The RNA helicase DDX3 induces neural crest by promoting AKT activity

Neural crest (NC) cells are multipotent stem cells that are induced to form at the neural plate border (the border of neural plate and epidermis; NPB) during vertebrate gastrulation. After the induction and subsequent maintenance, NC cells migrate out of the closing neural tube to specific destinations, where they differentiate into various types of cells and contribute to many tissues, such as the craniofacial structures, dental tissues, the peripheral nervous system, pigment cells and cardiac tissues (Bronner and Simões-Costa, 2016; Simões-Costa and Bronner, 2015; Stuhlmiller and García-Castro, 2012). In humans, impaired NC development can lead to birth defects that are collectively called neurocristopathies, including craniofacial disorders, congenital heart diseases and pigment defects (Dubey and Saint-Jeannet, 2017; Vega-Lopez et al., 2018).

Nearly all of our knowledge on the induction of NC was obtained from studies using Xenopus and other non-mammalian vertebrates such as chicks, owing to technical difficulties that have been encountered in mice (Barriga et al., 2015; Betters et al., 2018). These studies have uncovered a two-stage process for NC induction. First, the interplay of multiple signaling pathways activates the expression of transcription factors such as Pax3, Zic1 and Msx1 (the ‘NPB specifiers’), which induce NPB formation. These NPB specifiers subsequently induce other transcription factors, including Snai2, FoxD3 and Sox9 (the ‘NC specifiers’), to specify the NC fate within the NPB (Bronner and Simões-Costa, 2016; Simões-Costa and Bronner, 2015; Stuhlmiller and García-Castro, 2012). A major signaling pathway that induces the NC in non-mammalian vertebrates is canonical Wnt/β-catenin signaling (hereinafter referred to as ‘Wnt signaling’), which is required for both the initial formation of the NPB and the subsequent specification of the NC (Li et al., 2018; Prasad et al., 2019). Similarly, the induction of NC from human embryonic stem (ES) cells or induced pluripotent stem (iPS) cells also requires the activation of Wnt signaling (Gomez et al., 2019; Leung et al., 2016; Menendez et al., 2011; Mica et al., 2013), indicating that Wnt function in NC induction is conserved during vertebrate evolution.

The DEAD/H box RNA helicase DDX3 plays important roles in many cellular processes, including cell cycle regulation, differentiation, survival and apoptosis, primarily by regulating ATP-dependent RNA metabolism, such as RNA splicing, nuclear export and translation (He et al., 2018; Heerma van Voss et al., 2017; Sharma and Jankowsky, 2014). DDX3 contains a helicase ATP-binding domain and a helicase C-terminal domain, both of which are evolutionarily conserved and required for its RNA helicase activity. In addition, DDX3 has been shown to regulate Wnt signaling by stimulating the casein kinase CK1ε to phosphorylate the adaptor protein Dishevelled, resulting in enhanced β-catenin nuclear import. This mechanism is independent of the RNA helicase activity of DDX3, as mutants with the ATP-binding domain and part of the helicase C-terminal domain deleted can still activate Wnt signaling (Cruciat et al., 2013). There are two functional DDX3 homologs in humans, DDX3X and DDX3Y, which are localized on the X and Y chromosome, respectively. Although the two gene products share high sequence similarity, they have distinct expression patterns and function. DDX3Y is translated specifically in the testes and is necessary for male fertility, whereas DDX3X is expressed ubiquitously and accounts for most, if not all, DDX3 function in somatic tissues (Kotov et al., 2017; Snijders Blok et al., 2015). In rodents, there is at least one additional homolog of DDX3 that is actively expressed. The tissue specificity of DDX3X and DDX3Y is also different in rodents and primates, suggesting that these genes may have evolved differently in the two mammalian orders (Chang and Liu, 2010; Matsumura et al., 2019). In contrast, Xenopus tropicalis has only one ddx3 homolog, which is localized on an autosome, and the encoded protein is 91% identical to human DDX3X. Similar to human DDX3X, the X. tropicalis ddx3 mRNA is expressed ubiquitously throughout early development (Cruciat et al., 2013). Thus, X. tropicalis may be a suitable model for studying DDX3 function in somatic tissue development.

Recently, de novo mutations in DDX3X were reported to be a common cause of intellectual disability in humans (https://ddx3x.org) (Dikow et al., 2017; Kellaris et al., 2018; Lennox et al., 2020; Nicola et al., 2019; Snijders Blok et al., 2015; Wang et al., 2018). These mutations were found primarily in females, and there are strong indications that some of them lead to embryonic lethality in males (Nicola et al., 2019; Snijders Blok et al., 2015). Most of these mutations were predicted to result in decreased DDX3X activity, and the defects in female patients may be caused by haploinsufficiency or dominant-negative effects of the mutants; partial X inactivation of the DDX3X gene may also contribute to the phenotypes (Garieri et al., 2018; Lennox et al., 2020; Snijders Blok et al., 2015). Individuals with DDX3X mutations display various defects in central nervous system (CNS) development, including intellectual disability, autism spectrum disorder, microcephaly and corpus callosum hypoplasia. Most patients also have craniofacial disorders, ranging from mild dysmorphic facial features to severe oral-facial clefts (Dikow et al., 2017; Kellaris et al., 2018; Lennox et al., 2020; Nicola et al., 2019; Snijders Blok et al., 2015; Wang et al., 2018). Other common symptoms include congenital heart diseases, pigment defects and hearing/visual impairment (Snijders Blok et al., 2015; Wang et al., 2018); a recent report also suggests that these patients have high rates of neuroblastoma: a rare NC-derived childhood tumor (Lennox et al., 2020). Because this spectrum of non-CNS symptoms is typically caused by impaired NC development (Dubey and Saint-Jeannet, 2017; Vega-Lopez et al., 2018), we hypothesized that DDX3X mutations can lead to not only CNS defects but also neurocristopathies.

In this study, we have investigated the roles of DDX3 in NC development using X. tropicalis as a model. Knockdown (KD) of DDX3 in X. tropicalis embryos led to inhibited NC induction and hypoplasia in craniofacial cartilage. Wild-type human DDX3X, but not an RNA helicase-dead mutant, rescued the NC induction defects caused by DDX3 KD. In both X. tropicalis embryos and human cells, DDX3 KD reduced endogenous AKT activity and Wnt signaling. The latter could be attributed to decrease in AKT-mediated GSK3β phosphorylation and inhibition, which resulted in enhanced degradation of β-catenin. In addition, Snai1, another GSK3β substrate, was also reduced post-transcriptionally. The defects in NC induction caused by DDX3 KD could be rescued by a constitutively active form of AKT, or by co-expression of β-catenin and Snai1. Finally, DDX3 KD led to decreased levels of RAC1, a known downstream target of DDX3 and upstream activator of AKT, and exogenous RAC1 also rescued the craniofacial phenotypes caused by DDX3 KD in X. tropicalis embryos. Collectively, our data suggest a crucial role for DDX3 in regulating the AKT-GSK3β signaling axis through RAC1, and provide a mechanistic explanation for the similar neurocristopathies that have been observed in individuals with mutations in DDX3X, RAC1 or AKT-related genes.

We first performed in situ hybridization to determine the expression of ddx3 during early Xenopus development. As reported previously (Cruciat et al., 2013), ddx3 is expressed ubiquitously in the animal half of the embryos at early gastrula stages (Fig. S1A,A′). The expression was primarily found in the dorsal ectoderm at the end of gastrulation (Fig. S1B,B′), and in the neural plate with relatively high levels at the NPB during neurulation (Fig. S1C,D). At tailbud stages, ddx3 mRNA was detected in the head and somites (Fig. S1E). Thus, the expression of ddx3 is consistent with its potential function in CNS and NC development. To understand the roles of DDX3 in NC development and craniofacial morphogenesis, we used a new transgenic X. tropicalis line expressing enhanced green fluorescent protein (eGFP) driven by the snai2 promoter/enhancer. The strong eGFP expression in the differentiating cranial NC lineage allows direct observation of craniofacial cartilage development in live snai2:eGFP embryos (Li et al., 2019). Injection of a well-characterized translation-blocking DDX3 morpholino (MO), which targets the 5′-untranslated region (5′-UTR) of the X. tropicalis ddx3 mRNA (Cruciat et al., 2013), led to hypoplasia of craniofacial cartilage structures on the injected side, when compared with the uninjected side (Fig. 1A). The same phenotype was obtained with a second MO that targets a different site (MO DDX3-2), confirming that this phenotype is specific for DDX3 KD (Fig. 1A). These results suggest that, similar to human DDX3X, Xenopus DDX3 is also important for craniofacial morphogenesis.

To assess the effects of DDX3 KD on early NC development, we examined the expression of NPB and NC specifiers at the end of gastrulation (stage ∼12.5), when the induction of cranial NC is nearly complete. KD of DDX3 reduced the expression of snai2 and sox9, two NC specifiers, indicating that NC induction was inhibited (Fig. 1B,C). Interestingly, different NPB specifiers responded differently to DDX3 KD: although both pax3 and msx1 were downregulated, zic1 expression remained largely unchanged and was slightly expanded in some embryos (Fig. 1B). The expansion of zic1 was probably secondary to the reduction in PAX3, as KD of PAX3 causes a similar effect (Sato et al., 2005). Consistent with these results, injection of a plasmid encoding X. tropicalis DDX3 expanded the expression of snai2 and msx1 (Fig. S2). Because loss of either Pax3 or Msx1 can block subsequent NC specification (Monsoro-Burq et al., 2005), we conclude that DDX3 depletion inhibits NC induction by interfering with proper NPB formation, although we cannot rule out the possibility that DDX3 also regulates NC specification directly. Co-injection of wild-type human DDX3X mRNA restored the expression of all markers to normal levels (Fig. 1B,C), further confirming the specificity of the MO and conservation of DDX3 function in NC induction between frogs and humans. Nearly all missense DDX3X mutations associated with human birth defects affect evolutionarily conserved residues within the two RNA helicase domains, and some of these mutations have been shown to disrupt the RNA helicase activity of DDX3X (Lennox et al., 2020). We therefore tested whether DDX3 RNA helicase activity is needed for NC induction. As shown in Fig. 1C, the S382A/T384A (‘AAA’) mutant of human DDX3X, which cannot unwind RNA but retains the ATPase activity (Chao et al., 2006), did not rescue sox9 expression in DDX3 morphants, suggesting that the RNA helicase activity of DDX3 is important for its function in NC induction.

DDX3 is known to be a key regulator of Wnt/β-catenin signaling (Cruciat et al., 2013), and the differential responses of NPB specifiers to DDX3 KD (Fig. 1B) are consistent with previous reports showing that Wnt signaling induces the expression of pax3 and msx1 but not zic1 (de Crozé et al., 2011; Hong and Saint-Jeannet, 2007; Li et al., 2009). To investigate directly the roles of DDX3 in Wnt signaling during NC induction, we took advantage of a transgenic Wnt reporter line (Tran et al., 2010). Using this reporter line, we showed recently that there is strong endogenous Wnt signaling at the NPB during NC induction (Li et al., 2018). As expected, this Wnt activity was inhibited by KD of DDX3 (Fig. 2A). The transcription factor GBX2 is a direct Wnt target that mediates β-catenin-induced pax3 and msx1 expression during Xenopus NC induction (Li et al., 2009). Similarly, when human ES and iPS cells are induced to differentiate into NC cells, gbx2 is expressed before the NPB specifiers such as pax3 and msx1 in a β-catenin-dependent manner (Leung et al., 2016). In embryos injected with the DDX3 MO (‘morphants’), there was a reduction of gbx2 at stage ∼12 that can be rescued by the mRNA encoding wild-type human DDX3X (Fig. 2B). Conversely, ectopic DDX3 expanded the expression of gbx2 (Fig. S2), further supporting an important role for DDX3 in Wnt signaling during NC induction.

Activation of the Wnt signaling pathway starts with binding of the soluble Wnt ligands to cell-surface Frizzled receptors and LRP5/6 co-receptors, which leads to inhibition of glycogen synthase kinase 3 β (GSK3β)-mediated β-catenin phosphorylation and degradation. Consequently, β-catenin accumulates and enters the nucleus, where it interacts with LEF/TCF family transcription factors to control target gene expression (Loh et al., 2016; Nusse and Clevers, 2017). DDX3 has been shown to regulate Wnt signaling through an RNA helicase-independent mechanism (Cruciat et al., 2013). However, when compared with wild-type human DDX3X, the AAA mutant had decreased activity in rescuing gbx2 expression in DDX3 morphants (Fig. 2B). In the transgenic Wnt reporter embryos, this mutant also did not significantly restore the reduced Wnt activity caused by DDX3 KD (Fig. S3), suggesting that DDX3 RNA helicase activity is involved in Wnt signaling. This is consistent with previous studies showing that RK-33, a selective DDX3 RNA helicase inhibitor, inhibits Wnt signaling in multiple types of cancer cells (Heerma van Voss et al., 2017, 2015; Tantravedi et al., 2018). Because cancer cells usually harbor many mutations, which could interfere with the action of DDX3, we tested the effects of RK-33 in HEK293T cells, which do not contain any major mutations despite being somewhat tumorigenic (Stepanenko and Dmitrenko, 2015). As in cancer cells, RK-33 inhibits endogenous Wnt signaling in HEK293T cells; a similar effect was observed with a DDX3X siRNA (Fig. 2C,D). Together, these results point to an important role of DDX3 RNA helicase activity in Wnt signaling and NC induction.

We next examined if DDX3 can regulate Wnt signaling via a mechanism that is different from what was described previously. Interestingly, in both X. tropicalis DDX3 morphants and HEK293T cells treated with DDX3X siRNA, we detected decreased levels of total and active (unphosphorylated) β-catenin (Figs 2E, 3B,C), which could not be explained by the helicase-independent mechanism that affects β-catenin nuclear import only (Cruciat et al., 2013). To test whether DDX3 function in Wnt signaling is dependent on β-catenin degradation, we compared the effects of DDX3X siRNA on Wnt signal activities induced by wild-type β-catenin and a stabilized mutant that cannot be phosphorylated by GSK3β. As shown in Fig. 2F,G, KD of DDX3X or treatment with RK-33 effectively reduced the Wnt activity induced by wild-type β-catenin but not the stabilized mutant, suggesting that DDX3 can activate Wnt signaling by stabilizing β-catenin.

The stability of β-catenin is mainly controlled by GSK3β-mediated phosphorylation, and GSK3β activity can be regulated by upstream Wnt ligand-receptor interaction or other signal inputs. Among the latter, the serine/threonine kinase AKT can inhibit GSK3β activity via direct phosphorylation of the Ser9 residue (Cross et al., 1995; Fukumoto et al., 2001; Naito et al., 2005; Sharma et al., 2002). DDX3 has been shown to promote AKT activity in colon cancer cells, and mutations that cause gain of AKT activity in humans and mice lead to overgrowth phenotypes in the CNS and NC, which are opposite to those caused by loss of DDX3 function (Fig. 1) (Akgumus et al., 2017; Butler et al., 2005; Rivière et al., 2012; Wu et al., 2016). In addition, AKT and its upstream activator phosphatidylinositol 3-kinase (PI3K) regulate NC development in Xenopus and zebrafish, although the exact roles of PI3K and AKT in NC induction remain controversial (Bahm et al., 2017; Ciarlo et al., 2017; Figueiredo et al., 2017; Geary and LaBonne, 2018; Pegoraro et al., 2015). We therefore asked whether DDX3 KD interferes with the AKT-GSK3β signaling axis. AKT is activated through two consecutive phosphorylation events, the first occurring at Thr308 and the second at Ser473 (Manning and Toker, 2017). In HEK293T cells, siRNA-mediated KD of DDX3X inhibited both phosphorylation events without affecting total AKT levels (Fig. 3A). Similarly, MO-mediated DDX3 KD also blocked Ser473 phosphorylation in X. tropicalis (Fig. 3B,C); our pThr308 antibody did not pick up the endogenous pThr308 signal in frog embryos. Consistent with the reduced AKT activity, the phosphorylation of GSK3β at Ser9, but not the total GSK3β levels, was also reduced upon DDX3X KD or treatment with RK-33 in HEK293T cells (Fig. 3D,E). Because phosphorylation at Ser9 inhibits GSK3β activity, these results provide an explanation for the decrease in β-catenin level when DDX3 activity was reduced (Figs 2E, 3B,C). Besides β-catenin, the transcription factor Snai1 is also phosphorylated by GSK3β and subsequently targeted to degradation (Yook et al., 2005; Zhou et al., 2004). To test whether DDX3, which promotes the inhibitory phosphorylation of GSK3β, can stabilize Snai1, we expressed exogenous HA-tagged Snai1 in both X. tropicalis embryos and HEK293T cells. KD of DDX3 resulted in a clear reduction of the exogenous Snai1 protein in both cases (Fig. 3F,G), indicating that DDX3 upregulates Snai1 post-transcriptionally. Together, our data suggest that DDX3 is required for AKT activation and downstream GSK3β inhibition, leading to stabilization of β-catenin and Snai1 in Xenopus embryos and human cells.

The function of PI3K-AKT signaling in NC induction is controversial. Although treatment of gastrula-stage X. laevis embryos with a PI3K inhibitor reduces the expression of NC markers, a constitutively active form of PI3K can also reduce NC markers induced by Pax3 and Zic1 in isolated X. laevis animal caps, which are thought to contain pluripotent stem cells (Buitrago-Delgado et al., 2015; Geary and LaBonne, 2018; Pegoraro et al., 2015). However, PI3K has AKT-independent function (Faes and Dormond, 2015), and the mechanisms underlying the possible roles of AKT in NC induction remain largely unknown. To address these issues, we first determined the patterns of NPB and NC markers in embryos treated with AKT inhibitor IV (AKTi), which blocks AKT activity downstream of PI3K (Kau et al., 2003; Wang et al., 2006). Because inhibition of AKT activity at earlier stages caused gastrulation defects, we cultured the embryos in AKTi from stage ∼10 to ∼12.5. This transient inhibition of AKT activity reduced the expression of NC specifiers snai2 and sox9, as well as the NPB specifier msx1, but not the other NPB specifier zic1 (Fig. 4A). These effects are similar to those caused by DDX3 KD (Fig. 1B,C). As discussed above, msx1 but not zic1 is induced by Wnt signaling, and AKT can stabilize β-catenin by directly inhibiting GSK3β activity. We therefore hypothesized that AKT is required for activating the Wnt pathway: one of the most important signaling pathways in NC induction. To test this hypothesis directly, we treated the Wnt reporter embryos with AKTi, and indeed observed a reduction in the endogenous Wnt activity at the NPB (Fig. 4B). A similar reduction was obtained with the injection of an mRNA encoding a dominant-negative AKT mutant (dnAKT; Fig. 4C), confirming that AKT is a key regulator of Wnt signaling during NC induction.

To establish the relationship between DDX3 and its downstream targets in NC induction, we carried out a series of rescue experiments. AKT is recruited to the plasma membrane, where it is activated, through the interaction of its pleckstrin homology (PH) domain and the phosphorylated lipid products of PI3K (Manning and Toker, 2017). Replacement of the PH domain by a myristoylation signal results in a constitutively active form of AKT (caAKT) that functions independently of PI3K (Kohn et al., 1996). We found that overexpression of caAKT induced Wnt activity in HEK293T cells, and this activity was not affected by DDX3X KD (Fig. S4). This is in line with our hypothesis that AKT functions downstream of DDX3 to activate Wnt signaling. We tested further whether caAKT can rescue the phenotypes caused by DDX3 KD in X. tropicalis embryos. Indeed, co-injection of the caAKT mRNA restored snai2 expression and Wnt signaling at the NPB in DDX3 morphants (Fig. 5A,B), confirming the roles of AKT in mediating DDX3 function in Wnt signaling and NC induction. Although AKT is known to protect cells from apoptosis and promote survival (Manning and Cantley, 2007), a previous study shows that KD of PFKFB4, which is required for AKT activation, does not cause apoptosis in Xenopus embryos until stage ∼14 (Pegoraro et al., 2015). Similarly, we were unable to detect any apparent increase in apoptosis in stage ∼12.5 DDX3 morphants using terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assays (Fig. S5). These data, together with the lack of reduction in zic1 level at the NPB (Fig. 1B), indicate that the loss of NC markers upon DDX3 KD is not caused by death of NC cells or their precursors but is a bona fide inhibition of NC induction. We have shown that KD of DDX3 reduced the inhibitory phosphorylation of GSK3β at Ser9, and have identified β-catenin and Snai1 as two GSK3β substrates whose levels are regulated by DDX3 (Figs 2E and 3B-G). Because both of these GSK3β substrates are essential for NC induction in Xenopus (Aybar et al., 2003; LaBonne and Bronner-Fraser, 1998), we tested whether they are downstream effectors of DDX3 in NC induction. Ectopic expression of either β-catenin or Snai1 could partially rescue the NPB specifier msx1 and the NC specifier sox9, and a combination of both GSK3β substrates resulted in a nearly complete rescue (Fig. 5C). Based on these results, we conclude that DDX3 functions in NC induction by regulating AKT as well as β-catenin and Snai1.

We next asked how DDX3 activates AKT and the downstream signaling events. DDX3 can promote translational initiation of mRNAs with long and structured 5′-UTR through its RNA helicase activity, a function that is conserved from yeast to humans (Guenther et al., 2018; Sen et al., 2015). One of these mRNAs encodes RAC1, which can mediate DDX3 function in other physiological processes in mammals (Chen et al., 2015, 2016; Ku et al., 2018). Notably, RAC1 can activate AKT through its downstream effector p21-activated kinase (PAK) (Higuchi et al., 2008). In Xenopus embryos, a dominant-negative mutant of RAC1 reduces, whereas a constitutively active mutant expands, the expression of NC specifiers (Broders-Bondon et al., 2007). We therefore examined whether RAC1 functions downstream of DDX3 in Xenopus. To do this, we obtained an X. tropicalis rac1 cDNA with a 257-nucleotide 5′-UTR. Although the 5′-UTR of X. tropicalis rac1 did not show any significant sequence similarity to that of human RAC1, the prediction using the Mfold algorithm suggests that they contain similar secondary structures that are energetically stable (Fig. S6). Thus, the mechanism of translational regulation of RAC1 may be conserved. As expected, treatment of HEK293T cells with RK-33 or DDX3X siRNA reduced the levels of endogenous RAC1 (Fig. 6A,B). Although a similar ∼90% reduction in RAC1 protein was detected in X. tropicalis embryos with DDX3 KD (Fig. 6C,D), there was only a minimal decrease of rac1 mRNA in the same batches of injected embryos, as shown by quantitative reverse transcription polymerase chain reaction (RT-qPCR) using two separate pairs of primers (Fig. 6E), indicating that DDX3 also regulates RAC1 primarily through post-transcriptional mechanisms in X. tropicalis. Finally, we generated an X. tropicalis rac1 mRNA without the 5′-UTR, which did not cause any apparent defects when injected alone into an anterodorsal blastomere of eight-cell stage snai2:eGFP embryos (Fig. S7). Co-injection of this mRNA with DDX3 MO rescued the craniofacial defects (Fig. 6F), suggesting that RAC1 mediates DDX3 function in NC development.

First discovered in 2015, mutations in DDX3 have drawn a great deal of research interest as a frequent cause of intellectual disability in humans (Dikow et al., 2017; Kellaris et al., 2018; Lennox et al., 2020; Nicola et al., 2019; Snijders Blok et al., 2015; Wang et al., 2018). Although nearly all published studies on individuals with DDX3 mutations are focused on the CNS defects, we noted that the majority of these individuals also have abnormalities in craniofacial structures and/or other NC-derived tissues, which have not been addressed previously. We therefore hypothesized that DDX3 plays important roles in NC development, and provide here the first evidence supporting this hypothesis. Based on our data, we propose that the RNA helicase activity of DDX3 is required for efficient translation of RAC1, which activates AKT to phosphorylate and inhibit GSK3β. The inhibition of GSK3β leads to stabilization of β-catenin and Snai1: two proteins essential for NC induction (Fig. 6G).

Although DDX3 has been shown to activate Wnt signaling through an RNA helicase-independent mechanism (Cruciat et al., 2013), several lines of evidence suggest the existence of an additional helicase-dependent mechanism. Missense DDX3X mutations associated with human birth defects cause decrease in RNA helicase activity, and assays with zebrafish embryos indicate that these mutations impair the ability of DDX3X to activate Wnt signaling (Lennox et al., 2020; Snijders Blok et al., 2015). Furthermore, RK-33, a selective inhibitor of DDX3 RNA helicase activity, reduces Wnt signaling in several cancer cell lines (Bol et al., 2015; Tantravedi et al., 2018). Similarly, RK-33 inhibits Wnt signaling in HEK293T cells in our reporter assays, and the helicase-dead AAA mutant did not efficiently rescue the reduced expression of the NC marker sox9 or the Wnt target gene gbx2 in DDX3 morphants (Figs 1C, 2B,C). In addition, we uncovered a helicase-dependent effect of DDX3 KD on β-catenin levels, which is consistent with the downregulation of AKT activity. Thus, our data suggest that DDX3 can also regulate Wnt signaling through its RNA helicase activity.

The RNA helicase activity of DDX3 is needed for the efficient translation of some mRNAs with highly structured 5′-UTR (Guenther et al., 2018; Sen et al., 2015). A known direct target of DDX3 in mammals is RAC1 (Chen et al., 2015, 2016; Ku et al., 2018), which is extremely conserved throughout vertebrate evolution (100% identical between human and X. tropicalis orthologs). Interestingly, mutations in RAC1 cause craniofacial disorders and other potential NC and CNS defects in humans, which are highly similar to those defects caused by DDX3X mutations (Reijnders et al., 2017). In line with these observations, rac1 is expressed in the NC and CNS in Xenopus embryos, and has been implicated in Xenopus NC induction (Broders-Bondon et al., 2007; Lucas et al., 2002). KD of DDX3 leads to drastically reduced RAC1 protein but not mRNA levels in X. tropicalis embryos, and exogenous RAC1 rescues the craniofacial defects caused by DDX3 KD (Fig. 6C-F), suggesting that RAC1 mediates DDX3 function in NC induction. RAC1 can activate AKT through its immediate downstream effector PAK (Higuchi et al., 2008). Among the vertebrate PAK family members, four (PAK1-4) contain a CDC42/RAC interaction/binding (CRIB) motif within the N-terminal regulatory domain, which binds and inhibits the C-terminal kinase domain when PAK is inactive. The direct interaction of the CRIB motif with activated RAC1 leads to dissociation of the regulatory domain from the kinase domain and activation of PAK (Zhao and Manser, 2012). The freed kinase domain of PAK1 can serve as a scaffold to bind AKT to facilitate its membrane localization and activation through a kinase-independent mechanism (Higuchi et al., 2008), but a separate study suggests that PAK1 activates AKT by phosphorylating AKT directly at Ser473 (Mao et al., 2008). Hence, it would be of interest to test whether PAK also functions downstream of DDX3 in NC induction and, if so, whether this function is dependent on PAK kinase activity.

Previous studies suggested that the PI3K-AKT pathway can either promote or inhibit NC induction (Ciarlo et al., 2017; Figueiredo et al., 2017; Geary and LaBonne, 2018; Pegoraro et al., 2015). Here, we show that direct inhibition of AKT activity reduces endogenous Wnt signaling at the NPB, causing altered expression of NPB and NC markers that resembles loss of DDX3 or Wnt signaling (Fig. 4A-C). These data are more in line with the model that the PI3K-AKT pathway is required for NC induction (Ciarlo et al., 2017; Figueiredo et al., 2017; Pegoraro et al., 2015), and indicate that one mechanism through which AKT functions in NC induction is by regulating Wnt signaling, a major signaling pathway that is crucial for NC induction. It should be noted that the results shown by Geary and LaBonne were mainly obtained from gain-of-function assays in dissected animal caps (Geary and LaBonne, 2018), and are difficult to compare with the in vivo studies in Xenopus and zebrafish published by the other groups (Ciarlo et al., 2017; Figueiredo et al., 2017; Pegoraro et al., 2015). However, our results do not contradict the model that excessive PI3K/AKT activities inhibit NC induction. In fact, we found that high levels of ectopic DDX3 can cause reduction in NPB and NC markers (data not shown), suggesting that stringently controlled AKT activity is required for proper NC induction.

AKT-mediated phosphorylation of GSK3β at Ser9 inhibits GSK3β activity, leading to the stabilization of certain GSK3β substrates (Cross et al., 1995; Zhou et al., 2004). Although some studies show that AKT promotes Wnt signaling, likely by inhibiting GSK3β and stabilizing β-catenin (Fukumoto et al., 2001; Naito et al., 2005; Sharma et al., 2002), others suggest that AKT and Wnt act on two separate pools of GSK3β and do not crosstalk with each other (Ding et al., 2000; Ng et al., 2009). Besides inhibiting GSK3β, AKT has also been shown to activate Wnt signaling through other mechanisms. For example, PI3K/AKT signaling can upregulate the transcriptionally active β-catenin (containing unphosphorylated Ser37 and Thr41) in HEK293T and other cell lines, possibly by inducing the phosphatase PP2A to dephosphorylate these two residues (Persad et al., 2016). Thus, AKT may mediate DDX3-induced Wnt signaling through GSK3β-dependent and -independent mechanisms.

Snai1, another substrate of GSK3β, is generally believed to be stabilized upon AKT-mediated phosphorylation of GSK3β (Zhou et al., 2004). Here, we show that KD of DDX3 results in decreased levels of ectopically expressed Snai1 (Fig. 3F,G), and that ectopic Snai1 and β-catenin cooperatively rescue the NC induction defects caused by DDX3 KD in X. tropicalis embryos (Fig. 5C). These data suggest that DDX3-regulated AKT-GSK3β signaling axis induces NC through Snai1 in addition to β-catenin. Currently, most protocols for inducing the NC from human pluripotent stem cells include GSK3β inhibition, which is thought to activate Wnt signaling through β-catenin stabilization (Gomez et al., 2019; Leung et al., 2016; Menendez et al., 2011; Mica et al., 2013). In light of our results, it would be of interest to examine whether part of the NC induction abilities of GSK3β inhibitors can be attributed to the stabilization of Snai1.

Consistent with our model that DDX3, RAC1, AKT and β-catenin function in the same signaling cascade to regulate NC induction (Fig. 6G), mutations in RAC1, CTNNB1 (encoding β-catenin) and genes that directly control AKT activity (including AKT1, AKT3, PIK3R2, PIK3CA and PTEN) can all lead to craniofacial disorders and other potential neurocristopathies, such as pigment defects, in humans (Akgumus et al., 2017; Butler et al., 2005; Kharbanda et al., 2017; Reijnders et al., 2017; Rivière et al., 2012; Tucci et al., 2014). Our current efforts are therefore focused on investigating the effects of human mutations in DDX3X and these target genes on downstream signaling and NC induction. In addition, it is worth noting that individuals carrying mutations in these genes also display highly similar CNS defects, including intellectual disability, autism spectrum disorder, micro/macrocephaly and corpus callosum hypo/hyperplasia (Akgumus et al., 2017; Butler et al., 2005; Kharbanda et al., 2017; Reijnders et al., 2017; Rivière et al., 2012; Snijders Blok et al., 2015; Tucci et al., 2014; Wang et al., 2018). Hence, it is tempting to test whether the signaling cascade that we propose here (Fig. 6G) is also essential for neurogenesis and CNS development. Our study may lay the foundation for future work to understand the pathophysiology of the birth defects caused by mutations that interfere with this signaling cascade.

Constructs encoding dnAKT (Addgene #9031) (Ramaswamy et al., 1999), caAKT (Addgene #10841) (Kohn et al., 1996), FLAG-tagged human β-catenin (Addgene #16828) (Kolligs et al., 1999) and HA-tagged human Snai1 (Addgene #31697) (Kajita et al., 2004) were purchased from Addgene. Full-length cDNA clones for human DDX3X (Accession: BC011819), human RAC1 (BC050687) and X. tropicalis gbx2 (Accession: NM_001011472) were from GE-Dharmacon. The cDNAs for human DDX3X and RAC1 were subcloned into a pCS2+ expression vector with an in-frame HA tag, and the S382A/T384A (‘AAA’) mutants of DDX3 were generated by PCR using the mutagenesis primers CCACACTATGATGTTTGCTGCTGCTTTTCCTAAGGAAATAC (forward) and GTATTTCCTTAGGAAAAGCAGCAGCAAACATCATAGTGTGG (reverse). Constructs for the expression of X. laevis β-catenin and Snai1 and for in situ hybridization for snail2, sox9, foxd3, zic1 and msx1 were obtained in previous studies (Li et al., 2018; Wei et al., 2012, 2010). In vitro transcription was carried out as described to generate in situ hybridization probes and mRNA transcripts (Sive et al., 2000). Pharmacological inhibition of AKT and DDX3X was performed with AKT Inhibitor IV (Calbiochem 124011) and RK-33 (Selleckchem S8246), respectively. Antibodies that were used in this study include mouse anti-myc (DSHB 9E10, 1:500), mouse anti-HA (Sigma-Aldrich H9658, 1:1000), rabbit anti-DDX3 (Abcam ab151965, 1:1000), mouse anti-RAC1 (Abcam ab33186, 1:1000), rabbit anti-β-catenin (Sigma C2206, 1:2000), rabbit anti-active-β-catenin (Cell Signaling Technology 8814, 1:2000), mouse anti-GSK3β (Santa Cruz sc-53931, 1:1000), rabbit anti-piS9-GSK3β (Invitrogen PA1-4688, 1:1000), rabbit anti-AKT1 (Aviva Systems Biology AVARP06008_P050, 1:1000), rabbit anti-pS473-AKT (Cell Signaling Technology 4060S, 1:1000), rabbit anti-Xenopus-pS473-AKT (Cell Signaling Technology 9271S, 1:1000) and rabbit anti-pT308-AKT (Cell Signaling Technology 9275, 1:1000). Secondary antibodies that were used include HRP-conjugated rabbit anti-mouse (Sigma-Aldrich A9044, 1:7500) and goat anti-rabbit (Sigma-Aldrich A0545, 1:10,000). Mouse anti-β-actin (Sigma-Aldrich A5316, 1:15,000) was used as a loading control.

Wild-type X. tropicalis frogs were purchased from NASCO. The transgenic X. tropicalis Wnt reporter frogs were provided courtesy of Dr Kris Vleminckx (Flanders Institute for Biotechnology, Belgium), and the transgenic X. tropicalis snai2:eGFP line was generated as described previously (Li et al., 2019). Methods involving live animals were carried out in accordance with the guidelines and regulations approved and enforced by the Institutional Animal Care and Use Committees at West Virginia University and University of Delaware. Embryo were collected and injected with PLI-100A microinjectors (Harvard Apparatus) as described previously (Li et al., 2018). Morpholinos DDX3 (5′-CCGTGATTGATGCTT-3′) and DDX3-2 (5′-TTTCCACGGCCACATGACTCATAAC-3′) were designed and generated by Gene Tools. Alexa Fluor 555 (Invitrogen; for direct phenotype observation) or 488 (Invitrogen; for in situ hybridization) was co-injected as a lineage tracer. Injected embryos were sorted by the co-injected lineage tracer and cultured in 0.1× MBS to desired stages, when the embryos were imaged directly or processed for in situ hybridization, western blotting or RT-qPCR. Bright-field and fluorescence imaging was carried out using a Zeiss Axiozoom v16 epifluorescence microscope, and images of embryos were taken with an AxioCam MRc Rev3 camera. Western blotting and RT-qPCR were carried out as described below.

HEK293T cells (ATCC), recently authenticated and tested for contamination, were cultured in DMEM (ATCC) supplemented with 10% fetal bovine serum (Gibco) at 37°C with 5% CO₂. Cells were transfected with DDX3 or control siRNA (Dharmacon J-006874-06-0002 and D-001810-01-05) at 50% confluency or with plasmids at 70% confluency, using Lipofectamine 3000 (Invitrogen). For experiments that used both siRNA and plasmid, siRNA was transfected first, the media were changed and the plasmid was transfected on the following day. For TOP/FOPFLASH assays, cell were transfected and luciferase assays were carried out as described previously (Wei et al., 2010).

Embryos collected at desired stages were fixed in 4% paraformaldehyde for 24 h at 4°C, and in situ hybridization was conducted subsequently as described previously (Sive et al., 2000). TUNEL assays were carried out for fixed embryos using terminal transferase (New England BioLabs) and digoxigenin-11-dUTP (Roche) as described previously (Shi et al., 2009). lacZ mRNA was co-injected as a lineage tracer, and β-galactosidase activity was detected using 1 mg/ml Red-Gal (Millipore) in PBS buffer containing 5 mM K₃Fe(CN)₆, 5 mM K₄Fe(CN)₆, and 2 mM MgCl₂.

For western blotting, embryos or cells were lysed in ice-cold RIPA lysis buffer (Fisher Scientific) as described previously (Li et al., 2018). Approximately 15 µg protein from HEK293T cell lysates or 50 embryos equivalent was loaded in each lane of 12% SDS-polyacrylamide gels. Blots were detected with HRP-conjugated antibodies and chemiluminescence substrates (GE Healthcare) using a Bio-Rad ChemiDoc or LI-COR Odyssey imager. For RT-qPCR experiments, total RNA was isolated from 10 embryos using the RNeasy Mini Kit (Qiagen), and RNA quality was assessed by electrophoresis. Total RNA was reverse transcribed into cDNA using the iScript cDNA Synthesis Kit (Bio-Rad Laboratories) with DNase I (Qiagen) treatment. Quantitative PCR was performed on the Quant Studio 6 Flex (Applied Biosystems) using the qMAX Green Low Rox qPCR Mix (Accuris) under the following conditions: a 2 m initial denaturation at 95°C, followed by 40 cycles of 5 s at 95°C and 30 s at 60°C. The following primer sets were used for the quantitative PCR: rac1 first pair, CTTGGAATGCTCTGCCCTTA (forward) and TCTCTTTCGCTTCTTGACTGG (reverse); rac1 second pair, ACAGGAGGACTACGACAGATTA (forward) and GGTACCACTTAGCACGAACAT (reverse); gapdh, GTGACACTCATTCCTCCATCTT (forward) and GCTGTAGCCACACTCGTTAT (reverse). All samples were assayed in triplicate and data were processed using the Comparative C_T Method (Schmittgen and Livak, 2008).

Embryos were scored by comparing the injected side with the uninjected side of the same embryos. The percentage of normal and reduced phenotypes were calculated for injected embryos obtained from multiple independent experiments, and χ²-squared tests were performed to compare the percentage of embryos with normal phenotypes in different treatment groups. For craniofacial phenotypes, injected snai2-eGFP embryos were allowed to develop to stage ∼46 and scored for defects in head cartilage structures. Images of head cartilages (eGFP) were taken with a Zeiss Axiozoom.v16 epifluorescence microscope.

We thank Dr Kris Vleminckx for providing the transgenic Wnt reporter frogs, and Drs William Sellers, Richard Roth, Eric Fearon and Paul Wade for providing plasmids through Addgene.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.184341.reviewer-comments.pdf