Bmp signaling maintains a mesoderm progenitor cell state in the mouse tailbud

During gastrulation, cells from the epiblast migrate through the primitive streak and undergo epithelial-mesenchymal transition to form the mesodermal and endodermal layers. As the primitive streak regresses toward the caudal region of the embryo, mesoderm progenitors successively contribute to notochord, paraxial mesoderm (muscles and bones), intermediate mesoderm (urogenital system) and lateral plate mesoderm (heart, gut and vasculature). Towards the end of gastrulation, somitogenesis proceeds from a pool of mesoderm tailbud progenitors responsible for generating caudal somites and tail elongation (Bénazéraf et al., 2010; Neijts et al., 2014; Wilson et al., 2009). Mesoderm progenitor cells form the presomitic mesoderm (PSM), which is progressively segmented into metameric units under the control of the segmentation clock (reviewed by Hubaud and Pourquié, 2014). The molecular network governing the transition from mesoderm progenitors to somites is well documented (Neijts et al., 2014; Wilson et al., 2009). In mouse and chick embryos, this network involves a signaling core that includes Wnt3a, brachyury (T) and Fgf8, which regulate downstream effectors of the segmentation clock (Galceran et al., 2001; Yamaguchi et al., 1999) and prevent retinoic acid-mediated differentiation of mesoderm cells (Diez del Corral et al., 2003). In this system, Fgf signaling is thought to maintain PSM cells in a progenitor state by activation of the retinoic acid-inactivating enzyme Cyp26a1 (Diez del Corral et al., 2003; Wahl et al., 2007). As mesoderm cells exit the Fgf8 gradient in the PSM, an increase in retinoic acid promotes their differentiation (Diez del Corral et al., 2003). Fgf signaling further regulates the migration of PSM cells toward the rostral region, where Fgf8 acts as a repellent signal from the caudal trunk/tailbud region, and Fgf4 acts as an attractive signal in the anterior streak and node region (Yang et al., 2002).

Bmp4, a member of the TGFβ superfamily, is also involved in somite formation. In frog and chick embryos, both gain- and loss-of-function studies identified a role for Bmp4 and Bmpr1 in tailbud outgrowth and somite differentiation (Beck et al., 2001; Reshef et al., 1998; Row and Kimelman, 2009). In zebrafish, loss of Bmp results in ventral fin and tail mesoderm defects (Pyati et al., 2005; Stickney et al., 2007). However, similar comprehensive studies on the later role of Bmp signaling in axis elongation and tail somite formation in mice have been hampered by the essential role of Bmp4 during gastrulation (Winnier et al., 1995). The few Bmp4-deficient embryos reaching a later stage show an axis elongation defect, which supports a role in posterior trunk elongation and patterning (Winnier et al., 1995). Accordingly, inactivation of noggin, an inhibitor of Bmp4 in this system (Wijgerde et al., 2005), leads to defects in elongation and in somite differentiation (McMahon et al., 1998; Wijgerde et al., 2005). However, the role of Bmp signaling in tailbud mesoderm progenitors remains poorly understood at the cellular and molecular levels.

To better understand the regulation of mesoderm progenitor differentiation we took advantage of a BAC transgenic line expressing GFP under the control of the Pax2 locus. Although Pax2 gene inactivation does not affect somitogenesis or axial elongation, the Pax2 locus is specifically expressed in a graded fashion in the caudal region of the embryo (Bouchard et al., 2005, 2002; Kuschert et al., 2001). Using this transgene to isolate a population of tailbud mesoderm progenitor cells, we identified a crucial role for Bmp signaling in sustaining the entire network of mesoderm progenitor cells and preventing their premature differentiation into mesoderm derivatives such as muscle, cartilage, heart, vasculature and blood cells.

To better understand the maintenance of progenitor cell identity in the tailbud, we used the Pax2-GFP BAC transgenic line, which accurately expresses GFP in Pax2-expressing cells (Bouchard et al., 2005), including the nephric duct, tailbud, PSM and ventral mesoderm (Fig. 1A,B). We used fluorescence-activated cell sorting (FACS) to isolate GFP⁺ cells from the tailbud and cultured them on growth-arrested mouse embryonic fibroblasts (MEFs), which promotes their long-term survival (Fig. 1C,D). Interestingly, the vast majority of Pax2-GFP-expressing cells gradually lost their fluorescent signal over a period of 96 h, without significant cell death (Fig. 1E,F, Fig. S1A). We hereby refer to cells expressing the Pax2-GFP BAC transgene as Pax2-GFP^on and cells that downregulate GFP expression in primary cultures as Pax2-GFP^off. qRT-PCR analysis of cultured Pax2-GFP^off cells (separated from PKH26-labeled MEFs by FACS) showed a marked reduction in the expression of tailbud mesoderm markers T and Msgn1, suggesting that a signal was required to maintain the mesoderm progenitor cell fate in 2D culture (Fig. 1G). The few remaining Pax2-GFP^on cells in 96 h cultures expressed the renal lineage markers Pax8 and Gata3, indicating that a small number of renal cells were present in primary cell cultures but that the levels of Pax2-GFP were not significantly affected at 96 h in these cells (Fig. S1B).

To determine whether the tailbud progenitor cell fate was maintained by a secreted signal(s) from within this population, we plated Pax2-GFP^on cells in 3D Matrigel cultures at decreasing densities. These culture conditions allowed for the maintenance of Pax2-GFP expression when plated at high cell density (Fig. 1H). By contrast, Pax2-GFP^on mesoderm progenitor cells plated at lower densities led to a density-dependent decrease of Pax2-GFP signal, indicating that tailbud progenitor cell fate maintenance depends on the activity of a paracrine signal, which is likely to be diluted in 2D cultures (Fig. 1H).

To identify the paracrine signal involved in mesoderm progenitor cell fate, we undertook a candidate approach. We initially tested the capacity of Wnt3a, Fgf8, Fgf4 and Bmp4 to reactivate GFP expression in Pax2-GFP^off cells (after 96 h in culture), all of which have previously been implicated in axis elongation (Wilson et al., 2009). Of these, only Bmp4 led to a strong GFP reactivation after 12 h of treatment, whereas the other factors were not sufficient to trigger a detectable response (Fig. 2A,B, Fig. S2A-C). This suggests that Bmp is an important paracrine factor for maintaining the tailbud progenitor fate. To validate this possibility, we treated progenitor cells plated at a high density in Matrigel (which maintain Pax2-GFP expression) with the Bmp inhibitor noggin. In contrast to untreated cells that maintain Pax2-GFP at high density in Matrigel (Fig. 1H), noggin treatment led to a significant decrease in Pax2-GFP expression, further supporting the importance for Bmp in this process (Fig. S2D-G). We determined the tailbud expression domain of Bmp4 by in situ hybridization. Bmp4 is expressed in the ventral but not in the dorsal mesoderm (Fig. S2H). Together, these results suggest that a ventral mesoderm source of Bmp acts on dorsal mesoderm progenitor cells.

Given the important roles of both Fgf4 and Fgf8 during trunk elongation, we assessed the response to Bmp4 in combination with these signaling molecules at different time points (Fig. 2A). Interestingly, the combination of Fgf8 and Bmp4 shortened the timing of GFP re-expression to 6 h, down from 12 h with Bmp4 alone (Fig. 2B,C). By contrast, addition of Fgf4 abrogated the response to Bmp4 (Fig. 2B,C). These experiments suggest that Fgf8 facilitates the response to Bmp signaling, whereas Fgf4 seems to counteract Bmp.

To gain further insight into the transcriptional control of mesoderm progenitor cell fate, we performed whole-transcriptome RNA sequencing (RNA-seq) on Pax2-GFP^on mesoderm progenitors (untreated at 0 h), Pax2-GFP^off cells (untreated at 96 h), and on mesoderm progenitor cells at 96 h treated with Bmp4 alone (5 h and 11 h post-treatment) or in combination with Fgf8 (5 h post-treatment) or Fgf4 (11 h post-treatment) (Fig. 2A, Fig. S3A,B, Tables S1-S12). Pairwise comparisons of these transcriptional profiles were performed to identify a gene signature associated with Pax2-GFP^on cells (Fig. S3A,B, Table S13). Gene ontology analysis of the Pax2-GFP^on gene signature (which contained 170 genes) showed a statistical enrichment for mesoderm lineage markers, specifically those of presomitic mesoderm and tailbud progenitor cells (Fig. 3A). In fact, most known mesoderm progenitor-associated genes were differentially regulated between Pax2-GFP^on (untreated at 0 h) and Pax2-GFP^off (untreated at 96 h), and were found to be expressed in all Pax2-GFP-expressing cells (Fig. 3B). Remarkably, these mesoderm markers, which included Msgn1, T, Fgf8, Wnt5a, Wnt3a, Tbx6, Cdx2, Cdx4 and Nkx1-2, were also the most differentially regulated genes between untreated cells at 0 h and 96 h, and between Bmp4-treated (96+5 h) and untreated cells at 96 h (green dots, Fig. 3B,C,D, Table S15). These results indicate that the loss of Pax2-GFP expression is primarily associated with a loss of mesoderm progenitor cellular identity, and that Bmp4 alone is sufficient to re-establish the mesoderm progenitor gene signature (Fig. 3B-D, green dots). The Bmp4 response was supported by Fgf8 treatment, but was largely prevented by co-treatment with Fgf4 (Fig. 3B).

To confirm the RNA-seq results, we performed qRT-PCR analysis of key regulators of the mesoderm progenitor fate, which validated the loss of identity during the first 96 h in culture and the reactivation of these genes by Bmp4 (Fig. 3E). Hence, these results indicate that the maintenance of the mesoderm progenitor gene signature is largely dependent on the sustained activity of Bmp4, which can be modulated by Fgf signaling.

To better understand gene network activation by Bmp4, we performed qRT-PCR of key regulated genes in the presence of the translation inhibitor cycloheximide (CHX). This analysis identified three groups of mesoderm progenitor genes based on their response to Bmp4. Genes such as T and Nkx1-2 were unaffected by the presence of CHX, indicating that they are direct targets of Bmp signaling (Fig. 4A). Others, such as Msgn1 and Cdx4, were activated 3- to 4-fold in cells treated with Bmp4 plus CHX, but failed to reach the activation levels observed with Bmp4 alone (Fig. 4B). This suggests a direct regulation by Bmp signaling that is complemented by a Bmp-regulated coactivator. Finally, Pax2 expression was found to be strictly dependent on translation, indicating that it is an indirect target of Bmp signaling activity (Fig. 4C).

From these results, we conclude that Bmp signaling is sufficient for the expression of several key regulators of mesoderm progenitor fate, resulting in the maintenance of the underlying gene regulatory network.

To determine whether the maintenance of mesoderm progenitor cells by Bmp also occurs in the embryo, we treated ex utero mouse embryo cultures with the Bmp inhibitor noggin. Pax2-GFP embryos were dissected at E8.75 and maintained in rolling cultures for 36 h in the presence or absence of noggin. We observed a reduction of GFP signal intensity in noggin-treated versus control embryos (Fig. 5A-B′). We next visualized the effect of noggin by whole-mount in situ hybridization against T. In contrast to the high expression levels in control embryos, the addition of noggin to the culture medium resulted in a significant downregulation of T (Fig. 5C,D). The effect of Bmp inhibition was extended to the mesoderm progenitor markers Nkx1-2, Msgn1 and Pax2, as shown by qRT-PCR of noggin-treated and control tailbud cells. These experiments confirmed that Bmp signaling is required to maintain the mesoderm progenitor transcriptional program in vivo (Fig. 5E).

To identify the fate changes associated with the loss of mesoderm progenitor identity, we determined the gene signatures associated with Pax2-GFP^off cells using the pairwise comparisons performed previously (Fig. S3A,C, Table S14). Gene ontology analysis identified distinct signatures associated with muscle, heart, blood, bone and endothelial cell differentiation (Fig. 6A, Table S16). Genes associated with these ontological signatures were also among the most differentially regulated genes in Pax2-GFP^off (untreated at 96 h) compared with Pax2-GFP^on (untreated at 0 h) (colored dots, Fig. 6B). Treatment with Bmp4 alone or together with Fgf8 effectively downregulated these gene signatures, whereas Bmp4 and Fgf4 generally failed to abrogate their expression (Fig. 6C, Table S16). The Bmp4 transcriptional response was validated by qRT-PCR analysis on a subset of differentially regulated genes (Fig. 6D). Here again, the Bmp4 response was supported by Fgf8, whereas Fgf4 negatively affected the regulation of a subset of Bmp-responsive genes (Fig. 6D).

To test whether Bmp activity affected the neural and hindgut tailbud cell fates, we tested markers for these lineages by qRT-PCR. We observed a response of Sox2 and the hindgut markers Sox17 and Epha2 to Bmp4 in primary cultures but noggin treatment failed to validate a strong role for Bmp signaling in embryo cultures ex utero (Fig. S4A-D). Hence, Bmp signaling prevents the aberrant expression of cell differentiation markers primarily in tailbud mesodermal progenitor cells.

We next tested if the transcriptional upregulation of differentiation markers within mesoderm progenitors upon loss of Bmp signaling led to a distinct commitment to these fates. Immunohistochemical analysis of cells at 0 h showed little expression of markers such as Ter119 (Ly76; blood), Pax7 (skeletal muscle) and Gata4 (heart) (Fig. 7A). However, the expression of these differentiation markers was upregulated after 96 h in culture (Fig. 7B). Strikingly, Pax2-GFP^off cells at 96 h frequently upregulated more than one lineage marker, indicating an aberrant and dysregulated differentiation process (Fig. 7B,D). This anomalous differentiation was lost when cells were treated with Bmp4 (Fig. 7C,D).

To test if the inhibition of Bmp signaling can cause aberrant differentiation of mesoderm progenitors in the developing embryo, we performed ex utero embryo cultures in the presence of noggin for 48 h. Noggin-treated embryos showed abnormalities in the development of the somites and neural tube, and a shortened caudal trunk (Fig. S4E,F). Immunohistochemical analysis of noggin-treated tailbud sections showed atypical morphology, with condensed cell clusters (Fig. 7F) and ectopic expression of Gata4 and Pax7. These abnormal structures contrasted with the morphology of control embryos (Fig. 7E). From these results we conclude that Bmp plays a crucial role in maintaining mesoderm progenitors in an undifferentiated state, before commitment towards somitogenesis.

Axial elongation proceeds through the coordinated differentiation of progenitor cells to pattern the growing embryo. During this process, the formation of new somites is initiated in the caudal trunk region and involves the migration and differentiation of mesoderm progenitor cells. These progenitors upregulate somite differentiation markers as they exit the influence of caudal signals responsible for maintaining those cells in an undifferentiated state. In this report, we identified Bmp as a key signal responsible for mesoderm progenitor maintenance through the direct activation of several of the most important regulators of the mesoderm progenitor molecular network. We further identified a role for Bmp in preventing the aberrant activation of differentiation markers of mesoderm derivative tissues (muscle, heart, blood, bone and vasculature cell lineages). These experiments identify Bmp as a key regulator of mesoderm progenitor homeostasis in the tailbud.

The process of somite formation from mesoderm progenitor cells has been studied extensively (Bénazéraf et al., 2010; Neijts et al., 2014; Wilson et al., 2009). The gene regulatory network underlying this process relies on complex signaling and transcriptional events. The apex of this network includes Wnt signaling (e.g. Wnt5a and Wnt3a), the transcription factor T and Fgf signaling (e.g. Fgf4 and Fgf8), which enter a mutual regulatory loop and regulate downstream patterning modules such as the establishment of the Notch-mediated segmentation clock, the repression of the retinoic acid differentiation signal and the activation of caudal Hox and Cdx patterning genes (Aulehla and Pourquie, 2010; Deschamps and van Nes, 2005; Hubaud and Pourquié, 2014; Wilson et al., 2009). In the mouse, very little is known about the role of Bmp signaling in mesoderm progenitor specification and maintenance (McMahon et al., 1998; Wijgerde et al., 2005). This is largely due to the crucial role of Bmp4 at an early stage of gastrulation (Winnier et al., 1995). In zebrafish, constitutive activation of the Bmp receptor Bmpr1b in the tailbud leads to an expansion of the T-positive mesoderm progenitor zone (Row and Kimelman, 2009), while transgenic expression of a dominant-negative version of the same receptor leads to ectopic tail formation (Pyati et al., 2005).

Here, we used a primary culture system to assess the role of signaling molecules in mesoderm progenitor maintenance in the mouse tailbud. Mesoderm progenitor cells in culture rapidly lose their identity and spontaneously upregulate aberrant cell differentiation markers. Supplementing with Bmp was sufficient to reactivate the entire mesoderm progenitor gene network within 5 h after treatment. This rapid response suggests that Bmp signaling regulates most genes in the network either directly or early in the transcriptional cascade. Specifically, we found that key network regulators such as T, Nkx1-2, Msgn1 and Cdx4 are either fully or partially responsive to Bmp signaling in the absence of protein translation, indicating a direct transcriptional response to Bmp signaling. Of interest, Bmp signaling and T have been found to genetically cooperate in tailbud progenitor cell differentiation (O'Neill and Thorpe, 2013). T was also reported to interact with the Bmp signaling molecule Smad1 (Messenger et al., 2005). Hence, the immediate response gene T might cooperate with Bmp-Smad signaling to promote the full mesoderm progenitor transcriptional response.

The capacity of Bmp4 to activate the mesoderm progenitor cell fate appears to act via a paracrine signaling mechanism. Plating mesoderm progenitor cells at high density in Matrigel maintained reporter gene expression in a Bmp-dependent manner, as shown by noggin treatment. The fact that this response was lost at lower density argues against an autocrine mechanism. Interestingly, Bmp4 is expressed in the ventral tailbud mesoderm, while most mesoderm progenitor regulatory genes are expressed in the more dorsal region (Inman and Downs, 2006; McMahon et al., 1998; Yoon et al., 2000). Together, these observations suggest a model by which ventral Bmp sustains a dorsal mesoderm progenitor cell fate. Pax2-GFP was expressed in both the ventral Bmp-expressing cells and the dorsal PSM progenitor cells, allowing us to observe the interaction between both compartments.

The fact that our analyses did not identify an overt neural or endodermal signature between Bmp-treated and untreated cells might reflect a more prominent role of Bmp in activating the mesoderm progenitor response. Alternatively, it might reflect the origin of the cells isolated on the basis of Pax2-GFP expression, or a differential maintenance of certain cell types in our culture conditions. In either case, the strong and widespread mesodermal response of Pax2-GFP cells to Bmp treatment unequivocally identifies an important role for this pathway in the maintenance of mesoderm progenitor cell fate.

Our data suggest that Fgf4 and Fgf8 have differential roles during mesoderm progenitor maintenance. Whereas Fgf8 had a positive effect on the global response to Bmp signaling, Fgf4 negatively affected this response for a large proportion of Bmp-responsive genes. Such opposing roles for Fgf4 and Fgf8 have been observed in mesoderm progenitor cell migration (Yang et al., 2002). The opposite roles of Fgf4 and Fgf8 in Bmp4 signaling and mesoderm progenitor migration contrast with the functional redundancy observed in mesoderm patterning (Naiche et al., 2011; Boulet and Capecchi, 2012). Whereas single-knockout embryos form normal somites, Fgf4;Fgf8 double-knockout embryos show important axial elongation and somitogenesis defects accompanied by premature differentiation of PSM cells toward the somitic fate (Naiche et al., 2011; Boulet and Capecchi, 2012). Hence, it is possible that Fgf4 has a dual function during mesodermal cell maturation. It might act redundantly with Fgf8 in the PSM region, and contribute to the termination of Bmp signaling in the more anterior paraxial mesoderm region. The expression of Fgf4 in the primitive streak region, anterior of Fgf8 expression (Welsh and O'Brien, 2000; Boulet and Capecchi, 2012), is compatible with this possibility. However, a detailed understanding of the activity of individual Fgfs in tuning the Bmp transcriptional response requires further experimentation.

A striking observation we made is that the loss of mesoderm progenitor cell identity in vitro was accompanied by the upregulation of distinct gene signatures of muscle, vasculature, bone, heart and blood cell fates. Conversely, mesoderm progenitor cell fate reactivation upon Bmp4 treatment was accompanied by a sharp downregulation of these gene signatures. It is interesting to note that all of the aberrant gene signatures are mesoderm derivatives of the paraxial (muscle and bone/cartilage) and lateral plate splanchnic mesoderm (heart, vasculature and blood). The muscle and bone/cartilage gene signatures are likely to reflect a premature differentiation of mesoderm progenitor cells into somitic derivatives. This is in line with the observation that Bmp signaling needs to be downregulated for proper somite formation (O'Neill and Thorpe, 2013). The dynamic expression pattern of noggin in the tailbud is also consistent with such regulation of progenitor differentiation (Goldman et al., 2000; Ohta et al., 2007). In the mouse, noggin-deficient embryos show an increase in Bmp4 expression associated with posterior somite patterning defects, which can be rescued by lowering Bmp4 gene dosage (Wijgerde et al., 2005). In zebrafish embryos, Bmp downre­­gulation by specific inhibitors was also required for upregulation of the myogenic program (Reshef et al., 1998; Row and Kimelman, 2009). Hence, the loss of Bmp signaling in mesoderm progenitor cells seems sufficient to trigger part of the differentiation program associated with somite formation.

The distinct upregulation of heart, vasculature and blood gene signatures was more surprising as they are derivatives of another mesoderm compartment, namely the lateral plate mesoderm (Lopez-Sanchez et al., 2001; Psychoyos and Stern, 1996). Hence, Bmp signaling appears to maintain a competence of progenitor cells to derive paraxial mesoderm. In the absence of Bmp signal these cells upregulate other mesoderm markers in an aberrant manner, such that these cells fail to progress towards defined cell fates and instead accumulate in the caudal trunk.

Together, our results point to a crucial role for Bmp signaling in the maintenance of the tailbud mesoderm progenitor cell population in the mouse and clarify the structure of the regulatory network underlying axial elongation.

Pax2-GFP BAC transgenic mice were described previously (Bouchard et al., 2005; Pfeffer et al., 2002). Mice were kept in the C57BL/6 genetic background and genotyped using primers listed in Table S17. Appropriate matings were performed and detection of copulatory plugs was designated embryonic day (E) 0.5. Retired breeder rats were used for serum extraction (Takahashi et al., 2014) and acquired from the McGill Comparative Medicine and Animal Resource Centre. All animal experiments were performed in accordance with the ethical guidelines provided by Canadian Council of Animal Care.

Pax2-GFP embryos were dissected at E9.25-E9.5 and GFP-expressing tailbud cells were separated by FACS analysis as follows. GFP-expressing tailbuds were dissected from the region of the trunk of the embryo caudal to the last somite, followed by trypsinization at 37°C for 15 min to obtain a single-cell suspension. FACS sorting was performed based on granularity and size (SSC, FSC), viability using propidium iodide (PI) (Sigma, 1 μg/ml), and GFP using a FACSAria (BD Biosciences). Sorted PI⁻/Pax2-GFP⁺ cells were cultured in medium containing DMEM with 10% FBS and 1% penicillin-streptomycin in the presence of mitomycin C-treated (Sigma, 10 µg/ml) mouse embryonic fibroblast (MEF) feeder layers, which were obtained from wild-type C57BL/6 mice. The MEFs were stained with PKH26 Red Fluorescence Cell Linker Dye (Sigma, 2 μM) so that they could be sorted out from the Pax2-GFP⁺ tailbud cells after co-culture. Co-cultures were incubated at 37°C in 5% CO₂ for 96 h before treatment with recombinant proteins/growth factors recombinant human (rh) BMP4 (Sigma and USCN, 100 ng/ml), rhFGF8 (Sigma, 100 ng/ml), rhFGF4 (Sigma, 100 ng/ml) and/or recombinant mouse (rm) Wnt3a (Sigma, 100 ng/ml). For cells treated with cycloheximide (Sigma, 1 μg/ml), the treatment was undertaken simultaneously with the growth factor timecourse. Pax2-GFP⁺ cells were plated in 50:50 DMEM medium and Matrigel (Sigma) at different cell densities in the absence or presence of rmNoggin (Cedarlane, 100 ng/ml). Cell fluorescence and proliferation were measured every 3 h in IncuCyte-FLR imaging system (Essen BioScience). The average fluorescence intensity of Pax2-GFP⁺ tailbud cells at 0 h was set to 100% across different treatments and measured as arbitrary units (A.U.) over time.

Total RNA was obtained using the RNeasy Micro Kit (Qiagen) following the manufacturer's protocol. Mesoderm progenitor cells from Pax2-GFP animals (GFP-expressing or not) were sorted out from PKH26-stained MEFs by FACS as described above. Sequencing libraries were prepared by Genome Quebec Innovation Centre (Montreal, Canada) using the TruSeq Stranded Total RNA Sample Preparation Kit (Illumina, TS-122-2301) by depleting ribosomal and fragmented RNA, synthesizing first- and second-strand cDNA, adenylating the 3′ ends and ligating adaptors, and enriching the adaptor-containing cDNA strands by PCR. The libraries were sequenced using an Illumina HiSeq 2000 sequencer, with 100 nucleotide paired-end reads, generating ∼60 million reads per sample. The sequencing reads were trimmed using CutAdapt (Martin, 2011) and mapped to the mouse reference genome (mm10) using STAR aligner (version 2.4.0e) (Dobin et al., 2013), with default parameters, and annotated using the Gencode M2 (version M2, 2013) annotation (Mudge and Harrow, 2015). Htseq-counts [part of the HTSeq (Anders et al., 2015) framework, version 0.5.4p5] was used for expression quantification, and DESeq (Anders and Huber, 2010) for differential expression analysis. Heatmaps and scatter plots were generated using R (version 3.3.1). The Pax2-GFP^on gene signature consisted of genes showing significantly higher expression in any Pax2-GFP^on cells than in untreated cells at 96 h. The Pax2-GFP^off gene signature consisted of genes showing significantly higher expression in any Pax2-GFP^off cells than in untreated cells at 0 h.

Total RNA from cultured tailbud cells was extracted using the RNeasy Mini Kit (Qiagen) and reverse transcribed with MMLV (Invitrogen) according to manufacturers' procedures. qRT-PCR was performed using Green-2-Go Mastermix (BioBasic) on a Realplex2 Mastercycler (Eppendorf). Primers are listed in Table S17. All samples were obtained in technical triplicates, and transcript levels were standardized using B2m. Biological triplicates were run for both control and treated samples in all the experiments and expression levels were compared using the 2^−ΔΔCT method (Pfaffl, 2001). One-way ANOVA or unpaired two-tailed Student's t-tests were performed using Prism software (GraphPad). Data are presented as mean±s.d.; P<0.05 was considered statistically significant.

Ex utero live embryo cultures were established as previously described (Gray and Ross, 2011) using E8.75 Pax2-GFP embryos. Briefly, embryos were dissected out of the uterus and carefully separated from the decidua, keeping the yolk sac in place, and kept in DMEM medium with 10% FBS and antibiotics. Both vehicle-treated and rmNoggin-treated (Cedarlane,100 ng/ml) embryos were cultured at 37°C and 5% CO₂ in DMEM with 50% rat serum in glass bottles on a roller apparatus (LabQuake).

Immunofluorescence staining was performed as described previously (Stewart et al., 2013). Tailbud culture cells were plated on glass slides coated with Geltrex (Gibco). E8.75-E9.5 embryos were mounted in Optimal Cutting Temperature (OCT) compound and sectioned to obtain 10-12 μm sections. Antibodies and dilutions used are listed in Table S18.

Tissues for in situ hybridization were fixed in 4% paraformaldehyde, passed through a sucrose gradient and then embedded in OCT and sectioned at 12 μm. RNA probes for T (Herrmann, 1991) and Bmp4 (Wall and Hogan, 1995) were synthesized using T7 or SP6 RNA polymerase following the manufacturer's specifications (Roche) for in situ hybridization.

Bright-field whole-mount images were acquired with a Stemi 2000-C microscope (Zeiss) and sections were imaged using an Axioplan 2 microscope (Zeiss). Immunofluorescence images were acquired using either an Observer.Z1 (Zeiss) or an LSM3, LSM710 or LSM800 confocal microscope (Zeiss) at the Advanced BioImaging Facility of McGill University. Quantification and image analysis were performed using Zen (Zeiss) and Fiji (ImageJ) software.

We are grateful to members of the M.B. laboratory and Dr Yojiro Yamanaka for critical reading of the manuscript. We thank the McGill Advanced BioImaging Facility (ABIF) for technical support with confocal microscopy, and the Flow Cytometry Platform of the McGill University Life Sciences Complex for Cell Sorting.