Lhx3 and Lhx4 suppress Kolmer–Agduhr interneuron characteristics within zebrafish axial motoneurons

Progressive restriction of cell fate during development is a complex process that requires coordinated changes in gene expression. After terminal division of progenitors, unique combinations of transcription factors promote acquisition of post-mitotic characteristics that define individual cell types. However, cells closely related by lineage share a common regulatory history and express many of the same transcription factors. This is particularly true in the vertebrate ventral spinal cord, where different types of neurons derive from a single progenitor domain specified by a common set of transcription factors (Lewis, 2006).

To understand how the fates of closely related neurons are specified, we examined derivatives of the progenitor of motoneuron (pMN) domain in the zebrafish ventral spinal cord. The zebrafish pMN lineage is well-characterized and generates several alternative neuron classes (Park et al., 2004). Analysis of these neuron classes is simplified because each spinal hemisegment has only a few neuron types, the development of which can be followed in intact embryos (Fig. 1) (Lewis and Eisen, 2003; McLean and Fetcho, 2008). Between 9 and 15 h post fertilization (hpf), the pMN domain generates 3-4 early-born primary MNs (PMNs) (Myers et al., 1986), as well as two interneurons (INs), Kolmer–Agduhr′ (KA′) and ventral-longitudinally descending (VeLD) (Park et al., 2004), within each spinal hemisegment. Both PMNs and these two INs can be identified by soma position, axon morphology and gene expression (Bernhardt et al., 1990,, 1992; Kuwada et al., 1990). After 15 hpf, the pMN domain generates additional cell types: smaller, more numerous secondary MNs (SMNs); oligodendrocyte precursors; and a third class of INs, circumferentially descending (CiD) (Park et al., 2004). CiDs are morphologically and molecularly similar to excitatory V2a INs that arise from the dorsally adjacent p2 progenitor domain later in development (Batista et al., 2008; Kimura et al., 2008). At early stages these cells can be distinguished because CiDs are olig2-positive, whereas V2a INs are vsx1-positive (Kimura et al., 2006; Park et al., 2004). As in amniote vertebrates, zebrafish p2 progenitors also generate inhibitory V2b INs (Batista et al., 2008; Kimura et al., 2008).

After their terminal divisions, pMN-derived neurons express unique combinations of transcription factors from three different families. PMNs and SMNs are Islet⁺ Mnx⁺ Lhx3⁺; VeLDs are Islet⁻ Mnx⁺ Lhx3⁺; CiDs are Islet⁻ Mnx⁻ Lhx3⁺; KA′s are Islet⁻ Mnx⁻ Lhx3⁻ (Batista et al., 2008; Hutchinson and Eisen, 2006; Kimura et al., 2006; Seredick et al., 2012). Knockdown of either Islet or Mnx family members compromises aspects of MN identity, and PMNs acquire features of specific pMN-derived INs. For example, in the absence of Islet proteins, PMNs aberrantly express GABA and have ipsilaterally descending axons in the ventral longitudinal fasciculus, characteristic of VeLDs (Hutchinson and Eisen, 2006). By comparison, in the absence of Mnxs, one PMN subtype aberrantly expresses vsx2 and vglut and often has ipsilaterally descending axons in the lateral neuropil, which are characteristics of CiDs (Seredick et al., 2012). These results suggest that different transcription factor families expressed in post-mitotic MNs are dedicated to suppressing specific alternative IN programs within MNs. This prompted us to test the hypothesis that in PMNs, Lhx family members suppress characteristics of the remaining pMN-derived IN, KA′.

Lhx3 is an LIM homeodomain transcription factor that plays pivotal roles in MN and IN fates during neural development in both vertebrates and invertebrates (Thaler et al., 2002; Thor et al., 1999). In vertebrates, genome duplication created an Lhx3 paralog, Lhx4, that can substitute for Lhx3 in formation of V2a- and MN-promoting complexes (Gadd et al., 2011). Although mouse knockouts of either Lhx3 or Lhx4 are perinatally lethal (Li et al., 1994; Sheng et al., 1996), MNs are generated normally (Sharma et al., 1998). However, at cervical levels, MN subtype identity is altered in the combined absence of Lhx3 and Lhx4. Cervical MNs, the axons of which normally exit ventrally, acquire features of spinal accessory column (SAC) MNs, and their axons exit dorsally (Sharma et al., 1998). Importantly, the fate of MNs at other axial levels remains unknown. Despite intensive investigation of Lhx3 involvement in regulatory networks that promote MN, V2a and V2b fates (Joshi et al., 2009; Lee et al., 2012,, 2008,, 2009; Lee and Pfaff, 2003; Thaler et al., 2002), it remains untested whether Lhx3 or Lhx4 is necessary for ventral spinal cord IN specification.

Here, we show that zebrafish Lhx4 is expressed, like Lhx3, in MNs and INs derived from both pMN and p2 domains. Morpholino oligonucleotide (MO) knockdown of Lhx3 and Lhx4 revealed that, at thoracic levels, many cells fated to become MNs aberrantly acquire characteristics of KA′ INs. In addition, overexpression of Lhx3 in pMN-derived neurons and progenitors causes an increase in CiDs at the expense of KA′s. Moreover, Lhx3 and Lhx4 are necessary for specification of many late-developing V2a and V2b INs. These findings provide new evidence of roles for Lhx3 and Lhx4 in the fates of specific ventral spinal cord INs and demonstrate that these Lhx proteins suppress a competing KA′ IN program within both trunk level axial MNs and pMN-derived CiD INs.

Zebrafish lhx3 is expressed in pMN-derived PMNs, SMNs, VeLDs and CiDs (Appel et al., 1995; Batista et al., 2008), whereas KA′s are not reported to express lhx3. Lhx3 and Lhx4 cooperate to promote MN identity in mouse (Sharma et al., 1998); however, Lhx4 was uncharacterized in zebrafish. Thus, we determined the lhx4 expression pattern on embryos of various stages and compared it with lhx3. lhx4 expression begins at 11 hpf, the same time as lhx3 (Fig. 2A,B). Like lhx3, lhx4 is expressed in two medial cell stripes in the region that will form the ventral spinal cord. At 24 hpf, lhx4 is expressed in fewer cells than lhx3 (Fig. 2C,D). To understand the relationship between lhx3 and lhx4 expression, we assayed them simultaneously in Tg(nrp1a:GFP) embryos that label MNs (Sato-Maeda et al., 2008). At 16 hpf, all lhx4-expressing spinal cord cells co-express lhx3, whereas some lhx3-expressing cells do not express lhx4 (Fig. 2E, arrowheads). The position of these cells relative to the nrp1a:GFP transgene suggests that they are primarily pMN-derived MNs and INs that also express lhx3 (Appel et al., 1995). At 20 hpf, lhx3 and lhx4 are co-expressed in more dorsal cells. The relative locations of these more dorsally located co-expressing cells suggest they are p2-derived INs that express lhx3 (Batista et al., 2008; Kimura et al., 2008). At 24 and 28 hpf, lhx3 and lhx4 appear to be almost entirely co-expressed (Fig. 2G,H). Thus, at these stages, lhx4, like lhx3, is expressed in MNs as well as in pMN- and p2-derived INs. In addition, lhx4 is expressed in the pituitary and in a set of dorsal INs between the eyes (supplementary material Fig. S1). lhx4 expression in the pituitary is similar to that of lhx3; however, the dorsal IN population that expresses lhx4 does not express lhx3.

To examine expression of Lhx3 and Lhx4 proteins, we generated polyclonal antibodies (see Materials and Methods; supplementary material Fig. S2). To confirm that Lhx4 is expressed in MNs, we compared expression patterns of embryos double-labeled with anti-Lhx4 and anti-Islet (Korzh et al., 1993). At 18 hpf, anti-Lhx4 antibody labeled exclusively Islet⁺ cells, suggesting that initially only PMNs express Lhx4 (Fig. 2J). By contrast, at 18 hpf many cells are Lhx3⁺Islet⁻, consistent with Lhx3 expression in VeLDs, CiDs and V2as (Fig. 2I, arrowhead) (Appel et al., 1995; Batista et al., 2008). By 24 hpf, Lhx3 and Lhx4 continued to be expressed in all Islet⁺ cells, indicating expression in all PMNs and at least some SMNs (Fig. 2K,L). At this time, anti-Lhx4 antibody labeled many Islet⁻ cells in the ventral spinal cord, consistent with Lhx4 expression expanding into VeLDs, CiDs and V2as. These data indicate that Lhx4, like Lhx3, is expressed in PMNs, SMNs, VeLDs, CiDs and V2as but is excluded from medially located KA′s (Fig. 2Q).

To examine whether Lhx3 and Lhx4 are expressed in spinal cord progenitors similar to mouse and chick (Sharma et al., 1998; Tsuchida et al., 1994), we co-labeled transgenic embryos expressing GFP in pMN or p2 progenitors with anti-Lhx3/4 antibodies and a mitotic marker, anti-phospho Histone H3 (P-H3). We identified pMN progenitors using Tg(olig2:GFP) embryos that express GFP in pMN progenitors and their post-mitotic derivatives (Shin et al., 2003). We found that GFP⁺ Lhx3/4⁺ cells in 12, 16 and 20 hpf Tg(olig2:GFP) embryos never co-expressed P-H3 (Fig. 2M and data not shown). We confirmed that Lhx3 and Lhx4 are excluded from pMN progenitors by labeling proliferating cells with 5-ethynyl-2′-deoxyuridine (EdU) at 11 hpf, the time of peak PMN generation (Myers et al., 1986). When processed at 12 hpf, no Lhx3/4⁺ cells had incorporated EdU (Fig. 2N,P). We identified p2 progenitors, using Tg(vsx1:GFP) embryos that express GFP in progenitors just prior to their final division as well as in V2 INs (Kimura et al., 2008). By contrast to pMN progenitors, at 20 hpf some p2 progenitors were Lhx3/4⁺ and P-H3⁺ (Fig. 2O,P). These data show that Lhx3 and Lhx4 are expressed by p2 progenitors but are absent from pMN progenitors, including KA′ progenitors.

The altered axon projection patterns of cervical MNs in mouse embryos lacking Lhx3 and Lhx4 (Sharma et al., 1998) prompted us to examine zebrafish trunk MNs in the absence of these proteins. We labeled MO-injected embryos with zn1 and anti-Syt2b antibodies that reveal axons of two PMN subtypes, ventrally projecting caudal primary (CaP) and dorsally projecting middle primary (MiP) (Melancon et al., 1997; Myers et al., 1986; Trevarrow et al., 1990). Injection of either lhx3 or lhx4 MO alone had modest to no effect on PMN axon formation (Fig. 3A-C,E). However, co-injection of lhx3 and lhx4 MOs severely reduced PMN axon numbers (Fig. 3D,E). Reduction was sporadic along the length of the trunk and in many cases differentially affected CaPs and MiPs within a single trunk segment. SMN development was also compromised. lhx3 and lhx4 MO-injected embryos lacked Alcama (Fashena and Westerfield, 1999) expression on SMN somata and fasciculated axons (Fig. 3F,G). Live imaging of s1020t;Tg(UAS-E1b:Kaede) embryos revealed that SMN axons projecting to dorsal and ventral axial muscles adjacent to trunk somites 5-15 were variably present at 48 hpf. Motor nerves that were present were much thinner, suggesting fewer SMN axons per nerve (Fig. 3H,I). By contrast to Lhx3- and Lhx4-deficient mouse cervical MNs (Sharma et al., 1998), axons of Lhx3- and Lhx4-deficient zebrafish trunk MNs that left the spinal cord invariably exited at the ventral root.

The loss of trunk MN axons following Lhx3 and Lhx4 knockdown prompted us to test whether these MNs ever formed. Lhx3 and Lhx4 knockdown resulted in normal numbers of Islet⁺ trunk cells at 28 and 48 hpf, suggesting that both PMNs and SMNs are generated (Fig. 4A,C,G-J). This is similar to Lhx3 and Lhx4 mouse embryos in which cervical MNs are present, but have altered identities (Sharma et al., 1998). By contrast, Islet1 is required for Lhx3 and Lhx4 expression in mouse (Pfaff et al., 1996), whereas zebrafish Lhx3 and Lhx4 expression is unaffected by Islet1 knockdown (Fig. 4D,E). Together, these observations show that equivalent numbers of MNs are generated in the presence and absence of Lhx3 and Lhx4 and suggest that, in zebrafish, Islet1 expression in trunk MNs is regulated independently of Lhx3 and Lhx4. These data are consistent with our previous results showing that, in the absence of Islet1, prospective zebrafish trunk MNs retain Lhx3 expression but differentiate as INs (Hutchinson and Eisen, 2006).

To confirm the specificity of our MO phenotype, we performed a mosaic rescue experiment by injecting UAS:EGFP-P2A-lhx3 plasmid into the s1020t enhancer trap line in which GAL4VP16 is expressed under olig2 control (Wyart et al., 2009). This allowed us to label cells overexpressing Lhx3 and to evaluate the proportion of labeled MNs. In control embryos, MNs comprised 55% of labeled neurons (Table 1), whereas in embryos injected with lhx3 and lhx4 MOs MNs comprised only 31% of labeled neurons. Co-expression of Lhx3 with EGFP restored the percentage of labeled MNs to 55%.

The reduction of trunk motor projections in lhx3+lhx4 MO-injected embryos led us to examine the phenotype of cells otherwise fated to become MNs. As the loss of Islet or Mnx results in hybrid PMNs with VeLD or CiD characteristics, respectively (Hutchinson and Eisen, 2006; Seredick et al., 2012), we hypothesized that PMNs in embryos lacking Lhx3 and Lhx4 would have characteristics of KA′s, the other pMN-derived IN. To test this, we co-labeled embryos with antibodies against Islet and Gad and found that, in the absence of Lhx3 and Lhx4, many Islet⁺ ventral spinal cord cells co-expressed Gad (Fig. 5A,B) and GABA (data not shown). These cells also expressed chat (chata, choline O-acetyltransferase a – Zebrafish Model Organism Database), which encodes the synthetic enzyme for acetylcholine, their normal transmitter (Fig. 5C,D). These results provide evidence that, in the absence of Lhx3 and Lhx4, PMNs develop a hybrid identity in which they express both MN and IN characteristics.

To resolve the identity of MN-IN hybrids, we labeled pMN-derived cells by injecting UAS:EGFPCAAX plasmid into s1020t embryos. This allowed us to visualize morphologies and proportions of MNs, VeLDs and KA′s. In addition to labeled MiPs, CaPs, VeLDs and KA′s seen in control embryos (Fig. 5E,F), in the absence of Lhx3 and Lhx4 we observed classes of aberrant axon projections. Notably, all aberrantly projecting neuron classes acquired ipsilaterally ascending axons, a hallmark of pMN-derived KA′s. There were two types of INs with axons that ascended through lateral spinal cord (ascending interneurons, aINs): most of these INs had a proximal axon that descended through the medial spinal cord before turning laterally and ascending through the lateral neuropil (Fig. 5G), although some aINs had axons that ascended through the spinal cord without a proximal descending segment (Fig. 5H). The laterally positioned somata of these neurons distinguished them from medially positioned KA′ somata. A second class of aberrant neurons had bifurcating spinal axons, one ascending and the other one descending (bifurcating interneurons, bINs; Fig. 5I). A third class had MN-IN hybrid morphologies (MN-aIN; Fig. 5J). These cells had motor projections into the ventral myotome as well as ascending IN axons. The aberrant neurons with ascending axons were probably transformed MNs, as there were fewer MNs without a significant increase in either KA′s or VeLDs (Fig. 5K).

Despite ectopic expression of Gad and GABA in MNs in the absence of Lhx3 and Lhx4 (Fig. 5 and data not shown), the total number of GABAergic cells in the ventral spinal cord was unchanged (control=10.0±2.0 GABA⁺ cells/hemisegment; lhx3 lhx4 MOs=8.8±2.0 GABA⁺ cells/hemisegment). This observation raised the possibility that one or more types of pMN-derived IN failed to differentiate its normal transmitter phenotype in the absence of Lhx3 and Lhx4. VeLDs and KA′s are both GABAergic (Park et al., 2004). In lhx3+lhx4 MO-injected embryos, VeLDs expressed GABAergic properties normally (supplementary material Fig. S3), but far fewer KA′s expressed GABAergic properties (supplementary material Fig. S4). Because KA′s do not express lhx3 or lhx4, we hypothesized that these genes might affect KA′ differentiation cell non-autonomously. Previous studies showed that Notch promotes KA′ identity (Shin et al., 2007) and that PMNs express Notch ligands (Appel and Eisen, 1998; Haddon et al., 1998); thus, Lhx3 and Lhx4 might regulate expression or presentation of Notch ligands in PMNs to promote Gad expression in KA′s. We tested this idea by overexpressing the Notch intracellular domain (NICD) and by blocking Notch activity with DAPT. Notch overexpression resulted in supernumerary GABAergic KA′s, and DAPT treatment diminished the number of GABAergic KA′s (supplementary material Fig. S4), as observed in previous studies. Consistent with our hypothesis, overexpressing NICD in lhx3+lhx4 MO-injected embryos restored GABAergic KA′ numbers to wild-type levels. However, blocking Notch activity in lhx3+lhx4 MO-injected embryos enhanced reduction of GABAergic KA′s (supplementary material Fig. S4). Together, these results suggest that, although both Notch and Lhx3 and Lhx4 affect KA′ differentiation, they do so through parallel pathways.

Our data suggest that Lhx3 and Lhx4 inhibit KA′ characteristics in PMNs; thus, we predicted that misexpression of lhx3 would inhibit KA′ formation. To test this hypothesis, we injected UAS:EGFP-P2A-lhx3 plasmid into s1020t embryos to drive Lhx3 expression in olig2-expressing cells (Wyart et al., 2009). In addition to the typical MNs, VeLDs and KA′s labeled in control embryos, we observed two classes of INs with aberrant morphologies. One class had an ipsilateral, descending axon that emerged from a ventromedial soma, a position characteristic of KA′s (dKA′s; Fig. 6A). The other class had an axon that emerged from a ventromedial soma and grew circumferentially before descending through the lateral neuropil. This class also had an ascending collateral that arose from the proximal portion of the main axon (CiD-KA′; Fig. 6B). This characteristic CiD axon morphology was not observed in control embryos, presumably because CiDs typically arise later than the time at which we assayed (Park et al., 2004). The aberrant INs with descending axons were probably transformed KA′s, as fewer KA′s formed without significant increases in the proportions of either MNs or VeLDs (Fig. 6C). These results demonstrate that Lhx3 is sufficient to inhibit KA′ fate and support the idea that Lhx3 specifies CiDs.

Vertebrate p2 progenitors express Lhx3 and produce immature V2 INs that differentiate into two distinct types: V2as maintain Lhx3 expression, express Delta and differentiate as Vsx2⁺ excitatory INs; V2bs extinguish Lhx3 expression and require Notch activity to differentiate as Tal1⁺ Gata3⁺ inhibitory INs (Batista et al., 2008; Del Barrio et al., 2007; Kimura et al., 2008; Lundfald et al., 2007; Peng et al., 2007). Lhx3 overexpression is sufficient to induce ectopic V2a formation in chick (Tanabe et al., 1998; Thaler et al., 2002), but whether Lhx3 and Lhx4 are necessary to promote V2 development remains untested.

To address this question, we knocked down Lhx3 and Lhx4 in Tg(vsx1:GFP) embryos and scored vsx2 and tal1 expression in GFP⁺ neurons. At 24 hpf, we observed no difference in the number of V2a and V2b INs (Fig. 7A-D,K). However, in slightly older embryos we noticed a reduced number of vsx2⁺ cells in the absence of Lhx3 and Lhx4 (data not shown). Thus, we examined the number of V2as and V2bs at 48 hpf, when many more are present, using a Tg(vsx2:GFP) line (Kimura et al., 2006) to score V2as, or labeling Tg(vsx1:GFP) embryos with gata3 to identify V2bs. Whereas the number of immature Tg(vsx1:GFP) V2 neurons at 48 hpf was unchanged (Fig. 7I,J,M), the number of neurons expressing differentiated V2a and V2b markers was significantly reduced (Fig. 7E-H,L) by the absence of Lhx3 and Lhx4. These results suggest that Lhx3 and Lhx4 are dispensable for the earliest-born V2 INs, but are required for differentiation of later-born V2a and V2b INs.

Here, we have made three principal findings: first, Lhx3 and Lhx4 cooperate to inhibit a competing KA′ IN program within PMNs (Fig. 8). Second, Lhx3 segregates alternative fates of two pMN domain-derived INs. Third, Lhx3 and Lhx4 are necessary for formation of late-developing V2 INs. We discuss each of these findings in turn.

Lineage analysis revealed that PMNs and KA′s both derive from the pMN domain and can be siblings (Park et al., 2004). We determined that Lhx3 and Lhx4 act together to prevent PMNs from acquiring KA′ features, including transmitter expression and axonal projection.

Zebrafish MNs have an inherent propensity to acquire characteristics of pMN-derived INs. Knockdown of Islet1 results in excess VeLDs and eliminates motor projections (Hutchinson and Eisen, 2006), and knockdown of Mnxs causes MiP PMNs to acquire CiD features (Seredick et al., 2012). These results reveal the regulatory logic by which MN and pMN-derived IN fates are segregated (Fig. 8B): one family of transcription factors blocks each of the alternative pMN-derived IN programs in post-mitotic MNs. This mechanism efficiently resolves identities of distinct neurons, the fates of which diverge only after the final progenitor division.

Notably, knockdown of Lhx3 and Lhx4 does not result in a complete PMN-to-KA′ fate conversion. Resulting cells adopt a hybrid phenotype, expressing PMN markers while also acquiring KA′ characteristics, and in some cases even adopting a hybrid morphology. Formation of MN-IN hybrids is a consistent phenotype following knockdown of many zebrafish transcription factors, including those described above, as well as Nkx6.1 and Nkx6.2 (Hutchinson et al., 2007) and the Met receptor (Tallafuss and Eisen, 2008). These results suggest that none of these genes, with the possible exception of Islet, act as true master regulators of MN fate. Rather, each gene or gene family refines identity by suppressing the inherent potential of MNs to initiate genetic programs normally restricted to INs. Similar observations have been made in mouse. In Runx1 mutant mice, a subset of brachial level MNs co-express Islet1 and the IN marker Pax2 (Stifani et al., 2008). In Nkx6.1 Nkx6.2 double mutants, Islet1 and the IN marker Evx1 are transiently co-expressed (Vallstedt et al., 2001). In Mnx1 knockout mice, MNs transiently co-express the V2a marker, Vsx2 (aka Chx10) (Arber et al., 1999; Thaler et al., 1999). Interestingly, Vsx2 binds to an Mnx1 enhancer in vivo and blocks expression of an MN reporter in vitro (Lee et al., 2008), suggesting that a reciprocal strategy of blocking MN characteristics within INs might be required under some circumstances. Whether or not transcription factors are deployed in VeLDs and KA′s to suppress MN characteristics remains unknown.

Strikingly, brachial MNs in Lhx3 Lhx4 knockout mice convert to SAC MNs, the only class of spinal MN that normally lacks Lhx3 and Lhx4 expression (Sharma et al., 1998). SAC MNs are characterized by mediolateral soma positions and unusual lateral axon exit (Bravo-Ambrosio and Kaprielian, 2011). Our results in zebrafish show that, in the absence of Lhx3 and Lhx4, the few trunk motor axons that persist exit the spinal cord through the ventral root. These seemingly disparate results might be a consequence of the different axial levels analyzed. We restricted our analysis to spinal segments adjacent to somites 8-15. By comparison, Sharma et al. showed that Islet⁺ MNs adjacent to cervical vertebrae 6-8 acquire features of SAC MNs, which are normally located adjacent to cervical vertebrae 1-5 (Sharma et al., 1998). DiI injection of dorsal axial muscles failed to label MNs at thoracic levels in double mutant mice, but the projections these presumptive MNs made was not described. We suspect that regional differences dictate the characteristics adopted by MNs in Lhx3- and Lhx4-deficient vertebrate embryos. MNs at cervical levels probably adopt an SAC MN fate, whereas those at more caudal levels probably acquire KA′ IN features. KA′s are best characterized in anamniotes (Bernhardt et al., 1992; Dale et al., 1987); however, similar cells were described in rats and primates (Stoeckel et al., 2003; Vigh et al., 2004). Definitive resolution of these differences will require further analysis of relationships among these neurons.

Very little is known regarding fate segregation among pMN-derived INs. Our forced-expression experiments reveal that Lhx3 acts, at least in part, as a molecular switch to segregate CiD from KA′ fates. Our results and others (Appel et al., 1995; Batista et al., 2008) show that by 18 hpf Lhx3 is expressed in cells in addition to MNs in ventral neural tube; some of these cells are pMN-derived INs. This expression is maintained until at least 28 hpf. CiDs are derived from the pMN domain after 15 hpf, following PMN formation (Park et al., 2004). Together, these data indicate that Lhx3 is in the right place at the right time to promote CiD and inhibit KA′ fates. This idea is further corroborated by the differential ability of PMN subtypes to acquire CiD features in the absence of Mnxs (Seredick et al., 2012) as a consequence of subtle but important differences in temporal expression of Islet transcription factors. CaPs express either Islet1 or Islet2a continuously. Thus, in the absence of Mnxs, they maintain expression of Islets and Lhxs and retain MN characteristics. By comparison, MiPs transiently downregulate Islet (Appel et al., 1995; Hutchinson et al., 2007; Hutchinson and Eisen, 2006). In the absence of Mnxs, MiPs temporarily express only Lhx during the Islet-negative period, the combinatorial code for CiDs, and can acquire a hybrid CiD phenotype.

We were surprised to find that Lhx3 and Lhx4 regulate expression of GABAergic properties in KA′s, although KA′s do not express Lhxs. This cell non-autonomous effect is not mediated via Notch signaling, the most obvious possibility (Shin et al., 2007). There are many other possible pathways, including electrical signaling through gap junctions. Most neurons within the developing zebrafish ventral spinal cord are electrically coupled through the first day of development (Saint-Amant and Drapeau, 2001; Wyart et al., 2009), and electrical activity has been shown to promote acquisition of transmitter phenotype (Spitzer, 2012). Lhx MO-injected embryos do not move properly at early developmental stages, perhaps as a result of disrupted activity of Lhx⁺ MN and/or INs. It would be interesting to investigate this possibility in future experiments.

We show that all Lhx3⁺ Lhx4⁺ INs in the zebrafish spinal cord have descending axons, and that the intraspinal portion of the vast majority of MNs descend to the nearest adjacent ventral root. Even when forced expression of Lhx3 fails to drive CiD axon morphology in KA′s, it is often sufficient to redirect KA′ axons to extend into the caudal spinal cord. In mouse, forced expression of Lhx3 in Mnx1⁺ MNs re-directs many axons into axial muscle at the expense of projections to normal targets in the limb, sympathetic ganglia and intercostal muscles (Sharma et al., 2000). One apparent target of Lhx3 overexpression is Fgfr1 (Shirasaki et al., 2006). Attraction to Fgf8 was proposed to set the axon trajectory of the Lhx3⁺ medial motor column MNs that project to dermamyotome. Clearly, other signals contribute to medial motor column MN guidance at this choice point, as relatively few motor axons are misrouted in Fgfr1 conditional mutants. It is intriguing to consider that an Fgf gradient established in the neural tube (Diez del Corral et al., 2003) might contribute to the descending pathways taken by these Lhx3⁺ Lhx4⁺ neurons.

Previous work in chick demonstrated sufficiency of Lhx3 in promoting expression of Vsx2, a V2a fate marker (Tanabe et al., 1998; Thaler et al., 2002). Our observations reveal that Lhx3 and Lhx4 are necessary for differentiation of both V2a and V2b INs, without affecting the number of immature V2 INs formed. Curiously, some V2a and V2b INs continue to be generated in Lhx3- and Lhx4-deficient embryos. In zebrafish, the first-born V2a and V2b pairs form independently of Lhx3 and Lhx4, as do a significant fraction of the V2a and V2b INs generated by 48 hpf. Although we cannot rule out that incomplete knockdown contributes to persistence of V2a and V2b INs at 48 hpf, there might be parallel or redundant mechanisms that cooperate with Lhx3 and Lhx4 to promote V2 IN specification. One attractive candidate is Bhlhe22 (previously Bhlhb5), an Olig-related transcription factor, the expression of which in V2 progenitors and V2a INs overlaps with Lhx3 (Skaggs et al., 2011). Reminiscent of our results after Lhx3 and Lhx4 knockdown, loss of Bhlhe22 function in chick reduced the number of V2a and V2b INs.

Interestingly, even after Lhx3 and Lhx4 knockdown, zebrafish embryos remain touch-responsive; however, they cannot sustain swimming through 4 dpf (data not shown). This is consistent with preservation of early-born V2a INs driving the highest-frequency movements during the initial phase of the escape response, and reduced numbers of later-born V2a INs driving lower-frequency movements during subsequent phases of the escape response (McLean and Fetcho, 2009). These observations suggest the exciting possibility that Lhx3 and Lhx4 coordinate development of subclasses of V2a INs underlying the circuitry responsible for setting locomotor speed, a question that remains to be addressed in future studies.

Wild-type, Tg(nrp1a:GFP)^js12 (Sato-Maeda et al., 2008), Tg(olig2:EGFP)^vu12 (Shin et al., 2003), Tg(vsx1:GFP)^nns5 (Kimura et al., 2008), Tg(vsx2:EGFP)^nns1 (Kimura et al., 2006) and Et(-0.6hsp70l:GAL4VP16)^s1020t (hereafter referred to as s1020t), Tg(UAS-E1b:Kaede)^s1999t (Wyart et al., 2009) zebrafish (Danio rerio) embryos were collected from natural crosses and staged by hours post fertilization at 28.5°C (hpf) and gross morphology (Kimmel et al., 1995). All experiments were carried out in accordance with animal welfare laws, guidelines and policies and were approved by the University of Oregon Institutional Animal Care and Use Committee.

We isolated a full-length zebrafish lhx4 transcript (1171 base pairs; 390 amino acids) from 24 hpf zebrafish cDNA using the following primers: forward 5′-CACACGGCGAAAGAACTCACG-3′; reverse 5′-TTTGCC-CACACCGAACACTG-3′. In phylogenetic analyses the protein encoded by this gene clusters with other Lhx4 proteins, not with Lhx3 proteins, indicating that it is an Lhx4 homolog (data not shown). Like other Lhx3 and Lhx4 proteins, zebrafish Lhx4 has two very highly conserved LIM domains as well as a highly conserved homeodomain.

Embryos were fixed and processed for in situ hybridization or antibody staining as previously described (Seredick et al., 2012). RNA probes include: islet2a and lhx3 (Appel et al., 1995); chat (Tallafuss and Eisen, 2008); tal1 and gata3 (Batista et al., 2008); vsx2 (Kimura et al., 2006); and lhx4. Primary antibodies include: mouse anti-Alcama (ZIRC, zn5; 1:4000); rabbit anti-GABA (Sigma, A2052; 1:1000); rabbit anti-Gad (Abcam, ab11070; 1:500); mouse anti-GFP (Clontech, 632381; 1:1000; or Life Technologies, A11120; 1:1000); rabbit anti-GFP (Life Technologies, A11122; 1:1000); mouse anti-phospho-Histone H3 (Abcam, ab14955; 1:1000); mouse anti-Islet (DSHB, 39.4D5; 1:1000); mouse anti-Syt2b (ZIRC, formerly znp1; 1:1000); mouse zn1 (ZIRC; 1:200).

To label proliferating cells, embryos were treated with 10 mM EdU solution (Click-iT EdU Alexa Fluor 555; Invitrogen, C10338) in embryo medium plus 15% DMSO for 20 min on ice. After rinsing into embryo medium, embryos recovered for 40 min and were fixed and processed by immunohistochemistry. Incorporated EdU was detected according to manufacturer's instructions.

Antibodies against Lhx3 and Lhx4 were generated by immunizing rabbits with truncated proteins corresponding to amino acids 237-397 of Lhx3 and 241-389 of Lhx4. We generated three antibodies: anti-Lhx3 antibody labels nuclei in ventral spinal cord and pituitary, the same cell population labeled by lhx3 RNA probe. Anti-Lhx4 antibody labels fewer nuclei in ventral spinal cord than anti-Lhx3, as well as the pituitary and a dorsal IN population in the head, the same cell populations labeled by lhx4 RNA probe. A third antibody labels as many nuclei in the ventral spinal cord as anti-Lhx3, as well as the pituitary and the dorsal INs in the head labeled by anti-Lhx4; we named this antibody anti-Lhx3/4.

One- to two-cell stage embryos were injected with 2.5 nl of translation-blocking MOs (GeneTools) targeting lhx3 MO, 5′-GTTCTAAC-AACATTCTGGCGATAAA-3′ (125 µM), and/or lhx4 MO, 5′-GCAG-CACAGCCGCACTTTGCATCAT-3′ (350 µM). lhx3 and lhx4 MO injection specifically eliminated expression of their respective proteins (supplementary material Fig. S2). MO specificity was confirmed using mis-match control MOs [lhx3 MO 5-mis (5′-GTTGTAAGAACATT-GTGGCCATTAA-3′) and lhx4 MO 5-mis (5′-GCACCACCCCGGA-CTTTCCATGAT-3′)] and rescue of the MN phenotype by mosaic overexpression of Lhx3. MO-injected embryos were morphologically normal and lacked elevated cell death assessed by Acridine Orange staining.

To assess morphology of individual cells and determine proportions of pMN-derived neuronal classes, we injected UAS:EGFPCAAX or UAS:EGFP-P2A-lhx3 DNA along with Tol2 transposase RNA into s1020t embryos. Synonymous substitutions in lhx3 were introduced to prevent MO targeting of rescue RNA. Fluorescent embryos were selected and imaged live at 28-32 hpf. Alternatively, individual neurons in lhx3 and lhx4 MO-injected embryos were dye-labeled with 5% tetramethylrhodamine-dextran (D3308; Life Technologies) in 0.2 M KCl as previously described (Hale et al., 2011).

Images of embryos were captured on a Zeiss Axioplan using a digital camera, or a Zeiss Pascal confocal microscope. Adobe Photoshop was used to adjust image brightness and contrast.

Comparisons of means between groups were performed using two-sample unequal variance t-tests. An exact binomial test was used to assess changes in proportions of labeled MNs. Multiple comparisons were performed using a one-way ANOVA, followed by a post hoc Tukey–Kramer test. Analyses were performed using JMP 10 and Microsoft Excel. All data are reported as means±s.d.

We thank Chris Doe for comments on the manuscript. Joe Fetcho, Shin-Ichi Higashajima and Bruce Appel generously provided transgenic fish. The Islet monoclonal antibody developed by Thomas Jessell was obtained from the Developmental Studies Hybridoma Bank, developed under the auspices of the NICHD and maintained by The University of Iowa, Department of Biology, Iowa City, USA. We thank Ellie Melancon, Amanda Lewis, Jacob Lewis, Dayna Lamb and the University of Oregon Zebrafish Facility Staff for excellent fish husbandry and Ellie Melancon, Julia Ganz and Levi Simonson for help with some of the RNA in situ hybridization panels in Fig. 2.