Neural tube patterning by Ephrin, FGF and Notch signaling relays

Central pattern generators (CPGs) are discrete neuronal networks that drive rhythmic motor behaviors (Marder and Bucher, 2001; Wilson and Maden, 2005). Spinal cord CPGs control the swimming behavior of lampreys and fish and the repetitive movements of amniote limbs. They are composed of motoneurons and interneurons that are assembled into ‘motor pools’ innervating specific muscles. Understanding how neuronal subtypes interconnect to control locomotion has become an intensive area of research (Goulding, 2009).

Sea squirts, or ascidians, belong to the Urochordates, the sister group to the vertebrates (Delsuc et al., 2006). The tadpole larvae of Ciona intestinalis possess a typical chordate body plan with a central nervous system (CNS) composed of only ∼100 neurons (Meinertzhagen et al., 2004; Nicol and Meinertzhagen, 1991). Most of these neurons are dedicated to the swimming behavior of the tadpole, a crucial process as this is the only motile phase of the ascidian life cycle (Kajiwara and Yoshida, 1985; Mackie and Bone, 1976). The numerous advantages of Ciona as an experimental organism include a compact genome, invariant cell lineages, rapid development and simple transgenesis methods (Lemaire, 2011; Satoh, 2003). The cellular simplicity of the Ciona CNS makes it an ideal model for understanding the development and function of chordate-specific neuronal networks (Meinertzhagen et al., 2004; Meinertzhagen and Okamura, 2001).

The motor ganglion (MG) controls the swimming behavior of the tadpole. It is situated at the base of the tail in the trunk (hence the alternative term ‘trunk ganglion’) and was originally called the ‘visceral ganglion’ despite the absence of contacts with viscera (Grave, 1921). The MG contains five bilaterally symmetrical pairs of cholinergic neurons that project their axons along the muscle bands in the tail (Fig. 1A) (Horie et al., 2010; Takamura et al., 2002; Takamura et al., 2010). Together with commissural glycinergic inhibitory interneurons in the nerve cord, they drive the alternating left-right contractions of the tail that propel the tadpole forward. This behavior is retained by dissected preparations containing only the MG, nerve cord and tail (Brown et al., 2005; Nishino et al., 2010).

Each of the five pairs of MG neurons are morphologically distinct and express a particular homeodomain ‘code’ (Ikuta and Saiga, 2007; Imai and Meinertzhagen, 2007; Imai et al., 2009; Stolfi and Levine, 2011). We will hereafter refer to these pairs of neurons as single cells on one side of the embryo. The neurons of the MG, from anterior to posterior, are A12.239, A13.474, A11.118, A11.117 and A10.57 (Fig. 1F). A conserved motoneuron ‘code’ of Lhx3-Islet-Mnx suggests the occurrence of two primary motoneurons (A11.118 and A10.57) (Lee et al., 2008; Thaler et al., 2002; Thor et al., 1999). A11.118 forms prominent neuromuscular endplates (Fig. 1B). Putative interneurons A13.474 and A11.117 express Vsx (also known as Chx10), which specifies V2 spinal interneurons in vertebrates. A11.117 also expresses Pitx (Fig. 1C) (Stolfi and Levine, 2011), suggesting similarity to Pitx2+ cholinergic v0 interneurons in the vertebrate spinal cord (Zagoraiou et al., 2009). Lastly, the decussating A12.239 neuron expresses Dmbx (Fig. 1D) (Takahashi and Holland, 2004), Lhx1/5 and PouIV and might be related to an uncharacterized subset of v0 interneurons in the vertebrate spinal cord (Ohtoshi and Behringer, 2004).

Vertebrate neuronal subtypes are established along the dorsal-ventral (DV) axis by opposing bone morphogenetic protein (BMP) and Shh signaling gradients (Briscoe and Novitch, 2008; Briscoe et al., 1999; Ericson et al., 1996; Ericson et al., 1997; Lee et al., 2000). This mechanism does not pattern the Ciona MG, despite conserved expression of BMP2/4 and Shh in the Ciona neural tube (Hudson et al., 2011). We recently presented evidence that fibroblast growth factor (FGF) and Notch signaling specify different MG neuronal subtypes (Stolfi and Levine, 2011). Here, we present evidence that Ephrin/Eph-mediated inhibition of FGF signaling in progenitors of the posterior MG leads to the localized activation of Notch signaling in the anterior MG. This FGF-Notch signaling relay leads to sequential homeodomain codes along the anterior-posterior (AP) axis that are similar to the DV patterning domains seen in the vertebrate spinal cord. Thus, divergent signaling strategies are used to specify related neuronal subtypes in the Ciona MG and vertebrate spinal cord.

Templates for in vitro transcription of RNA probes are from the Ciona intestinalis Gene Collection Release 1 (http://ghost.zool.kyoto-u.ac.jp/indexr1.html) (Satou et al., 2005; Satou et al., 2002), with plate/well identifiers cross-referenced to annotation on the ANISEED database (Tassy et al., 2010): HesB (04g10), Ngn (29n04), Mnx (42o14), Lhx1/5 (44f09), Lhx3 (25m02), En (28a10), PouIV (32g05), Pax3/7 (42e20), Delta2 (16e18), SoxB1 (01k09). Template for EphrinAb probe was made by cloning the full-length coding sequence (see below) into the same backbone as the other gene collection clones. Templates for Pax6, Nkx6, Dmbx and Vsx riboprobes were constructed previously (Stolfi and Levine, 2011).

En reporter was constructed using a fragment –2085 to +24 bp from the translation start of En. FGF8/17/18. Dmbx, Vsx and Msx reporters were described previously (Russo et al., 2004; Stolfi and Levine, 2011). Dmbx minimal enhancer from C. intestinalis was PCR-amplified with the following primers: Dmbx min F: TTTGTGACGTCACGCCTTAAC; Dmbx min R: AGTACTATGACGTTACAATCC. Dmbx minimal enhancer from C. savignyi was PCR-amplified with the following primers: Dmbx-Cs min F: CACTTTCCGAATCAACGTCC; Dmbx-Cs min R: TTGTCTGTCTCTGCGTACGTC.

Eph3ΔC was constructed by removing the intracellular C-terminus of Eph3, leaving amino acid (aa) residues 1-579 (Picco et al., 2007). Eph1ΔC (aa 1-616), Eph2ΔC (aa 1-612), Eph4ΔC (aa 1-611) and Eph-likeΔC (aa 1-683) were similarly designed and constructed. Nomenclature is according to the list of RTK genes on the Ghost database (Satou et al., 2005). Constitutively active Eph3 (caEph3) was made by fusing the C-terminal transmembrane and intracellular domains (aa 512-1027) of Eph3 to the self-dimerizing N-terminus (aa 1-391) of Drosophila Torso receptor tyrosine kinase (Sprenger and Nusslein-Volhard, 1992). EphrinAb, Pax3/7 and Pax6 were cloned from a tailbud embryo cDNA library. Coding sequences of Ngn, SoxB1 and all other Ephrins [EphrinAa (41n20), EphrinAc (07i06), EphrinAd (01j20)] were taken from collection clones. Ngn::WRPW was constructed by fusing the bHLH domain (aa 84-173) to a Groucho-recruitment WRPW motif (aa 326-337) from Drosophila Hairy (Wainwright and Ish-Horowicz, 1992). Pax3/7::WRPW was made by fusing the N-terminal paired box and homeodomain (aa 3-253) to WRPW. Pax3/7::VP16 was made by fusing aa 3-253 to the transactivation domain of the Herpes virus VP16 activator (aa 413-490) (Rusch and Levine, 1997). Dmbx::WRPW was made by fusing the first 116 amino acid residues from Dmbx to the WRPW motif. dnFGFR, caFGFR, Su(H)-DBM and Delta2 constructs have been described (Davidson et al., 2006; Hudson and Yasuo, 2006; Pasini et al., 2006; Shi and Levine, 2008).

Animals were obtained from Pillar Point Marina (Half Moon Bay, CA, USA) or ordered from M-Rep (San Diego, CA, USA). Embryos were manipulated as previously described (Christiaen et al., 2009) and grown at 20°C, unless otherwise stated.

Control embryos for fluorescent protein-based assays were electroporated with lacZ plasmids to control for transfection load. Control embryos for in situ hybridizations were electroporated with GFP/mCherry plasmids. For experiments in which increased dose of perturbation was desired, plasmid concentration was simply increased, with transfection load equalized by co-electroporation with lacZ plasmids.

Images were photographed using a Zeiss AxioImager A.2 upright compound microscope, a Leica SP2 upright confocal microscope or a Zeiss700 inverted confocal microscope.

U0126 (Promega) and DAPT (Enzo Life Sciences) were resuspended in DMSO and diluted in seawater to 10 μM and 100 μM, respectively. Both were added to seawater immediately prior to the start of treatment, to minimize precipitation. Control embryos were treated with DMSO alone.

Fluorescence in situ hybridization was performed as described (Beh et al., 2007). Two-color in situ hybridizations were performed by labeling one probe with digoxigenin-UTP and the other with fluorescein-UTP. Sequential TSA-plus (Perkin-Elmer) tyramide-based detection using peroxidase domain (POD)-conjugated anti-digoxigenin and anti-fluorescein Fab fragments (Roche) was performed with a 10 minute 0.01 N HCl inactivation step between the first probe detection and incubation with the second POD conjugate. β-Galactosidase detection was carried out using mouse anti-β-galactosidase primary (1:1000, Promega #Z378) and Alexa Fluor-conjugated secondary antibodies (1:1000, Invitrogen).

For dpMAPK detection, embryos were fixed in a PIPES-sucrose-FA buffer (2 M sucrose, 0.5 EGTA, 0.5 M PIPES, 10% formaldehyde) for 30-60 minutes at room temperature. Fixed embryos were quenched with 0.3% H₂O₂ in 0.3% horse serum/PBS/Tween for 30 minutes and incubated with monoclonal mouse anti-dpERK primary antibody (1:500, Sigma #M9692). Embryos were then incubated with biotinylated horse anti-mouse IgG (Vector Laboratories #BA-2000) and avidin-biotin-peroxidase complexes using Vectastain Elite ABC Reagent R.T.U. kit (Vector Laboratories), according to the manufacturer’s recommendations. TSA-plus amplification was carried out as recommended. Co-detection of β-galactosidase was performed using rabbit anti-β-galactosidase (1:1000, Invitrogen #A-11132), coupled to Alexa Fluor-conjugated secondary (1:1000).

When counterstaining certain embryos, fluorophore-conjugated phalloidin or phallicidin (1:200-1:500, Invitrogen) was used to reveal cell shape or Hoechst (1:1000-1:4000, Invitrogen) to stain nuclei.

A Pax3/7 fragment lacking the disordered C-terminal region (Pax3/7ΔC, residues 3-259) was cloned into pRSETB (Invitrogen) for expression in Escherichia coli. pRSETB-Pax3/7ΔC was transformed into BL21-CodonPlus cells (Agilent) and grown in 2 liters of TB. Expression was induced at optical density (OD) 0.6 with 100 μM IPTG and grown for an additional 16 hours at 16°C. Cells were harvested by centrifugation and resuspended in Ni-A Buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 10% glycerol, 0.5 mM TCEP, 25 mM imidazole, 50 mM AEBSF). Following sonication at 4°C, lysate was cleared by centrifugation at 15,000 g for 45 minutes, and (His)₆-Pax3/7ΔC was eluted with Ni-B Buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 10% glycerol, 0.5 mM TCEP, 350 mM imidazole) from 5 ml His-Trap Nickel Column (GE) using AKTA-Explorer FPLC (GE). Fractions containing (His)₆-Pax3/7ΔC were identified by SDS-PAGE, pooled and purified by size exclusion chromatography using analytical S200 Column in Buffer-200 (200 mM NaCl, 20 mM HEPES pH 7.5, 5% glycerol and 0.5 mM TCEP). (His)₆-Pax3/7ΔC eluted at a molecular weight of 31 kDa. Fractions containing soluble, monomeric (His)₆-Pax3/7ΔC were pooled and concentrated to 2 mg/ml.

Oligonucleotides (26-mer) containing the wild-type (GCGCGAAAT) or mutant (GGCCGAAAT) Pax3/7 binding site (Elim Biopharmaceuticals) and their respective complements were annealed by mixing 10 pmol of each oligo in a buffer containing 10 mM Tris pH 7.5 and 100 mM NaCl. The reaction was heated to 95°C for 5 minutes and allowed to cool slowly to room temperature over 1 hour. The oligonucleotides were labeled with γ-³²P-ATP and Optikinase (USB) following the manufacturer’s instructions. The reactions were spun twice through P-6 columns (Bio-Rad) to remove free nucleotides.

EMSAs were performed using standard procedures. For non-competition assays, purified Pax3/7ΔC was serially diluted to create stocks of 10 μM to 304 pM. Ten microliters of each stock was mixed with 10 μl of buffer containing labeled oligonucleotide for final protein concentrations ranging from 5 μM to 152 pM. The binding reaction conditions were 20 mM HEPES pH 7.6, 125 mM KCl, 0.2 mM EDTA, 0.5 mM DTT, 4 mM MgCl₂, 0.1 nM oligonucleotide (∼2500 cpm). Reactions were incubated at room temperature for 45 minutes and run on a 5% native polyacrylamide gel (60:1 acrylamide:bis acrylamide) in 0.5× TBE for 2 hours at 180 V, 4°C. Gels were dried and visualized using a Typhoon phosphorimager (GE).

For the assays containing non-specific competitor DNA, reactions were set up and performed as above with the following exceptions: reactions contained motif-containing oligonucleotide at 0.1 nM and 2.5 ng/μl poly[dI-dC] (1500-fold excess) and the final protein concentrations in each reaction ranged from 62.5 μM to 2.5 nM in twofold steps.

The invariant lineages that produce the five pairs of MG neurons have been described in detail (Cole and Meinertzhagen, 2004). The A9.30 lineage produces four of the five neurons, whereas the A10.57 motoneuron arises from the more posterior A9.29 lineage (Fig. 1F). These neurons exhibit partially overlapping domains of expression of Engrailed (En), Pax3/7, Pax6, Nkx6 and Lhx3 along the AP axis, similar to the DV patterning of the vertebrate spinal cord (Fig. 1E) (Briscoe et al., 2000; Imai et al., 2009).

Localized expression of BMP and Shh establishes these DV homeodomain codes in the vertebrate spinal cord. Although BMP and Shh exhibit similar DV expression profiles in the Ciona neural tube, they are not required for MG neuronal specification (Hudson et al., 2011). Instead, FGF and Notch signaling appear to control this process. We recently showed that targeted expression of a dominant negative FGF receptor (dnFGFR) causes the A9.30 lineage to produce only A11.117 interneurons (the posterior-most identity within the A9.30 lineage). Moreover, targeted expression of a dominant negative form of Su(H), the transcriptional effector of Notch signaling, resulted in the duplication of A12.239 decussating neurons (Stolfi and Levine, 2011). In the present study, we sought to determine how FGF and Notch signaling work in concert to pattern the Ciona MG.

Progressively older embryos were treated with the irreversible MEK (MAPKK) inhibitor U0126 (Fig. 2A,B). Expression of En, Mnx and Vsx was monitored using two-color fluorescence in situ hybridization, to assess the fate of A9.30 daughter cells (A10.60 and A10.59) and granddaughter cells (A11.120, A11.119, A11.118, A11.117). En is expressed in A11.120 and A11.119, whereas Mnx and Vsx are expressed in A11.118 and A11.117, respectively. Early application of U0126 [6 hours post-fertilization (hpf)] causes Vsx to be expressed in all four granddaughter cells rather than being restricted to the posterior-most A11.117 neuron. We interpret this result as the transformation of A10.60 into A10.59, followed by the conversion of A11.118 to the A11.117 fate (Fig. 2C).

By contrast, treatment of older embryos with U0126, at 7 and 8 hpf, does not abolish En expression. Instead, Vsx is ectopically expressed only in A11.118, at the expense of Mnx (transformation of the A11.118 motoneuron into a duplicated A11.117 interneuron) (Fig. 2A-C). This result suggests that the initial specification of the A10.60 and A10.59 progenitor cells is normal, but there is a conversion of A11.118 into an A11.117 fate. Treatment at 9 hpf does not alter expression of En, Vsx nor Mnx. Together, the preceding results suggest that differential MAPK activity is first required for the delineation of the A10.60 and A10.59 fates, and subsequently required for distinguishing A11.118 and A11.117 subtypes. As U0126 is an irreversible inhibitor, treatment at 6 hpf affects both fate choices, resulting in four A11.117-like cells (Fig. 2B,C). Consistent with these observations, di-phosphorylated MAPK (dpMAPK) can be detected in A10.60 and A11.118 of wild-type embryos (supplementary material Fig. S1). Interestingly, Mnx expression is expanded within the A9.29 lineage upon U0126 treatment (Fig. 2A). It is normally restricted to the A10.57 motoneuron, the posterior daughter cell of A9.29. Thus, the specification of the A10.57 motoneuron might also depend on differential regulation of MAPK activity.

Ephrins have been shown to mediate inhibition of FGF/MAPK in the Ciona embryo (Picco et al., 2007; Shi and Levine, 2008). Ephrins are membrane-anchored ligands that interact with Eph receptor tyrosine kinases. Ephrin/Eph signaling suppresses MAPK in receiving cells, inhibiting FGF- and MAPK-dependent processes. This has been shown to act at the level of Ras, downstream of the FGF receptor but upstream of MAPK (Elowe et al., 2001; Miao et al., 2001). EphrinAb is expressed in A9.29, which is situated just posterior to A9.30 (Fig. 3A) (Hudson et al., 2007). Thus, EphrinAb+ A9.29 descendants continuously contact the posterior-most cell of the A9.30 lineage. As such, EphrinAb is a strong candidate for the positional cue for MAPK inhibition in A10.59 and A11.117.

To determine whether Ephrin is essential for differential MAPK activity in the A9.30 lineage, a truncated form of the Eph3 receptor that sequesters Ephrin ligand was selectively expressed in this lineage using FGF8/17/18 regulatory sequences (FGF8/17/18>Eph3ΔC) (Picco et al., 2007). The entire lineage is converted to En-expressing cells, suggesting ectopic activation of MAPK+ and the transformation of A10.59 into A10.60 (Fig. 3B-D). By contrast, ectopic activation of Ephrin-Eph signaling (via a constitutively activated form of Eph3) abolishes En expression, suggesting the reciprocal transformation: A10.60 into A10.59 (supplementary material Fig. S2B). Similarly, treatment of FGF8/17/18>Eph3ΔC-electroporated embryos with U0126 also abolishes En expression (supplementary material Fig. S2C), confirming that Ephrin signaling acts upstream of, and antagonistic to, MAPK to control cell fate choice.

Ectopic expression of EphrinAb (via FGF8/17/18>EphrinAb) inverts the position of A11.117 and A11.118 fates (Fig. 3B,E). Under this condition, the dividing A10.59 cell encounters higher levels of EphrinAb from its anterior A10.60 sister cell. According to the model of Ephrin-MAPK antagonism, this is expected to result in an inversion of A11.118/A11.117 asymmetric MAPK activation and cell fate (explained in supplementary material Fig. S3). Furthermore, the first fate choice (A10.60 versus A10.59) is not affected because Ephrin cannot downregulate MAPK in cis. Overexpression of the other four Ciona EphrinA ligands did not cause this phenotype (data not shown). The preceding results suggest that EphrinAb in the A9.29 lineage is the positional cue that inhibits MAPK in A10.59 and A11.117.

The first indication of a role for Delta/Notch signaling in patterning the anterior half of the MG came from targeted inhibition of Notch using a dominant-negative form of Su(H) [FGF8/17/18>Su(H)-DBM]. These embryos exhibit a duplication of the A12.239 neuron, shown by OFF-ON-OFF-ON expression of Dmbx, PouIV and Lhx1/5 in the A10.60 lineage (Fig. 4A; supplementary material Fig. S4A) (Stolfi and Levine, 2011). Similar results are obtained with the γ-secretase inhibitor DAPT, which inhibits proteolytic cleavage of activated Notch receptors (supplementary material Fig. S4B). The duplication of A12.239 identity appears to result from the conversion of A11.119 into an A11.120-like progenitor upon inhibition of Notch signaling (Fig. 4B).

Further evidence for this fate transformation (A11.119 into an ectopic A11.120 progenitor cell) was obtained by visualizing the expression of Pax3/7, a key marker of A11.120 and its descendants (Fig. 4C). Pax3/7 is initially expressed in A11.120, and subsequently in the daughter cells of A11.120, with enhanced expression in A12.239 (posterior daughter cell of A11.120). Pax3/7 expression is not normally detected in A11.119 or its descendants. However, in embryos electroporated with FGF8/17/18>Su(H)-DBM (inhibition of Notch signaling), Pax3/7 expression is expanded into the A11.119 lineage (Fig. 4C). Further evidence for the transformation of A11.119 into a duplicated A11.120 progenitor cell was obtained by investigating the expression of marker genes for A11.119, including SoxB1, Pax6 and HesB. These markers are lost upon targeted inhibition of Notch signaling via FGF8/17/18>Su(H)-DBM (supplementary material Fig. S4C). By contrast, there was only a modest reduction of Nkx6 expression in A11.119, suggesting that it might lie downstream of A11.119 specification (supplementary material Fig. S4C).

We next sought to identify a ligand that could activate Notch in A11.119. Delta2 is selectively transcribed in A10.59 and its descendants, A11.118 and A11.117 (Fig. 4D,E). Delta2 expression is also seen later in A11.120 (Fig. 4E). Perhaps Delta2 ligand on the surface of A11.118 activates Notch in the neighboring A11.119 cell. By contrast, A11.120 is not in contact with a Delta2+ cell and, according to the model of lateral inhibition, initiates Delta2 expression (Doe and Goodman, 1985; Fortini, 2009).

To determine whether Delta2 expression in the posterior MG (e.g. A10.59) is required for Notch activation in A11.119, we blocked A10.59 specification by transformation with the FGF8/17/18>Eph3ΔC transgene (constitutive MAPK activation in A10.60 and A10.59). Early Delta2 expression in A10.59 is lost (Fig. 4D) and Delta2 is ectopically expressed in A11.119 (Fig. 4E). As a result, Pax3/7 and Dmbx are de-regulated, as the cells are left to resolve their Notch signaling status through lateral inhibition without the positional cue normally provided by A10.59/A11.118 (Fig. 4E,F). If Notch signaling is inhibited in this background by co-electroporation with FGF8/17/18>Su(H)-DBM, all cells of the lineage activate Dmbx, as expected (supplementary material Fig. S4D).

To test whether Delta2 is sufficient to impinge on A11.119/A11.120 fate, we mis-expressed it using the En driver (En>Delta2). En expression is initiated in both daughter cells of A10.60, but is strongly upregulated in A11.119 and downregulated in A11.120. Electroporation of En>Delta2 results in a proportion of embryos showing inverted expression of Pax3/7, suggesting a conversion of A11.120 into A11.119 and vice versa (supplementary material Fig. S4E). As a result, Dmbx reporter expression is similarly altered (supplementary material Fig. S4F). Taken together, these results suggest that a Delta2-mediated relay from the posterior MG patterns the anterior MG; Delta2 expression in the posterior A9.30 lineage is dependent on the inhibition of FGF signaling by localized EphrinAb (supplementary material Fig. S5).

We hypothesized that the localized expression of Pax3/7 in A11.120 might be required for specification of its daughter cell A12.239, as ectopic expression of Pax3/7 in A11.119 correlates with ectopic activation of marker genes normally expressed in A12.239: Lhx1/5, PouIV and Dmbx. Moreover, inhibition of Pax3/7 via morpholino injection results in loss of PouIV and Lhx1/5 expression (Imai et al., 2009). Thus, it would appear that Lhx1/5, PouIV and Dmbx are directly or indirectly co-regulated by Pax3/7.

This was tested by mis-expression of Pax3/7 throughout the A9.30 lineage with the FGF8/17/18 driver (FGF8/17/18>Pax3/7). The resulting embryos exhibit ectopic expression of the Dmbx>GFP reporter gene throughout the A9.30 lineage (Fig. 5A). Strikingly, the mis-expression of Pax3/7 also results in the formation of supernumerary decussating neurons (Fig. 5B). Similar results were obtained with a heterologous Pax3/7::VP16 transcriptional activator (data not shown), whereas a repressor form of Pax3/7 (Pax3/7::WRPW) strongly represses Dmbx (Fig. 5A). Thus, Pax3/7 appears to function as an activator that either directly or indirectly activates Dmbx expression in A12.239.

To determine whether Pax3/7 directly activates Dmbx, we analyzed the sequence of a minimal ∼300 bp enhancer from the Dmbx 5′ regulatory region that is sufficient to mediate low levels of restricted expression in the A12.239 decussating neuron (Fig. 5D). A putative Pax3 binding site (Lagha et al., 2008) was identified within this minimal enhancer (Fig. 5C). Mutation of this site abolishes the activity of the enhancer (Fig. 5D). The Pax3/7 binding site is also conserved in the related species C. savignyi and is essential for the activity of the C. savignyi minimal enhancer in transgenic C. intestinalis embryos (Fig. 5E).

The preceding results suggest that Pax3/7 directly activates Dmbx expression in the A12.239 neuron. Additional evidence was obtained by in vitro binding assays. Gel shift assays were performed with recombinant Ciona Pax3/7 protein and double-stranded DNA oligonucleotides containing either the normal or mutant version of the Pax3/7 binding motif in the presence of a 1500-fold mass excess of the nonspecific competitor poly dI-dC. Under these conditions, Pax3/7 exhibited a 3.5-fold increase in binding to the wild-type motif compared with the mutant motif (Fig. 5F). This appears modest when contrasted with the dramatic effect the mutation has on the minimal Dmbx enhancer in vivo. However, mutation of the Pax3/7 site within the context of the ‘full-length’ Dmbx enhancer results in a similarly moderate decrease in activity (expression seen in 12/100 mutant versus 52/100 wild type; data not shown), suggesting that the minimal enhancer is particularly sensitized to binding site mutations. We conclude that Pax3/7 directly activates Dmbx expression.

Pax3/7 is also expressed in additional cell types, most conspicuously in neural plate border cells (supplementary material Fig. S6A). We hypothesized Pax3/7 is cooperating with the neurogenic bHLH factor Neurogenin (Ngn) to activate Dmbx in the MG. Ngn is transiently expressed in MG precursors and is maintained in A12.239 (supplementary material Fig. S6B,C) Expression of a repressor form of Ngn (Ngn::WRPW) abolishes Dmbx expression (Fig. 5G). Furthermore, mis-expression of Ngn in the neural plate border, where Pax3/7 is normally present, is sufficient to activate Dmbx in these cells (Fig. 5H). Thus, Ngn and Pax3/7 cooperatively activate Dmbx and specify an A12.239 decussating MG interneuron fate.

The preceding results suggest that Pax3/7 expression in A11.120 activates Dmbx (and perhaps additional makers) in A12.239 (posterior daughter cell of A11.1120) but not A12.240 (anterior daughter cell of A11.120). What inhibits Dmbx expression in A12.240? We reasoned that this was linked to the choice between neuronal and non-neuronal fate. In contrast to the A12.239 neuron, A12.240 is an undifferentiated cell that expresses the neural progenitor marker SoxB1 and gives rise to four ependymal cells (i.e. quiescent stem cells) (Horie et al., 2011).

Delta/Notch-mediated lateral inhibition is frequently invoked as a determinant of neuron/non-neuron cell fate decisions (Fortini, 2009; Pierfelice et al., 2011). Downregulation of Notch signaling in the anterior MG results in occasional ‘twin’ neurons. That is, both A12.240 and A12.239 undergo neurogenesis and express Dmbx. However, this effect is not enhanced by increasing levels of Notch inhibition (Fig. 6E; supplementary material Fig. S7). We thereby sought to identify other signaling processes that might influence the asymmetric fates of A11.120 daughter cells.

Involvement of MAPK in distinguishing A12.240 from A12.239 was suggested by immunostaining for dpMAPK, which revealed MAPK activation in A12.240 but not A12.239 (Fig. 6A,B). FGF/MAPK is known to inhibit neural differentiation in vertebrates (del Corral et al., 2002; Mathis et al., 2001). We hypothesized that FGF/MAPK signaling is required for SoxB1 expression and repression of neurogenesis in A12.240. To test this idea, FGF/MAPK signaling was perturbed by late addition of the MEK inhibitor U0126, in order to avoid disruption of earlier MAPK-dependent patterning (Fig. 6C,D). SoxB1 expression is abolished in A12.240 (Fig. 6C) and there is ectopic activation of Dmbx, resulting in twin neurons (Fig. 6D). Axonal projections from these ‘twins’ are often impaired, although they still extend contralaterally. The ectopic decussating neuron sometimes projects its axon anteriorly instead of towards the tail (supplementary material Fig. S8).

This twinning phenotype was mimicked by targeted expression of dnFGFR using the En driver (Fig. 6D). Moreover, combined inhibition of both FGF signaling (En>dnFGFR) and Notch signaling [FGF8/17/18>Su(H)-DBM] results in the transformation of all four descendants of A10.60 into Dmbx+ decussating neurons (Fig. 6D). In these experiments, inhibition of Notch signaling results in the transformation of A11.119 into A11.120, and both sets of daughter cells lack MAPK activation and express Dmbx. These results suggest that FGF-MAPK signaling inhibits neurogenesis in A12.240 as well as in ectopic A12.240-like cells.

Mis-expression of Dmbx using the En driver unexpectedly recapitulated the twinning phenotype (Fig. 6D). Upon electroporation with either En>dnFGFR or En>Dmbx, there is a loss of dpMAPK staining in A12.240 (Fig. 6F). Dmbx1 is thought to act as a repressor in vertebrates, and we found that a repressor form of Dmbx (Dmbx::WRPW) recapitulates the activity of the full-length protein (n=51/69, data not shown). Thus, we hypothesize that Dmbx can promote its own expression by repressing FGF/MAPK signaling.

In vertebrates, SoxB1 genes promote a neural progenitor identity but can inhibit neuronal differentiation by interfering with proneural gene activity (Bylund et al., 2003; Graham et al., 2003; Holmberg et al., 2008). To test the role of SoxB1, we overexpressed it using the En driver (Fig. 6G). This results in loss of Dmbx reporter expression, consistent with the conversion of A12.239 to A12.240. Electroporation of En>SoxB1 also overrides the ‘twinning’ phenotype of En>dnFGFR or En>Dmbx. Co-electroporation with En>Ngn partially rescues Dmbx expression, suggesting that SoxB1 might function downstream of FGF signaling to inhibit Ngn activity.

It is conceivable that Delta/Notch signaling itself contributes to asymmetric activation of MAPK in A11.120 daughter cells. First, we observed a moderate ‘twinning’ effect upon Notch inhibition (Fig. 6E; supplementary material Fig. S7). Second, Delta/Notch lateral inhibition effectors Ngn and HesB show mutually exclusive expression patterns, which is consistent with the lateral inhibition model: Ngn is expressed in A12.239, whereas HesB is expressed in A12.240 (supplementary material Fig. S6B). Delta/Notch signaling could bias Ngn and Pax3/7 expression to A12.239. Ngn and Pax3/7 cooperatively activate Dmbx, which, in turn, downregulates MAPK activity, thereby repressing SoxB1 expression and promoting cell-cycle exit. Thus, we propose that a Notch-FGF relay, mediated by Dmbx, is operating to determine asymmetric neuronal differentiation (summarized in Fig. 7B).

We have presented evidence that the specification of Ciona MG neuronal subtypes depends on short-range Ephrin, FGF and Delta/Notch signals. The A9.30 blastomere expresses FGF8/17/18, which serves as an intrinsic source of FGF signaling within the lineage (Imai et al., 2009). Differential MAPK activity within the A9.30 lineage depends on EphrinAb in the neighboring A9.29 lineage (summarized in Fig. 7A). Inhibition of MAPK in the posterior A10.59 cell leads to expression of a Delta2 ligand. This, in turn, triggers Notch signaling in the posterior daughter of A10.60 (A11.119). The activation of Notch signaling in A11.119 restricts the expression of Pax3/7 to the anterior-most cell of the lineage, A11.120. Pax3/7 appears to be a direct activator of Dmbx expression in the posterior daughter of A11.120 (A12.239). Differential Notch and MAPK activity in the A11.120 daughter cells promotes Dmbx expression and neurogenesis in A12.239 (Notch^OFF, MAPK^OFF), and SoxB1 expression and maintenance of the progenitor state in A12.240 (Notch^ON, MAPK^ON). Perturbation of different components of these pathways at different times and in different cells results in tadpoles with altered configurations of MG neurons (Fig. 7C).

Delta/Notch signaling, propagated by lateral inhibition, could be providing the spatial cue for differential MAPK+ activity in the A12.240 and A12.239 sister cells. It is possible that A12.240 experiences Notch signaling leading to restricted expression of Ngn in A12.239, and localized activation of Dmbx by the combination of Ngn and Pax3/7. Dmbx somehow inhibits MAPK activity, thereby repressing SoxB1 expression in A12.239 and promoting differentiation (Fig. 7B). In zebrafish, Dmbx1a has been shown to promote cell-cycle exit and differentiation of neurons (Wong et al., 2010). It will be interesting to see whether this is occurring in the context of Delta/Notch lateral inhibition and whether it functions to counteract FGF and SoxB1, as seen in the Ciona MG.

The patterning of the Ciona MG depends on extracellular positional cues that are distinct from those operating in the vertebrate spinal cord. The MG is patterned by a succession of short-range Ephrin, FGF and Delta/Notch signaling, whereas the spinal cord depends on long-range Shh and BMP gradients. Despite these distinct spatial patterning mechanisms, there are similarities in the gene expression profiles of the neuronal progenitors. This is apparent in both the AP-DV correspondence of transcription factor expression profiles as well as the apparent conservation of combinatorial transcriptional codes for motoneurons and interneuron subtypes.

Despite the contrast in short-range versus long-range positional cues between the two systems, certain signaling strategies might be conserved. In the vertebrate spinal cord, downregulation of FGF signaling is required for neurogenesis (del Corral et al., 2002), and Notch has been shown to regulate neuronal subtype diversification (Peng et al., 2007), suggesting that FGF and Notch signaling could be used for short-range ‘fine-tuning’ of cell fates within broad DV patterning domains. Furthermore, retinoic acid (RA)-dependent Hox1 expression in both the Ciona anterior MG (Imai et al., 2009; Nagatomo and Fujiwara, 2003) and the vertebrate hindbrain (Marshall et al., 1992) raises the intriguing possibility that ascidians have consolidated DV and AP patterning of the CNS along a single axis.

Beyond the mechanisms underlying patterning and cell fate choice, the Ciona MG and vertebrate spinal cord might share gene regulatory networks controlling axon guidance and neuronal connectivity. We have shown that Pax3/7 regulates the expression of Dmbx and Lhx1/5 expression in A12.239 and is sufficient to impose a contralateral projection when mis-expressed in other MG neurons. In vertebrates, Pax3 and Pax7 are required for ventral commissure formation in the spinal cord (Mansouri and Gruss, 1998). Lhx1/5 is a known marker of commissural dorsal spinal cord interneurons (Gowan et al., 2001; Reeber et al., 2008), though it remains to be seen whether Dmbx1+ spinal cord neurons are also decussating (Gogoi et al., 2002; Ohtoshi and Behringer, 2004). Thus, ascidians and vertebrates might share a conserved Pax3/7-dependent regulatory network to specify a decussating interneuron identity.

We thank N. Kam, O. Hobert, S. Amacher, J. Gerhart, L. Feldman and members of the Levine laboratory for comments; and W. Shi for sharing Eph3ΔC and Eph4ΔC constructs.