Amniotic ectoderm expansion in mouse occurs via distinct modes and requires SMAD5-mediated signalling

Animals classified as amniotes, including reptiles, birds and mammals, develop extra-embryonic tissues and organs that perform respiration-related, nutritional and protective functions during embryogenesis: the chorion, yolk sac, amnion and allantois. Alongside their supportive role, extra-embryonic tissues take part in the development of the early embryo proper by providing signalling cues and spatial continuity (Horn and Panfilio, 2016; Mesnard and Constam, 2010; Shao et al., 2016). The amnion is the innermost extra-embryonic tissue that delineates the fluid-filled amniotic cavity, providing space for the embryo to move freely. The amnion protects the embryo against traumas, infections and toxins (Schmidt, 1992). Despite its location at the extra-embryonic–embryonic interface, the amnion has classically not been considered a driving force in shaping the early embryo proper. However, in mice, the expansion of the amniotic cavity is crucial for the reorientation of notochord cells along the anterior-posterior axis and for the global morphogenesis of the early embryo (Imuta et al., 2014). Furthermore, in chickens the physical expansion of (pro)amnion is required for the correct positioning of the head (de Melo Bernardo and Chuva de Sousa Lopes, 2014). In humans, impaired amnion integrity, premature rupture of amnion and amniotic bands often cause preterm birth and fetal malformations, including amputations and fetal compression syndromes (Menon and Richardson, 2017; Opitz et al., 2015). Despite the importance of amnion, its developmental origin and the molecular regulation of amniogenesis remain ill-defined.

In the mouse, amnion and chorion development are intimately related. Early during gastrulation, at embryonic day (E) 6.5-7.0, extra-embryonic mesoderm emerging from the most posterior part of the primitive streak (further called ‘streak’) accumulates in the incipient fold at the posterior extra-embryonic–embryonic junction. This amniochorionic fold gives rise to both amnion and chorion, but initially consists only of (prechorionic) extra-embryonic ectoderm. The epiblast-derived amniotic ectoderm contributes relatively late, and only to the floor of the fold. The cuboidal chorionic ectoderm thus greatly exceeds the squamous amniotic ectoderm portion of the fold at all stages (Pereira et al., 2011). The extra-embryonic mesodermal layer, still shared by the future amnion and chorion, develops from the accumulating posterior mesoderm and lines the expanding exocoelomic cavity. When the fold is fully expanded, the chorionic walls of the proamniotic canal fuse at the extra-embryonic–embryonic junction at the embryo's anterior side, the anterior separation point (Pereira et al., 2011). This causes amnion-chorion separation, and partition of the proamniotic cavity into the amniotic, exocoelomic and ectoplacental cavities at E7.5. Shortly after amnion-chorion separation, foregut invagination begins, marking the onset of ventral folding morphogenesis. The flexible amniotic membrane envelops the entire embryo only on completion of ventrolateral morphogenesis and body wall closure (Schmidt, 1992).

Several studies indicate that bone morphogenetic protein (BMP) signalling at the extra-embryonic–embryonic interface is important for amnion-chorion separation and amnion development in the mouse (reviewed in Pereira et al., 2011). Binding of BMPs to their receptors triggers activation by phosphorylation of SMAD1, SMAD5 and SMAD9, which co-regulate BMP-responsive gene expression (Katagiri and Watabe, 2016). SMAD5-deficient embryos develop severe morphogenesis defects (Chang et al., 1999; Yang et al., 1999), amnion-chorion separation is delayed and the anterior separation point is shifted posteriorly (Bosman et al., 2006). After closure of the proamniotic canal, a striking aggregate of cells develops anteriorly in the amnion (Chang et al., 1999). These aggregates ectopically express the mesoderm inducer Nodal, in addition to T, Fgf8 and Eomes, all established streak mesoderm markers (Pereira et al., 2012). The aggregates are variable in size, position and relative composition at E8.5-E9.0. In contrast to the avascular wild-type (WT) amnion, the aggregates become vascularized and develop red blood cells and primordial germ cell (PGC)-like cells (Bosman et al., 2006; Chang et al., 1999), all posterior streak derivatives. BMP-SMAD signalling is active during amniogenesis as shown in a BRE::LacZ reporter mouse that monitors transcriptional activity of phosphorylated BMP-SMAD proteins (Cajal et al., 2014; Dobreva et al., 2012). Reporter activity is present in the amniotic ectoderm component of the amniochorionic fold, the separated amnion, the anterior midline region distal to the extra-embryonic/embryonic junction and, from amnion-chorion separation onwards, in the lateral junctions of amnion with surface ectoderm. In other species, BMP-SMAD signalling is also crucial for amnion-like tissue development, e.g. in amniogenesis in human pluripotent stem cell cultures (Shao et al., 2016), and in defining amnion and serosa or amnioserosa in different insect species (Rafiqi et al., 2012; Horn and Panfilio, 2016). Furthermore, prominent amniotic expression of BMP4 has been reported in cynomolgus monkeys (Sasaki et al., 2016).

Little is known about the developmental origin of the amnion, particularly the amniotic ectoderm (Kinder et al., 2001; Lawson et al., 1991). To remedy this, we made a prospective lineage analysis by labelling single epiblast cells (Lawson et al., 1991) in the proximal half of WT mouse embryos before and early in gastrulation (E6.2-E6.5). We identified four distinct types of clonal growth pattern that expand amniotic ectoderm differently and with different magnitude. Additionally, we investigated the earliest steps of abnormal amniogenesis in mice deficient in SMAD5 by sequencing the mRNA (RNA-seq) of individual control and mutant amnion samples, and by chimera analysis. Correlating these data with the clonal analysis supports the conclusion that the impaired amniogenesis in Smad5 mutants results from the two main cell populations in amniotic ectoderm being affected by the SMAD5 deficiency.

To gain insight into the location of amnion progenitors and to trace their contribution in amniotic ectoderm, we labelled single epiblast cells in the proximal half of WT embryos at prestreak (PS) and early-streak (ES) stages and analysed the resulting clones. Two properties of single-cell labelling followed by short-term retrospective lineage analysis were exploited in this approach. First, labelling by iontophoretic injection enables analysis of clones with known spatial and temporal origins. It gives fine-resolution fate maps and the opportunity to reconstruct morphogenetic change from patterns of clonal expansion during a period of rapid growth. Second, lineage analysis of prospectively labelled clones allows the clonal history to be inferred within the cell layer of origin. Thus, the time trajectory within the amniotic ectoderm, which is continuous with the epiblast within which the clone arose, can be traced with confidence retrospectively.

The fate maps of PS and ES stages show that the amniotic ectoderm is derived mainly from mixed fate progenitors in the proximal anterior and anterolateral epiblast, whereas amniotic mesoderm progenitors, also of mixed fate, lie more posteriorly and distally. These findings are graphically displayed in Fig. 1. From the triangular-shaped region spanning the anterior midline that generated amniotic ectoderm, only a minority of labelled descendants (8%, 245/3032 from 61 clones) were in amniotic ectoderm. Not all epiblast cells in this region gave rise to amniotic ectoderm, but the composition of the clones was otherwise indistinguishable from that of their amnion-contributing neighbours. This showed an anterior-posterior transition in the fate map from predominantly surface ectoderm to extra-embryonic mesoderm (Fig. 1, comparing upper and lower panels). The amniotic ectoderm therefore develops from a relatively small number of descendants of proximal-anterior and anterolateral epiblast that, before and at the onset of gastrulation, is mainly prospective surface ectoderm and extra-embryonic mesoderm.

The detailed findings from the 28 clones that contributed to amniotic ectoderm are presented in Fig. 2, Table 1 and Fig. S1. Overall, four clone types were distinguished on the basis of their distribution within the amniotic ectoderm (Fig. 2) and clone composition (Table 1).

Type I clones originated in a very restricted region of extreme proximal anterior/anterolateral epiblast, abutting the extra-embryonic ectoderm: they were mixed clones with descendants mainly in extra-embryonic structures, posterior streak and PGCs (Table 1). Type I clones were found at PS , but not at ES, stage (see Fig. S2 for possible explanations). These clones spread in the amniotic ectoderm from the anterior separation point to the posterior edge of the amnion, indicating that they contributed to the amnion from its inception: the position of the most anterior cells of these clones reflected the origin of the progenitors in association with presumptive chorionic ectoderm. The first amniotic cell of the clone would have been an extreme posterior (proximal) epiblast or posterior streak cell in contact with the floor of the early amniochorionic fold; it then divided, leaving one daughter associated with the prechorionic ectoderm, the other with the epiblast/streak. Succeeding divisions expanded the amniotic portion of the fold. Clones #1 and #2 must have gone through two cell generations in the epiblast before contributing one cell (clone #1), or one followed by two (clone #2), to the emergent amnion between E6.8 and E7.1 (Fig. S1A) at midstreak (MS) to late-streak (LS) stages [see Fig. 3.5 of Lawson and Wilson (2015) for correlation of age and stage]. Subsequently, the clones expanded through the following three or two generations, respectively (Fig. S1A). The distribution pattern of Type I clones indicates that these clones represent the founding population of the amniotic ectoderm until amnion closure.

Type II clones originated over a larger area of proximal-anterior epiblast than Type I. These mixed clones contributed mainly to extra-embryonic mesoderm, posterior streak and PGCs with generally only a small minority of descendants in amniotic ectoderm (Table 1). The contribution was limited to posterior amnion (Fig. 2), and was slightly later than Type I and over a longer period (E7.2-E8.1) (Fig. S1B). Hence, Type II descendants moved into the established primordium and remained as short files that had not extended further than 15% of the anterior-posterior length of the amnion by early somite stages (Fig. 2). At the LS stage, the amnion ectoderm consists of only about 20 cells [EMA10 eMouseAtlas (Armit et al., 2017)] and, given the behaviour of Type II clones, the founding population associated with the chorionic ectoderm is probably complete by this time. Therefore, the founding population of the amniotic ectoderm appears to be both small and established within a narrow time window.

Type III clones originated in the proximal quadrant of epiblast spanning the anterior midline i.e. in prospective surface ectoderm, which was the main clonal component. Single, small inclusions of labelled cells in the periphery of the amnion were found between the axial level of the midbrain and the node (Fig. 2). The size of the inclusion expanded with the rest of the clone. The inferred clonal histories (Fig. S1C) indicate that the inclusion of two cells would have occurred between E7.3 and E7.7 [expected early preheadfold (EPHF) to late headfold (LHF)], i.e. in the period of rapid amnion expansion during and after closure, onset of axial elongation from the node, expansion of the neural folds and initiation of the foregut. This suggests that inclusion into the periphery of the amnion from embryonic ectoderm is usually limited to a period when considerable adjustments to turgor in the amniotic and yolk sac cavities occur, and when the rostral half of the embryo is growing rapidly.

Type IV clones originated in extreme proximal anterior to anterolateral epiblast, spanning the midline. They were generated between ES+ and MS stages within the same area as the earlier Type I clones, but were exceptional in being lineage restricted with all descendants in amniotic ectoderm (Table 1, Fig. 2). The pattern of clonal spread at late prehead fold (LPHF)/early headfold (EHF) indicated that early descendants entered the amnion at, or close to, the anterior midline. They not only expanded the amniotic ectoderm anterior to the anterior separation point, but also contributed to the main body of amniotic ectoderm laid down by the original Type I founding population (Fig. 3A,B). Descendants in somite-stage embryos were found concentrated mainly between the axial levels of heart and hindbrain.

An interesting mixed type clone (#28) contributed to the amnion from both anterior and posterior. The inferred history (Fig. S1D) indicated that the ES progenitor would have divided twice before one cell became restricted to amniotic ectoderm (behaving further as Type IV), and the remaining three cells behaved as Type II, contributing further descendants to posterior amnion, posterior streak and allantois. Together, these results indicate that extreme proximal descendants of PS- or ES-stage epiblast that remain anterior become part of a relatively late population of exclusively amniotic ectoderm precursors that expand the amnion specifically from the anterior.

In brief, Type I descendants initiate amniotic ectoderm. Until closure of the proamniotic canal, Type I and II descendants, both from mixed fate progenitors, expand the amniotic ectoderm. Around the time of proamniotic canal closure and after amnion-chorion separation, the amniotic ectoderm becomes extensively expanded directly from the anterior by the late-contributing, lineage-restricted Type IV descendants. Type III descendants expand the amnion only at the periphery. The behaviour and timing of the different groups that constitute the amniotic ectoderm are summarized in Figs 3B and 8A,B.

In normal early somite-stage mouse conceptuses, amnion appears as a translucent and stretched tissue. In contrast, most Smad5 knockouts (KOs) are distinguished by the presence of a variable aggregate of cells in the amnion. Additionally, mutants have various ventral folding defects (Chang et al., 1999) (Fig. 4A). Amniotic ectoderm is cuboidal before aggregate formation, in contrast to the squamous control amnion, and the junction between amniotic and embryonic ectoderm is anteriorly less pronounced in all Smad5 mutants (Fig. 4B).

We investigated the molecular differences between WT and Smad5 mutant amnion by RNA-seq analysis. Amnions were microdissected (Fig. 4C,D) from control (n=6), and stage-matched littermate Smad5 knockout (n=6) embryos ranging between EPHF and three-somite (3S) stages (Table S1), before major morphological differences become apparent. We synthesized and amplified cDNA from individual amnion samples. Absence of contamination with surrounding tissues was confirmed by quantitative reverse-transcription PCR (RT-qPCR) marker analysis in control samples (Fig. S3A).

The variance-normalized counts of the RNA-seq data (Table S1) were used for expression visualization, clustering and differential gene expression analysis. Unsupervised clustering of all 12 samples by principal component analysis (PCA) resulted in clustering of five of six control samples and the segregation of the six knockout samples into two distinct groups, which we named KO-SetA and KO-SetB (Fig. 5A). Ctrl6.1 was determined to be an outlier and omitted from further analysis. Gene set enrichment analysis (GSEA) of the top differentially expressed genes (adjusted P-value <0.05) showed that KO-SetA samples were significantly enriched for categories such as ‘gastrulation’, ‘mesoderm development’ and ‘cell fate specification’, whereas gene sets like ‘integrin binding’ and ‘regulation of cell adhesion’ were under-represented (Fig. S3B,C, Table S2, Supplementary Materials and Methods). Conversely, KO-SetB samples were enriched in categories such as ‘placenta development’ and ‘extra-embryonic ectoderm’, whereas ‘extracellular matrix’ and ‘epithelial-to-mesenchymal transition (EMT)’ sets were downregulated.

Closer examination of samples with the KO-SetA signature (Fig. 5B,C, Tables S3 and S4) confirmed and expanded our former in situ hybridization results on ectopic expression of Nodal, T, Lefty2, Wnt3 and Fgf8 in Smad5 mutant amnion (Pereira et al., 2012). Furthermore, KO-SetA samples significantly overexpressed several markers related to EMT, streak and streak-derived mesoderm (Fig. 5B,C, Table S3). They include the streak mesoderm-related Fst (57-fold increase in KO-SetA compared with control), Zic3 (79-fold); the Nodal co-receptor Tdgf1 or Cripto (22-fold); and the Nodal target genes Fgf5 (71-fold), Mixl1 (14-fold) and Mesp1 (18-fold). Remarkably, several extra-embryonic ectoderm markers were robustly enriched in samples with KO-SetB signature (Fig. 5B,C, Tables S3 and S4). These included genes such as Elf5 (391-fold more in KO-SetB compared with control), Esrrb (202-fold), Sox2 (15-fold), Sox21 (107-fold) and Cldn3 (146-fold).

The two KO sets shared some features. Under-representation of genes encoding ECM components, such as collagens, and downregulation of amnion-enriched transcripts like Postn (Dobreva et al., 2012), Tfap2a and Twist1 (Bosman et al., 2006) and Flrt3 (Egea et al., 2008), suggested loss of amnion identity (Fig. 5C, Fig. S3B,C, Table S3 and S4). Eomes – a streak marker and target gene of Nodal signalling (Arnold et al., 2008), and also a marker for extra-embryonic ectoderm (Ciruna and Rossant, 1999) – was found enriched in both KO sets. Target genes of BMP signalling, such as Smoc2, Postn (Inai et al., 2008) and Flrt3 (Tomás et al., 2011), were under-represented in both KO sets (Table S3); however, differential expression of other target genes of BMP/SMAD5, BMP ligands and modulators of BMP signalling were specifically observed in KO-SetB samples, e.g. Msx1, Hand1 and noggin (Nog).

The distribution of KO-SetA and KO-SetB expression profiles was validated by RT-qPCR in a larger cohort of 25 E7.5 littermate control-knockout amnion sample pairs. We selected four KO-SetA (Fgf5, Nodal, T and Lefty2) and four KO-SetB (Elf5, Essrb, Cldn4 and Sox2) genes enriched in one but not the other set (Fig. 5C, Table S5). One of 25 pairs showed differential expression of only one of the selected markers, and was therefore not categorized. The majority of the pairs (64%) segregated either into a KO-SetA (48%) or KO-SetB (16%) signature (Fig. 5D). Several pairs (32%) showed an intermediate signature with two or more markers (over)expressed from each set. Collectively, the data demonstrate the presence of two transcriptionally distinct signatures in the amnion of Smad5 mutants. The high prevalence of the intermediate signature suggests transition from a KO-SetB to a KO-SetA category (progressive loss of ExEc inclusion features because of ExEc transdifferentiation into an aggregate), or co-existence of mixed KO-SetA and KO-SetB phenotypes.

Subsequently, we documented where the top overexpressed markers of KO-SetA and KO-SetB localize in mutant amnion. Ectopic T, Nodal and Wnt3 expression, and Nodal- and β-catenin-mediated signalling in amniotic ectoderm of Smad5 mutants was previously documented (Pereira et al., 2012). Here, we show the expression of a new SetA marker, Fgf5. Fgf5 is normally expressed in the epiblast at ES stage, including in the region in which amniotic ectoderm precursors reside. Its expression is lost anteriorly in the late streak early allantoic bud (LSEB) and Fgf5 is absent in normal amnion (Li et al., 2013). We found Fgf5 expressed ectopically in nine of ten Smad5 EPHF-EHF embryos. The ectopic expression was prevalent in the anterior embryonic ectoderm and continuous with the anterior amniotic ectoderm, consistent with the continuous cuboidal ectoderm shown in Fig. 4B. Its range of extension into amniotic ectoderm was variable, but could reach as far posteriorly as the abnormally positioned anterior separation point (Fig. 6A).

Next, we assessed the localization of a KO-SetB marker, Elf5, an extra-embryonic ectoderm-specific transcription factor (Donnison et al., 2005) and Oct4/Pou5f1, normally detected in amniotic ectoderm at EPHF stages (Downs, 2008). In E7.5 control embryos, amniotic ectoderm was Oct4⁺ and Elf5^–, while extra-embryonic ectoderm showed Elf5 presence but no Oct4 (Fig. 6B, Ctrl). This pattern was present in the majority of the mutants (20/25) (Fig. 6B, KO-SetA). However, in a fraction of the mutants (5/25), a portion of ectoderm, continuous with the amniotic ectoderm, was strongly Elf5⁺ and Oct4^– (Fig. 6B, KO-SetB). This pattern did not correlate with a specific developmental stage (data not shown). The ectopic presence of Elf5, but also of Eomes (Fig. S4), in amniotic ectoderm in some mutants is suggestive of inclusion of extra-embryonic ectoderm, although de novo transcript induction could not be excluded at this stage.

Based on (1) the delayed closure of the proamniotic canal and amnion-chorion separation in Smad5 knockout embryos (Fig. 6); (2) the two different KO set signatures (Fig. 5); and (3) the robust expression of extra-embryonic ectoderm enriched transcripts in the KO-SetB samples (Fig. 5), we reasoned that a deficit in the amniotic ectoderm could result in compensatory inclusion of nonamniotic epiblast tissue and/or trailing extra-embryonic ectoderm into the mutant amnion.

Tetraploid (4n) complementation assays were performed to determine whether an extra-embryonic ectoderm inclusion occurs in the mutant amnion, and whether such inclusions transdifferentiate into aggregates (Fig. 7A). Donor embryonic stem cells (ESCs) in ESC↔4n embryo chimeras normally give rise to all epiblast-derived tissues, including the extra-embryonic mesoderm and amniotic ectoderm. Conversely, tetraploid descendants are restricted to trophectoderm derivatives, including chorionic extra-embryonic ectoderm and extra-embryonic endoderm lineages. The segregation of descendants of ESCs and tetraploid cells in a chimera thus enables a distinction to be made between cells of epiblast or extra-embryonic ectoderm origin. GFP⁺ WT and Smad5 knockout ESCs (Bosman et al., 2006) were aggregated with GFP⁻ tetraploid embryos (Fig. 7A).

Amniotic ectoderm and amniotic mesoderm were indeed GFP⁺ in control chimeras (WT;GFP⁺ ESC↔4n WT) (Fig. 7B, Fig. S5A). The majority of the knockout chimeras (Table 2) had uniformly GFP⁺ amniotic ectoderm, like the other epiblast-derived tissues. In most of the E7.5 knockouts (16/21), much of the amniotic ectoderm was morphologically indistinct from the anterior embryonic ectoderm (Fig. 7C,D, Fig. S5B). Many of these embryos (Table 2) displayed a ‘head-out’ phenotype, with the anterior epiblast/future head region protruding outside the visceral yolk sac (Fig. 7C, Fig. S5E) as if anterior amniotic ectoderm was lacking. In a few KO chimeras (3/35, 9%) the GFP⁺ SMAD5-deficient posterior amniotic ectoderm appeared undersized and stretches of tetraploid (GFP⁻) extra-embryonic ectoderm were trapped in the amnion (Fig. 7E, Fig. S5G-H).

While nearly all conventional Smad5 knockout embryos develop an anterior aggregate in the amnion beyond the four-somite stage (Pereira et al., 2012), only 1/14 E8.5 knockout chimeras had an anterior aggregate (Fig. 7D). A more posterior thickening was observed in 5/14 knockout chimeras (Fig. S5C,F). Amniotic aggregates were always GFP⁺, confirming their embryonic origin. Therefore, Elf5⁺ extra-embryonic ectoderm inclusion is not a transient state of all Smad5 mutants, but represents a possible alternative consequence of a posterior amniotic ectoderm deficit in the absence of SMAD5.

An amniotic ectoderm deficit could result from alterations in apoptosis or cell proliferation. TUNEL analysis revealed little cell death in mutant or control conceptuses with no evidence of increased apoptosis in mutants (Fig. S6), in agreement with apoptosis-related gene sets being unaltered in mutants in GSEA. Quantification of proliferation in the ectoderm of the amniochorionic fold by phospho-histone 3 (PH3) detection revealed no significant difference in mitotic cells between control and knockout embryos (Fig. S7A,B).

Altered cell shape, ECM assembly and/or cell-cell and cell-matrix interactions could affect the squamous architecture and expansion of the amniotic ectoderm in mutants. GSEA analysis confirmed robust downregulation in ‘collagen and ECM’-related gene sets in mutants (Fig. S3B,C, Table S2). This differential expression could be caused by absence of SMAD5 in amniotic ectoderm, but could also reflect the relative under-representation of amniotic ectoderm over trapped nonamniotic tissue. We examined the spatial protein distribution of the ECM component collagen IV (ColIV). Normally, the ColIV matrix intertwined between amniotic ectoderm and mesoderm cells at E7.5 and E8.5 (Fig. S7A,B). In mutants, ColIV deposition seemed unaffected in regions in which the amnion was a smooth, thin bilayer. ColIV was, however, absent from ectoderm in regions with thickened amniotic ectoderm or an aggregate, and seemed only anchored by amniotic mesoderm (Fig. S7A,B). Postn, an ECM protein that directly interacts with fibronectin, collagen I (ColI) and ColIV (Conway et al., 2011; Horiuchi et al., 1999), had a similar distribution to ColIV in mutants (Fig. S7C). The differential presence of collagens and Postn in the amniotic aggregate supported amnion identity being partially affected. This might result from altered properties of the anterior amniotic ectoderm in controlling shape and stretchability (anterior amniotic ectoderm misdifferentiation), or inclusion of nonamniotic epiblast-derived tissue (epiblast inclusion).

Clonal analysis of normal amnion development showed that only a few descendants of some cells in a restricted area of the early proximal anterolateral epiblast contribute to amniotic ectoderm development. These descendants expand the amnion in distinct modes. Type I cells initiate amniotic ectoderm development and, together with Type II cells, expand this tissue from the posterior (Fig. 8A,B). Both originate in the epiblast adjacent to the extra-embryonic ectoderm, which is a potent source of BMP4 and BMP8B (Ying et al., 2000; Lawson et al., 1999), in a region that contributes to allantois, PGCs, extra-embryonic mesoderm and posterior embryonic mesoderm (Fig. 1; Lawson and Hage, 1994). Type I and II progenitors have co-descendants in exactly these posterior tissues, and all these tissues are delayed and reduced in Smad5 mutants (Bosman et al., 2006). Type III clones contribute facultatively, from the periphery, to amniotic ectoderm, and are derived from a region (presumptive surface ectoderm) that requires BMP signalling (Madabhushi and Lacy, 2011). Strikingly, Type IV progenitors reside in the epiblast spanning the anterior midline of the extra-embryonic–embryonic junction at ES+/MS, an area that is then precisely overlain by a restricted source of BMP2 in the anterior visceral endoderm (Madabhushi and Lacy, 2011). Overall, the identified progenitor regions are in close proximity to BMP sources, the relationship being closest for the very restricted, sequentially present Type I and IV progenitors that generate the bulk of the amniotic ectoderm.

When relating the clonal analysis results to the Smad5 mutant embryo analyses (Fig. 8B-D), the consistently delayed and posteriorized proamniotic canal closure is compatible with a posterior amniotic ectoderm deficiency resulting from undersized or delayed BMP4-dependent amniotic ectoderm founding population(s) [Type I (and II)]. The inclusion of chorionic extra-embryonic ectoderm in the amniotic environment of KO-SetB and some tetraploid chimeras showed that the trailing chorionic portion of the amniochorionic fold might become trapped in the amnion at proamniotic canal closure (Fig. 8C).

In the majority of the mutants, the KO-SetA mutants, the phenotype seems to implicate primarily the Type IV group (Fig. 8D). In these mutants, the anterior extra-embryonic–embryonic border is obscure, and the anterior ectoderm is nonsquamous. One possibility is that the Type IV progenitor population expands normally in the amniotic ectoderm territory but fails to acquire the typical expanded squamous morphology of amniotic epithelium. Alternatively, under-representation of Type IV descendants is compensated by inclusion of neighbouring anterior surface ectoderm into the amniotic environment. The RNA-seq analysis did not directly support such an event as there was no major differential expression of surface ectoderm markers, such as Dlx5 or Aldh1a3 (Qu et al., 2016); however, low abundant transcripts might have been under-represented in our study. Conversely, the persistent expression of the early epiblast marker Fgf5 in the anterior extra-embryonic–embryonic junction region of the mutants is suggestive of a stalled epiblast differentiation (atypical anterior ectoderm). In the absence of specific molecular markers of amniotic ectoderm in ES/MS embryos, the answer to whether the KO-SetA signature tissue is caused by misdifferentiated Type IV descendants or atypical surface ectoderm could potentially be obtained from prospective lineage analyses at ES+ to LS stages in Smad5 mutants.

Further support for SMAD5 being a major signal mediator of BMP2 in the proximal anterolateral region of the embryo comes from mice that lack Bmp2 ubiquitously or conditionally in the anterior visceral endoderm (Zhang and Bradley, 1996; Madabhushi and Lacy, 2011). Such embryos show defects that are reminiscent of and equally variable as those in Smad5 mutants (Bosman et al., 2006; Chang et al., 1999) and tetraploid chimeras; i.e. proamniotic canal closure defects, severely impaired anterior midline epiblast tissues and ventral folding morphogenesis defects. However, an aggregate with posterior streak mesoderm potency in the anterior part of the amnion seems to form exclusively in Smad5 mutants. This suggests that SMAD5 deficiency compromises additional BMP signalling pathways, e.g. BMP4, but potentially also BMP5 that is produced in embryonic mesoderm flanking the anterior extra-embryonic–embryonic interface (Solloway and Robertson, 1999). Alternatively, intracellular non-SMAD5 cascades could be disturbed. Indeed, in cell culture experiments, loss of SMAD5 can result in gain of Nodal-SMAD2/3 signalling (Pereira et al., 2012). Indications of such alterations in the Smad5 mutants are the robust ectopic expression of Nodal and gain of Nodal signalling (expression of Nodal-SMAD2/3 target genes Lefty2, Eomes, Cripto, Wnt3) in the nonsquamous cells in anterior amniotic ectoderm (Pereira et al., 2012) and in the KO-SetA RNA-seq results.

The chimera study showed that the amniotic aggregate is of epiblast origin, thus excluding that extra-embryonic ectoderm inclusions (KO-SetB) transdifferentiate into an aggregate with characteristics of posterior streak mesoderm (KO-SetA mutants). However, the intermediate expression of selected KO-SetB and KO-SetA markers in some mutant amnion samples (8/25) is compatible with ‘mixed’ or variable phenotypes. Whether a mutant develops into a SetA or SetB phenotype, or ‘anterior’ and ‘posterior’ amniotic ectoderm defects co-exist, might be a stochastic event. When the amniotic ectoderm does not stretch fast enough around the time of proamniotic canal closure, the mechanical forces at the extra-embryonic–embryonic interface that balance turgor in the exocoelom and (pro)amniotic cavities might ‘drag’ tissue from the conceptus into the amniotic ectoderm from wherever there is least resistance (as normally in Type III clone behaviour). Interestingly, an extra-embryonic ectoderm inclusion in the amnion has also been described for Pagr1a knockouts (Kumar et al., 2014). Pagr1a encodes a protein associated with PAX-interacting protein 1 (Paxip1) and is predominantly expressed in the chorion. Its absence results in downregulation of Bmp2 in the anterior extra-embryonic–embryonic junction.

The driving force behind the cuboidal-to-squamous transition of amniotic ectoderm still remains unclear. Interestingly, BMP-SMAD signalling has been reported as a driver of organogenesis by regulating cell shape and interpreting mechanical forces at the tissue level (Nerurkar et al., 2017; Poduri et al., 2017; Wang et al., 2012). It is likely that the increase in turgor associated with the dynamic expansion of the exocoelomic and amniotic cavities, the folding morphogenesis of the embryo following amnion-chorion separation, and/or alterations in cell-matrix and cell-cell interactions each contribute to significant changes in cytoskeletal rearrangements. Our study relates BMP-SMAD signalling to these events in the amniotic ectoderm.

Collectively, the amnion RNA-seq approach and chimera study have shown that SMAD5 mediates spatial cues crucial for the establishment of the amniotic ectoderm. The founding population of the amniotic ectoderm in mouse embryos deficient in SMAD5 is probably too small or delayed, and the stalled cuboidal-to-squamous transition of anterior amniotic ectoderm does not keep pace with exocoelom expansion and axial growth, hindering ventral folding morphogenesis. The complex and variable amnion phenotypes can be attributed to SMAD5 loss of function affecting different progenitor populations for amniotic ectoderm. These arise sequentially from a distinct region of the anterior proximal epiblast exposed to BMP4 and BMP2, and expand amniotic ectoderm initially posteriorly and then from the anterior. Our study also supports a new role for the flexible amnion, in addition to its known protective function, one in accommodating the shaping of the embryo proper.

[C57BL6×CBA.Ca] F1 and F2, and some Dub:ICR mouse (Mus musculus) embryos were used (Table S6). Terminology of embryo staging was as in Lawson and Wilson (2015). For the nominal age at injection, embryonic day was counted, with E0 being the middle of the dark period during overnight mating; 0.1 day was subtracted to allow for the developmental delay as a result of dissection and labelling. Culture of E6 PS stage and E6.5 ES stage embryos and the iontophoretic injection of single epiblast cells were as reported previously (Beddington and Lawson, 1990; Lawson et al. 1991; Perea-Gomez et al. 2001). In brief, clones were generated by iontophoretically injecting one epiblast cell/embryo with either a mixture of 7.4% horse radish peroxidase (HRP, EIA grade, 10814407001, Roche) and 2.6% lysinated rhodamine dextran (LRDX, M_r 10×10³, D1817, Molecular Probes) or with 10% LRDX alone, both in 0.05 M KCl, using 1-3 nA depolarising current for 10-20 s, with a duty cycle of 500 ms/s and frequency of 1 s. An additional injection of 5-10 s into an extra-embryonic visceral endoderm cell in the same focal plane provided a marker clone in the same radial position as the injected epiblast cell. Embryos were cultured for 1 day or 1.5+ days, stained for HRP where relevant, fixed with 5% paraformaldehyde in PBS, dehydrated and cleared in BABB, when labelled cells were counted. The location and number of labelled cells were confirmed on 7 µm serial sections of the material embedded in glycol methacrylate (Technovit). PGCs were identified on the basis of their characteristic pattern of alkaline phosphatase activity, in sections of embryos labelled only with LRDX (Ginsburg et al., 1990; Lawson and Hage, 1994).

Clonal descendants in the amniotic ectoderm were plotted on an appropriate template using information from the whole mount and sections to normalize the plot. Amnion templates were derived from 3D embryo reconstructions in the eMouseAtlas (Armit et al., 2017), specifically EMA:224 (TS 10b, LSEB/EPHF), EMA:17 (TS 11, LPHF/EHF), EMA:220 (TS 12a, 2S) and EMA:218 (TS 12b, 5S). Templates were created from measurements over the surface of the amnion in both sagittal and frontal orientation. Because of the extreme curvature of the amnion by the 5S stage, the 2D template for this stage is a compromise between the linear dimensions over the surface and the shorter periphery of the base of a dome. Template dimensions were adjusted for 20% shrinkage during histological processing of the embryos used in the eMouseAtlas.

Smad5-deficient mouse embryos (within decidua or isolated) from heterozygous Smad5^−/+ crosses (Smad5^tm1Zuk) (Chang et al., 1999) of mixed C57BL/6J×CBA background were collected in ice-cold PBS at E7.5-E8.5. Amnion was stepwise microsurgically isolated from EPHF- to 2S-stage embryos using tungsten needles as shown in Fig. 4C,D, and frozen on dry ice. PCR genotyping was performed on leftover embryonic material. For genotyping of embryos sectioned within the decidua, side sections were mounted on PALM membrane slides (415190-9041-000, Zeiss), stained with Cresyl Violet and subjected to laser capture microdissection (LCM) on a PALM MicroBeam laser capture system (Zeiss) to isolate embryonic tissue for genotyping. DNA for PCR was isolated after on-cap lysis in LCM lysis buffer (0.5 M EDTA pH 8.0, 1 M Tris pH 8.0 and Igepal CA-630). Sex was determined by Jarid1c (Kdm5c)/Jarid1d (Kdm5d) and Sry/IL3 PCR (Table S7). In all experiments, stage-matched Smad5^−/+ or Smad5^+/+ littermates were used as controls to Smad5^−/− embryos. All animal experiments were approved by the ethical commission of KU Leuven (P097/2008; P077/2014; P251/2014). The care and use of mice complied with the national and institutional regulatory standards of the Hubrecht Laboratory (The Netherlands), The University of Edinburgh (UK) and KU Leuven (Belgium).

Primers for RT-qPCR on cDNA from single amnion samples (Table S7) were designed in maximal proximity to the 3′ ends of target transcripts, to compensate for the potential 3′ bias that might be generated during amplification (see ‘Single amnion cDNA amplification’ section). Samples were processed on a LightCycler480 Real-Time PCR System with LightCycler 480 SYBRGreen I Master Mix (Roche, 4707516001) in technical duplicates. Reference genes for amniotic tissue were chosen using GeNorm. For the analysis in Fig. 5D, expression was normalized to two reference genes – Gapdh and Psmd4 – using the ΔΔCt method and qbase+ software (Biogazelle). For the analysis shown in Fig. S3A, expression was normalized to Psmd4, and presented with the ΔCt method, which reflects the Ct difference between the reference and the test transcript (in Fig. S3A, the higher ΔCt, the higher the expression).

cDNA for RT-qPCR and RNA-seq was prepared from individual E7.5 amnions and amplified following adaptation of a procedure originally designed for RNA-seq of single-cells (Tang et al., 2009) on the SOLiD platform (‘Double-stranded cDNA synthesis from a single cell’, Applied Biosystems demonstrated protocol). Briefly, reverse transcription and second-strand cDNA synthesis was performed in the sample lysate without RNA isolation, using SuperScript III Reverse Transcriptase (Invitrogen) and poly-dT universal primers. This was followed by two rounds of PCR amplification (18 and 12 cycles, respectively) and column purifications (QIAquick, Qiagen), generating up to 3 kb coverage at the 3′ end of all poly-adenylated transcripts. The resulting cDNA was used for RT-qPCR and RNA-seq. The cDNA yield for the 12 RNA-seq samples is shown in Table S1.

Between 30 and 60 million reads were generated per library; 20%-27% of them mapped uniquely to the genome (Table S1). Raw colorspace reads were mapped to the mouse genome GRCm38.83 with TopHat2 software (v2.1.1) (Kim et al., 2013; Trapnell et al., 2009) using colorspace compatible parameters: Bowtie 1 (v1.1.2.0), --color --quals and coverage-search algorithm. Transcript coordinates were extracted from the Genome Reference Center reference annotation with Gffread from the Cufflinks v2.1.1 suite and merged to gene coordinates with mergeBed from the Bedtools v2.17.0 toolkit (Kim et al., 2013; Quinlan and Hall, 2010; Trapnell et al., 2011). GC content and gene length were derived from the gene coordinates. The numbers of aligned reads per gene were summarized using featureCounts v1.5.3 with parameters ‘-Q 0 -s 0 -t exon -g gene id’ (Liao et al., 2014). We removed 22,528 genes for which all samples had less than 1 count per million. As such, we continued with raw counts for 16,033 genes.

Raw counts were imported in R package DESeq2 1.18.1 (Love et al., 2014) for library size normalization and differential expression analyses. The counts were subjected to variance stabilizing transformation (VST) (Anders and Huber, 2010). The variance normalized counts were used for gene expression visualization and clustering. Clustering of all 12 samples was performed by principal component analysis (PCA). In DESeq2, a negative binomial generalized linear model (GLM) was fitted against the normalized counts. Differential expression was tested for with a Wald statistical test, also implemented in the DESeq2 package. The resulting P-values were corrected for multiple testing with Benjamini–Hochberg to control the false discovery rate (FDR). Differentially expressed genes were further filtered using FDR≤0.05 and |log2 FC|≥1. Expression heat maps were generated in MultiExperiment Viewer (MeV) using VST values. GSEA analysis is described in the Supplementary Materials and Methods.

Embryos and decidua were fixed overnight in 4% paraformaldehyde in PBS, embedded in paraffin and sectioned at 6 µm. Fixed chimeric embryos were pre-embedded in 1.5% agarose prior to paraffin embedding and sectioning. Fluorescent and chromogenic antibody staining was performed on an automated platform (Ventana Discovery, Ventana Medical Systems) or manually using standard procedures. We used antibodies raised against Oct3/4 (1:100, ab19857), Eomes (1:250, ab23345), collagen IV (1:100, ab6586), phospho-histone 3 (1:100, ab5176) and GFP (1:100, ab13970) from Abcam; and Elf5 (1:40, sc-9645) and periostin (1:100, sc-67233) from Santa Cruz Biotechnology. Differently labelled secondary antibodies raised in donkey or goat were used (1:300, Jackson ImmunoResearch). Omitting the primary antibody served as a negative control. Nuclei were counterstained with DAPI (1:1000, D3571, Invitrogen) or Haematoxylin. TUNEL assay was performed using a Fluorescein In Situ Cell Death Detection kit (Roche) according to the manufacturer's instructions. Proliferation data were statistically analysed with Prims 6 software (GraphPad). The nonparametric Mann–Whitney U-test was used for comparisons of the two groups.

Whole-mount in situ hybridization was performed as previously described (Pereira et al., 2011). The Fgf5 cDNA template used for riboprobe synthesis and digoxigenin labelling was generated by PCR (Table S7). Probe hybridization was at 60°C.

Chimeric embryos were produced as described (Eakin and Hadjantonakis, 2006). Briefly, tetraploidy was achieved by electrofusion of two-cell stage CD1 embryos. WT and Smad5 knockout mouse ESCs (Bosman et al., 2006) were transduced with lentivirus containing eGFP gene under the control of elongation factor 1 alpha promoter. Two tetraploid zona pellucida-free morula embryos were aggregated with GFP⁺ ESCs to produce either Smad5^−/−;GFP⁺ ESC↔4n WT or WT;GFP⁺ ESC↔4n WT chimeras, respectively knockout and WT chimeras. Blastocysts were transferred into the uteri of pseudopregnant females. Embryos were microdissected, imaged for GFP fluorescence and fixed. In total, 35 knockout and five WT chimeras were embedded in agarose and paraffin. Sections were subjected to immunofluorescence against GFP.

We are grateful to Z. Zhang and L. Vermeire for chimera production; R. Janky and VIB Nucleomics Core (www.nucleomics.be) for mapping and bioinformatics analysis; A. Francis, E. Seuntjens and E. Radaeli for immunosupport (InfraMouse, Hercules infrastructure ZW09-03); C. Hertz-Fowler's team (Liverpool) for SOLiD sequencing; J. Korving and P. Lanser for methacrylate sections, E. Maas for mouse husbandry; C. Verfaillie for the lentiviral unit; and A. Camus, M. Bialecka, R. Dries, L. Umans and D. Davidson for discussions.