The LIM and POU homeobox genes ttx-3 and unc-86 act as terminal selectors in distinct cholinergic and serotonergic neuron types

The development of the nervous system is a multistep process that employs a series of sequentially acting regulatory factors that successively restrict and determine cellular fates. During the process of terminal differentiation, individual neuron types acquire specific, hard-wired features that are maintained by the neuron type throughout the life of the animal. A number of transcription factors have been identified that initiate and maintain specific terminal differentiation programs in the developing nervous system (Hobert, 2011). For example, in mouse, the Nurr1 (Nr4a2) transcription factor initiates and maintains the terminal differentiation program of dopaminergic neurons in the midbrain (Smidt and Burbach, 2009), whereas the Pet1 transcription factor initiates and maintains the terminal differentiation program of serotonergic neurons (Liu et al., 2010). However, few neuronal transcription factors are expressed exclusively in only one specific neuronal cell type (Gray et al., 2004; Lein et al., 2007). For example, in addition to being expressed in midbrain dopaminergic neurons, Nurr1 is expressed in other non-dopaminergic neuronal cell types in which its function is not well understood, such as the adult olfactory bulb, specific cortical areas and the hippocampus (Zetterström et al., 1996). The expression of a given transcription factor in distinct neuronal populations poses the fundamental question of whether there are underlying common themes in the activity of the transcription factor in distinct neuronal cell types.

We have undertaken a systematic, in-depth comparison of the activity of two transcription factors in the development of several distinct neuronal cell types in the nematode C. elegans, examining whether there are indeed conceptual similarities in the activities of a given transcription factor in distinct neuron types. We used, as a starting point, a member of the LIM homeobox gene family, an ancient family of neuronal patterning genes that display complex expression patterns in the nervous system of many different species, from invertebrates to vertebrates (Hobert and Westphal, 2000; Simmons et al., 2012; Srivastava et al., 2010). One unifying theme is their expression in terminally differentiating neurons (Hobert and Ruvkun, 1998; Moreno et al., 2005). We focus here on the ttx-3 LIM homeobox gene, which is the sole C. elegans member of the Lhx2/9 subclass of LIM homeobox genes. In vertebrates, Lhx2 is expressed in multiple neuronal cell types and is required for the differentiation of olfactory sensory neurons (Hirota and Mombaerts, 2004; Kolterud et al., 2004), the specification of cortical neuron fate (Mangale et al., 2008) and the differentiation of thalamic neurons (Peukert et al., 2011). Whether there is a common theme in the function of Lhx2 in these distinct neuronal cell types is not known.

The C. elegans Lhx2/9 ortholog ttx-3 is exclusively expressed in a small number of neurons in distinct head ganglia (Altun-Gultekin et al., 2001). ttx-3 null animals display broad differentiation defects in the cholinergic AIY interneuron class. AIY interneurons of ttx-3 null mutants are generated and still express pan-neuronal features, but fail to express scores of terminal identity markers that define the functional properties of AIY, including genes required to synthesize and package acetylcholine, genes encoding neuropeptide receptors, various types of ion channels and many others (Altun-Gultekin et al., 2001; Hobert et al., 1997; Wenick and Hobert, 2004). TTX-3 exerts this control through direct binding to a cis-regulatory motif shared by all of its target genes. ttx-3 expression is turned on in the neuroblast that generates AIY and its expression is maintained throughout the life of the neuron through an autoregulatory feedback loop (Bertrand and Hobert, 2009) to ensure persistent expression of its target genes. A number of transcription factors have been described in the C. elegans nervous system that display similar broad-ranging effects on the terminal differentiation programs executed by the neurons in which they are expressed. These transcription factors have been called ‘terminal selectors’ (Hobert, 2008; Hobert, 2011). It is still an open question how broadly the terminal selector concept applies throughout the nervous system; that is, how common it is that many distinct and functionally unrelated identity features of a specific neuron type are directly co-regulated by a transcription factor or a combination of transcription factors.

Here, we investigate the role of ttx-3 in two additional neuron classes in which it is normally expressed, namely the cholinergic AIA interneuron class and the serotonergic NSM neuron class. We find in all three neuron classes that there is a common theme of ttx-3 function in that it is broadly required to induce many distinct and functionally unrelated terminal identity features of the respective neuron class. Yet the downstream targets of ttx-3 in these neuron classes are distinct and are determined by the cooperation of ttx-3 with a distinct set of transcription factors in different neuron classes. One of these factors is the POU homeobox gene unc-86, which is required together with ttx-3 to control the identity of the serotonergic NSM neurons. unc-86 in turn cooperates with the ARID-type transcription factor cfi-1 to control many terminal identity features of the cholinergic IL2 sensory and URA motor neurons. Our studies therefore provide further support for the terminal selector concept and show that, in combination with other regulatory factors, one factor can serve as terminal selector in distinct neuronal cell types regulating distinct neuronal differentiation programs.

A ttx-3 reporter gene that contains the ttx-3 locus together with a few kilobases upstream but no downstream sequences (ttx-3^promA::gfp; Fig. 1) was previously shown to be continuously expressed in five distinct neuronal cell types: the cholinergic AIY and AIA interneuron classes, the ASI and ADL chemosensory neuron classes and a previously uncharacterized neuronal pair in the pharyngeal nervous system (Altun-Gultekin et al., 2001). Transient expression was observed in the AIN and SMDD neurons at embryonic stages (Bertrand and Hobert, 2009). A fosmid reporter construct, which contains more than 30 kb surrounding the ttx-3 locus and which rescues the AIY differentiation defect of ttx-3 mutant animals, mirrors the expression of the smaller, locus-restricted reporter construct in the AIY, the AIA, the AIN and the pharyngeal neuron class (Fig. 1). Based on position, morphology and colabeling with the NSM marker mgl-1::mCherry, we identified the pharyngeal neurons that express ttx-3 as the NSM neuron pair. The NSM neurons are serotonergic, neurosecretory cells that are thought to be involved in sensing food (Albertson and Thomson, 1976; Harris et al., 2011; Horvitz et al., 1982).

There are also notable differences in the expression pattern of the fosmid reporter and the smaller reporters. First, expression in the AIN neurons is maintained throughout development with the fosmid reporter, whereas it is restricted to embryos with smaller reporters (Bertrand and Hobert, 2009). Second, the expression in amphid sensory neurons is markedly different. In larval and adult animals, the fosmid reporter is expressed in the ASK neuron class, whereas the smaller reporters are expressed in the ADL and ASI sensory neurons (Fig. 1).

Previous studies have shown that ttx-3 expression in the AIY interneuron pair is controlled by a distal initiator element ∼1 kb upstream of the ttx-3 locus and a maintenance element in the second intron of the ttx-3 locus (Bertrand and Hobert, 2009; Wenick and Hobert, 2004). We find that the expression of ttx-3 in the NSM and AIA is controlled via regulatory elements present in the seventh intron of the ttx-3 locus (Fig. 1). As mentioned above, ttx-3 expression is maintained throughout the life of the AIA and NSM neurons, but maintained expression of a ttx-3 reporter gene construct (ttx-3^intron7::gfp; Fig. 1A) in the AIA and NSM neuron types does not require ttx-3 gene activity (Fig. 1C).

We focused our analysis of ttx-3 mutants on the cholinergic AIA interneurons and the serotonergic NSM neurons, which both continuously express ttx-3 throughout their lifetime. We have previously reported that expression of the marker of cholinergic identity, unc-17 (vesicular ACh transporter), as well as the expression of an orphan G protein-coupled receptor (GPCR), sra-11, is reduced in the AIA neurons of ttx-3 mutants (Altun-Gultekin et al., 2001). We extended this analysis by examining the expression of seven additional markers of terminal AIA fate: the choline reuptake transporter encoded by cho-1; the metabotropic glutamate receptor mgl-1; the ionotropic glutamate receptor glr-2; the neuropeptides flp-2 and ins-1; the receptor tyrosine kinase scd-2; and the receptor guanylyl cyclase gcy-28d. Each of these markers is expressed in terminally differentiated AIA interneurons and several of them have previously been implicated in AIA interneuron function (Shinkai et al., 2011; Tomioka et al., 2006). The expression of each of these seven markers is affected in the AIA neurons of ttx-3 mutants (Fig. 2). Their expression in other neuron types is unaffected in ttx-3 mutants, with the exception of two markers that are also downregulated in NSM neurons (mgl-1, scd-2, as described below). ttx-3 is likely to act cell-autonomously since the AIA differentiation defects are rescued in transgenic ttx-3 mutant animals that express ttx-3 cDNA under control of the ins-1 promoter (supplementary material Table S1).

AIA neurons remain present in the ttx-3 null mutant, as assessed by the weak but recognizable expression of some terminal differentiation genes (Fig. 2). However, their normally unipolar neurite morphology appears disrupted; ectopic branches can be observed to emanate from the cell body and the main neurite appears blebbed in ttx-3 mutants (supplementary material Fig. S2). The AIY interneurons, which have a unipolar axon morphology similar to that of AIA interneurons in wild-type animals, display similar morphological defects in ttx-3 mutants (Hobert et al., 1997). The expression of terminal identity markers that label several distinct neuron types that are lineally related to AIA is not altered (data not shown) (Altun-Gultekin et al., 2001), suggesting that the AIA neuron pair might remain in an undifferentiated state, rather than switching to an alternate fate. Based on a more extensive cell fate marker analysis, a similar conclusion was previously drawn about the fate of the AIY neuron class in ttx-3 mutants (Altun-Gultekin et al., 2001). Taken together, our fate marker and morphological analyses indicate that ttx-3 broadly affects the AIA terminal differentiation program. These effects are comparable to the previously described broad effects that loss of ttx-3 has on the terminal differentiation of AIY interneurons.

On a mechanistic level, ttx-3 operates in a distinct manner in the AIA versus AIY neurons since it operates with distinct co-factors and through distinct cis-regulatory elements. The co-factor of ttx-3 in AIY, the ceh-10 homeobox gene (Altun-Gultekin et al., 2001), is not expressed in AIA neurons, and AIA neurons display no differentiation defects in ceh-10 null mutants (two markers tested). Moreover, the cis-regulatory motifs through which ttx-3 acts to control AIY versus AIA identity are distinct. In the AIY neurons, ttx-3 acts on its many target genes through a cis-regulatory motif, termed the ‘AIY motif’, that provides a cooperative binding site for a TTX-3-CEH-10 heterodimer (Wenick and Hobert, 2004). Mutation of the AIY motif in a locus that is expressed in AIY and AIA neurons, the cholinergic cho-1 locus, results in a severe reduction in expression in the AIY interneurons but not in the AIA interneurons (Fig. 3A).

In the AIA neurons, by contrast, ttx-3 acts through a distinct cis-regulatory signature, which we deciphered through a mutational analysis of the cis-regulatory control regions of three AIA-expressed, ttx-3-dependent terminal differentiation genes: mgl-1, ins-1 and cho-1. We generated transgenic animals that express nested, shorter versions of these three reporters and identified a 259 bp element in the cho-1 promoter, a 74 bp element in the mgl-1 promoter and a 68 bp element in the ins-1 promoter that are sufficient to direct gfp expression to AIA neurons (Fig. 3A-C). Examining these elements for common patterns, we noted that all these elements contain a shared and phylogenetically conserved G(A/G)ATC motif (Fig. 3D). Mutating this motif in the context of any of the three promoters resulted in a reduction of AIA expression of the respective reporter (Fig. 3A-C). In the case of mgl-1, two G(A/G)ATC motifs are present in the minimal promoter; mutation of either causes an intermediate reduction in reporter gene expression, and mutation of both motifs results in complete loss of expression (Fig. 3A-C).

Since G(A/G)ATC does not match the consensus binding site for a LIM homeodomain transcription factor such as TTX-3, we also examined the minimal reporters for the presence of conserved TAAT motifs, which comprises the core consensus site for LIM homeodomain transcription factors (Berger et al., 2008). We indeed found several TAAT motifs in the three cis-regulatory modules and for each of them we identified a TAAT motif that, when mutated, affected reporter gene expression in vivo (Fig. 3A-C). These TAAT motifs can be assembled into a larger sequence matrix, TAATTNGA (Fig. 3D). In two cases, mutation of the TAATTNGA alone affected reporter gene expression, whereas in the third case (cho-1) a complete loss of expression can only be observed upon simultaneous mutation of both the GAATC motif and the TAAT-containing motif (Fig. 3A). The residual AIA expression of a cho-1 reporter construct in which the GAATC motif is mutated, but the TAAT motif is left intact, is abolished in ttx-3 mutants (data not shown), consistent with ttx-3 operating through the TAAT motif.

We examined whether the TAATTNGA motif is indeed a TTX-3 binding site using gel shift assays with bacterially produced TTX-3 protein and probes derived from the mgl-1 and cho-1 locus. We found that TTX-3 is able to bind these sites in vitro (Fig. 3E). Deletion of the TAAT site that is required for reporter gene expression in vivo resulted in the loss of TTX-3 binding in vitro (Fig. 3E).

The combination of G(A/G)ATC and TAAT motifs might define a cis-regulatory signature that is generally required for gene expression in AIA neurons, since we found a combination of these two motifs to be present in the cis-regulatory control regions of the other four ttx-3-dependent terminal AIA markers (Fig. 3D). Taken together, these data show that AIA identity features are co-regulated by a shared cis-regulatory signature that is controlled by TTX-3 and an as yet unknown co-factor.

We next analyzed the effect of loss of ttx-3 on the terminal differentiation program of the serotonergic NSM neurons, a neuron type that has not previously been examined in ttx-3 mutants. Many terminal identity markers of NSM have been described, including the battery of genes that are required to synthesize, package and reuptake serotonin: tph-1/TPH (tryptophan hydroxylase), cat-4/GTPCH (GTP cyclohydrolase), cat-1/VMAT (vesicular monoamine transporter), bas-1/AAAD (aromatic amino acid decarboxylase) and mod-5/SERT (serotonin reuptake transporter) (Fig. 4A) (Jafari et al., 2011; Ranganathan et al., 2001; Sze et al., 2002). Previous expression analysis of a vesicular glutamate transporter, eat-4, suggested that NSM might use the neurotransmitter glutamate (Lee et al., 1999). However, a fosmid-based eat-4 reporter does not show expression in NSM neurons (Serrano-Saiz et al., 2013) (supplementary material Fig. S1A).

To broaden the spectrum of available terminal markers, we analyzed the expression of other C. elegans orthologs of enzymes involved in monoaminergic transmitter metabolism (Fig. 4A) and identified another NSM-expressed terminal marker, ptps-1 (Fig. 4C; supplementary material Fig. S3). In addition to examining these serotonin (5HT)-related markers, we also examined the expression of three metabotropic neurotransmitter receptors (mgl-1, mgl-3, dop-3), three neuropeptides (nlp-13, flp-4, nlp-3), a glycoprotein hormone alpha subunit (flr-2) and a receptor tyrosine kinase (scd-2). All of these genes are expressed throughout the life of the NSM neurons. As mentioned above, scd-2 and mgl-1 are also expressed in AIA neurons, where their expression is affected by ttx-3. We find that the expression of five of these 14 NSM terminal identity markers is either partially or completely eliminated in the NSM neurons of ttx-3 null mutants (Fig. 4C, Fig. 5, Table 1). ttx-3 is likely to act cell-autonomously since we can rescue the NSM differentiation defects by driving ttx-3 cDNA under the control of a cat-1 promoter fragment, which is expressed in a subset of monoaminergic neurons of C. elegans (supplementary material Table S1).

We recently reported that the effects of the loss of a terminal selector type transcription factor in dopaminergic neurons can be partially compensated for by other, co-expressed terminal selectors (Doitsidou et al., 2013). Therefore, we considered the possibility that the lack of an impact of ttx-3 loss on nine out of 14 NSM markers could be due to the activity of compensatory terminal selector type transcription factors. We sought to identify such a factor, focusing on two homeodomain transcription factors previously shown to be expressed in NSM, namely the empty spiracles homolog ceh-2 and the POU homeobox gene unc-86 (Aspöck et al., 2003; Finney and Ruvkun, 1990). We observed no NSM differentiation defects in ceh-2 null animals (data not shown), but we observed striking NSM differentiation defects in unc-86 mutants. Loss of unc-86 was previously shown to affect the expression of tph-1 and cat-1 in NSM neurons, but without effect on 5HT antibody staining (Sze et al., 2002). Other differentiation features of NSM neurons had not previously been examined in unc-86 mutants. Upon examining the expression of all 14 markers of NSM fate in unc-86 null mutants, we found that the expression of eight is partially or completely eliminated (Figs 4, 5, Table 1).

To examine whether unc-86 directly affects the expression of these terminal identity features, we analyzed the cis-regulatory control regions of four of them: tph-1, bas-1, cat-1 and cat-4. Through mutational analysis, we defined small (∼200 bp) elements that still yielded expression in the NSM neurons (Fig. 6) and, within each of these elements, we identified predicted POU homeodomain binding sites (Rhee et al., 1998). We introduced mutations into these sites in the context of two loci (tph-1 and bas-1) and found that these mutations resulted in a loss of reporter gene expression in vivo (Fig. 6A,B). Gel shift analysis further confirmed that these POU homeodomain sites indeed bind bacterially produced UNC-86 protein in vitro (Fig. 6E).

We noted that terminal markers of NSM identity that were severely affected in unc-86 mutants tended to be those that were weakly or unaffected in ttx-3 mutants; vice versa, markers unaffected in ttx-3 mutants tended to be affected in unc-86 mutants (Table 1). Even though this observation might simply indicate that unc-86 and ttx-3 act completely independently of one another, we considered the possibility that unc-86 and ttx-3 might collaboratively control NSM identity but that their relative importance may be distinct for different target genes. To investigate this possibility, we examined unc-86; ttx-3 double-null mutants and found that markers that are either partially or unaffected in ttx-3 and unc-86 single mutants are more strongly affected in the double mutant (Figs 4, 5, Table 1). This also holds for 5HT antibody staining, which is not affected in either single mutant but completely abrogated in the ttx-3; unc-86 double mutant (Fig. 4D), probably owing to the combined effect that both genes have on the expression of the 5HT reuptake transporter mod-5. As summarized in Table 1, nine of the 15 tested identity features (14 reporter genes and 5HT antibody staining) are affected by both ttx-3 and unc-86, with effects either visible in both single mutants, or as a non-additive, synergistic effect revealed in the double mutant. As described below, there are also synergistic effects of ttx-3 and unc-86 on NSM morphology. In six of the 15 cases, either ttx-3 or unc-86 already has completely penetrant effects (Table 1). Taken together, these data argue that unc-86 and ttx-3 jointly control terminal NSM differentiation. The mechanistic basis of the cooperation is unclear at present because we have so far not been able to identify functional TTX-3 binding sites in terminal NSM identity marker genes.

To further examine potential interactions of unc-86 and ttx-3, we investigated whether they affect each others expression. We find that continuous expression of unc-86 in NSM neurons depends on unc-86 itself [autoregulation of unc-86 was also previously noted (Baumeister et al., 1996)], but not on ttx-3 (supplementary material Fig. S1B,C). Vice versa, ttx-3 expression in NSM neurons is not affected in unc-86 or in unc-86; ttx-3 mutants (data not shown).

Apart from affecting the expression of terminal identity markers, loss of unc-86 and ttx-3 also results in specific effects on the morphology of the NSM neurons. During embryonic stages, these neurons normally extend a neurite posteriorly toward the nerve ring, which then bifurcates to form a ventral and a dorsal neurite (Axäng et al., 2008) (Fig. 4B). We observe that in unc-86(n846) mutants the NSM somas are correctly positioned but there are significant defects in outgrowth of the ventral neurite (61% of animals show outgrowth defects; n=31). By contrast, in ttx-3(ot22) mutants, the primary defect observed in ventral neurite outgrowth is the formation of aberrant bifurcations (41% of animals show such defects; n=39). Both ttx-3 and unc-86 single mutants also show defects in dorsal axon termination [25% (n=32) of ttx-3 mutants and 29% (n=34) of unc-86 mutants].

In early larval stages, ventral NSM neurites begin to extend elaborate arbor structures onto the nerve ring target field (Axäng et al., 2008). These axon arborizations require the NSM-expressed netrin receptor UNC-40/DCC, which is tightly localized to puncta within the main shaft of the NSM neurite and at the tips of axon arbors (Nelson and Colón-Ramos, 2013). These arbor structures persist into the adult stage and contain presynaptic sites, as assessed with a rab-3 marker (Nelson and Colón-Ramos, 2013). In ttx-3 mutants, these ultrastructural features are unaffected, but unc-86 mutants display a highly penetrant defect in axon arborization (Fig. 7A,C). Furthermore, unc-86 mutants display defects in the dynamic regulation of UNC-40 localization (Fig. 7B,D). In wild-type animals, UNC-40::GFP is diffusely distributed at the L1 stage and becomes localized to bright puncta in the NSM neurite and at the tips of axon arbors as axons are arborizing at the L4 stage. By the adult stage, UNC-40::GFP again becomes diffusely distributed. However, a significant fraction of unc-86 mutant NSMs retain a juvenile-like pattern of UNC-40 localization during the adult stage, in which UNC-40::GFP remains localized to bright puncta (Fig. 7B).

We observe synergistic morphological defects in unc-86(n846); ttx-3(ot22) double mutants. Ventral neurites never reach the middle of the pharyngeal isthmus and are often truncated immediately following the guidance decision to turn posteriorly (100% premature ventral neurite termination, n=18; Fig. 7E). Furthermore, neurites contain large anterior swellings not seen in wild-type animals (33% contain additional anterior swellings, n=18; Fig. 7E). The morphological appearance of NSM neurites in unc-86; ttx-3 mutants is reminiscent of the normal morphology of M3 neurons (Albertson and Thomson, 1976), which are lineally related to NSM (Sulston et al., 1983). M3 neurons are glutamatergic (Lee et al., 1999) and we indeed find that in unc-86 mutants the vesicular glutamate transporter eat-4 is ectopically expressed in NSMs (supplementary material Fig. S1A).

Apart from our description of unc-86 terminal selector function in the serotonergic NSM neurons, unc-86 had previously been described to broadly affect the terminal differentiation program of other serotonergic (Sze et al., 2002) as well as glutamatergic (Duggan et al., 1998; Serrano-Saiz et al., 2013) neurons. We asked whether unc-86 might affect the terminal differentiation program of neurons that use yet another neurotransmitter system. We turned to the six IL2 sensory neurons that are involved in nictation behavior (Lee et al., 2012). The IL2 neurons express unc-86 throughout their lifetime and have been inferred to be cholinergic (Lee et al., 2012). We corroborated the cholinergic identity of the IL2s by finding that reporter fusions to the unc-17/cha-1 locus and to the choline reuptake transporter cho-1 are expressed in IL2 neurons (Fig. 8A). The expression of these two key markers of cholinergic identity is eliminated in unc-86 mutants (Fig. 8A). Expression of the nicotinic acetylcholine receptor subunit des-2 is also lost in the IL2 neurons of unc-86 mutants (Treinin et al., 1998).

In addition to these cholinergic markers, we examined the expression of other genes previously shown to be expressed in IL2 neurons, namely the unc-5 netrin receptor, the guanylyl cyclase gcy-19, the kinesin klp-6 and the Notch ligand lag-2 (which is expressed in IL2 neurons at the dauer stage) (Leung-Hagesteijn et al., 1992; Ortiz et al., 2006; Ouellet et al., 2008; Peden and Barr, 2005). The expression of all of these terminal markers of IL2 identity is eliminated in IL2 neurons of unc-86 mutants (Fig. 8A). IL2 neurons also fail to take up dye in unc-86 mutants (Tong and Bürglin, 2010), suggesting morphological defects. The IL2 neurons are nevertheless generated in unc-86 mutants, as assessed by intact expression of the pan-neuronal marker rab-3 and the pan-sensory marker osm-6 (50 animals were scored for each marker).

unc-86 is expressed in two additional cholinergic neuron classes in the anterior ganglion besides the IL2 sensory neurons, namely the URA motoneurons [which are synaptically connected to the IL2 neurons (White et al., 1986)] and the URB interneurons. We found that cholinergic identity was also strongly affected in both URA and URB neurons of unc-86(n846) loss-of-function mutants (supplementary material Fig. S4).

Since none of the previously known co-factors of unc-86 [mec-3 for touch neurons (Duggan et al., 1998) and ttx-3 for NSM neurons (this paper)] is expressed in IL2, URA or URB neurons, unc-86 is likely to act with another co-factor in IL2 neurons. cfi-1 is an ARID transcription factor previously shown to be co-expressed with unc-86 specifically in IL2 and URA neurons (Shaham and Bargmann, 2002). Loss of cfi-1 results in ectopic expression of identity markers for the CEM neuron in IL2 and URA neurons (Shaham and Bargmann, 2002), which prompted us to investigate whether cfi-1 might also positively control their cholinergic identity. We find that the cholinergic identity of both IL2 and URA neurons is affected in cfi-1(ky651) loss-of-function mutants, albeit not as strongly as in unc-86 null mutants (Fig. 8A; supplementary material Fig. S4). To investigate whether unc-86 and cfi-1 genetically interact, we examined non-additive synergistic interactions of the two genes using a hypomorphic unc-86 allele, n848. Animals carrying this allele show mild IL2 and URA differentiation defects, but in combination with the cfi-1(ky651) mutant allele there are strong synergistic, i.e. non-additive, defects in IL2 and URA differentiation (Fig. 8A; supplementary material Fig. S4). We conclude that unc-86 and cfi-1 cooperate to control IL2 and URA identity.

Two main conclusions can be drawn from the data presented in this paper. First, our data provide general support for the terminal selector concept. Second, our data show that a given transcription factor can operate as a selector of terminal neuron identity in distinct neuronal cell types and that this is achieved through cooperation with distinct co-factors (summarized in Fig. 8B). In other words, individual neuronal cell types use distinct combinatorial codes of terminal selectors, and individual components of the code are reused in distinct combinations in different cell types.

The terminal selector concept was initially proposed based on a relatively small number of C. elegans transcription factor mutant phenotypes (Hobert, 2008). In each of these mutant backgrounds, a neuronal cell is born and expresses pan-neuronal features but fails to adopt neuron type-specific identity features. Importantly, terminal differentiation is very broadly affected in terminal selector mutants, such that not only functionally linked features (such as enzymes and transporter in a neurotransmitter synthesis/transport pathway), but also seemingly completely independent differentiation features that have no obvious biochemical connection (e.g. sensory receptors, neuropeptides and ionotropic neurotransmitter receptors) fail to be expressed. That the removal of an individual transcription factor results in such broad defects could not necessarily be assumed since transcriptomic approaches generally show that individual cell types expresses several dozen transcription factors (e.g. Etchberger et al., 2007). This could be interpreted to mean that the identity features of a neuron are regulated in a piecemeal manner, rather than being ‘mastered’ by a single transcription factor or a small combination thereof (Hobert, 2011). Two major questions raised by the terminal selector concept were how broadly it applies to different cell types in the C. elegans nervous system and how it applies to transcription factors expressed in distinct neuron types.

Here, we have shown that the terminal differentiation programs of very distinct neuron types - a cholinergic interneuron (AIA), a serotonergic sensory/motor neuron (NSM) and cholinergic sensory and motor neuron classes (IL2 and URA) - are controlled by distinct combinatorial codes of transcription factors. These factors regulate many distinct identity features of these distinct neuron types, ranging from neuropeptides to neurotransmitter synthesis pathway genes to neurotransmitter receptors and other signaling molecules.

In the case of the cholinergic AIA interneuron, we found that the expression of every tested terminal differentiation marker is affected in ttx-3 mutants. Since the available AIA marker collection essentially represents a random snapshot of terminal markers that characterize AIA identity, one might extrapolate the regulatory impact of ttx-3 on each one of these genes to the many hundreds, if not thousands, of genes that are expressed in AIAs, such that ttx-3 is likely to affect a very large number of them. The estimated very broad effect of ttx-3 on AIA identity is consistent with what we observed for the cholinergic AIY interneuron, in which ttx-3 mutation also affects the expression of all known identity features (Altun-Gultekin et al., 2001; Wenick and Hobert, 2004). Even though both neuron types have similar morphologies, are cholinergic, and are directly postsynaptic to various sensory neurons, AIY and AIA have different functions (Hobert et al., 1997; Shinkai et al., 2011; Tomioka et al., 2006), connect to a different spectrum of synaptic partners (White et al., 1986) and express distinct gene batteries. Yet, in both cases, ttx-3 very broadly affects the differentiation of each neuron type.

The distinct target gene specificities of ttx-3 in AIA and AIY neurons can be explained by neuron type-specific co-factors and by ttx-3 acting through distinct cis-regulatory motifs. AIY-expressed genes display a characteristic cis-regulatory signature that is recognized by a combination of the TTX-3 and CEH-10 homeodomain proteins (Wenick and Hobert, 2004). As we have shown here, AIA-expressed genes share a distinct cis-regulatory signature that is composed of two separate motifs located in close proximity, one a TTX-3 binding site and the other a binding site for a presumptive TTX-3 co-factor. This is analogous to the situation in the AIY interneuron class, in which TTX-3 and CEH-10 operate through a bipartite motif (the ‘AIY motif’) composed of a TTX-3 and a CEH-10 binding site (Wenick and Hobert, 2004). Genes that are expressed in both AIY and AIA neurons (e.g. cho-1) contain a modular assembly of both the AIY and AIA cis-regulatory signature.

Similar to the ttx-3-dependent control of the central cholinergic interneurons AIY and AIA, the mouse LIM homeobox gene Lhx7 is required for the terminal differentiation of cholinergic striatal interneurons (Lopes et al., 2012). As with other terminal selector transcription factors, Lhx7 function appears to be continuously required to maintain cholinergic identity. Co-factors that operate together with Lhx7 are currently not known. Lhx7 is expressed in many other neurons in the CNS. It will be interesting to determine whether Lhx7 also operates as a terminal selector in these other neuron types.

ttx-3 activity is not restricted to cholinergic neurons. We find that ttx-3 is also a key regulator of serotonergic neuron identity. The activity of ttx-3 in the serotonergic NSM neuron class is, however, distinct from that of AIA and AIY. Whereas the expression of several NSM-expressed effector genes is completely eliminated in ttx-3 mutants, the expression of some effector genes is only partially affected or not affected at all. In cases in which only partial or no effect was observed, joint removal of another homeobox gene, unc-86, resulted in much stronger or complete loss of effector gene expression. Vice versa, the expression of effector genes that are unaffected in expression in unc-86 mutants is lost in either ttx-3 mutants or in the ttx-3; unc-86 double mutant. Taken together, elimination of both of the POU/LIM homeobox genes unc-86 and ttx-3 has profound effects on NSM identity, paralleling the profound effect that another POU/LIM homeobox combination (unc-86 and mec-3) has on touch neuron differentiation (Duggan et al., 1998). How unc-86 and ttx-3 interact to control NSM differentiation is currently unclear. Both genes are continuously expressed in NSM neurons, but do not regulate the expression of each other. Based on the synergistic nature of the effect of joint ttx-3 and unc-86 removal on the expression of some target genes (no or limited effect in single mutants, complete loss in double mutant), we propose that both transcription factors act jointly on common target gene promoters. For some target genes, the loss of one regulatory factor can be completely or partly compensated for by the other regulatory factor; in other cases, such compensation is not possible. unc-86 and ttx-3 might therefore not always act in a strict cooperative sense, but rather act independently on target gene promoters. There is already a notable precedent for such a mechanism, as we recently found that a combination of three different transcription factors controls dopaminergic neuron identity. For some target genes, individual transcription factor mutants display very limited effects, but double mutants strongly affect target gene expression (Doitsidou et al., 2013). In the case of NSM, we cannot however rule out the possibility that some genes are exclusively regulated by unc-86 whereas others are exclusively regulated by ttx-3.

Apart from demonstrating ttx-3 terminal selector function in distinct neuron types, we have also shown here that the POU homeobox gene unc-86 can similarly act as a terminal selector in distinct neuron types. A role of unc-86 in the differentiation of serotonergic and glutamatergic touch neurons has been described previously (Desai et al., 1988; Duggan et al., 1998; Sze et al., 2002; Serrano-Saiz et al., 2013). We show here that unc-86 also controls the terminal differentiation programs of three distinct cholinergic neuron types. Two of these cholinergic neuron types are synaptically connected and form a simple sensory-to-motor circuit (White et al., 1986). The role of unc-86 in controlling cholinergic IL2 sensory neuron specification is reminiscent of, and might even be homologous to, the function of the POU homeobox gene acj6 in controlling expression of the cholinergic gene locus in Drosophila olfactory neurons (Lee and Salvaterra, 2002). The ARID-type transcription factor cfi-1 cooperates with unc-86 to control the cholinergic identity of IL2 and URA neurons. Although neuronal differentiation functions have been reported for the cfi-1 homolog dead ringer (retained - FlyBase) in Drosophila (Ditch et al., 2005), the functions of vertebrate orthologs (Arid3 genes) in the nervous system remain to be explored.

For a list of strains and transgenes and notes on their generation see supplementary material Table S2.

Young adult animals were fixed in 4% paraformaldehyde overnight and then treated with 5% β-mercaptoethanol overnight followed by 1000 units/ml collagenase (Sigma-Aldrich) treatment. Rabbit anti-serotonin whole serum (Sigma-Aldrich, S5545) was used at 1:100 dilution. Worms were then washed and incubated with Alexa Fluor 555 donkey anti-rabbit IgG (1:1000; Life Technologies, A-31571).

DNA sequences were subcloned into pPD95.75 expression vector (Addgene). For some smaller constructs, PCR products were directly amplified from subcloned constructs that have the same 3′ end of the promoter sequences. DNAs for injection were PCR amplified to eliminate vector backbone, gel purified and then injected as complex arrays (10 ng/μl) with digested rol-6(d) (3 ng/μl) as injection marker, or plasmid mix was directly injected [50 ng/μl together with 100 ng/μl rol-6(d)].

Full-length unc-86 cDNA was cloned into the pET-21b His tag expression vector (EMD Millipore) and transformed into BL21(DE3) pLysS bacteria (Novagen). Protein expression was induced using 1 mM IPTG for 4 hours at 37°C and batch purified using Ni-NTA resin (Qiagen) under denaturing conditions as described (Wenick and Hobert, 2004). TTX-3 was purified and electrophoretic mobility shift assays (EMSAs) were performed as described (Wenick and Hobert, 2004). Probe sequences are listed in supplementary material Table S3.

We thank Q. Chen and B. Alarcon for expert assistance in generating transgenic strains, V. Reinke for providing the ttx-3 fosmid reporter, E. Serrano for eat-4 reporters, members of the worm community for providing reporter genes and members of the O.H. laboratory for comments on the manuscript.