The serpin PN1 is a feedback regulator of FGF signaling in germ layer and primary axis formation

The question of how the body plan acquires proper proportioning of its germ layers and subdivides the primary axis into head, trunk and tail structures is of fundamental importance in developmental biology. In Xenopus, mesoderm formation occurs after the midblastula transition, when signals from the vegetal endoderm, primarily members of the Nodal-related class of the TGFβ family and Fibroblast growth factors (FGFs), induce mesoderm from competent ectoderm in the adjacent marginal zone (De Robertis et al., 2000; Kimelman, 2006). At the advanced gastrula stage, posterior mesoderm secretes signals such as Wnts and FGFs, but also Nodals and BMPs, that convert head into trunk and tail structures (Niehrs, 2004; Pera et al., 2014). The ectoderm and head are considered as default fates that form in the absence of growth factor signaling.

FGFs stimulate mesoderm induction and posterior development via the Extracellular signal-regulated kinase (Erk) pathway (Amaya et al., 1991; Umbhauer et al., 1995). Fgf4 and the T-box transcription factor Xenopus brachyury (Xbra) engage in a positive feedforward loop that causes amplification of the mesoderm-inducing and posteriorizing signal (Isaacs et al., 1994; Schulte-Merker and Smith, 1995). Although negative-feedback mechanisms for FGF signals exist in the mesoderm (Böttcher and Niehrs, 2005), it is not clear how the ectoderm protects itself against self-propagating FGF signals. Whether FGF signals need to be suppressed to allow head formation has not convincingly been demonstrated.

Proteoglycans through their heparan sulfate (HS) chains capture FGFs at the cell surface, regulate their extracellular transport to target cells and participate in complex formation with their receptors (Yu et al., 2009; Matsuo and Kimura-Yoshida, 2013). We previously described an autoinductive loop of FGF and the secreted serine protease HtrA1 that leads to the mobilization of FGF/proteoglycan complexes and long-range FGF signaling during mesoderm induction and posteriorization in Xenopus embryos (Hou et al., 2007). FGFs stimulate HtrA1 transcription also in the chick (Ferrer-Vaquer et al., 2008) and zebrafish (Kim et al., 2012). It is apparent that the proteolytic activity of HtrA1 needs to be regulated to protect the integrity of proteoglycans and ensure proper FGF signaling in the embryo.

In a direct screen for secreted proteins in early Xenopus embryos, we identified Protease nexin-1 (PN1) (Pera et al., 2005). PN1, which is also known as Glial-derived nexin or SerpinE2, is a serine protease inhibitor (serpin) that contains an exposed reactive center loop (RCL) that covalently binds to and blocks proteases such as thrombin, plasminogen activator, trypsin, urokinase and factor XIa (Baker et al., 1980; Stone et al., 1987; Knauer et al., 2000). Serpins inhibit target proteases through a suicide substrate mechanism, by which the protease cleaves the RCL at the process site and forms a covalent acyl-enzyme complex that causes irreversible inhibition of the protease (Olson and Gettins, 2011; Li and Huntington, 2012). PN1 null mice exhibit increased proliferation in the postnatal cerebellum (Vaillant et al., 2007), but no apparent early developmental defects have been described. In Xenopus embryos, overexpression of PN1 inhibits convergence extension movements and the expression of mesendodermal markers (Onuma et al., 2006). Here, we show that PN1 gene activity is positively regulated by FGF signals and is crucial for the formation of ectoderm and head structures. We report that PN1 binds and inhibits HtrA1, regulates the turnover of Syndecan 4 (Sdc4) and controls the range of FGF/Erk signaling. Thus, our study uncovers a novel important role of PN1 as negative-feedback regulator in the extracellular regulation of the HtrA1-FGF axis during germ layer and primary axis development in Xenopus.

In a screen for secreted proteins from gastrula stage Xenopus laevis embryos, we previously isolated five non-redundant full-length cDNA clones that encode the serpin family member PN1 (Pera et al., 2005). PN1 appeared as a 45 kDa protein in the supernatant of transfected and [³⁵S]methionine/cysteine-labeled HEK293T cells (Fig. 1A). Using NCBI BLAST, we identified a second X. laevis PN1 (here referred to as PN1.b) that differs slightly from the originally identified PN1.a homeolog by 4% in derived amino acid sequence (supplementary material Fig. S1).

We compared the gene expression of PN1.b, the previously described PN1.a (Pera et al., 2005; Onuma et al., 2006) and the secreted serine protease HtrA1 (Hou et al., 2007). RT-PCR and whole-mount in situ hybridization analyses reveal maternal and zygotic transcription of PN1.a and PN1.b, whereas HtrA1 is robustly expressed only after the midblastula transition (Fig. 1B-K′; supplementary material Fig. S2A-H). At stage 10, all three genes were expressed in the animal cap and marginal zone (Fig. 1C-E; supplementary material Fig. S2C,D). At stage 13, they overlapped in the anterior neural plate, while HtrA1 showed an additional unique domain in the posterior mesoderm (Fig. 1F-H′). In tailbud embryos, transcripts of both PN1 homeologs were found in the notochord (Fig. 1I-J′). Interestingly, PN1.b was also robustly expressed in migrating neural crest cells, the ear placode and spinal cord, where PN1.a was less abundant or absent, suggesting differential regulation at this stage. HtrA1 transcripts were localized in the notochord region and migrating neural crest cells (Fig. 1K,K′).

The gene expression patterns of PN1 and HtrA1 partially overlap with those of several FGFs and FGF receptors (Lea et al., 2009) and coincide with sites of FGF/Erk activation (Christen and Slack, 1999) in the marginal zone, anterior neural plate, notochord, ear placode and branchial arches, suggesting a possible interaction. In mRNA-injected embryos and animal cap explants at stage 10, Fgf4 activated whereas the dominant-negative FGF receptor-1 construct XFD (Amaya et al., 1991) downregulated PN1.a and PN1.b transcription (Fig. 1L-O; supplementary material Fig. S2I-K), suggesting that FGF signals stimulate the expression of both PN1 homeologs.

We injected synthetic mRNA of PN1.a (hereafter called PN1) into Xenopus embryos (Fig. 2). It was previously shown that marginal injection of PN1 causes exogastrulation (Onuma et al., 2006). We observed that, upon animal injection of PN1 mRNA, embryos underwent normal gastrulation but developed increased head structures and a shortened trunk-tail axis (Fig. 2A,B; supplementary material Fig. S3A,B). The eyes of PN1-injected tadpoles exhibited an expansion of the optic fissure known as coloboma. We obtained similar results with PN1.b mRNA (supplementary material Fig. S3C,E). In post-neurula embryos, PN1 mRNA widened the expression domains of the telencephalic marker Foxg1 and the posterior midbrain marker En2, expanded Fgf8 in the hatching gland, frontal plate and the midbrain-hindbrain boundary, and delayed the splitting of the Rax-positive eye field into distinct bilateral domains (Fig. 2D-I). PN1 mRNA also enlarged expression of the heart marker Nkx2-5 and the anterior domain of Sizzled in the ventral mesoderm (Fig. 2J,K,P,Q). Marginal injection of PN1 mRNA caused loss of the pan-mesodermal marker Xbra in early gastrula embryos (Fig. 2L,M) (Onuma et al., 2006). At the neurula stage, animally injected PN1 mRNA led to a reduction and posteriorward retraction of the neuronal differentiation marker N-tubulin and the trunk mesoderm marker Papc, whereas the anterior ectoderm marker Otx2 was expanded (Fig. 2R,S,V,W; supplementary material Fig. S7A,B), underscoring its anteriorizing activity.

To investigate the impact of the serpin-specific RCL on PN1 activity, we generated a point mutation in PN1 (PN1pm), in which the critical arginine and serine residues at its process site are changed to proline (R362P and S363P; Fig. 3A). Microinjection of PN1pm mRNA caused only mild reduction of Xbra (Fig. 2O) (Onuma et al., 2006) and did not affect N-tubulin, Papc or Otx2 expression (Fig. 2U,X), supporting the importance of an intact RCL for PN1 patterning activity.

We realized that several effects of PN1 overexpression, including coloboma, loss of mesodermal Xbra expression and reduction of N-tubulin-positive neurons, are similar to phenotypes resulting from knockdown of HtrA1 (Hou et al., 2007) or suppression of FGF receptor activities (Amaya et al., 1991; Hardcastle et al., 2000). Moreover, anteriorization also results from inhibition of canonical Wnt signaling, e.g. by the secreted Wnt antagonist Dkk1 (supplementary material Fig. S3D,F) (Glinka et al., 1998). We now show that an antisense morpholino oligonucleotide (MO) against HtrA1 (HtrA1-MO), XFD mRNA and Dkk1 mRNA reduce Papc and posteriorly expand Otx2 expression (Fig. 2Y-AA), indicating that inhibition of HtrA1 and FGF resembles the effect of Wnt antagonism and promotes head formation. HtrA1 mRNA rescued PN1-induced suppression of mesodermal fate, reduction of neuronal differentiation and stimulation of anterior development (Fig. 2C,N,T,BB). Fgf4 mRNA also compensated PN1-induced anteriorization (Fig. 2CC). Expression of pCS2-Wnt3a (referred to as Wnt3a-DNA) reverted the expansion of Otx2 but, unlike Fgf4, failed to rescue reduction of Papc expression by PN1 (Fig. 2DD). These results indicate that HtrA1 overexpression can rescue all PN1 activities and that PN1-induced anteriorization depends on FGF and, at least in part, on Wnt signals.

To determine whether PN1 and HtrA1 physically interact, we performed co-immunoprecipitation with proteins overexpressed in Xenopus embryos (Fig. 3). We generated FLAG-tagged constructs of PN1, PN1pm and the deletion mutant PN1ΔC, which lacks the C-terminal RCL and serpin signature motif (Fig. 3A; supplementary material Fig. S1) (Onuma et al., 2006). Immunoblot analysis revealed that FLAG-PN1 co-precipitated efficiently with HtrA1 and that HtrA1, in turn, co-precipitated with FLAG-PN1 (Fig. 3B). HtrA1 co-immunoprecipitated with FLAG-PN1pm less efficiently than with FLAG-PN1 and not at all with FLAG-PN1ΔC (Fig. 3C), suggesting that PN1 binds via its C-terminal RCL to HtrA1.

We next explored the protein domains of HtrA1 that interact with PN1, using a series of HtrA1 mutant constructs (Fig. 3D). FLAG-PN1 co-immunoprecipitated with HtrA1ΔN and HtrA1ΔPDZ-Myc to a similar degree as with wild-type HtrA1 and HtrA1-Myc (Fig. 3E, lanes 1, 4-6) but failed to immunoprecipitate with HtrA1ΔTry (Fig. 3E, lane 2). This suggested that the trypsin domain, but not the IGF-binding domain, kazal-type serine protease inhibitor domain or the PDZ domain, of HtrA1 is important for the binding to PN1. Since FLAG-PN1 co-immunoprecipitated with equal levels of HtrA1 (S307A) and HtrA1 (Fig. 3E, lanes 1 and 3), the catalytic serine residue in the active site of the enzyme is unlikely to undergo a covalent linkage with the RCL of PN1. Moreover, co-immunoprecipitated PN1 and HtrA1 were detectable in SDS-PAGE as individual proteins (45 kDa and 55 kDa, respectively) but not as an SDS-stable 1:1 complex that resists the boiling and reducing conditions, further supporting that PN1 and HtrA1 may interact through non-covalent binding. This differs from published records, which show that PN1 forms stable covalent bonds with target proteases (Baker et al., 1980; Stone et al., 1987; Knauer et al., 2000). Together, our data suggest a non-covalent serpin-serine protease complex, in which the C-terminal RCL of PN1 specifically binds to the trypsin domain of HtrA1 (Fig. 3F).

In mRNA-injected embryos, FLAG-PN1 suppressed HtrA1-induced anencephaly and ectopic tail structures, mesoderm induction and neuronal differentiation, whereas FLAG-PN1pm had only little and FLAG-PN1ΔC no rescuing effect (supplementary material Fig. S4). Thus, the ability of PN1 to counteract HtrA1 activities (PN1>PN1pm>PN1ΔC) reflects the strength of their interaction and suggests that an intact RCL is required for PN1 to antagonize HtrA1.

We performed loss-of-function studies to further investigate the role of PN1 in the embryo (Fig. 4). We used two non-overlapping MO sequences that target the translation initiation site of the PN1.a gene (PN1.a-MO1+2) and one MO directed against the PN1.b gene (PN1.b-MO) (Fig. 4A). Immunoblot analysis showed that PN1.a-MO1+2 and PN1.b-MO robustly blocked protein biosynthesis of PN1.a-FLAG and PN1.b-FLAG, respectively, in mRNA-injected Xenopus embryos, whereas a non-specific standard control MO (designated co-MO) had no effect (Fig. 4B). By contrast, translation of a non-targeted FLAG-PN1 mRNA was not reduced by PN1.a-MO1+2 nor PN1.b-MO, underscoring their specificity. Using a polyclonal antibody against PN1, we were able to demonstrate that a combination of PN1.a-MO1+2 and PN1.b-MO (collectively termed PN1-MO hereafter) significantly reduced endogenous PN1 protein levels in Xenopus gastrula embryos (Fig. 4C, lanes 1 and 2). The endogenous PN1 protein from uninjected embryos ran at the same level as overexpressed PN1 protein from mRNA-injected siblings (Fig. 4C, lanes 1 and 3), validating the anti-PN1 antibody.

Animal injection of PN1-MO caused a slight delay of blastopore closure during gastrulation, severe microcephaly and enlargement of the proctodeum at tailbud stage, and loss of eye and shortening of tail structures in tadpole embryos (Fig. 4D,E and Fig. 5U,V; supplementary material Fig. S5). Individual knockdown of PN1.a (PN1.a-MO1+2) or PN1.b (PN1.b-MO) led to mild microcephaly (Fig. 4F,G). Analysis of molecular markers confirmed that targeting each homeolog alone has weaker effects (supplementary material Fig. S6) than their compound knockdown (Fig. 4I and Fig. 5B; supplementary material Fig. S7D), underscoring the redundant functions of the two PN1 genes. These and the following phenotypes were specific, as the co-MO (Figs 4-7; supplementary material Figs S5, S7-S10), a random-MO (Fig. 5H,S,AA) and a PN1-5mm-MO, which contains five base mismatches with the PN1.a and PN1.b target mRNAs (Fig. 5I,R,BB), had no effect, and the effects of the PN1-MO were reverted by co-injection of non-targeted FLAG-PN1 mRNA (Fig. 5G,Z and Fig. 6M).

PN1 morphant embryos showed depletion of En2 and reduction of Foxg1, Rax, Fgf8 and Isthmin expression at the boundary between mid and hindbrain (Fig. 4H-O). Injected PN1-MO also reduced Nkx2-5 and Sizzled expression in the anterior heart mesoderm, while Sizzled expression in the ventroposterior mesoderm was expanded (Fig. 4P-S). These findings suggest an important role of PN1 in promoting anterior development.

We further investigated germ layer formation and anteroposterior axis development in PN1-depleted embryos (Fig. 5). Excessive mesoderm formation and posteriorization have previously been reported for overexpression of HtrA1 (Hou et al., 2007) and Fgf4 (Isaacs et al., 1994). At the early gastrula stage, animally injected PN1-MO, HtrA1 mRNA and Fgf4 mRNA caused ectopic Xbra expression and reduction of ectodermal Foxi1e expression in the animal hemisphere (Fig. 5A-D′). More careful analysis of hemisections revealed that the expansion of Xbra and concomitant loss of Foxi1e expression in response to PN1-MO, HtrA1 and Fgf4 are restricted to deep cells, while the outer layer of the animal cap never showed ectopic Xbra signals (supplementary material Fig. S8A-D′). To test whether the mesoderm expansion in PN1-depleted embryos was due to elevated HtrA1 and FGF receptor activities, we removed the function of this protease and of FGFR1 in the PN1 morphant background. Animal injection of both HtrA1-MO and XFD mRNA restored the normal distribution of Xbra and rescued Foxi1e expression in PN1 morphant embryos (Fig. 5E-F′; supplementary material Fig. S8E-F′), suggesting that PN1 promotes ectoderm and suppresses mesoderm in an HtrA1- and FGF-dependent manner.

At the neurula stage, PN1-MO reduced Otx2 and expanded Papc and Xcad3 expression anteriorly (Fig. 5K,T; supplementary material Fig. S7C,D). In comparison, both HtrA1 and Fgf4 mRNAs translocated the border between Otx2 and Papc expression anteriorward (Fig. 5L,M). Sectioned embryos revealed that the mesodermal marker Papc did not spread to the overlying ectoderm upon PN1-MO, HtrA1 mRNA and Fgf4 mRNA injection (supplementary material Fig. S9). Wnt3a-DNA robustly reduced Otx2 (Min et al., 2011), but did not anteriorly expand Papc expression (Fig. 5P). HtrA1-MO, XFD mRNA, the dominant negative FGF receptor-4a mRNA dXFGFR4a (Hongo et al., 1999) and Dkk1 mRNA reverted PN1-MO-induced posteriorization (Fig. 5N,O,Q; supplementary material Fig. S7E,F) and restored head structures in PN1 morphant tailbud embryos (Fig. 5W,Y), underscoring the need not only to inhibit Wnt, but also to suppress HtrA1 and FGF receptor activities in head formation. Together, our data suggest that the function of PN1 to promote anterior head development relies on endogenous HtrA1, FGF and zygotic Wnt signaling.

Since TGFβ/Nodal signals promote mesoderm induction (Kimelman, 2006) and BMP and maternal Wnt signals promote dorsoventral development (De Robertis et al., 2000; Niehrs, 2004), we asked whether PN1 affects these pathways. Smad2 and Smad1/5/8 are specific signal transducers of Nodal and BMP signals, respectively (Faure et al., 2000). Immunoblot analysis of injected Xenopus embryos showed that neither PN1 mRNA nor PN1-MO induced any changes in the C-terminal phosphorylation of Smad2 at blastula stage (Fig. 6A) and of Smad1/5/8 at gastrula stage (Fig. 6B). The failure of PN1 to misregulate the expression of Sox17a (a Nodal target; Hudson et al., 1997), Vent2 (a BMP target; Hata et al., 2000), Gsc (a Nodal and Wnt target; Watabe et al., 1995) and Xnr3 (a Wnt target; McKendry et al., 1997) in gain- and loss-of-function experiments (supplementary material Fig. S10) further supports the notion that PN1 does not affect early signaling by members of the TGFβ family or maternal Wnt signals during dorsoventral patterning. The observation that PN1 inhibits transcription of Xbra (Fig. 2N and Fig. 5B) and Xcad3 (Fig. 5T), which are immediate early targets of the FGF pathway (Latinkic et al., 1997; Isaacs et al., 1998), underscores that PN1 may antagonize FGF signals.

To validate that PN1 regulates FGF signaling, we used the doubly phosphorylated form of Erk (dpERK) as specific readout of FGF signaling (Christen and Slack, 1999). Immunohistochemical analysis of early gastrula embryos showed that PN1 mRNA, but neither PN1pm nor PN1ΔC, blocked endogenous Erk activation in the marginal zone (Fig. 6C-F). HtrA1 mRNA-induced expansion of dpERK into the animal hemisphere (Fig. 6G) (Hou et al., 2007) was blocked by co-injection of PN1 mRNA, less by PN1pm and not at all by PN1ΔC (Fig. 6H-J). PN1-MO induced animalward expansion of dpERK signals, an effect that was reverted by HtrA1-MO and XFD mRNA (Fig. 6K,L,N,O), further demonstrating that PN1 restricts Erk activation in the ectoderm through inhibiting HtrA1 and FGF receptor signaling. At the late gastrula stage, robust dpERK signals were observed in the posterior mesoderm (Fig. 6P) (Christen and Slack, 1999). Injected PN1 mRNA and pCS2-PN1 (referred to as PN1-DNA), which is expressed only after the midblastula transition, blocked posterior Erk activation, whereas pCS2-PN1pm (PN1pm-DNA) had no effect (Fig. 6Q-S). By contrast, PN1-MO expanded dpERK signals towards the anterior pole (Fig. 6T,U), and co-injected HtrA1-MO reverted this phenotype (Fig. 6V), suggesting that PN1 protects the anterior territory of the embryo against posteriorizing FGF/Erk signals via inhibition of HtrA1. We conclude that PN1 promotes ectoderm and head development by antagonizing HtrA1 and FGF/Erk signaling (Fig. 6W).

We previously introduced an animal cap sandwich assay to demonstrate that HtrA1 stimulates the long-range distribution of FGF signals in the extracellular space (Hou et al., 2007). Here, we investigate the impact of PN1 on FGF paracrine signaling (Fig. 7A-K). An inducer animal cap injected with nlacZ mRNA as a lineage tracer was recombined with a responder cap. Following 6 h culture, the sandwiches were stained with X-Gal and analyzed by in situ hybridization for expression of the FGF target Xbra (Fig. 7A). An Fgf4 and HtrA1 mRNA-injected inducer cap caused robust elongation and widespread Xbra expression in the adjacent responder cap (Fig. 7B,C) (Hou et al., 2007). Concomitant overexpression of PN1, but not PN1pm, blocked the effects by HtrA1 and Fgf4 (Fig. 7D,E). Whereas co-MO and PN1-MO injection alone had no effect (Fig. 7F,G), a combination of Fgf4 mRNA and co-MO triggered moderate elongation and Xbra signals close to the interface with the injected inducer cap (Fig. 7H). Notably, Fgf4 mRNA and PN1-MO caused pronounced elongation of the uninjected responder cap and robust Xbra expression even at distance from the Fgf4 source (Fig. 7I). The presented data are statistically significant (Fig. 7J,K) and suggest an essential role of PN1 in regulating FGF paracrine signaling.

Christen and Slack (1999) reported an elegant experiment to analyze the spread of FGF proteins in Xenopus embryos. Following injection of digoxigenin (DIG)-labeled Fgf4 mRNA into one animal blastomere at the 32-cell stage, a ring of dpERK-positive cells is induced in the animal cap around the exogenous FGF source after the midblastula transition (Fig. 7L). It was previously shown that this Erk activation outside of the source area is due to diffusion of FGF proteins (Christian and Slack, 1999). At stage 8.5, ectopic Erk activation spanned about one cell diameter around the Fgf4 mRNA injection site (Fig. 7M). Upon co-injection of Fgf4 and HtrA1 mRNAs, signals spread approximately eight cell diameters from the source (Fig. 7N). This eightfold increase in diameter (a 64-fold increase in area) of the activated dpERK patch indicated that HtrA1 facilitates FGF diffusion. HtrA1-mediated expansion of Erk activation was strongly reduced by co-injection of PN1 mRNA, but only a little by PN1pm mRNA (Fig. 7O,P). By contrast, microinjection of PN1-MO augmented the diameter of Fgf4-induced Erk activation almost threefold compared with co-MO (Fig. 7Q,R). Co-injection of HtrA1-MO limited the spread of Erk signals in PN1 morphant embryos (Fig. 7S), suggesting that PN1 functions to restrict FGF diffusion in an HtrA1-dependent manner. The observation that co-injection of XFD mRNA completely blocked Fgf4-induced dpERK expression in PN1-deficient embryos (Fig. 7T) confirms that FGF signals account for the expansion of Erk activation. A quantification of the data is presented in Fig. 7U and suggests a key role for PN1 in regulating FGF diffusion.

In Xenopus, the transmembrane proteoglycan Sdc4 is co-expressed with PN1 in the early animal hemisphere of blastula and anterior neural plate of neurula embryos (Muñoz et al., 2006; Kuriyama and Mayor, 2009). Moreover, Sdc4 is a proteolytic target of HtrA1 (Hou et al., 2007). Overexpression of HA-PN1 increased protein levels of FLAG-Sdc4 (Fig. 7V, lanes 2 and 3) and reverted HtrA1-induced breakdown of FLAG-Sdc4 proteins (Fig. 7V, lanes 4 and 5). By contrast, PN1-MO reduced protein levels of FLAG-Sdc4 (Fig. 7W). The results suggest that PN1 might stabilize Sdc4 in the embryo and protect this proteoglycan from degradation by the HtrA1 protease.

In this study, we identified PN1 as an important feedback regulator of FGF signaling during ectoderm formation and head development. The finding that the transcription of two Xenopus PN1 homeologs is positively regulated by FGF signals is consistent with previous reports that FGF2 stimulates murine PN1 expression in primary mid/hindbrain and cultured cerebellum cells (Küry et al., 1997; Vaillant et al., 2007), supporting the contention that PN1 might be an FGF target gene. PN1 is a member of the serpin family of serine protease inhibitors, with an RCL that participates in an unusual non-covalent binding to the proteolytic trypsin domain of HtrA1. PN1 blocks the activity of HtrA1 and prevents it from degrading the proteoglycan Sdc4. We show that PN1, via inhibition of HtrA1, regulates FGF diffusion and paracrine signaling in the early embryo.

Our functional study of PN1 and its link to the HtrA1-FGF axis reinforce the conclusion that germ layer formation and anteroposterior axis development are interrelated processes that share common molecular principles (Fig. 8A,B). In the early Xenopus gastrula, FGF/Erk signals and the transcription factor Xbra mutually activate each other (Isaacs et al., 1994; Schulte-Merker and Smith, 1995), and HtrA1 stimulates FGF/Erk signaling in the marginal zone (Hou et al., 2007), causing an expansion of mesoderm at the expense of ectoderm in the inner layer of the animal cap. We suggest that the animalward spread of mesodermal fate is blocked by FGF-activated PN1, which, through direct binding, inhibits HtrA1 and thereby restricts mesodermalizing FGF/Erk signaling (Fig. 8A). In support of this model, overexpression of PN1 blocks endogenous dpERK and Xbra expression in the marginal zone. An intact RCL domain is crucial for PN1 to inhibit HtrA1-induced Erk activation and mesoderm formation in the animal hemisphere. By contrast, downregulation of PN1 induces ectopic Xbra and reduction of ectodermal Foxi1e expression in deep cells of the animal cap, which could be the result of a morphogenetic movement defect or direct cell fate change. The observation that ectopic Erk activation and expansion of mesoderm at the expense of ectoderm cells in PN1 morphant gastrula embryos can be reverted by both HtrA1-MO and XFD mRNA provides evidence that PN1 protects ectoderm and suppresses mesoderm via inhibition of HtrA1 and FGF/Erk signals.

At the early neurula stage, two fields have segregated along the primary body axis, with the anterior neural plate showing colocalization of PN1 and HtrA1 and the posterior mesoderm exhibiting an autocatalytic feedback of HtrA1, FGF/Erk and Xbra (Fig. 8B). PN1 blocks HtrA1 in the anterior domain and restricts HtrA1-FGF/Erk-Xbra activity to the posterior part of the embryo. In support of this conclusion, PN1 overexpression expands anterior at the expense of posterior marker genes in an HtrA1- and FGF-dependent manner, and zygotic PN1 suppresses posterior Erk activation, whereas PN1 knockdown has the opposite effect and expands expression of the FGF-responsive Xcad3 gene. Previous studies have highlighted roles of FGF signals (Amaya et al., 1991) and HtrA1 (Hou et al., 2007) in posterior trunk/tail formation, but whether their activities need to be suppressed to allow head development has not conclusively been demonstrated. Here, we show that both XFD mRNA and HtrA1-MO expand the anterior ectoderm marker Otx2 at the expense of the posterior mesoderm marker Papc. Moreover, the anteriorward expansion of Erk activation and reduction of anterior Otx2 expression in PN1-depleted embryos is reverted by downregulation of FGF receptor and HtrA1 activities. Thus, FGF/Erk needs to be inhibited to allow head formation, and PN1 protects the head from this posteriorizing signaling. Taken together, PN1 is an antagonist of the HtrA1-FGF axis and fulfills important functions to maintain the ectoderm and head as default states.

We suggest a scenario in which PN1 controls long-range FGF signaling by regulating HtrA1-mediated cleavage of proteoglycans (Fig. 8C). Upon release from the cell, FGFs are bound by the HS chains of cell surface proteoglycans that restrict their diffusion (Yu et al., 2009). HS is abundant in the Xenopus blastula and neurula (Yamada et al., 2009). HtrA1 triggers cleavage of proteoglycans, including Sdc4 (Hou et al., 2007), and mobilizes FGFs complexed to its soluble ectodomain, thereby converting proteoglycans from negative diffusion regulators into shuttles that transport FGFs through the extracellular space. Notably, HS-bound FGFs are protected against degradation by secreted serine proteases (Saksela et al., 1988; Sommer and Rifkin, 1989). We show that, in the animal cap of stage 8.5 embryos, HtrA1 can expand the range of Erk activation up to eight cell diameters outside of an artificial Fgf4 source, suggesting that this protease is able to facilitate the diffusion of FGF proteins over a distance of 300 µm. As proteoglycans are important co-receptors of the FGF-FGFR signaling complex (Matsuo and Kimura-Yoshida, 2013), their unrestricted degradation might cause a collapse of FGF signaling. Moreover, HtrA1 can degrade free FGF8 protein (Kim et al., 2012), further underscoring the need to control HtrA1 proteolytic activity. PN1 binds to HS, which attracts the serpin to the cell surface and elevates protease inhibition (Li and Huntington, 2012). We showed that, through physical interaction, PN1 inhibits HtrA1 and stabilizes Sdc4. PN1 restricts HtrA1-stimulated FGF diffusion and fulfills a necessary function in controlling paracrine FGF/Erk signaling in the Xenopus embryo.

Proteoglycans via their HS chains bind to other growth factors in addition to FGFs, such as members of the TGFβ family, Wnts and Hh signals (Häcker et al., 2005). Yet in Xenopus, PN1 seems not to affect signaling intermediates and targets of the TGFβ pathways, nor target genes of the maternal Wnt/β-catenin pathway. Our findings that Dkk1 overexpression rescues posteriorization in PN1 morphant embryos, and that aspects of the anteriorizing effect of forced PN1 expression can be compensated by Wnt3a-DNA, supports at least a partial involvement of zygotic Wnt signaling in PN1 action during primary axis development. A previous study indicated that the posteriorizing effect of FGF signaling can be indirectly mediated by Wnt, e.g. by secondary induction of Wnt3a or Wnt8 (Kazanskaya et al., 2000). In addition, Sdc4 can positively regulate the Wnt/planar cell polarity pathway, which in turn inhibits Wnt/β-catenin signaling (Muñoz et al., 2006; Astudillo et al., 2014). Our observations that PN1 negatively regulates FGF signals and stabilizes Sdc4 protein in an HtrA1-dependent manner are consistent with a possible indirect inhibition of zygotic Wnt/β-catenin signaling by PN1. In mouse, PN1 inhibits Shh signaling in the postnatal cerebellum and in prostate cancer (Vaillant et al., 2007; McKee et al., 2012). However, Hh antagonism is unlikely to explain the PN1 activities described here because Hh signals do not induce mesoderm and – similar to PN1 – inhibit neuronal differentiation and promote anterior development in the early Xenopus embryo (Lai et al., 1995; Franco et al., 1999; Min et al., 2011).

Is the PN1-HtrA1-FGF interaction that we present here for germ layer and primary axis formation relevant for other aspects of development and disease? In the mouse postimplantation embryo, a crucial function of secreted serine proteases has recently been demonstrated for HSPG cleavage and spreading of Fgf4/8 proteins in the extra-embryonic ectoderm (Shimokawa et al., 2011). PN1 is expressed in the adjacent ectoplacental cone (Mansuy et al., 1993), suggesting a possible role for the local retention of FGF signals in these derivatives of the trophectoderm. Furthermore, HtrA1 is implicated in cancer, arthritis, age-related macular degeneration and Alzheimer's disease (Clausen et al., 2011). To our knowledge, PN1 is the first in vivo inhibitor of HtrA1 to be described. Thus, the functional interaction of PN1, HtrA1 and FGF signals, as established here for the Xenopus embryo, could be relevant for the formation of an early mammalian cell lineage and help to develop new therapies for human diseases.

Unless indicated otherwise, MOs, mRNAs or DNAs were injected four times animally at the 2- or 4-cell stage. Co-MO, random-MO, PN1-5mm-MO, PN1.a-MO1+PN1.a-MO2 (30 ng each), PN1.b-MO or PN1-MO (20 ng each of PN1.a-MO1, PN1.a-MO2, PN1.b-MO) were injected at a total of 60 ng per embryo. For single injections with 100 pg nlacZ mRNA as lineage tracer, a quarter of the above indicated MO amounts were used. Injected amounts of DNA per embryo were: pCS2-Wnt3a, 30 pg; pCS2-PN1, 100 pg; pCS2-PN1pm, 100 pg. For MO sequences, details of X. laevis PN1 and HtrA1 expression constructs and mRNA preparation see the supplementary materials and methods and Table S1.

Xenopus embryos and explants were prepared, cultured, microinjected and processed through Red-Gal staining, whole-mount in situ hybridization, and whole-mount immunostaining with anti-dpERK antibody (Cell Signaling Technology, 9101; 1:250) as described (Pera et al., 2015). The preparation of DIG-labeled Fgf4 RNA and double staining of DIG with NBT/BCIP (Roche) followed by dpERK with Fast Red (Roche) were performed according to Christen and Slack (1999). RT-PCR was performed as previously described (Hou et al., 2007) using the primers listed in supplementary material Table S2.

A peptide of X. laevis PN1 (amino acids 198-212, PENTKKRTFHGPDGK-Cys) was synthesized and used as antigen to raise a polyclonal rabbit antibody (Proteogenix, Oberhausbergen, France). Details of immunoprecipitation and immunoblotting are provided in the supplementary materials and methods.

We thank E. Amaya, I. Dawid, E. De Robertis, R. Harland, H. Isaacs, C. Niehrs, J. Slack, D. Weinstein, A. Zaraisky and J. Larrain for plasmids, and O. Wessely and U. Häcker for critical review of the manuscript.