C. elegans ZAG-1, a Zn-finger-homeodomain protein, regulates axonal development and neuronal differentiation

Cell signaling and cell-intrinsic processes orchestrate the specification and differentiation of distinct neuron types(Edlund and Jessell, 1999). For example, a gradient of sonic hedgehog establishes the expression pattern of different transcriptional regulators along the dorso-ventral axis in the developing, vertebrate ventral spinal cord. The unique complement of factors in each domain determines, in part, the subsequent fate adopted by each neuron. Presumably, it also initiates transcriptional cascades that generate the appropriate characteristics for an individual neuron type, such as axonal structure, synaptic connections and differentiated gene-expression profile. Although factors, such as LIM-homeodomain and ETS-domain proteins, which contribute to the establishment of axonal-projection patterns and connectivity, have been identified in several organisms (reviewed by Hobert and Westphal, 2000; Shirasaki and Pfaff, 2002),the repertoire of factors needed to generate attributes for most neuron types is largely unknown.

The nervous system of C. elegans consists of relatively few neurons but it contains a great diversity of neuron types. The 302 neurons of the adult hermaphrodite are grouped classically into 118 classes, based on position, axonal morphology and connectivity(White et al., 1986). Subsequent studies indicate that individual neurons in some classes exhibit different gene-expression profiles, revealing the potential for a still greater number of distinct neuron types(Troemel et al., 1999; Yu et al., 1997). Several genes controlling the formation of neuron-specific characteristics and functional properties have been discovered. Some genes specify the fate of a neuron from an alternative or default state: mutation of the LIM-homeodomain gene lim-4 causes AWB to adopt an AWC-like fate(Sagasti et al., 1999); and mutation of the forkhead domain gene unc-130 induces ASG to adopt an AWA-like fate (Sarafi-Reinach and Sengupta, 2000). Other genes define the characteristics of different neuron types: the homeodomain gene unc-42 regulates axonal development and neuron-type-specific expression of several glutamate and chemosensory receptors (Baran et al.,1999; Brockie et al.,2001); the homeodomain gene unc-30 controls differentiation of GABAergic D-type motor neurons and unc-30 mutants have defects in axon pathfinding, synaptic connectivity and expression of the glutamic acid decarboxylase UNC-25 and GABA vesicular transporter UNC-47(Eastman et al., 1999; Jin et al., 1994).

δEF1/ZFH-1 Zn-finger-homeodomain proteins are distinguished by two arrays of highly similar C2H2-type Zn-finger domains and a centrally located homeodomain, and can act as transcriptional repressors through recruitment of the corepressor C-terminal binding protein (CtBP). Although Drosophila has one δEF1/ZFH-1 homolog(Fortini et al., 1991),vertebrates have two, one of which is most similar to the eponymous chickδEF1 (Funahashi et al.,1993) (also known as Nil-2-a, TCF8, ZEB, BZP, AREB6, MEB1 and ZFHEP) (Sekido et al., 1996),the second, SIP1 (also known as ZEB-2), can associate with Smad proteins(Postigo and Dean, 2000; Verschueren et al., 1999). Drosophila zfh-1 is expressed in embryonic mesoderm, mesodermally derived tissues and motor neurons (Lai et al., 1991), and is needed for development of gonadal mesoderm,heart and other tissues derived from mesoderm(Broihier et al., 1998; Lai et al., 1993; Su et al., 1999). MouseδEF1 is expressed in notochord, somites, limb, neural-crest derivatives and some CNS regions, and is required for thymus and skeletal development(Higashi et al., 1997; Takagi et al., 1998). AlthoughδEF1 is a negative regulator of muscle differentiation in vitro(Postigo and Dean, 1997),mouse δEF1 knockouts lack muscle and CNS defects. Mouse SIP1 is needed for development and migration of specific neural-crest cells(Van De Putte et al.,2003).

Here we report the genetic and molecular characterization of the lone C. elegans δEF1/ZFH-1 homolog, ZAG-1. zag-1 is essential for many aspects of neuronal differentiation and is expressed widely and dynamically during formation of the nervous system and in some muscles. ZAG-1 represses its own expression by interacting with conserved sequences in its promoter and, possibly, introns. We propose that zag-1 acts as a transcriptional repressor during the late stages of neuronal differentiation to establish several neuron-specific characteristics, including correct cell position and axon structure, and proper expression of cell-surface proteins,transmembrane receptors, ion channels, and biosynthetic enzymes and reuptake transporters for neurotransmitters.

Worms were raised at 20°C and cultured as described by Brenner(Brenner, 1974). The following strains were used. LGIV: lin-1(e1275), vab-2(e96), dpy-13(e184),unc-86::gfp(kyIs179) (Gitai et al.,2003); LGV: him-5(e1490), sra-6::gfp(oyIs14)(Sarafi-Reinach et al., 2001), nmr-1(akIs3) (Brockie et al.,2001); LGX: lin-15(n765); undetermined linkage: lin-11::gfp(mgIs21) (Hobert et al., 1998), unc-25::gfp(juIs75)(Jin et al., 1999).

We recovered pag-3 and zag-1 alleles in a screen for mutants with defects in PVQ, which will be described in detail elsewhere. In brief, sra-6::gfp(oyIs14) parental generation (P0) animals were treated with either ethyl methanesulfonate (EMS) or N-ethyl-N-nitrosourea(ENU) (De Stasio et al.,1997), F2 progeny that exhibited behavioral defects, such as an uncoordinated movement (Unc) or egg-laying defective (Egl) phenotype, were transferred individually to new plates and their progeny examined using epifluorescence microscopy to identify mutants with PVQ cell-fate specification, axonal development and other defects. We picked Uncs and Egls to enrich for animals with potential abnormalities in the structure or function of the nervous system.

Five pag-3(zd48, zd49, zd63, zd111 and zd120)and two zag-1(zd85 and zd86) alleles were recovered following EMS treatment and one pag-3(zd124) allele was found following ENU treatment. All six pag-3 mutations mapped to LGX and failed to complement pag-3(ls20) for the Unc phenotype. Two-factor crosses of zd85 and zd86 indicated linkage to lin-1 IV. Complementation testing indicated that zd85 and zd86 are allelic and defined a new gene, zag-1. We performed three-factor crosses to map zag-1further. From zag-1(zd85)/lin-1 dpy-13 heterozygotes, 24 out of 27 Lin non-Dpy recombinants segregated Zag-1 animals and from zag-1(zd86)/lin-1 dpy-13 heterozygotes, 20 out of 21 Lin non-Dpy recombinants segregated Zag-1 animals. From vab-2/lin-1 zag-1(zd85)heterozygotes, 2 out of 40 Zag non-Lin recombinants segregated Vab-2 animals and 30 out of 35 Lin non-Zag recombinants segregated Vab-2 animals. Thus, zag-1 maps between vab-2 and dpy-13.

Germline transformation (Mello and Fire, 1995) of zag-1 mutants was performed by co-injecting test DNA (10-50 μg ml^-1) and sur-5::SUR-5-GFP reporter DNA (50 μg ml^-1), pTG96,which labels most nuclei with GFP (Yochem et al., 1998). Transformed F1 and F2 animals were identified by GFP expression from pTG96. The cosmids F28F9, F47C12, W02C12, W03D2, F49F1,R07C12 and R08D10 were tested, either individually or in pools, for rescue of the Unc phenotype of zag-1(zd85) animals. Three out of five extrachromosomal array lines containing F28F9 and none of the lines harboring the other cosmids were rescued. Subclones of F28F9 were constructed using pBluescript II KS(-). The 10 kb KpnI-SalI genomic fragment(pSK55; position 3,851,820-3,861,821 of chromosome IV) rescued (3/3 lines), whereas an 8.8 kb PstI-PstI fragment (pSK54;position 3,852,925-3,861,764) failed to rescue (0/2 lines).

Transcriptional and translational reporter lines were generated using GFP-expression vectors provided by A. Fire. DNA sequences upstream of each gene were amplified from N2 genomic DNA using PCR, and cloned in-frame into an appropriate GFP vector. For dat-1::gfp, DNA sequences -3671 to +1 were amplified (+1 is first base of ATG) and cloned into SphI and MscI sites of pPD95.77. For talin::gfp, DNA sequences -2070 to +6 were amplified and cloned into SphI and BamHI sites of pPD95.77. For tph-1::gfp, DNA sequences -1748 to +9 were amplified and cloned into HindIII and MscI sites of pPD95.81. For unc-129::gfp, DNA sequences -3087 to +6 were amplified and cloned into XbaI and MscI sites of pPD95.75. Sequences were included in downstream primers to generate a BamHI, MscI or StuI (unc-129::gfp) site. glr-1::gfp, mec-4::gfpand odr-2::cfp were derived from previously described reporters(Chou et al., 2001; Lai et al., 1996; Maricq et al., 1995) by replacement of `wild type' GFP sequences with GFP S65C or CFP (GFP Y66W,N146I, M153T and V163A) coding sequences with synthetic introns.

Using PCR and standard cloning methods, we introduced an XbaI site at the end of exon 7 (TCTACCTAG changed to TCTAGATAG)and fused GFP gene and unc-54 3′ untranslated region (UTR)sequences from pPD95.77 to zag-1 genomic sequences. zag-1::ZAG-1-GFP is predicted to produce a full-length ZAG-1-GFP fusion protein that includes a substitution of arginine for the C-terminal threonine. We made zag-1::ZAG-1(213)-GFP, which is expected to generate a ZAG-1(213)-GFP fusion protein containing the N-terminal 213 amino acids of ZAG-1, by cloning the GFP gene and unc-54 3′UTR sequences from pPD95.77 into the BamHI site in exon 2.

We engineered a SmaI site at codon 7 (GCCATGC changed to GCCCGGG) and generated zag-1::gfp transcriptional reporters using pPD95.81 and the following zag-1 upstream genomic fragments: pSK73, -4030 (KpnI) to +21; pSK62, -2925 (PstI)to +21; pSK109, -1623 (SnaBI) to +21; pSK110, -736 (MscI) to+21; and pSK111, -321 (EcoRV) to +21. We mutated specific CACCT motifs present in zag-1::gfp promoter (pSK109) using the Promega GeneEditor in vitro Site-Directed Mutagenesis System to create pSK183, -1114A(CAGGTG changed to CAGATG) (E box); pSK184, -918A(AGGTG changed to AGATG); pSK185, -895T(CACCT changed to CATCT); and pSK186, -918A, -895T. Mutations were confirmed by sequencing.

Extrachromosomal arrays were generated by germline transformation of lin-15(n765) animals with GFP-reporter DNA (50 μg ml^-1)and lin-15(+) gene (pSK1) (50 μg ml^-1), and stable chromosomal integration was induced by treatment with trimethyl psoralen (TMP)and UV. lin-15(n765) mutants exhibit a temperature-dependent,multivulva phenotype that can be rescued by pSK1(Clark et al., 1994). The two zag-1 translational reporters (zdIs39, zdIs40) were created using the dominant rol-6(su1008) roller marker (pRF4); a similar expression pattern was observed using lin-15(+) marker. Each integrant was backcrossed three times with N2. The strain name, reporter name,allele designation, linkage group (if known) and plasmid name for transgenics generated in this study are: MT4003, glr-1::gfp(zdIs3) IV, pSK35;MT4005, mec-4::gfp(zdIs5) I, pSK39; MT4010, odr-2::cfp(zdIs10)V, pSK37; MT4013, tph-1::gfp(zdIs13) IV, pSK41; MT4015, talin::gfp(zdIs15) V, pSK57; MT4021, zag-1::gfp(zdIs21) IV,pSK62; MT4031, dat-1::gfp(zdIs31) X, pSK101; MT4039, zag-1::ZAG-1-GFP(zdIs39), pSK71; MT4040, zag-1::ZAG-1(213)-GFP(zdIs40), pSK67; MT4041, zag-1(-918A, -895T,4.0 kb)::rfp(dsred2)(zdIs41), pSK209; and MT4042, unc-129::gfp(zdIs42) IV, pSK127.

Images were acquired using a Hamamatsu Orca CCD camera and a Leica DMRE microscope, and edited using Improvision Openlab and Adobe Photoshop.

We determined the DNA sequence of the longest predicted EST clone, yk312a9,of F28F9.1. DNA sequences were determined using the ABI PRISM BigDye Terminators Cycle Sequencing kit and either an ABI PRISM 310 Genetic Analyzer or MJ Research BaseStation 51 DNA Fragment Analyzer. The genomic organization of F28F9.1 based on the sequence of yk312a9 is different from the hypothetical gene structure in WormBase release WS99 (17 April, 2003)(http://wormbase.org);we believe the yk312a9-based structure is correct because it is derived from cDNA sequence and the open reading frames are highly conserved in the related nematode C. briggsae.

To find alterations associated with zag-1 mutations, the coding and splice-junction regions were amplified from N2, zd85 and zd86 genomic DNA using PCR and flanking primers. The DNA sequences of the amplified products were determined directly using either internal sequencing primers or the PCR primers and a cycle-sequencing protocol. zd85 and zd86 each contained a single point mutation within the F28F9.1 gene and the N2 F28F9.1 sequence was identical to that reported by the C. elegans Sequencing Consortium.

We discovered zag-1 in a screen for mutations that disrupt PVQ development. PVQ is a pair of interneurons located in the lumbar ganglia in the tail, and each extends a single axon that runs within the ventral nerve cord (VNC) to the nerve ring in the head(White et al., 1986). An sra-6::gfp (serpentine receptor class A) reporter is expressed in PVQ and in amphid neurons ASH and ASI (Troemel et al., 1995), which allows direct observation of these neurons in living animals using epifluorescence microscopy(Fig. 1A). We treated sra-6::gfp(oyIs14) animals with the mutagens EMS and ENU and identified mutants with PVQ defects (see Materials and Methods). Expression of sra-6::gfp was seen uniformly in ASH and ASI but was observed rarely in both PVQs in eight mutants (Fig. 1B, Table 1, data not shown). Six mutations (zd48, zd49, zd63, zd111, zd120 and zd124) are pag-3 alleles. We reported previously that PVQ is present in pag-3 mutants but fails to express sra-6::gfp(Cameron et al., 2002).

The remaining two alleles (zd85 and zd86) defined a new gene, zag-1 IV. Both are recessive and exhibit a similar,uncoordinated, behavioral phenotype. zag-1 worms were active but,typically, had a kinked appearance and serious difficulty moving backward. The penetrance and expressivity of the PVQ sra-6::gfp expression defect were incomplete (Table 1). PVQ axons, when detected, followed their normal trajectory and 67 out of 68 reached the nerve ring. We examined zag-1 animals using DIC microscopy and observed neuronal nuclei in positions characteristic of PVQ,which indicated that the lack of GFP-labeled neurons resulted from either reduction or elimination of sra-6::gfp expression rather than the absence of PVQ.

lin-11::gfp(mgIs21) has faint PVQ expression in young larvae (Hobert et al., 1998),permitting visualization of cell bodies but not axons. We detected both PVQ cell bodies in zag-1; lin-11::gfp larvae (21/21 for mgIs21,19/19 for zd85; mgIs21 and 23/23 for zd86; mgIs21),confirming the presence of PVQ and revealing the proper regulation of lin-11::gfp expression. Together, these observations indicate that zag-1 is needed to specify some features of PVQ-cell fate, such as the expression of sra-6, but not lin-11.

The HSN serotoninergic motor neurons are born in the tail and migrate to flank the gonad in the midbody during embryogenesis(Desai et al., 1988; Sulston et al., 1983). HSN matures during the fourth larval (L4) stage: the nucleus and nucleolus enlarge and a distinctive structure, called the `hood', forms around the nucleus. The HSN axons go round the vulva, enter the VNC and continue to the nerve ring.

Serotoninergic neurons express the tryptophan hydroxylase TPH-1(Sze et al., 2000). tph-1::gfp(zdIs13) expressed GFP in serotoninergic neurons ADF, HSN,NSM and male-specific CP and R(1,3,9) (Fig. 1C). GFP expression in HSN was observed first during late L4 and continued throughout adulthood, consistent with the immunological detection of serotonin that starts in young adults(Desai et al., 1988). We found that HSN GFP expression was rare or absent in zag-1 tph-1::gfp adult animals, whereas expression was unchanged in ADF, NSM, CP and R(1,3,9)(Fig. 1D, Table 1). unc-86::gfp(kyIs179) displays GFP expression in HSN during larval development (Gitai et al.,2003). We examined zag-1(zd85); unc-86::gfpanimals and observed both HSNs in their proper midbody location (20/20 for kyIs179 and 18/18 for zd85; kyIs179). Thus, the HSNs are born, migrate to the midbody and express unc-86 but not tph-1 in zag-1 mutants.

odr-2 encodes a GPI-linked, cell-surface protein that is expressed by several sensory neurons, interneurons and motor neurons but not HSN(Chou et al., 2001). CFP-expressing HSNs were never observed in wild-type odr-2::cfpadults (n=50), whereas CFP-expressing HSNs were detected in all zag-1; odr-2::cfp adults (zd85; 34/40, 2 HSNs, 6/40, 1 HSN and zd86; 38/40, 2 HSNs, 2/40, 1 HSN). The morphology of HSN cell bodies and axons was typically abnormal in zag-1 mutants visualized using either odr-2::gfp or unc-86::gfp. The HSNs retained a generic neuronal appearance and, often, failed to form their characteristic hood. Although most HSN axons extended ventrally and entered the ventral nerve, their trajectory was aberrant and they often branched extensively;ectopic axons also projected from cell body(Fig. 1E-G). In summary, zag-1 mutants have defects in multiple aspects of HSN differentiation, including axon pathfinding, hood formation and tph-1expression. HSN development appears wild type until midlarval stages: the HSNs are born, migrate to the midbody and correctly express unc-86 and not odr-2.

Although the axons of most neurons in C. elegans are relatively simple and unbranched, the axons of a few neurons have well-defined,reproducible, branched structures (White et al., 1986). The dopaminergic neurons ADE and PDE have axons with stereotypic branched structures and can be visualized using a GFP reporter to the dopamine transporter gene dat-1(Nass et al., 2001). ADE is a pair of ciliated neurons located behind the second bulb of the pharynx. Each ADE projects a process ventrally that extends to the ventral ganglion and a process anteriorly that splits to a form a ciliated ending and a branch that enters the ring neuropil laterally (Fig. 2A,I). We found that the anterior process always formed a branch to the ring neuropil in wild-type dat-1::gfp(zdIs31) animals (47/47 ADEs) but this was generated infrequently in zag-1; dat-1::gfpanimals (8/100 ADEs zd85; zdIs31, 17/100 ADEs zd86, zdIs31)(Fig. 2B,I). The ciliated branch of the anterior process and the ventral process were unaffected,indicating that zag-1 mutants have a specific defect in branch formation rather than a general defect in ADE-axon outgrowth.

PDE is a pair of ciliated neurons located in the posterior body. Each PDE projects a process dorsally with a ciliated ending, and a process ventrally that enters the VNC, splits and extends anteriorly and posteriorly(Fig. 2C,I). Because the PDEL and PDER axons run together in the right ventral nerve bundle, we scored the extent of growth of the longest process. The posterior branch of the ventral PDE process typically extended ∼3/4 of the distance between PDE cell body and anus in wild-type dat-1::gfp animals (38/47 full length), but occasionally it was shorter (9/47 partial length) or failed to extend (2/47). By contrast, most zag-1; dat-1::gfp animals lacked the posterior branch (zd85; 3/100 full-length, 35/100 partial length, 62/100 no extension, and zd86; 1/100 full length, 26/100 partial length, 73/100 no extension) (Fig. 2D,I). The posterior branch was rarely redirected anteriorly. The anterior branch was slightly shorter than wild type in ∼1/2 of the zag-1; dat-1::gfpanimals and the ciliated dorsal process was unchanged. We conclude that zag-1 is required for the extension of the anteriorly directed axon branches of ADE and the posteriorly directed axon branches of PDE.

The HSN axons often form one or more short branches near the vulva(White et al., 1986). The HSN axons detected in zag-1(zd85) tph-1::gfp animals failed to branch at the vulva (0/17 HSNs) and were sometimes short or slightly misdirected (5/17),whereas most HSNs in wild-type tph-1::gfp animals formed a branch near the vulva (25/31) and were rarely short or misdirected (2/200). Most HSN axons in zag-1 mutants had severe guidance defects viewed using either odr-2::cfp or unc-86::gfp (see above). The few GFP-expressing HSNs detected in zag-1(zd85) tph-1::gfp animals presumably represented a population with sufficient zag-1 activity to promote tph-1::gfp expression and the formation of a nearly correct axon in most cases. However, the complete absence of stereotypic branches at the vulva argues that zag-1 is needed to shape multiple, discrete HSN-axon features, such as the extension of an axon with an appropriate nerve ring trajectory and formation of branches at the vulva.

The six DD and 13 VD motor neurons are located in the VNC. Each extends a process anteriorly that forms a collateral to the dorsal nerve cord and divides and projects both anteriorly and posteriorly (DD also extends a short process posteriorly) (Fig. 2I). DD1 and VD2 commissures extend on the left side and, typically, fasciculate,whereas DD2-6 and VD1,3-13 commissures extend on the right side. Many DD and VD axons terminated their growth prematurely, were misdirected and branched abnormally in zag-1 mutants; the anteriorly directed branches adjacent to dorsoventral commissures were also often missing(Fig. 2I, data not shown). In addition, many DD and VD commissures extended on the wrong side. DD1 and VD2 commissures were fasciculated and extended on the left side in wild-type unc-25::gfp(juIs75) animals (n=63) and, infrequently, a single, additional commissure extended inappropriately on the left side(12/63). By contrast, 74% of zag-1(zd85); unc-25::gfp animals had one or more inappropriate commissures on the left side (19/50 one, 11/50 two, 6/50 three and 1/50 four). Conversely, DD1 and VD2 often extended on the right side(5/50 both, 35/50 one). Thus, zag-1 is needed for the formation of anteriorly directed DD and VD axon branches as well as for several aspects of DD and VD axon guidance.

Several interneurons and motor neurons, including the command interneurons AVA, AVB, AVD, AVE and PVC that extend processes along the VNC, express the GLR-1 glutamate receptor (Hart et al.,1995; Maricq et al.,1995), whereas a subset of glr-1::gfp-expressing neurons,AVA, AVD, AVE, AVG, PVC and RIM, express the NMR-1 glutamate receptor(Brockie et al., 2001). We found that all zag-1 glr-1::gfp(zdIs3) animals (40/40 zd86)had three or more aberrant axons emanating from either the nerve ring or tail that zigzagged and branched profusely along the length of the animal, but 1 out of 30 wild-type glr-1::gfp animals had a single, short,inappropriate lateral process that exited the nerve ring(Fig. 2E,F). Ectopic, neuronal expression was detected in the tail and the nerve ring was disorganized and defasciculated. The VNC was also defasciculated in zag-1; nmr-1::gfpanimals, but the growth of ectopic axons in lateral positions was uncommon. Thus, most of the aberrant processes seen in zag-1 glr-1::gfp animals likely arose from glr-1::gfp-expressing neurons that do not express nmr-1::gfp. zag-1 mutations also caused variable expression of nmr-1::gfp in PVC and ectopic expression in an unpaired tail neuron. These results indicate that zag-1 is required for specification and axon guidance of interneurons and motor neurons that express glr-1and nmr-1, and to prevent misexpression of glr-1 and nmr-1.

AVG, an interneuron located in the retrovesicular ganglion, pioneers the right ventral nerve bundle and is thought to provide cues that promote the proper assembly and organization of VNC (R. M. Durbin, PhD thesis, University of Cambridge, UK, 1987). Ablation of the parent of AVG led to inappropriate growth of DD and VD commissures on the left side as well as extension of many longitudinal axons on the wrong side of the VNC. We found that AVG sometimes terminated its growth prematurely in zag-1; odr-2::cfp animals (14/40 zd86), but it extended completely in wild-type odr-2::cfpanimals. As such, the disorganization of VNC and extension of DD and VD commissures on the wrong side might result in part from defects in AVG outgrowth.

The six touch receptor neurons, ALM, PLM, AVM and PVM, are needed for the gentle touch response (Chalfie et al.,1985) and express the MEC-4 ion channel subunit(Lai et al., 1996). The two ALMs are born and migrate to characteristic anterior body positions during embryogenesis (Sulston et al.,1983). In wild-type mec-4::gfp(zdIs5) animals, 49 out of 50 ALM cell bodies were found ∼3/4 of the way between the second bulb of pharynx and the vulva in young adults and 1 out of 50 was found halfway or less. By contrast, 12 out of 49 ALMs in zdIs5; zd85 and 11 out of 45 ALMs in zdIs5; zd86 animals were positioned halfway or less, and many had a variable dorsoventral position, indicating that zag-1 is needed for ALM to complete its posterior migration. Occasionally, mec-4::gfpexpression in ALM was not seen in zag-1 mutants (50/50 zdIs5, 49/50 zdIs5; zd85 and 45/50 zdIs5;zd86).

Each ALM projects a short axon posteriorly and an axon anteriorly that runs at least to the first bulb of pharynx and forms a branch that enters neuropil and contacts the AVM branch. We found that all anterior ALM axons (50/50) had a wild-type trajectory in mec-4::gfp(zdIs5), whereas only 9 out of 50 were wild type in zdIs5; zd86 animals(Fig. 2G-I). Most anterior ALM axons (41/50) grew as far as the nerve ring and turned to join the neuropil. The nerve ring branch appeared tangled and often failed to contact the AVM branch, which was also disorganized. Anterior ALM processes in zdIs5;zd85 animals were similarly defective. The posterior ALM process was typically absent in wild-type mec-4::gfp and mec-4::gfp;zag-1 animals, which likely represents natural variation. These observations indicate that zag-1 is required to form the ALM-axon extension past the nerve ring. Thus, zag-1 is needed for multiple phases of ALM development, including cell migration, axonal development and mec-4 expression. mec-4::gfp expression and axon morphology of PLM, the posterior homolog of ALM, were essentially wild type in zag-1 mutants.

AVM and PVM are descendants of neuroblasts QR and QL, respectively. QR and its descendants migrate to specific positions in the anterior body, whereas QL and its descendants migrate to stereotypic positions in the posterior body. AVM and PVM were found in their correct locations in zag-1 animals but mec-4::gfp expression was altered. PVM mec-4::gfpexpression was often eliminated (25/25 zdIs5, 15/25 zdIs5;zd85 and 10/25 zdIs5; zd86) but AVM mec-4::gfpexpression was largely unaffected (25/25 zdIs5, 24/25 zdIs5;zd85 and 24/25 zdIs5; zd86). Inappropriate expression of odr-2::cfp was also observed occasionally in PVM (data not shown). AVM and PVM axons enter the VNC and run anteriorly, AVM continues past the first bulb of pharynx and PVM stops in the anterior body. AVM forms a branch that splits, enters the neuropil and contacts ALM branches. Only 8 out of 24 AVM processes extended past the first bulb of pharynx in mec-4::gfp;zag-1(zd86) compared to 25 out of 25 in wild-type mec-4::gfp(Fig. 2G-I). Sixteen out of 24 AVM axons turned to enter the nerve ring after passing the second bulb of pharynx and lacked an anteriorly directed lateral extension, similar to the phenotype displayed by ALM axons. AVM branches were disorganized and often failed to make contact with ALM processes. PVM axons appeared wild type. Therefore, zag-1 mutations had differential effects on AVM and PVM. AVM usually expressed mec-4 and had distinct axon-branch-formation defects, whereas PVM often failed to express mec-4 and had a wild-type axon structure when detected.

unc-25 encodes the GABA biosynthetic enzyme glutamic acid decarboxylase and is expressed in all 26 GABAergic neurons(Jin et al., 1999). The unc-25::gfp(juIs75) reporter, which has 1.8 kb of sequence 5′of the ATG and coding sequences for the first 13 amino acids of UNC-25,expresses GFP in a subset of GABAergic neurons: DD, VD and RME. GFP expression in these neurons was significantly higher in zag-1; unc-25::gfp than in wild-type unc-25::gfp animals. Tryptophan hydroxylase catalyzes the first step in serotonin biosynthesis and tph-1 is expressed in serotoninergic neurons (Sze et al.,2000). Although HSN lacked tph-1::gfp expression in zag-1 mutants, expression in ADF, NSM, CP and R(1,3,9) was comparable to wild type. However, unlike wild type, faint GFP expression was detected in VC4 and VC5 in zag-1 mutants (35/100 VC4 and 57/100 VC5 zd85 zdIs13). VC4 and VC5 are detected weakly and sporadically with serotonin antisera and express the vesicular monoamine transporter cat-1(Duerr et al., 1999). Therefore, in zag-1 mutants, GFP expression in VC4 and VC5 apparently reflects upregulation of undetectable wild-type tph-1::gfpexpression, rather than inappropriate expression. Last, dat-1 encodes a dopamine transporter homolog that presumably facilitates dopamine reuptake(Nass et al., 2001). The dat-1::gfp(zdIs31) strain had moderate GFP expression in the eight dopaminergic neurons, ADE, PDE and CEP. Expression was greatly enhanced in these neurons by zag-1 mutations. Together, these results indicate that zag-1 activity downregulates the expression of neurotransmitter biosynthetic and reuptake genes.

We cloned zag-1 by genetic mapping and transformation rescue of its mutant phenotype (Fig. 3A,B). zag-1 maps between vab-2 and dpy-13 on chromosome IV (see Materials and Methods). We tested cosmids within this 469 kb interval by germline transformation and identified a single cosmid, F28F9, that completely rescued the uncoordinated phenotype of zag-1(zd85) animals. A 10 kb KpnI-SalI subclone of F28F9, which encompasses the hypothetical gene F28F9.1, also rescued. We analyzed the nucleotide sequences of the predicted coding regions and splice junctions of F28F9.1 from wild-type and zag-1 mutants and found that both zd85 and zd86 are G:C to A:T transitions that generate termination codons in exon 5. Together, these data indicate that zag-1 is F28F9.1.

We determined the DNA sequence of a full-length cDNA (provided by Y. Kohara) to establish the zag-1-gene structure(Fig. 3C). zag-1 has seven exons, a five-nucleotide 5′ UTR and, at most, a 444-nucleotide 3′ UTR. Spliced leader 1 sequences in cDNA indicated that the zag-1 transcript is trans-spliced. On the basis of the DNA sequence, zag-1 encodes a 596-amino-acid protein with five C2H2-type Zn-finger domains, two at the amino end and three at the carboxyl end, and a single homeodomain located in the middle (hence zag-1,Zn-fingerhomeodomain, axon guidance) (Fig. 3D, Fig. 4A). The N and C-terminal Zn-finger-domain arrays are highly similar to each other, and to the Zn-finger domains in Drosophila ZFH-1 and vertebrateδEF1 and SIP1 proteins (Fig. 4B). Although the total number of Zn-finger domains varies in these proteins, the overall structure, namely, Zn-finger-domain clusters positioned at the N and C terminals and a centrally located homeodomain, is conserved and is a defining feature of the δEF1/ZFH-1 protein family. The Zn-finger-domain clusters of δEF1/ZFH-1 proteins bind to the consensus sequence CACCT, the E-box sequence CACCTG and tandem arrays of these motifs (Ikeda and Kawakami,1995; Remacle et al.,1999; Sekido et al.,1997), indicating that ZAG-1 can also probably bind to CACCT sequences. The ZAG-1 homeodomain is most similar to those present in LIM homeodomain proteins, such as C. elegans LIM-4 and MEC-3, and the second homeodomain present in the C. elegans ZFH-2-related protein encoded by ZK123.3 (Fig. 4C). ZAG-1 lacks sequences characteristic of the Smad-interacting domain found in vertebrate SIP1 orthologs (Verschueren et al., 1999).

δEF1/ZFH-1 proteins share a conserved motif, PXDLS, needed for association with C-terminal binding proteins, CtBPs. The transcriptional corepressor CtBP is an NAD-dependent dehydrogenase that interacts with several DNA-binding proteins to mediate transcriptional repression (reviewed by Chinnadurai, 2002; Turner and Crossley, 2001). ZAG-1 has a CtBP-interaction motif (PLDLT) indicating that ZAG-1 can probably recruit CtBP and repress transcription.

The zd85 and zd86 mutations generate termination codons in exon 5 of ZAG-1. zd85 alters the glutamine codon for residue 380(CAG → TAG) and zd86 changes the tryptophan codon for residue 369 (TGG → TAG). Both mutants are predicted to lack the C-terminal Zn-finger domains but retain the N-terminal Zn-finger domains, the homeodomain and the CtBP-interaction motif, and are likely to retain partial ZAG-1 function (see Discussion).

The C. elegans Sequencing Consortium has released preliminary genomic sequence of the nematode C. briggsae(www.wormbase.org),allowing identification of a zag-1 ortholog. The C. elegansand C. briggsae ZAG-1 proteins are similar (88% identity)(Fig. 4A). The predicted amino acid sequences of the Zn-finger domains, homeodomain and CtBP-interaction motif are identical, whereas other regions contain conservative substitutions and short deletions and insertions. The genomic organization of the two zag-1 genes is similar and the positions of introns are identical(Fig. 3C, Fig. 4A). In addition to coding region, sequence conservation is present upstream of the ATG and in several introns.

We generated zag-1 GFP translational and transcriptional transgenes to examine the expression pattern of zag-1(Fig. 5A). We fused 9.3 kb of zag-1 genomic sequence (which included 4 kb of upstream sequence) to GFP coding and unc-54 3′UTR sequences to obtain a construct that produced a full-length ZAG-1 protein tagged with GFP at its C-terminus(Fig. 5A). This zag-1::ZAG-1-GFP construct rescued zag-1 mutants (data not shown), indicating that the fusion protein is functional and expressed in a spatial and temporal pattern sufficient for rescue.

Examination of the zag-1::ZAG-1-GFP(zdIs39) strain revealed relatively faint GFP expression in neuronal nuclei from mid to late embryogenesis and during the L1 stage (Fig. 6A,B). We saw widespread expression in head and tail regions containing differentiating neurons beginning around the comma stage that diminished as embryogenesis continued. Expression remained in a few neurons at hatching and was typically nonexistent by midlarval stages. We also detected transient expression in the postembryonic Pn.a neuroblasts and their descendants during the L1. We conclude that zag-1 is expressed transiently in most or all neurons during embryogenesis and in the Pn.a-derived ventral cord neurons during the L1, and that zag-1expression coincides generally with the time period that neurons extend axons and undergo terminal differentiation.

We constructed a translational fusion to exon 2 that expressed a GFP chimeric protein containing the first 213 amino acids of ZAG-1. zag-1::ZAG-1(213)-GFP (zdIs40) animals displayed a pattern of neuronal expression similar to that of zag-1::ZAG-1-GFP(zdIs39)animals, indicating that sequences between the end of exon 2 and 7 were unnecessary for generating the observed expression pattern. The GFP signal was detected in neuronal nuclei, showing that the first 213 amino acids of ZAG-1 are sufficient for nuclear translocation.

We examined the expression patterns of transcriptional GFP reporters containing 0.3-4.0 kb of upstream zag-1 sequence using integrated or multiple, independent extrachromosomal array-containing lines to investigate further the regulation of zag-1 expression(Fig. 5A). In contrast to the expression patterns displayed by the translational reporters, using the 1.6 kb, 2.9 kb and 4.0 kb upstream fragments we observed bright, persistent, GFP expression in a subset of neurons in the head and tail that began during midembryogenesis and continued throughout larval development and adulthood(Fig. 6C). These three fragments produced largely similar neuronal expression patterns, although the 4 kb fragment directed expression in additional neurons in the head, including the command interneurons. By contrast, the 0.3 kb and 0.7 kb fragments yielded only weak expression in a few neurons in the nerve ring. These data imply that sequences between -1.6 kb and -0.7 kb (where the first nucleotide of ATG is+1) are required for most of the expression seen in the characteristic subset of head and tail neurons.

We observed anal depressor and intestinal muscle expression starting around hatching using either the 2.9 kb or 4.0 kb 5′ fragment. No expression was seen using 1.6 kb or shorter fragments, indicating that sequences between-2.9 kb and -1.6 kb are needed for expression in these enteric muscles. The 2.9 kb fragment produced transient expression in body-wall muscles, starting before the comma stage and ending during midlarval development. The 0.3 kb,0.7 kb, 1.6 kb and 4.0 kb fragments did not generate expression in body-wall muscle, suggesting that sequences between -2.9 kb and -1.6 kb promote body-wall-muscle expression and that sequences between -4.0 kb and -2.9 kb repress this.

The 4.0 kb zag-1::gfp transcriptional and zag-1::ZAG-1(213)-GFP translational reporters share the same 4 kb upstream sequences but differ because zag-1::gfp includes the coding sequences for the first seven amino acids of ZAG-1 (as do all the transcriptional reporter constructs) and zag-1::ZAG-1(213)-GFPincludes sequences for exon 1, intron 1 and most of exon 2. Thus, exon 1 through exon 2 sequences are needed to confer the transient, widespread, weak GFP expression that is characteristic of the two translational reporters. Although there are differences in the structure and localization of the fusion proteins, we believe that transcriptional mechanisms mediated by sequence elements in intron 1 are most likely to underlie the reduced, transient expression, as detailed below. Together, these results indicate that zag-1 is expressed broadly in the nervous system and in selected muscles. The different expression patterns of the translational and transcriptional constructs reveal the presence of multiple regulatory elements in the promoter and in the gene that either upregulate or downregulate zag-1 expression.

Expression of zag-1 in neurons and muscle is reminiscent of the neural and mesodermal expression of Drosophila zfh-1 and vertebrateδEF1. zfh-1 mutations perturb development of gonadal mesoderm,heart and other mesodermally derived tissues, and δEF1 is a negative regulator of muscle differentiation in vitro. We examined the morphology and development of body-wall muscle using the talin::gfp(zdIs15)reporter, which produces high GFP levels in all body-wall muscles starting during embryogenesis. GFP expression and muscle morphology were similar in wild-type talin::gfp and zag-1; talin::gfp animals. TGFβ UNC-129 is expressed in dorsal but not ventral body-wall muscles(Colavita et al., 1998); the body-wall expression pattern of unc-129::gfp(zdIs42) was unaffected by zag-1 mutations. The morphology of anal depressor and intestinal muscles was unchanged in zag-1 mutants when observed using a zag-1::gfp reporter. Thus, although zag-1 is expressed in anal depressor, intestinal and body-wall muscles, we failed to detect defects in either muscle development or differentiation in zag-1 mutants using these reporters.

We examined zag-1 GFP reporter expression in zag-1mutants to investigate whether ZAG-1 regulated its own expression. We found that zag-1 mutations affected neuronal expression patterns of the 1.6 kb, 2.9 kb and 4.0 kb upstream fragments but not the 0.3 kb and 0.7 kb fragments (Fig. 5A). In particular, we found that zag-1 mutations induced bright GFP expression in the Pn.a-derived VNC motor neurons that began during midL1 and continued throughout adulthood (Fig. 6E). We also observed at least seven neurons in the tail that expressed GFP compared to only PVQ and PVT in wild type, as well as many additional GFP-expressing neurons in the head, including the command interneurons. Therefore, zag-1 activity represses zag-1::gfpexpression in many neurons via sequences located between -1.6 kb and -0.7 kb. The finding that the 4.0 kb upstream fragment also directed expression in the command interneurons in wild-type animals indicates the presence of multiple,positive and negative regulatory elements for command interneuron expression. zag-1 mutations did not alter zag-1::gfp expression in anal depressor, intestinal and body-wall muscles. The expression pattern of the zag-1::ZAG-1(213)-GFP reporter, which did not rescue zag-1,was also unaffected by zag-1 mutations. The inability of zag-1 mutations to alter zag-1::ZAG-1(213)-GFP expression indicates that either the repression mediated by sequences in exon 1 through exon 2 is independent of zag-1 activity or our two zag-1mutants retain partial activity.

δEF1/ZFH-1 proteins bind to the consensus sequence CACCT, the E box sequence CACCTG and tandem arrays of these motifs(Ikeda and Kawakami, 1995; Remacle et al., 1999; Sekido et al., 1997). Examination of zag-1 genomic sequences from C. elegans and C. briggsae revealed several blocks of identity in addition to conserved sequences in coding regions. Several, conserved CACCT motifs are present in the 0.9 kb region that is essential for ZAG-1-mediated repression of zag-1::gfp expression and in the first, third and fourth introns(Fig. 3C, Fig. 5B). Because previous studies have shown that altering the CACCT sequence to CATCT either greatly reduced or eliminated binding of either N or C-terminal δEF1/ZFH-1 Zn-finger-domain clusters (Remacle et al.,1999), we introduced these mutations into the 1.6 kb promoter fragment to test whether these sites influenced zag-1 expression. We altered the conserved E box (CACCTG) at -1114 and the two CACCT sites at -918 and -895, either singly or together, and examined GFP-reporter expression. Alteration of the E-box site had no obvious effect on expression, whereas mutation of either CACCT motif had a moderate effect. Both -918A and -895T promoter mutants produced reproducible expression in the command interneurons and, occasionally, generated expression in one or two ventral cord motor neurons in animals that contained extrachromosomal arrays. In the CACCT double mutant (-918A, -895T), the pattern of expression was strikingly similar to that observed in zag-1 zag-1::gfp mutants, including expression in command interneurons and most or all ventral cord motor neurons. The 4.0 kb fragment containing the -918A and -895T CACCT mutations also produced an expression pattern comparable to that observed in a zag-1-mutant background (Fig. 6D-F). Based on these results and published studies of the binding of δEF1/ZFH-1 proteins, we conclude that the tandem CACCT site is a ZAG-1 binding site and that ZAG-1 directly represses expression via this site.

zag-1 activity establishes several neuronal characteristics, such as cell position, axonal structure and gene-expression profile. Although zag-1 mutations confer various defects on sensory, motor and interneurons, common or related phenotypes are evident. These include the absence of stereotypic axon branches and upregulation of neurotransmitter biosynthetic and reuptake genes. zag-1 functions less to define neuron identity per se and more to generate features characteristic of a particular type of neuron. The specificity and selectivity of zag-1phenotypes for each neuron type suggests that zag-1 acts in combination with other cell-type-specific factors to control differentiation.

SRA-6 is a candidate chemosensory receptor, based on its predicted seven transmembrane domain topology and expression in amphid sensory neurons ASH and ASI (Troemel et al., 1995). sra-6::gfp provides an ideal indicator of PVQ development and differentiation, although sra-6 function in interneuron PVQ is unclear. zag-1 is required for specific elements of PVQ differentiation: LIM homeodomain gene lin-11 expression and axonal development appear wild type, whereas sra-6 expression is either reduced or eliminated. Because there is no evidence that ZAG-1 acts as a transcriptional activator, regulation of sra-6 expression is,presumably, indirect. PVQ sra-6 expression also requires PAG-3, a Zn-finger-domain protein that functions in neural differentiation and cell lineage (Cameron et al., 2002; Jia et al., 1997). Although zag-1 and pag-3 are coexpressed in other neurons, they do not share other known phenotypes.

zag-1 mutations disrupt several late differentiation features of HSN, including axon pathfinding, hood formation and tph-1 expression. Early development of HSN appears unaffected because HSNs migrate to their correct, midbody position, express unc-86 and do not misexpress odr-2. zag-1 HSN phenotypes are remarkably similar to those of egl-45, sem-4 and unc-86(Desai et al., 1988). The Zn-finger-domain gene sem-4 controls neuronal and mesodermal development (Basson and Horvitz,1996) and the POU-homeodomain gene unc-86 regulates neuronal-cell lineage and differentiation(Finney et al., 1988). egl-45, sem-4 and unc-86 act in a genetic pathway containing both parallel and overlapping branches that controls HSN development(Desai et al., 1988). zag-1 likely functions downstream of or in parallel to unc-86 because zag-1 mutations do not affect unc-86::gfp expression.

Ectopic HSN odr-2 expression in zag-1 adults reveals that zag-1 activity blocks expression of nonHSN genes as well as promotes HSN differentiation and proper gene expression. Regulation of odr-2,which encodes a GPI-linked cell-surface protein, is neuron-type specific; that is, zag-1 mutations induce misexpression of odr-2 in HSN and, occasionally, PVM but do not otherwise appear to alter the odr-2expression pattern. Similarly, zag-1 is needed for proper expression of the glutamate receptors GLR-1 and NMR-1 and the ion channel MEC-4, and to prevent misexpression of GLR-1 and NMR-1.

zag-1 activity downregulates the expression of neurotransmitter biosynthetic and reuptake genes. Although zag-1 does not determine which neurons express dat-1, tph-1 and unc-25, it modulates expression levels in the appropriate neurons, acting, perhaps, as a `gain'switch. zag-1 expression appears to be restricted to embryos and L1s,but upregulation of expression is still apparent in adults, which indicates that either ZAG-1 is present but undetectable in adults or ZAG-1-established expression levels are maintained.

zag-1 is essential for the formation of stereotypic axon branches of several neuron types. zag-1 mutations block the generation of anteriorly directed branches of ADE, ALM and AVM and the posteriorly directed branch of PDE. Except for lacking these branches, the axon trajectories of these neurons appear wild type and full-length, indicating an explicit defect in the development of axon branches rather than a deficiency of axon outgrowth. Although HSN, DD and VD axon-branching defects are coupled with broader pathfinding errors, we believe that these defects reflect a specific role of zag-1 in branching and pathfinding. The dorso-ventrally oriented ADL branches were unaffected by zag-1 mutations (data not shown), indicating that zag-1 is not required for all axon-branching patterns.

The dramatic guidance, branching and fasciculation defects of many glr-1::gfp-expressing interneuron and motor neuron axons illustrate further the role of zag-1 in creating axonal structures. Exuberant extension and branching might reflect either a guidance defect or an inability to terminate outgrowth at appropriate targets. The defasciculation of the nerve ring and ventral nerve cord and premature termination of the AVG axon reveal that zag-1 is also required for nerve bundle formation and axon outgrowth.

The mature axonal morphology of C. elegans neurons has been described in great detail, although the assembly of these axonal structures is less well understood. For example, several nonexclusive ways to sculpt the mature ALM axon structure are possible: ALM growth cone might first extend to the first bulb of pharynx and a collateral branch might later enter neuropil;ALM growth cone might first project to nerve ring, turn and enter neuropil and a collateral branch might later project anteriorly to first bulb of pharynx;and ALM growth cone might extend to nerve ring, split and form both branches concurrently. The latter two scenarios are consistent with the observation that zag-1 mutants lack anterior but not neuropil branches. We analyzed axonal structures in young adult hermaphrodites and limited analysis of larvae revealed similar axonal defects. Our current results indicate that zag-1 mutations block the formation of axon branches but do not rule out the possibility of inappropriate axon-branch retraction or pruning. Real-time analysis is needed to resolve the axon assembly pathway of ALM and other neurons, but this is technically difficult at present. However,regardless of assembly pathway, zag-1 activity is needed to shape the mature axonal configuration.

HSN has two projections during the L2 and L3, one directed ventrally that will ultimately reach nerve ring and be maintained in adult, and one directed dorsally that is, presumably, retracted later because it is not found in adults (Garriga et al., 1993). The HSNs often have two processes in zag-1 mutants, supporting the notion that zag-1 activity sculpts the mature axonal morphology by promoting the formation of new branches, preventing extension of inappropriate processes and eliminating immature structures.

Interactions with vulval cells control HSN-axon branching(Garriga et al., 1993) and interactions with BDU guide AVM branch into the mature neuropil(Walthall and Chalfie, 1988). Similar cellular interactions are believed to guide the formation of other axon branches. zag-1 might act in either a signaling or responding neuron to coordinate branch formation and pathfinding interactions. The specificity of axon-branch defects in zag-1 mutants indicates the existence of a branch-formation program that functions in combination with a primary axon-guidance program to define a particular axonal-projection pattern. Such branching programs might function either during or subsequent to formation of primary axons and entail the regulation of genes that generate,recognize and transduce spatial and temporal branching signals. The branching defects of ADE, PDE, ALM and AVM might represent a failure to activate a branching program, whereas defects of HSN, VD and glr-1::gfp-expressing neurons might reflect coexpression or misexpression of primary axon-guidance and branching programs.

ZAG-1 is a δEF1/ZFH-1 Zn-finger-homeodomain protein. Mutation of either zag-1 or two, conserved CACCT sites in the zag-1promoter upregulates zag-1::gfp expression in many neurons, including ventral cord motor neurons and command interneurons. As Zn-finger-domain arrays from δEF1/ZFH-1 proteins bind CACCT sequences, we conclude that ZAG-1 represses zag-1 expression by binding directly to this tandem CACCT site. Although many direct target genes of δEF1/ZFH-1 and SIP1 have been identified (reviewed by van Grunsven et al., 2001), our results provide the first example thatδEF1/ZFH-1 proteins regulate their own expression. In addition, our findings that both CACCT sequences are needed for complete repression are consistent with studies of Remacle et al.(Remacle et al., 1999), which show that SIP1 binds as a monomer and contacts one CACCT site with one Zn-finger-domain cluster and the other CACCT(G) site with the second cluster.

δEF1-knockout mice have defects in skeletal and thymus development,whereas a δEF1^δC727 mutant, which expresses a δEF1 protein lacking the C-terminal Zn-finger-domain cluster like the two ZAG-1 mutant proteins, has defects in only thymus development(Higashi et al., 1997; Takagi et al., 1998). These results suggest that δEF1^δC727 retains sufficient activity to promote skeletal development and that δEF1/ZFH-1 binding sites can be divided into two classes: those that require the C-terminal Zn-finger-domain cluster and those that do not. We infer that our two ZAG-1 mutants also retain activity and regulate the subset of zag-1 target genes that possess the second class of ZAG-1 binding site. Conserved CACCT(G) motifs in the first, third and fourth introns represent candidate ZAG-1-binding sites; however, zag-1 mutations fail to alter expression of zag-1-ZAG-1(213)-GFP, which contains intron 1 sequences. These observations indicate that these sequences are not ZAG-1-binding sites and that other factors mediate repression or that these sequences belong to the second class of ZAG-1-binding site. Our conclusion that the truncated ZAG-1 proteins retain activity is consistent with the observation that a zag-1 deletion mutant is not viable because of a feeding defect (Wacker et al.,2003).

Our GFP-transgene studies indicate that zag-1 is expressed transiently in most or all neurons during embryogenesis and in the Pn.a-derived ventral cord neurons during the L1. In general, the time of zag-1 expression coincides with the period that neurons extend axons and undergo terminal differentiation. The analysis of the zag-1-promoter region indicates that zag-1 is also expressed in anal depressor, intestinal and body-wall muscles, although no obvious muscle defects were observed. Similarly, although Drosophila zfh-1 is expressed in motorneurons, defects in motorneuron development have not been identified in zfh-1 mutants. It is speculated that the lack of neuronal and muscle defects in δEF1-knockout mice reflects a redundancy with SIP1, which is also expressed in these tissues(Postigo and Dean, 2000). Examination of zag-1-null mutants and additional muscle-differentiation markers is needed to investigate further the role of zag-1 in muscle development

Numerous transcriptional regulators have been identified that control the generation, specification and differentiation of the 302 neurons of C. elegans. Mutation of these factors cause a variety of defects that affect different neuron types as well as different aspects and phases of neuronal development. Although ZAG-1 shares some activities and characteristics with known factors, the overall role of ZAG-1 in neuronal development is unique. For example, zag-1 mutants have HSN-differentiation defects that are similar to unc-86 mutants and PVQ-differentiation defects that are similar to pag-3 mutants. However, unlike zag-1 mutations, unc-86 and pag-3 mutations also cause some daughter cells to reiterate the lineage of their mother, demonstrating that UNC-86 and PAG-3 control the generation of particular neuron types. Touch-receptor neurons exhibit none of their unique differentiated features in LIM-homeodomain gene mec-3 mutants (Way and Chalfie,1988) but display only a subset of defects, such as errors in mec-4 expression, cell migration and axon-branch formation, in zag-1 mutants. Thus, ZAG-1 is not involved in generating neurons or in defining every trait of a particular neuron type. In contrast to lim-4 and unc-130 mutations, which cause specific neurons to adopt alternative or default cell fates(Sagasti et al., 1999; Sarafi-Reinach and Sengupta,2000), zag-1 mutations do not induce cell-fate transformations but, rather, prevent the acquisition of a subset of neuron-type characteristics. The function of zag-1 is most similar to lim-6, unc-30 and unc-42, which establish select aspects of late differentiation such as axonal development, synaptic connectivity and neuron-type-specific gene expression (Baran et al., 1999; Brockie et al.,2001; Hobert et al.,1999; Jin et al.,1994). However, LIM-6, UNC-30 and UNC-42 have more restricted patterns of expression compared to ZAG-1, which appears to be expressed in most or all neurons. Thus, ZAG-1 acts as a global regulator of neuronal differentiation and is the first transcription factor to be identified that controls axon-branch formation.

Although most of the identified regulators are likely to function as transcriptional activators, ZAG-1 acts as a transcriptional repressor. Other factors involved in late-neural differentiation that repress transcription directly include the homeodomain UNC-4 and Groucho-like corepressor UNC-37,which specify synaptic choice (Winnier et al., 1999). ZAG-1 might function as a temporal switch during neuronal development; that is, ZAG-1 might initiate late differentiation by turning off a `late-differentiation repressor' and/or genes involved in early differentiation. The ectopic expression of glr-1, nmr-1 and odr-2 in zag-1 mutants indicates that ZAG-1 also blocks the adoption of inappropriate neuronal characteristics. As with specification of neuron fate, combined actions of both transcriptional activators and repressors are needed to establish the characteristics of terminal differentiation. Last, the observation that a presumptive ZAG-1-binding site is conserved between C. elegans and C. briggsae indicates that a bioinformatics strategy can be used to identify potential direct gene targets of ZAG-1, which will provide further insight into zag-1function.

We thank Katrina Sabater and Ray Squires for generating plasmid constructs and integrated transgenic strains, and B. Prasad, P. Sengupta and members of the Clark laboratory for comments on manuscript. We are grateful to C. Bargmann, A. Chisholm, J. Chou, A. Coulson, M. Driscoll, A. Fire, Z. Gitai, M. Han, O. Hobert, Y. Jin, Y. Kohara, V. Maricq, T. Sarafi-Reinach and P. Sengupta for providing strains and/or clones. We are eternally indebted to the C. elegans Sequencing Consortium (Sanger Institute and Genome Sequencing Center, Washington University, St Louis) for providing C. elegans and C. briggsae genomic sequence. Some nematode strains used in this work were provided by the Caenorhabditis Genetics Center. This work was supported by grants from the Alfred P. Sloan Foundation, the March of Dimes, the New York City Council Speaker's Fund and the NIH/NINDS (R01 NS39397).