Cilia-driven asymmetric Hedgehog signalling determines the amphioxus left-right axis by controlling Dand5 expression

Bilateral symmetry is a fundamental feature of many animal phyla that are generally called bilateria. Most bilateria deviate from bilateral symmetry in a predictable manner known as directional left-right (L-R) asymmetry (Palmer, 1996). The establishment of L-R asymmetry during early embryogenesis is crucial for the correct positioning and morphogenesis of internal organs (Sutherland and Ware, 2009). The mechanism that initiates L-R asymmetric organogenesis is one of the central enigmas in developmental biology. In vertebrates, L-R asymmetry originates at a transient midline structure known as the L-R organizer (LRO), which forms at the posterior end of the notochord during the early somite stages (Blum et al., 2009, 2014). Each LRO cell harbours a motile monocilium that rotates clockwise to generate a leftward fluid flow (nodal flow) that acts as the breaking mechanism of L-R symmetry in mouse, Xenopus and zebrafish (Nonaka et al., 1998; Schweickert et al., 2007; Essner et al., 2005). Recent studies have also shown a role for motile cilia in the L-R asymmetric establishment of urochordate and sea urchin embryos (Nishide et al., 2012; Tisler et al., 2016). This result thus indicates that cilia-based symmetry breaking is a synapomorphy of the deuterostome. However, whether any invertebrate deuterostome harbours a homologue of the LRO remains to be investigated.

In most vertebrate embryos, nodal flow is an initial driving force for symmetry breaking, and several genes involved in L-R asymmetry development are expressed asymmetrically as a result of the flow (Okada et al., 2005; Blum et al., 2009). Among these genes, Dand5 (also known as Cerberus; Cerl2 in mouse, Charon in zebrafish and Coco in frog) first shows asymmetric expression around the LRO in an R>L manner (Schweickert et al., 2010; Nakamura et al., 2012). As Dand5 is a repressor of Nodal, Nodal near the LRO is next to show L>R asymmetry, which in turn induces an asymmetric signalling cascade in the left lateral plate mesoderm, directing the embryo to develop an L-R asymmetric pattern (Schweickert et al., 2010; Grimes and Burdine, 2017; Hashimoto et al., 2004). Although the role of nodal flow in L-R symmetry breaking in vertebrates has been addressed, how nodal flow is sensed and transduced into asymmetric gene expression remains unknown.

To date, two models, the two-cilia model and the morphogen model, have been proposed to explain the mechanism by which nodal flow breaks L-R symmetry. The two-cilia model posits that there are two classes of cilia on the LRO: one is composed of motile cilia located at the centre of the LRO and essential for generating the leftward flow, and the other comprises immotile cilia, present at the periphery of the LRO and crucial for sensing the flow (McGrath et al., 2003; Tabin and Vogan, 2003). The two-cilia model is also based on the proposition that the shear stress generated by the flow is different on the left side than it is on the right side of the LRO, and such a difference, after being sensed by the immotile cilia, results in the elevation of the calcium signal and asymmetric expression of the Nodal gene on the left side (McGrath et al., 2003; Tabin and Vogan, 2003). This model explains many phenomena found in vertebrates with mutant cilia and L-R defects, but it has recently been challenged by both in vitro experiments and in silico simulations (Delling et al., 2016; Ferreira et al., 2017). The morphogen model posits that the nodal flow is crucial for transporting the morphogen ‘X’ to the left side of the LRO, where the ‘X’ is sensed by receptors in the cells or cilia, resulting in asymmetric gene expression (Tanaka et al., 2005; Tabin and Vogan, 2003). Nodal vesicular parcels (NVPs) that carry Sonic hedgehog (Shh) and retinoic acid (RA) have been found to be transported preferentially to the left side of the mouse LRO by the flow (Tanaka et al., 2005). However, no obvious asymmetry of Shh or retinoid signalling has been detected at the mouse LRO (Tabin, 2006), and knocking out the Shh or Raldh2 (a gene involved in RA synthesis; also known as Aldh1a2) genes in mice caused no effects on the initial L-R determination (Tsukui et al., 1999). To date, the asymmetric distribution of Shh or RA has not been reported for LROs in other vertebrate species. Therefore, to determine whether Shh or RA has a role as the morphogen ‘X’, further investigation is required.

Among living invertebrate deuterostomes, amphioxus is of particular importance in studying the molecular mechanisms of vertebrate embryogenesis owing to its basal position within the phylum Chordata and its simple, but vertebrate-like, genome structure, embryogenesis and body plan (Holland et al., 2004; Putnam et al., 2008). In terms of L-R asymmetry, amphioxus shows pronounced L-R morphology at the larval stage, with the mouth and pre-oral pit forming on the left side, gill slits and a major part of the endostyle and club-shaped gland presenting on the right side (Soukup, 2017). Moreover, in contrast to the sophisticated phenotypes found in vertebrates, knocked out (or knocked down) or mis-expressed Dand5, Nodal and/or Lefty genes in amphioxus usually results in two kinds of bilateral defects: two-left (left-isomerism) and two-right (right-isomerism) phenotypes (Li et al., 2017; Soukup, 2017). Importantly, amphioxus also utilizes a signalling cascade composed of Dand5, Nodal, Lefty and Pitx in the patterning of the L-R asymmetry (Li et al., 2017). These features and their outcomes indicate that amphioxus is a useful and informative model for investigating the origin and mechanism of nodal flow. Interestingly, we recently found that Dand5 transcription in amphioxus embryos relies on the Hedgehog (Hh) gene (Hu et al., 2017). This finding indicates a potential role for Hh in the regulation of the asymmetric expression of Dand5 in amphioxus. In this report, we first investigate whether Hh signalling in amphioxus is asymmetrically activated using in situ hybridization and then clarify the role of Hh signalling in the regulation of asymmetric Dand5 expression. Furthermore, we investigate the role of cilia in asymmetric Hh signalling formation and initial symmetry breaking by disrupting the function of the cilia through genetic manipulation, brief high-salt treatments and a MC treatment. Together, the findings identify cilia movement and asymmetric Hh signalling as the two key regulators for the establishment of amphioxus L-R asymmetry, and we used this conclusion to propose a novel mechanism for L-R symmetry breaking in amphioxus.

Our recent results have shown that Hh signalling is required for Dand5 transcription in amphioxus (Hu et al., 2017). However, whether Hh signalling is asymmetric in developing embryos and whether this possible asymmetry is the cause of asymmetric Dand5 expression remain unknown. Hh (ligand), Ptch (receptor), Smo (signal transducer) and Gli (transcription factor) are the key components of the Hh signalling pathway. To address the questions raised, we previously analyzed Hh and Ptch gene expression during several stages of amphioxus embryos with the whole-mount in situ hybridization (WISH) method and detected no obvious asymmetric expression signal for either of these two genes (Hu et al., 2017). In the present study, we reanalyzed the expression pattern of these two genes and extended the analysis to the Smo and Gli genes. Specifically, to observe the staining signal more clearly, we conducted serial transverse section analysis on embryos after the WISH analysis was completed. Four sequential stages of embryos after the G5 stage (in the late gastrula stage, according to Hirakow and Kajita, 1991) were analyzed, with the observation continuing for an additional 90 min (until approximately the N1 stage, according to Hirakow and Kajita, 1994), during which time, the symmetric Dand5 expression was converted to show asymmetric expression (Fig. 1Aa-Ad). The results showed that in both the whole embryos (Fig. 1Ba-Dd) and in the transverse sections (Fig. 1Bc′-Dc′), Hh was symmetrically expressed in the axial mesoderm and gut endoderm, and Smo and Gli were symmetrically expressed in the paraxial mesoderm. However, in the anterior paraxial mesoderm of some examined embryos, very weak asymmetric expression (L<R) was detected for the Ptch gene (Fig. 1Ea-Ed). Transverse sections confirmed this observation (Fig. 1Ea′-Ed′). At the G5 stage, among the fourteen examined embryos, six (42.8%) showed L<R Ptch expression in the anterior mesoderm (Fig. 1Ea′), and at the subsequent three examined stages, the percentage of the embryos showing asymmetric Ptch expression increased to 73.9% (17/23), 90.9% (20/22) and 100% (16/16), respectively (Fig. 1Eb′-Ed′) (Fig. S1). As Ptch and Gli are targets of the Hh signalling pathway in vertebrates (Hooper and Scott, 2005; Lee et al., 1997), we analyzed the expression of Ptch and Gli in Hh^−/− embryos at the early neurula stage of amphioxus to determine whether they are also regulated by Hh signalling. Compared with that of the WT/Hh^+/− embryos, Ptch expression was dramatically decreased in the Hh^−/− embryos, but the Gli expression was not obviously affected (Fig. S2). This result indicates that Ptch is regulated by Hh signalling, whereas Gli is not, at least not at this stage of amphioxus embryonic development. From this, and the observed asymmetric expression of the Ptch gene, we suggest that Hh signalling is asymmetric in the early neurula stage of amphioxus embryos. Importantly, we also observed that the initial time and the region of asymmetric Ptch expression resembled those of the Dand5 gene (Fig. 1Aa-Ad,Ac′,Ea′-Ed′,F). This observation, together with previous findings (Hu et al., 2017), indicates that asymmetric Hh signalling might provide an initial cue for asymmetric Dand5 expression.

To further address the relationship between Hh signalling and Dand5 expression, we examined whether increasing Hh signalling activity could induce the ectopic expression of Dand5. Ptch is a target and negative regulator of Hh signalling in vertebrates and flies. We anticipated that, if this were the same case in amphioxus, mutation of the Ptch gene in amphioxus would result in upregulation of Hh signalling and Ptch expression. To test this hypothesis, we examined Ptch expression in Ptch mutant embryos at the N1 stage (approximately three somites at this stage). Unexpectedly, we found no obvious difference in Ptch expression between the Ptch^−/− embryos and their siblings (Fig. S3). Moreover, we found that injection of Hh mRNA into the Ptch mutants led to the upregulation of Ptch expression (Fig. S3). These results indicate that the Ptch mutant might not be a null mutant or that Hh may transduce Hh signalling via other receptors in amphioxus. Consistent with this speculation, the Ptch^−/− mutants expressed the Dand5 gene normally (L<R) at the N1 stage and exhibited no L-R defects at the larval stage (Hu et al., 2017).

We then injected amphioxus embryos with mRNAs coding the Hh protein or a constitutively active form of the Smo protein (caSmo) to increase Hh signalling. We found that both injections could effectively upregulate Hh signalling, as indicated by increased Ptch expression (Fig. S4). The injection of caSmo mRNA also changed the Dand5 expression pattern. In the uninjected control embryos, Dand5 was expressed predominantly on the right side (right side-specific or L<R) (Fig. 2A,B), whereas in the caSmo mRNA-injected embryos, 41.7% (10/24) showed bilaterally symmetric Dand5 expression (Fig. 2E) and 4.2% (1/24) displayed reversed Dand5 expression (L>R) (Fig. 2F), although the remaining 54.1% (13/24) expressed Dand5 normally in a L<R manner (Fig. 2C,D). Consistent with these findings, 20% (10/50) of the caSmo mRNA-injected embryos presented with the right-isomerism phenotype (Fig. 3C-E) at the larval stage. This phenotype was not presented by the uninjected control embryos (Fig. 3A,B,E). We also injected caSmo mRNA into the right- and left-side blastomere immediately after the first cleavage, as shown in Fig. 2O. The results showed that the right-side injection had no effect on the Dand5 expression pattern (R>L, similar to that of the uninjected control embryos) (Fig. 2G-J), but the left-side injection caused L=R and L>R Dand5 expression in 22% (8/36) and 5.5% (2/36) of the embryos, respectively (Fig. 2M,N), although the remaining embryos expressed the gene normally in a right side-specific or L<R manner (Fig. 2K,L). Combined, these data demonstrate that upregulating Hh signalling on the left side could induce ectopic Dand5 expression.

To determine whether asymmetric Hh signalling is the cause of asymmetric Dand5 expression, we recovered Hh signalling in one side of the Hh^−/− mutants by injecting caSmo mRNA at the two-cell stage and then examined the Dand5 expression at the early neurula stage. No Dand5 expression signal appeared in the Hh^−/− embryos (15/15) (Fig. 3G). However, caSmo mRNA injection on the right side of the Hh^−/− mutant led to 87.5% (7/8) embryos with right side-specific Dand5 expression (Fig. 3H). Remarkably, among the Hh^−/− embryos injected on the left side, 70% (7/10) showed Dand5 expression on the left side only (Fig. 3I). We verified this result by injecting Hh mRNA into the right and left sides of the Hh^−/− mutant embryos. The results showed that 80% (4/5) of right-side injected Hh^−/− embryos expressed Dand5 on the right and 75% (6/8) of left-side injected embryos expressed Dand5 on the left (Fig. 3K,L), whereas none of the nine uninjected Hh^−/− embryos showed Dand5 expression (Fig. 3J). Together with the data described above, the results indicate that asymmetric Hh signalling directs asymmetric Dand5 expression in amphioxus.

We demonstrated above that the left-side injection of Hh mRNA could induce Dand5 expression on the left side in the Hh^−/− embryos. However, when we injected Hh mRNA into the Hh^−/− embryos at the one-cell stage, most showed an L<R Dand5 expression pattern but not a bilaterally symmetric pattern (Fig. S5). Moreover, a higher proportion of wild-type embryos injected with caSmo mRNA at the unfertilized egg stage had the right-isomerism phenotype (10/50=20%) than the proportion of offspring of the Hh^+/− heterozygotes injected with Hh mRNA at the unfertilized egg stage (6/62=9.7%; Fig. 3E,F). We speculated that the discrepancy between the results of the caSmo and Hh mRNA injections might be caused by an asymmetric transport of the Hh protein to the right side. Because no commercial antibody against amphioxus Hh protein is available, we injected amphioxus unfertilized eggs with mRNA coding a Myc-tagged Hh protein and examined the Hh fusion protein distribution using whole-mount immunofluorescence (WIF) with a Myc antibody. The injection rescued Dand5 expression and the L-R morphology of the Hh^−/− embryos (Fig. S6), suggesting that the Hh-Myc fusion protein works as an endogenous Hh protein. Theoretically, the fused protein would be detected widely throughout the injected embryos. However, in contrast to this expectation, we observed an asymmetric staining signal (L<R) in the anterior paraxial mesoderm in 40% (2/5) of G5-stage gastrulae, in 100% (6/6) of the R0 neurulae (final gastrulae or early neurulae at stage G7 according to Hirakow and Kajita, 1991; embryos start rotating at this stage) and in 75% (6/8) of the R30 neurulae (30 min after initial rotation, at approximately the N1 neurula stage) (Fig. 4D-F″). This pattern is reminiscent of the endogenous expression pattern of Ptch and Dand5 mRNA (Fig. 1). No staining signal was detected in the uninjected embryos (Fig. 4A-C). Further immunoelectron microscopy analysis revealed that the Hh-Myc protein was localized in the extracellular interface of the ectoderm and paraxial mesoderm (Fig. S7). These results suggest that the asymmetric distribution of the Hh protein is likely the cause of the asymmetric Hh signalling demonstrated above.

The L-R asymmetric distribution of the Hh protein in amphioxus embryos raises the issue of whether the amphioxus embryo has a nodal flow similar to that in vertebrates. To address whether cilia are required for L-R asymmetric development in amphioxus, we generated mutants of the Kif3a or Foxj1 genes, which are required for ciliogenesis in vertebrates (Marszalek et al., 1999; Tamakoshi et al., 2006; Yu et al., 2008), using the TALEN method.

The Kif3a mutants had a 20 bp deletion of ∼110 bp downstream of the start codon (Fig. S8A). At the larval stage, compared with their heterozygous (Kif3a^+/−20) and wild-type siblings (Fig. S8B,D,D′), the Kif3a^−20/−20 mutants developed an enlarged mouth and fewer cilia in the pre-oral pit, but they presented with normal L-R asymmetry (Fig. S8C,E,E′). A comparison between Kif3a^−20/−20 and wild-type (Kif3a^+/+) early neurulae did not reveal any obvious differences in either the pattern or the length of the cilia on the epithelial ectodermal (EE) cells, neural ectodermal (NE) cells or the gastrocoel roof plate (GRP) cells (Fig. S8G-J,K). The inability of the Kif3a mutation to effect early ciliogenesis is likely caused by its strong maternal expression (Fig. S8L). Two mutant lines in the Foxj1 coding region were generated: one with a 1 bp deletion at position +318 bp and one with a 13 bp deletion at the beginning of the forkhead domain (Fig. S9A,B). However, neither of these mutants showed abnormalities in ciliogenesis or the L-R asymmetric axis. The cilia lengths on the EE, NE or GRP cells at the early neurula stage were similar to those of their wild-type siblings (Fig. S9C-F). The cilia movement in the Foxj1^−1/−1 mutants had no apparent defect (Movie 1). The transcriptome data for different developmental stages (Fig. S9H) and the in situ hybridization (Aldea et al., 2015) revealed negligible Foxj1 mRNA expression in the fertilized eggs and embryos until the gastrula stage, indicating that the inability of Foxj1 to disrupt ciliogenesis is not due to maternal expression.

To remove the amphioxus embryo cilia, we then tried a physical method in which we treated the embryos with high-salt (HS) seawater, a method that had been applied to sea urchin embryos (Auclair and Siegel, 1966; Tisler et al., 2016). The treatment removed most cilia from the embryos, as indicated by the evidence from the immunofluorescence analysis in which an antibody against acetylated-α-tubulin was used (Fig. S10A-D). After the treatment, the expressions of Brachyury (a marker of posterior mesoderm and of the presumptive notochord; Holland et al., 1995) and Netrin (a marker of the floorplate of the presumptive neural tube and the notochord; Shimeld, 2000) were unaffected compared with the untreated control embryos (Fig. S11A-D′). This finding indicates that the midline structures (including the neural plate and notochord) were formed normally in the treated embryos. However, the treatment caused 54.5% (12/22) of the embryos to lose Dand5 expression (Fig. 5C-E), and 78.1% (25/32) and 53.8% (28/52) of the embryos, respectively, to express Nodal and Pitx bilaterally (Fig. 5F-K). In agreement with this finding, 48.6% (17/35) of the treated embryos displayed a left-isomerism phenotype at the larval stage (Fig. 5A-B′). This observation demonstrates that cilia are required for the establishment of L-R asymmetry in amphioxus.

MC is a nontoxic viscous reagent used for inhibiting LRO cilia movement in vertebrates (Schweickert et al., 2007; Shinohara et al., 2012). To clarify the role of cilia movement without disrupting ciliogenesis, we blocked cilia movement by culturing embryos from the mid-gastrula to the R30 or three-somite stage in seawater containing 2% MC. The MC could enter the archenteron through the blastopore, as indicated by the presence of ink particles (added to the MC) in the archenteron of the three-somite embryos (Fig. 6A), and effectively inhibit cilia movement and neurula rotation (Movie 2), but no apparent effect on the expression of Brachyury or Netrin was found at the early neurula stage (Fig. S11E-H′). At the larval stage, 56.6% (48/85) of the treated larvae displayed L-R asymmetry defects (Fig. 6B). Among them, 83.3% (40/48) showed a left-isomerism phenotype, 12.5% (6/48) showed a right-isomerism phenotype, and the remaining 4.2% (2/48) exhibited an L-R reversal phenotype (Fig. 6B). Dand5 expression along the L-R axis was randomized by the treatment, with 44% (11/25) presenting as normal (L<R), 24% (6/25) being absent, 24% (6/26) presenting a bilateral phenotype (L=R), and 8% (2/25) showing reversed phenotype (L>R) (Fig. 6C-G). The expression levels of Nodal and Pitx were also altered in 65.5% (19/29) and 67.7% (21/31) of the treated embryos, respectively, and were either expressed bilaterally (L=R), were not expressed (or had decreased expression) or were expressed in the reverse pattern (L<R) (Fig. 6H-Q). We also analyzed the effect of MC treatment on Ptch expression to determine whether asymmetry of Hh signalling was affected. In the control embryos, Ptch is asymmetrically expressed in an L<R pattern in the paraxial mesoderm of the early neurulae (7/7; Fig. 6R). However, after MC treatment, 42.9% (6/14) and 14.2% (2/14) of the early neurulae showed bilateral (but decreased) and L>R Ptch expression in the paraxial mesoderm, respectively (Fig. 6T,U), although the remaining 42.9% (6/14) of them displayed a normal L<R Ptch expression (Fig. 6S). These results indicate that cilia movement is necessary for the establishment of asymmetric Hh signalling and L-R asymmetry development in amphioxus.

Neurula rotation driven by motile cilia is required for the L-R patterning in the ascidian Halocynthia roretzi (Nishide et al., 2012). To verify that the L-R defect induced by the MC treatment was caused by the inhibited cilia movement, but not the neurula rotation, we fixed the embryos at the G4 stage (later mid-gastrula stage) by threading them through a very fine glass needle (Fig. S12A,B) and released them at the three-somite stage. Ninety-nine embryos were analyzed, and six of them showed L-R defects at the larval stage (five larvae developed a left-isomerism phenotype, and one exhibited a right-isomerism phenotype) (Fig. S12C). This finding was not significantly different from that of the control (P=0.1431, Fisher exact test), in which two of the 124 embryos examined showed a left-isomerism phenotype (Fig. S12C). This result indicates that cilia movement, but not embryo rotation, is essential for the L-R asymmetric development in amphioxus.

In mice, NVPs containing the Shh protein can be asymmetrically transported across the node by the cilia-driven flow (Tanaka et al., 2005). To determine whether cilia movement is also involved in the asymmetric distribution of the Hh protein in amphioxus, we injected the embryos with Hh-myc mRNA and then cultured them with 2% MC as described above. Immunofluorescence analysis showed that 70% (7/10) of the injected embryos cultured in regular seawater had right-side enrichment of the Hh-Myc protein in the anterior-right paraxial mesoderm (Fig. 7A-B″). However, all (11/11) of the embryos injected with the Hh-myc mRNA and cultured in 2% MC failed to accumulate the secreted Hh-Myc protein in the right paraxial mesoderm (Fig. 7C-D″). In addition, the Hh mRNA-injected embryos cultured in 2% MC had a higher frequency of bilateral Dand5 expression than those cultured in regular seawater (Fig. S13). These results indicate that cilia movement plays an essential role in the asymmetric transportation of the Hh protein.

HS seawater-treated embryos showed similar phenotypes (lost Dand5 expression and bilateral Nodal expression; Hu et al., 2017) to those acquired by the Hh-knockout embryos. Indeed, Ptch expression was significantly downregulated in the HS-treated amphioxus embryos at the early neurula stage (Fig. S10E-G). These findings raise the possibility that the L-R defect induced by the HS treatment might be a secondary effect of impaired Hh signal transduction. To address this possibility, we investigated whether Hh signal transduction relies on cilia in amphioxus in a manner similar to the signal transduction in sea urchins and vertebrates. In vertebrates, the translocation of the Smo protein from the cytoplasm to the cilium serves as a marker for Hh signal activation (Huangfu et al., 2003; Warner et al., 2014), which is initiated through a ciliary localization motif in the Smo protein. We identified a similar ciliary localization motif in the amphioxus Smo protein. The results from the immunofluorescence analysis on the embryos injected with mRNA encoding caSmo-Myc and CLDcaSmo-Myc (carrying mutations in the ciliary localization motif) showed that, although the caSmo-Myc fusion protein was conspicuously detected on the cilia on the neuroectodermal and paraxial mesoderm cells (Fig. 8B-B″ and Fig. S14C,D) where endogenous Ptch expression was also detected (Fig. 1Ea′-Ed′), CLDcaSmo-Myc failed to locate to any cilia (Fig. 8C-C″). No Smo-Myc protein (red signal) was observed in the uninjected embryos (Fig. 8A-A″). Furthermore, Ptch expression was significantly increased in the caSmo-myc mRNA-injected embryos but not in the CLDcaSmo-myc mRNA-injected embryos, with respect to the levels expressed by the uninjected embryos (Fig. 8D-G). These results suggest that Hh signal transduction in amphioxus relies on cilia, and the abnormal phenotypes of L-R asymmetry caused by removing cilia in the HS treatment might be the result of a dysfunction in Hh signalling.

L-R asymmetry formation in vertebrates involves three major steps: symmetry breaking by cilia-driven nodal flow, asymmetric activation of gene expression and asymmetric organogenesis. Using the TALEN-based genome editing and heat shock promoter-driven overexpression method, we recently demonstrated that asymmetric genes, such as Dand5, Nodal, Lefty and Pitx, are also required for amphioxus L-R development, similar to their cognates in vertebrates (Li et al., 2017). In addition, we also found that the Hh gene is required for Dand5 expression in amphioxus (Hu et al., 2017). In this study, we identified cilia movement and asymmetric Hh signalling as two key regulators for the establishment of amphioxus L-R asymmetry. Based on these findings, we constructed a comprehensive model (Fig. 9) to explain how amphioxus L-R asymmetry is initially established. Our model suggests that, at late gastrula (G5) and early neurula (N1) stages, cilia movement transfers Hh protein from the anterior axial mesoderm, in which it is synthesized and secreted, preferentially to the anterior-right paraxial mesoderm. This right-biased Hh protein causes asymmetric Hh signalling and further induces Dand5 expression in the anterior-right paraxial mesodermal cells. However, it is still unknown whether Hh signalling regulates Dand5 transcription directly or not. A canonical binding element for the Hh signalling transcription factor Gli has been deduced in the upstream region of the Dand5 transcription initiation site (Hu et al., 2017). However, the element does not appear to be essential for Dand5 transcription, as amphioxus embryos without it (Fig. S15A) show normal Dand5 expression and L-R patterns (Fig. S15C,D). Nevertheless, the result does not exclude the possibility that Hh signalling regulates Dand5 expression directly, as other Gli-binding sites in the Dand5 promoter may exist.

The Hh signalling pathway is one of the key pathways that is essential for metazoan embryonic development. Its involvement in L-R asymmetry has been reported in mice, chickens and sea urchins (Meyers and Martin, 1999; Tsukui et al., 1999; Zhang et al., 2001; Levin et al., 1995; Warner et al., 2016). In chicken, Shh is expressed to the left of Hensen's node, where it functions as a left determinant by inducing asymmetric expression of Nodal and Pitx genes in the left side of the embryo (Levin et al., 1995). In contrast, in mice, Shh is bilaterally expressed in either the node or in other regions of the embryo, and knocking it out leads to the bilateral expression of the Nodal, Lefty2 and Pitx genes (Meyers and Martin, 1999; Tsukui et al., 1999). According to this finding, Meyers and Martin (1999) suggested that Shh acts as a right determinant by repressing the expression of left determinants on the right side, whereas Tsukui et al. (1999) posited that Shh is required to establish a physical barrier in the midline that prevents the spread of left determinants to the right side. Consistent with the latter idea, the expression of the Nodal inhibitor Lefty1 in the midline is lost in Shh^−/− embryos (Tsukui et al., 1999). A later study revealed that the involvement of Hh signalling in mouse L-R asymmetry development is achieved by maintaining and propagating Nodal expression in the lateral plate mesoderm (Tsiairis and McMahon, 2009). Similar to that in mice, Hh signalling in sea urchins is required to maintain and propagate late-phase Nodal signalling activity (Warner et al., 2016). However, the necessity of Hh signalling for L-R formation has not been demonstrated in zebrafish, as zebrafish mutants lacking Shh, Smo and/or other signalling components show normal L-R morphogenesis (Schauerte et al., 1998; Chen et al., 2001), although mis-expression of the Shh gene in zebrafish embryos causes L-R patterning defects (Schilling et al., 1999). We have previously shown that the Hh signalling pathway is required for amphioxus L-R asymmetry development because it controls Dand5 expression (Hu et al., 2017). In this study, we provide further evidence that Hh signalling is asymmetrically activated (L<R), which might provide an initial cue for asymmetric Dand5 expression in amphioxus embryos.

As discussed above, the mechanism by which Hh signalling regulates L-R asymmetry development in amphioxus appears to be different from that in other deuterostomes. However, it should be noted that NVPs containing the Shh protein have also been shown to be asymmetrically transported across the nodes of mouse embryos (Tanaka et al., 2005). It is currently unclear whether asymmetric Shh distribution in the mouse node is crucial for transducing nodal flow signal to induce asymmetric Dand5 or Nodal expression, as no mouse mutants conditionally lacking Hh signalling activity in the LRO have been analyzed. Mice have multiple paralogous genes with redundant roles for most members of the Hh signalling pathway, and Hh signalling can regulate L-R patterning in mice at different levels, as explained above. These complex networks make examining the Hh signalling function in the mouse LRO extremely difficult. For sea urchin embryos, which have a single Hh gene (Walton et al., 2006), the involvement of Hh signalling in Nodal expression regulation in the right-side ectoderm (late phase) has been demonstrated using morpholino, cyclopamine (a Smo inhibitor) and antibody (against kinesin-II, to block ciliogenesis)-mediated knockdown methods (Warner et al., 2016). However, this study does not necessarily preclude the possibility that Hh signalling can regulate the early phase of Nodal expression in the tip of the archenteron, as incomplete loss-of-function and nonspecific effects from the use of the methods described above may have had an influence. Interestingly, Ptch and Smo are co-expressed in the archenteron of sea urchin embryos, similarly to Nodal (Warner et al., 2014, 2016), which makes the above speculation more reasonable. Therefore, more studies are needed to reveal whether Hh signalling regulates asymmetric Dand5 expression or Nodal expression in species in addition to amphioxus in the future.

Bilateral symmetry breaking of vertebrate embryos relies on cilia-driven nodal fluid flow at the L-R organizer (Okada et al., 2005; Schweickert et al., 2007; Tabin and Vogan, 2003). The timing of the initial cilium incorporation into the L-R asymmetry is an area of interest to developmental biologists. A recent study found that cells in the archenteron possess motile cilia in sea urchin embryos, and the removal of the cilia using HS seawater resulted in abnormal Nodal expression and L-R defects (Tisler et al., 2016). Similarly, in ascidian H. roretzi, blocking neurula rotation driven by the motile cilia also disrupts asymmetric Nodal expression and L-R morphology (Nishide et al., 2012). In the sea urchin and H. roretzi, no Dand5 genes have been identified, and Nodal is the earliest gene asymmetrically expressed. Amphioxus embryos are heavily ciliated from the gastrula stage (Hirakow and Kajita, 1991). At the early neurula stage, the cilia begin to beat and drive the embryo to rotate within the fertilization envelope. We showed that removing cilia or blocking cilia movement in amphioxus embryos from the late mid-gastrula (G4) to the early neurula stages disrupts the asymmetric activation of Hh signalling and Dand5 expression. Given that amphioxus embryo rotation occurs after asymmetric Dand5 and Ptch expression is first detected and that blocking embryo rotation (without disturbing cilia movement) does not significantly affect L-R patterning, we speculate that cilia movement, but not embryo rotation, is responsible for amphioxus symmetry breaking. Together, these findings indicate that the involvement of motile cilia in symmetry breaking most likely emerged, at least, before the divergence of deuterostomes.

Although nodal flow was first identified in mouse embryos in 1998 (Nonaka et al., 1998), the mechanism by which nodal flow affects L-R patterning remains unclear. Several models have been proposed to explain this process, two of which (the two-cilia model and the morphogen model) have attracted the most attention. The two-cilia model posits that the motile cilia within the node generate the flow and the immotile cilia at the node edge sense the mechanical force from the flow. The morphogen model suggests that the nodal flow transports a certain substance called a morphogen preferentially to the left side of the node, which is then sensed by the morphogen receptors to initiate expression of the left-side genes such as Nodal, Lefty and Pitx2 (Tabin and Vogan, 2003). Consistent with the two-cilia model, Yoshiba et al. showed that the immotile cilia of perinodal crown cells could sense the flow in a Ca²⁺ channel protein (e.g. Pkd2)-dependent manner to induce the asymmetric expression of Cerl2 and Nodal in the node (Yoshiba et al., 2012). However, this finding could also be interpreted as evidence for the morphogen hypothesis, as the targets sensed by the immotile cilia can be either a mechanical force or a morphogen. Indeed, Tanaka et al. showed that the addition of recombinant Shh or Indian hedgehog into the embryo culture medium could elevate the left-specific Ca²⁺ level that had been suppressed by blocking FGF signalling (Tanaka et al., 2005). This suggests a role for Hh signalling in triggering an initial asymmetric Ca²⁺ elevation on the left side. Our findings in the present study, however, favour a combination of these two models. We showed that cilia of the amphioxus embryos were not only involved in transporting the Hh protein (morphogen) preferentially to the anterior-right paraxial mesoderm, but also required for Hh signal transduction. These two actions of the cilia eventually result in R>L Hh signalling and Dand5 expression in the anterior-right paraxial mesoderm. Surprisingly, we did not detect an obvious Hh fusion protein in the apical (ventral) side of the axial mesoderm or in the right paraxial mesoderm. The possible explanation for this is that the Hh protein in these domains was washed out during the WIF analysis, or the asymmetric Hh distribution may be an indirect consequence of cilia movement. As all cilia present in early neurulae, including those on the neural ectodermal cell on which caSmo is predominantly detected, are motile (see discussion below and Movie 3), we speculated that the cilia responsible for Hh signal transduction in amphioxus embryos are motile, similar to those in the sea urchin embryos (Warner et al., 2014).

The LRO in vertebrate and sea urchin embryos is derived from mesoderm and is the first place at which asymmetric gene expression occurs (Nakamura and Hamada, 2012; Tisler et al., 2016). Cells in the LRO also have motile monocilia, with an average length of 5 μm, which rotate to generate nodal flow (Okada et al., 2005; Schweickert et al., 2007). In amphioxus, the earliest asymmetric signals (such as those associated with Hh distribution, or Dand5 and Ptch expression) occur in the anterior-right mesoderm, which suggests that the GRP (mesoderm-derived) located between the anterior paraxial mesoderm in early amphioxus neurulae might be analogous to the vertebrate LRO (Li et al., 2017; Blum et al., 2014). However, monocilia on most GRP cells of the amphioxus embryo of the four sequential stages (from the G5 stage to 90 min after this stage) are much longer (∼10 μm) than those on the vertebrate LRO (Fig. S16), and they do not rotate; they only beat very weakly (Movie 3). We also tried to determine whether a unidirectional flow can be found at the GRP region by adding fluorescent beads to the GRP explant, but this experiment failed because the beads always gathered into large masses in the sea water (Fig. S17). Monocilia on the NE cells in early neurulae are of similar length (2-5 μm) to those on the vertebrate LRO, but they also move differently (they beat but do not rotate; Movie 3). The last class of monocilia observed in amphioxus early neurulae are those on EE cells; however, they also beat but not rotate and are longer than the cilia on the vertebrate LRO. Therefore, none of the three types of monocilia found in amphioxus embryos satisfies every criteria to match the cilia of the vertebrate LRO. This leaves unanswered the questions about whether the amphioxus embryo has an LRO-like structure and its possible position.

Taken together, our data show that motile cilia are essential for symmetry breaking in amphioxus through the asymmetric transport of Hh protein and the transduction of related asymmetric signals to induce asymmetric Dand5 expression. Therefore, cilia-driven symmetry breakage likely represents an evolutionarily conserved mechanism in deuterostomes.

Amphioxus (Branchiostoma floridae) were obtained from a stock maintained by Dr Jr-Kai Yu (Institute of Cellular and Organismic Biology, Academia Sinica, Taipei, Taiwan) and raised under previously described conditions (Li et al., 2012). Gametes were obtained by thermal shock (Li et al., 2013); embryos were raised as previously described (Liu et al., 2013). Hh mutants were generated as described in a previous report (Wang et al., 2015). Kif3a, Foxj1 and Dand5 mutants were generated using the TALEN method as previously described (Li et al., 2014, 2017). Homozygous mutants were generated by crossing the heterozygous animals. Genotyping of the live embryos was performed as previously described (Li et al., 2017), and genotyping of the embryos fixed with fixative was conducted with the same protocol except an extra 30 min wash in 500 ml of filtered sea water or PBS was added before lysis to remove the fixative. All embryos were staged according to the naming system described in Hirakow and Kajita (1991; 1994), unless otherwise stated. Primers adopted for gene fragment amplification are shown in Table S1.

Coding sequences of the Smo and Hh genes were amplified from cDNA libraries based on amphioxus embryos. PCR products were cloned into the pGEM-T-Easy vector (Promega) for probe synthesis or into the pXT7 vector for mRNA synthesis. A G-to-T mutation at the 1556 bp site downstream of the Smo start codon (resulting in a Trp-to-Leu change) was made to construct a constitutively active form of the protein according to the method of Xie et al. (1998) using PCR. A 6×myc tag was fused to the C terminus of the Hh and to the caSmo coding sequences using a Gibson assembly cloning kit (New England Biolabs). A form of caSmo (CLDcaSmo) with a defective ciliary localization motif was constructed according to Huangfu et al. (2003) using the PCR method. All constructs were confirmed using DNA sequencing. In vitro mRNA synthesis was conducted using a T7 mMESSAGE mMACHINE kit (Ambion). The primers used in the vector constructs described above are shown in Table S1.

Unfertilized eggs were injected and fertilized as previously described (Liu et al., 2013). For blastomere injection at the two-cell stage, embryos were de-enveloped ∼20 min after fertilization with a finely pulled plastic Pasteur dropper and then transferred at the two-cell stage to a 3 cm×4 cm coverslip coated with 4 μl 0.025 mg/ml polylysine. The injection was conducted essentially in a manner identical to the injection of an unfertilized egg. Embryos injected on the left side or right side were separated at the G5 stage and visualized under a fluorescence stereomicroscope (Olympus, SZX10).

Embryos at the desired developmental stages were fixed overnight with 4% PFA-MOPS-EGTA (wt/vol) at 4°C, transferred into 70% ethanol (vol/vol) and maintained at −20°C for use. Smo or Gli digoxigenin (DIG)-labelled antisense riboprobes (Roche) were synthesized with Sp6 or T7 RNA polymerase (Promega). Hh, Ptc, Dand5, Nodal, Lefty and Pitx RNA probes were generated as previously described (Li et al., 2017; Hu et al., 2017). WISH was performed as previously described (Yu and Holland, 2009). The stained embryos were photographed under an inverted microscope (Olympus, IX71), or they were embedded in agar-paraffin, sectioned to a thickness of 6 μm and photographed under an upright microscope (Zeiss, AX10).

For Myc or acetylated tubulin staining, the embryos were blocked in a solution containing 5% goat serum, 0.1% Triton X-100 and 0.1% bovine serum albumin (BSA) in PBS for 1 h and incubated overnight with a primary antibody against Myc (Santa Cruz Biotechnology, 9E10, 1:100) or acetylated tubulin (Sigma-Aldrich, T7451, 1:1000 or Cell Signaling Technology, 5335, 1:1000) at 4°C. Immunofluorescence signals were detected with a secondary antibody (Alexa Fluor 594-conjugated goat anti-mouse antibody, A-11032, or Alexa Fluor 488-conjugated goat anti-rabbit antibody, A-11034, Invitrogen, 1:400). DAPI (Solarbio, D8200, 1 μg/ml in PBST) was adopted for nuclear staining. Fluorescent images were acquired using a confocal microscope (Zeiss, LSM780).

Embryos for scanning electron microscopy were fixed overnight with fresh 2.5% glutaraldehyde in filtered seawater at 4°C, washed 3×10 min in 0.1 M PBS (pH 7.4), and dehydrated in an ethanol series to 100% ethanol. Samples were then dried using the critical point drying method, sprayed with platinum, placed on conductive tape, and observed under a JSM-6390 scanning microscope (JEOL).

The embryos used for immunoelectron microscopy were fixed overnight with 4% PFA-MOPS-EGTA at 4°C, dehydrated in an ethanol series, embedded in LR White resin (Sigma-Aldrich, 62661) and sectioned at 70 nm. The sections were washed 3× in 0.01 M PBS (pH 7.4), blocked in a solution containing 1% BSA, 0.05% Triton X-100, 0.05% Tween 20 in 0.01 M PBS (pH 7.4) for 20 min and incubated overnight with primary antibody against Myc (Santa Cruz Biotechnology, 9E10, 1:100) at 4°C. After being washed with ddH₂O, the sections were incubated with goat-anti-mouse IgG antibody conjugated with 10 nm colloidal gold (Electron Microscopy Sciences, 25128, 1:50) for 3 h, washed with ddH₂O, post-fixed in 2.5% glutaraldehyde in PBS (pH 7.4), stained in 2% uranyl acetate (pH 5.1) and observed under a Hitachi HT-7800 transmission electron microscope.

Embryos were de-enveloped at the G4 stage (late mid-gastrula stage) and then treated with 0.83 M NaCl seawater for 60-90 s to remove the cilia. As the embryos can recover cilia very quickly, they were also treated every 15-20 min until they reached the N1 stage. For the MC treatment, the embryos were de-enveloped at the G4 stage and then cultured in filtered seawater containing 2% MC (wt/vol) until R30 (half-hour after the initial neurula rotation stage) or the five-somite stage.

The embryos were de-enveloped at G4 stage, moved to seawater containing 0.5% MC (wt/vol) and threaded through a very fine glass needle and mounted on glass slides with double-sided foam adhesive (as shown in Fig. S12). They were released and transferred to filtered seawater at the three-somite stage and analyzed at the desired stages.

Embryos were de-enveloped at the G5 stage and cut with injection needles to obtain explants. The embryos or explants were then transferred to a 35-mm glass-bottom dish with 20-mm micro-wells filled with seawater containing 0.5% MC, and they were observed under a Zeiss Axio Imager 2 Research Microscope or a precision DeltaVision-OMX Super-Resolution Microscope (GE OMX V4).

The average optical density of the in situ hybridization signal was evaluated using Image-Pro Plus 6.0 software (Media Cybernetics). Cilia length statistics were determined using Image-Pro Plus 6.0 software. Significance was calculated using independent samples t-tests or a Fisher exact test with IBM SPSS Statistics version 19 (IBM Corporation).

We thank Dr Sebastian Shimeld for his helpful comments and language editing on the manuscript, Dr Chengtian Zhao for providing us with the acetylated tubulin antibody, and the three anonymous reviewers for their critical comments on the manuscript. We are grateful that Dr Luming Yao and Dr Caiming Wu helped us with sample preparation and electron microscopy observation and Dr Qingfeng Liu assisted us in confocal microscope observation.