory signal in the medial endoderm or surrounding mesenchyme, such as BMP4, dominates over SHH signaling in that middle region of the 3rd pp. SHH and BMP4 work in opposition in many developmental contexts to pattern tissues and organs, and there is evidence to support a similar role in patterning thymus and parathyroid fate within the 3rd pp (Moore-Scott and Manley, 2005; Gordon and Manley, 2011). We found that forced activation of the SHH signaling pathway in either NC mesenchyme or pouch endoderm did not inhibit Bmp4 expression within the pouch or in the surrounding mesenchyme. Therefore, BMP4 signaling could act to prevent ectopic parathyroid fate specification and Gcm2 expression in the central and ventral pouch. BMP4 dominance over SHH signaling within the pouch is also consistent with the establishment of both thymus fate and Foxn1 expression in the most ventral pouch, even in the presence of enforced expression of constitutively activated Smo. Clearly, it is only the ventral-most cells in the prospective thymus domain that are resistant to SHH-mediated activation of Tbx1 expression, which allows/enables these cells to maintain Foxn1 expression, even in the presence of SHH signaling. Whether BMP4 or another, unknown, dominant signal from the ventral 3rd pp and/or NC mesenchyme antagonizes SHH signaling remains to be determined.

In conclusion, our data show that cells in the 3rd pp endoderm are sensitive to both direct SHH signaling within the pouch and indirect SHH signaling from the NC mesenchyme. The most striking phenotype was obtained when SHH signaling was ectopically induced in the ventral 3rd pp endoderm. Despite the concomitant ventral expansion of Tbx1 and suppression of Foxn1 expression, Gcm2 was not turned on; furthermore, the most ventral cells were completely insensitive to ectopic SHH signaling. These data show that even high levels of ectopic SHH signaling cannot completely change the organ-specific fates of cells within the 3rd pp. Therefore, a robust network of signaling and transcriptional mechanisms exists, which are likely to include SHH, BMP, FGF and TBX1, that collaborate to establish organ-specific fates within the developing 3rd pp.

Smo^fx, Rosa26SmoM2, Wnt1Cre and Foxa2Cre^ERT2 mice were from The Jackson Laboratory and used to generate Wnt1Cre;Smo^fx (NC cell-specific deletion), Foxa2Cre^ERT2;Smo^fx (endoderm-specific deletion), Wnt1Cre;Rosa26SmoM2 (NC cell-specific activation) and Foxa2Cre^ERT2;Rosa26SmoM2 (endoderm-specific activation) embryos and littermate controls. Littermate controls shown in figures are Cre negative. Cre-positive and Cre-negative controls were assessed in all experiments; no effects of Cre expression alone were observed (data not shown). R26Rβgal reporter mice were from The Jackson Laboratory. Colonies were maintained on a mixed C57BL6/J;129Sv genetic background.

The day of the vaginal plug was designated embryonic day (E) 0.5. Embryos were staged by somite number and morphological cues. Genotyping of yolk sac-derived DNA employed published primers (Danielian et al., 1998; Soriano, 1999; Long et al., 2001; Jeong et al., 2004; Park et al., 2008). Pregnant females from tamoxifen-inducible Cre lines were injected with tamoxifen (Sigma; 3 mg per 40 g body weight) on E5.5. All experiments were approved by the UGA Committee for Animal Use in Research and Education.

Embryos for frozen sections were fixed in 4% paraformaldehyde (PFA) for 20 min (E10.5) or 30 min (E11.5-E13.5), washed in PBS, then 20% sucrose, and embedded in OCT (Sakura). Sections were cut on a cryostat at 8 µm. Embryos collected for paraffin sections were fixed for 1 h, washed in PBS, dehydrated through an ethanol gradient, permeabilized in xylene, and embedded in paraffin. Embryos collected for in situ hybridization (ISH) were kept cold and RNase-free reagents were used. Sections for immunostaining were cut on a Leica RM2155 microtome at 8 µm and for ISH at 12 µm.

Whole-mount and paraffin section ISH were performed as described (Manley and Capecchi, 1995; Moore-Scott and Manley, 2005) using mutant embryos and littermate controls. Each probe was analyzed on a minimum of two or three embryos per stage. Probes for Fgf8 (Crossley and Martin, 1995), Ptch1 (Goodrich et al., 1996), Tbx1 (Chapman et al., 1996), Gcm2, Foxn1, Bmp4 (Gordon et al., 2001) and Fgf10 (Bellusci et al., 1997) were previously described.

Immunostaining was performed on paraffin-embedded or frozen tissue fixed in 4% PFA. Paraffin sections were washed in xylene and rehydrated through an ethanol gradient to distilled H₂O. For antigen retrieval, tissue was boiled in AR buffer (10 mM sodium citrate pH 6, 0.05% Tween 20) for 30 min and allowed to cool for 20 min. Slides were incubated overnight at 4°C in 100 µl 5% donkey serum and 0.05% Triton-X100 in PBS containing primary antibodies. Slides were washed in PBS and incubated with secondary antibodies in PBS for 1 h at room temperature in the dark. Slides were washed three times in PBS, with the second wash containing DAPI. Slides were mounted with Fluorogel (EMS). Frozen sections were washed in PBS then incubated with primary and secondary antibodies as above. Primary antibodies were: goat anti-FOXN1 (1:200, Santa Cruz sc-23566, G-20), rabbit anti-GCM2 (1:200, Abcam ab64723), rabbit anti-TBX1 (1:100, Abcam ab18530), rabbit anti-cleaved caspase 3 (1:200, Cell Signaling #9661), rat anti-BrdU (1:10, AbD Serotec OBT0030CX) and rat anti-E-cadherin (1:200, Invitrogen 13-1900). Secondary antibodies were DyLight conjugated (1:1000, Jackson ImmunoResearch).

Pregnant female mice were injected intraperitoneally with 5-bromo-2'-deoxyuridine (BrdU, 50 mg/kg body weight; Sigma) 90 min before embryo collection. Embryos were processed for frozen sectioning as above. Sections were fixed in ice-cold acetone for 2 min, treated with 2 M HCl for 30 min, then incubated with rat anti-BrdU antibody (AbD Serotec, diluted 1:10 in PBS containing 10% donkey serum) overnight at 4°C. Sections were washed three times in PBS, and incubated with a DyLight-conjugated donkey anti-rat secondary antibody (1:1000, Jackson ImmunoResearch) for 30 min.

Thymus-parathyroid primordia were isolated from E12.0 Gcm2-EGFP mouse embryos (Condie, 2016) and washed in PBS. Tissue was digested to a single-cell suspension using 2 mg/ml hyaluronidase, 0.7 mg/ml collagenase and 0.05 mg/ml DNase (Sigma) for 10 min at 37°C. Cells were centrifuged, washed in medium (DMEM with 10% FBS, 10% glutamine, 10% pen/strep) then pelleted again. Almost all medium was removed and the resulting slurry pipetted onto a floating filter in a 6-well plate containing culture medium. Two days later, the reaggregate was transferred to a v-bottomed 96-well plate containing culture medium, incubated for 3 days, harvested, and fixed in 4% PFA for 30 min. Tissues were processed through a sucrose gradient and embedded in OCT. Frozen sections (10 µm) were cut and stained with anti-E-cadherin (Invitrogen) and DAPI. To confirm single-cell dissociation, reaggregated cells were collected after 24 h on the floating filter, processed, sectioned, and stained with DAPI.

Embryos for X-gal staining were processed as described (Gordon et al., 2001). Stained embryos were paraffin embedded and sectioned (10 µm).

Cell counting was performed manually using images of serial sections taken on a Zeiss Axioplan microscope, using the events feature in AxioVision Rel.4.8 software (Zeiss).

3D reconstructions were generated from serial sections using SurfDriver WinSurf 4.3 software.

For phenotypic analysis of mutants we used an initial sample size of three embryos (six primordia; n values refer to primordia number). Additional embryos were analyzed as needed to achieve statistical significance for any observed phenotype. Significance was determined using an unpaired t-test. Controls and mutants were processed in parallel for all experiments; littermate controls were used when possible, otherwise controls and mutants were from the same colony. All results from experiments in which one or more embryos or samples had a technical failure (tissue damage, weak or uneven staining) were discarded.

We thank Brian Condie for helpful discussion and suggestions; and Michelle Dookwah for assisting with the work on 3rd pp separation.