The C. elegans gene pag-3 is homologous to the zinc finger proto-oncogene gfi-1

We are interested in how patterns of gene expression are established during development. The touch receptor neurons of the nematode C. elegans have been extensively characterized and provide a defined group of cells with which to address this problem (Chalfie, 1993; Chalfie and Au, 1989; Chalfie and Sulston, 1981; Chalfie and Thomson, 1982; Mitani et al., 1993; Xue et al., 1992, 1993). The genes that specify the touch cells apparently act in a combinatorial fashion, either to generate touch cells or to restrict their generation (Mitani et al., 1993). unc-86 and lin-32 encode transcription factors that are required to generate the touch neuron lineages (Chalfie, 1993). UNC-86 protein is thought to activate mec-3, which encodes a LIM-type homeodomain protein that is also required for the specification of the touch cell fate (Way and Chalfie, 1988, 1989). mec-3 is expressed in ten cells: the six touch receptor neurons, the two FLP neurons and the two PVD neurons (Way and Chalfie,1989). The MEC-3 and UNC-86 proteins are thought to form a heterodimer that activates genes required for touch cell function such as mec-4 and mec-7 (Chalfie, 1993). Negative regulation appears to be important for restricting the touch receptor fate to six cells: lin-4 mutant animals misexpress mec-4 and mec-7 in the PVD neurons; egl-44 and egl-46 mutant animals misexpress mec-4 and mec-7 in the FLP neurons and sem-4 mutant animals misexpress mec-4 and mec-7 in the PHC cells in the tail (Mitani et al., 1993; Basson and Horvitz, 1996). In addition, the celldeath genes ced-3 and ced-4 help to restrict the number of touch neurons, as mutations of those genes result in extra touch receptor cells (Mitani et al., 1993).

By staining C. elegans with fluorescent antibodies under conditions that do not kill embryos in gravid hermaphrodites (Jia et al., 1996; Xie et al., 1995), we screened for mutations that alter the expression of a mec-7lacZ fusion gene. mec-7 encodes a β-tubulin that is highly expressed in the touch receptor neurons and is weakly expressed in the FLP, PVD and BDU neurons (Hamelin et al., 1992; Mitani et al., 1993). We isolated several mutations that affected the PAttern of Gene expression (pag) of mec-7lacZ (Jia et al., 1996).

Mutations in pag-3 result in misexpression of touch neuron genes in the BDU neurons (Jia et al., 1996). In addition to the mec-7lacZ reporter gene, the mec-7, mec-2gfp, mec-4lacZ and mec-9lacZ genes are also expressed in the BDU neurons of pag-3 mutant animals. Mutations in pag-3 do not appear to affect mec-3lacZ expression, but the expression of mec-7 in the BDU neurons of pag-3 mutants does require mec-3(+) (Jia et al., 1996). pag-3 mutants also have motility defects: they are lethargic and show a reverse kinker uncoordinated (Unc) phenotype. This suggests that the PAG-3 protein is important for the development of other cells in addition to the BDU neurons and is involved in locomotion. Three pag-3 alleles have been identified. All of the pag-3 alleles are recessive to the wild-type allele and cause the Unc phenotype and misexpression of touch neuron genes in the BDU neurons. The ls20 and ls64 alleles cause similar phenotypes, whereas the ls65 allele results in a slightly weaker Unc phenotype. pag-3 was mapped to the right arm of the X-chromosome between unc-3 and unc-7. Here we report the cloning and sequence analysis of pag-3, its pattern of expression and mosaic analysis of its function.

C. elegans was cultured as described by Brenner (1974) and genetic nomenclature follows the guidelines of Hodgkin (1995). The wildtype strain was N2 var. Bristol. pag-3 alleles ls20, ls64 and ls65 were described in Jia et al. (1996). Mutations used in this study were: unc (UNCoordinated), pag (PAttern of reporter Gene expression abnormal), daf (abnormal DAuer Formation), osm (defective avoidance of high OSMotic strength) and sup (SUPpressor). The alleles used were: LGIII, unc-93(e1500); LGX, unc-3(e151), daf-6(e1377), pag-3(ls20), unc-7(e5), osm-1(p808), sup-10(mn219) and sup-10(n983). The free duplication mnDp14 and the integrated mec-2gfp array uIs9 were used in the mosaic analysis. The extrachromosomal pag-3lacZ array lsEx24 was used in analysis of pag-3 expression in pag-3 mutant animals.

Germ-line transformation was done by the procedure of Mello et al. (1991). To allow identification of the transgenic lines, plasmid pRF4 containing the semidominant rol-6(su1006) allele, which confers a roller phenotype, was co-injected at 80 ng/μl. The following YACs and cosmids were tested: Y70D2, Y37H4, R10G8, C27C12, T28G8, T28D11, M03F9, C18B12, K11C2 and F09B12. YAC DNA was injected at about 5 ng/μl. Cosmid DNA was isolated by the alkaline lysis method (Sambrook et al., 1989) and was injected as single cosmids or in pairs at 15-30 ng/μl for each cosmid. Phage DNA was isolated by the plate lysate method for the rapid analysis of bacteriophage λ isolates (Sambrook et al., 1989). Phage DNA was injected alone or in pools of 5-10 phage at 5-10 ng/μl for each phage. Restriction fragments from the rescuing phage YJ105 were purified on gels and injected at 5 ng/μl.

We separated Y37H4 YAC DNA by pulse field gel electrophoresis (Chu et al., 1986; Schwartz and Cantor, 1984), with a 1% low melting point agarose gel in 0.5× TBE at 4°C (Sambrook et al., 1989). The gel slice containing the YAC DNA was digested with agarase, and the DNA was either precipitated for microinjection or concentrated with a Centricon 10 microconcentrator (Amicon). We constructed a phage library in the λGEM-11 vector (Promega) by the method of Whittaker et al. (1993). About 400 ng of YAC DNA was partially digested with MboI in the presence of dam methylase. The MboI sticky ends were partially filled in with dATP and dGTP, ligated into the phage vector, packaged and plated in strain LE392 following the manufacturer’s protocols. We recovered 110 phage with an average insert size of about 20 kb.

C. elegans (0.2-2.0 g) were grown on egg-yolk plates as previously described (McDermott et al., 1996), cleaned by sucrose flotation as described (Sulston and Hodgkin, 1988) and frozen in liquid nitrogen. Total RNA was isolated by the method of Maes and Messens (1992) with an additional chloroform extraction. The RNA samples (20 μg) were treated with glyoxal-dimethyl sulfoxide, electrophoresed in 10 mM phosphate buffer (pH 6.5) and transferred to a MagnaCharge (MSI) nylon membrane (Ausubel et al., 1993). The blots were prehybridized for 2 hours and then hybridized overnight at 65°C in Church hybridization buffer (Church and Gilbert, 1984). The blots were washed twice in 2× SSPE, 0.1% SDS, then twice in 0.1× SSPE, 0.1% SDS at 50°C for 15 minutes each. Probes against pag-3 RNA were made from a 1.0-kb pag-3 cDNA fragment generated by PCR as described below; probes against act-1 RNA were made from a 210 bp fragment from the 3′-untranslated region of act-1 isolated from the plasmid pT7-T3-18-103 (Krause et al., 1989). A Prime-It RmT Random Primer Labeling kit (Stratagene) was used. A Phosphor Imager screen and Image Quant software (Molecular Dynamics) were used to measure hybridization intensity.

We screened 1.5×10⁶ plaques of a mixed stage C. elegans cDNA library (Stratagene), but did not recover any pag-3 cDNA clones, so we generated overlapping pag-3 cDNA fragments by the reverse transcription-polymerase chain reaction (RT-PCR) from wild-type and all three pag-3 mutant strains. cDNAs were synthesized from 20 μg total RNA with M-MuLV reverse transcriptase (New England Biolabs). The reactions were carried out at 37°C for 1 hour with 50 pmoles of an oligo(dT) primer [5′-GACTCGAGTCGACATCGA(T) ₁₇-3′]. The 5′cDNA fragments were amplified by PCR with the trans-spliced leader 1 (SL1) primer (5′-GGTTTAATTACCCAAGTTTGAG-3′) and L1, a primer specific to the zinc finger region (see below and Fig. 1). 3′cDNA fragments were generated by 3′RACE (Frohman et al., 1988) with the U1 and U3 primers and the same oligo(dT) primer used in the reverse transcription reactions. Amplifications were performed with Taq DNA polymerase (Perkin-Elmer Cetus) under standard conditions. The PCR fragments were ligated into the pGEM-T vector (Promega).

A 5 kb XhoI-KpnI genomic DNA fragment containing pag-3 was ligated into pBluescript (Stratagene) to generate pYJ5 and was partially sequenced. Subclones containing unidirectional deletions for sequencing were generated with the Erase-a-Base kit (Promega) with exonuclease III. Sequencing was performed with the Sequenase 2.0 kit (US Biochemicals) and T7 and SP6 primers. Gaps in the sequence were filled in with the following oligonucleotide primers:

U1, 5′-TGTGAATACTGTGGAAAAAGAT-3′; U2, 5′-TCTAACCTTATCACCCACAC-3′;

U3, 5′-CTGTTGAATCGTTACTCTCG-3′; U4, 5′-GCGAGGAAATGGAGGAGTGA-3′; U5, 5′-ATGCGTGTCTGCTTCTCAAA-3′; U6, 5′-GGAGTGATGATGAAGAGGAG-3′; U7, 5′-CCTGAGGAGTATGGCAATGG-3′; L1, 5′-TAAGGTTAGAAGACTGACTGAA-3′; L2, 5′-ATCAATTTTACCAACATACCTG-3′;

L3, 5′-TGAATCAGAAGGTGTGTAGA-3′; L4, 5′-CCATACTCCTGAGGTATTGC-3′; L5, 5′-ACTTTCTCCTCTTCATCATC-3′; L6, 5′-TAACATCGCAATACCATAAA-3′.

Overlapping pag-3 cDNA fragments were generated as described above. Both strands of the cDNA fragments were sequenced. For each mutant allele, a single RT-PCR-generated clone was sequenced, and another two independent PCR products were sequenced to confirm each mutation. Sequences were assembled with the DNAsis program (Hitati Software Engineering). Sequence analyses were done with the Wisconsin package of programs (Genetics Computer Group, 1994). The sequence of the wild-type cDNA was deposited in GenBank (Accession # U63996).

To construct a pag-3lacZ fusion gene, a partially digested 1.1-kb XbaI fragment from the plasmid pYJ5 spanning the first four exons and extending 400 bp upstream was cloned into the C. elegans expression vector pPD21.28 (Fire et al., 1990) to generate the plasmid pGX1. (The 5′ XbaI site was from the pBluescript vector.) An overlapping 2.2 kb XbaI fragment from the phage YJ105 with a 3′ endpoint just upstream of the first exon was inserted into pPD21.28 to produce pGX2. A 1.8 kb PstI-XhoI fragment from pGX2 was inserted into pGX1 to generate the plasmid pGX3. (The PstI site was from pPD21.28.) Thus, pGX3 contains approximately 2.2 kb of the pag-3 promoter sequence, and 700 bp downstream encoding the amino-terminal 142 amino acids of the PAG-3 protein, which is fused to the lacZ coding region of pPD21.28 (see Fig. 4). The same sequences were introduced into the green fluorescent protein (gfp) expression vector TU#61 (Chalfie et al., 1994) to generate a pag-3gfp fusion gene.

Plasmid DNA was purified with the Wizard miniprep DNA purification system (Promega) and injected into the syncytial gonad of wild-type adult hermaphrodites as previously described (Mello et al., 1991). The rol-6(su1006)-containing plasmid pRF4 was co-injected as a marker for identifying transgenic animals. Each plasmid was injected at a concentration of 100 μg/ml. Transgenic animals were fixed and stained for β-galactosidase activity (Xie et al., 1995) with DAPI (4,6-diamidino-2-phenylindole) to visualize the positions of nuclei. Alternatively, fixed animals were stained with rat polyclonal anti-β-galactosidase antiserum and rhodamine-conjugated secondary antibody (Xie et al., 1995). lacZ- positive cells were identified on the basis of their positions relative to other nuclei and by their axonal morphologies. The relative positions and axonal morphologies of the BDU interneurons, the touch neurons, the VA and VB motor neurons and the AVF interneurons have been described (Sulston, 1976; Sulston and Horvitz, 1977; Sulston et al., 1983; White et al., 1986).

Animals mosaic for pag-3 were generated with duplication mnDp14 and identified by the uptake of fluorescein isothiocyanate (FITC) in their phasmid and amphid neurons. The FITC staining protocol was described by Hedgecock et al. (1985). Strain EA185 (unc-93; pag-3 unc-7 sup-10 osm-1; mnDp14) was made from the following strains: EA92 (pag-3; him-5), SP1490 (daf-6 unc-7 sup-10; mnDp14) and SP1528 (unc-93; unc-3 unc-7 sup-10 osm-1; mnDp3). Strain EA181 (daf-6 pag-3) was constructed from EA138 (unc-3 pag-3) and SP1490. Strain EA186 (daf-6 pag-3 sup-10) was constructed from strain EA181 and EA96 (lin-15 sup-10). Strain EA187 (pag-3; mnDp14) was constructed from EA92 and SP1490. Strain EA188 (pag-3 sup-10 osm-1) was constructed from strain SP1491 (unc-7 sup-10 osm-1; mnDp3) and EA128 (pag-3 lin-15). Strain EA189 (daf-6 pag-3 sup-10; mnDp14) was constructed from strain EA186, SP1490 and EA81 (pag-3). Strain EA197 (pag-3 sup-10 osm-1; mnDp14) was constructed from strain EA188, SP1490 and EA81. Strain EA208 (pag-3 sup-10 osm-1; uIs9; mnDp14) was constructed from EA197 and TU2162 (uIs9).

We cloned the pag-3 gene by genetic mapping and germ line transformation rescue experiments as summarized in Fig. 1. pag-3 was mapped between two previously cloned genes, unc-3 and unc-7, on the right arm of the X-chromosome about 0.6 map units to the right of unc-3 (Jia et al., 1996). We transformed pag-3(ls20) animals with yeast artificial chromosomes (YACs) and cosmids in that interval and tested the transgenic lines for rescuing activity. The 350-kb YAC Y37H4 rescued the pag-3 Unc phenotype; however, none of the cosmids in that region rescued pag-3. This suggests that pag-3 lies between cosmids K11C2 and F09B12 in the only region not covered by cosmids. We made a phage library from the YAC DNA and transformed pag-3(ls20) animals, first with pools of phage DNA and then with DNA from single phage. Phage YJ105 rescued the Unc phenotype. Restriction fragments from YJ105 were tested for rescuing activity, and a 5-kb XhoI-KpnI fragment rescued both the Unc and the misexpression phenotypes in all transformed lines. A smaller fragment, which has a 1.0 kb deletion from the XhoI end, failed to rescue pag-3 mutants. Southern analysis indicated that pag-3 is a single-copy gene (data not shown).

We sequenced portions of the genomic XhoI-KpnI fragment and found two regions homologous to the zinc fingers of the mammalian proto-oncogene product GFI-1. We used oligonucleotide primers to these regions to generate overlapping pag-3 cDNA fragments by RT-PCR extending 3′ to the poly(A) tail and 5′ to the trans-spliced leader 1 (SL1) sequence found on many C. elegans mRNAs. Oligonucleotides to the SL1 sequence and to a zinc finger-encoding exon (L1) were used to generate a 5′ fragment of the cDNA, and oligonucleotides U1 and U3 were used with the oligo(dT) adapter primer to generate 3′ fragments for sequencing (Fig. 1C). The combined fragments correspond to a cDNA of 1609 bp (Fig. 1C). Comparison of the genomic DNA and cDNA sequences indicated that there are ten exons spanning 4 kb (Fig. 1C). Because the pag-3 transcript is trans-spliced to SL1 at the 5′ end, the transcriptional start site is unknown, but presumably lies within the 421 bp upstream of the mRNA 5′ splice site in the rescuing XhoI-KpnI genomic fragment. A polyadenylation signal was found 15 bp upstream of the poly(A) tail of the cDNA fragments, and about 500 bp from the 3′ end of the genomic fragment. Thus, the rescuing genomic fragment apparently contains the entire pag-3 gene.

The pag-3 cDNA encodes a protein with five C2H2-type zinc fingers (Fig. 2). A single open reading frame generates a predicted protein of 336 amino acids when read from the first AUG codon. As shown in Fig. 3, the 127 amino acids spanning the zinc fingers of the PAG-3 protein are 77% identical and 82% similar to zinc fingers 2-6 of the rat GFI-1 protein (Gilks et al., 1993). The fingers match the consensus Y/FXCX2CX3Y/FX8HX3H, and the third and fourth linkers match the consensus TGEKYX for the Krüppel class of zinc finger proteins (Schuh et al., 1986). No other similarities were identified between PAG-3 and GFI-1 outside the zinc finger region. The GFI-1 open reading frame terminates after the zinc finger region, whereas the predicted PAG-3 protein extends an additional 76 amino acids beyond the fifth zinc finger (Fig. 3). Although no other defined motifs were identified in PAG-3, two other regions are worth noting. There is a proline-rich region from amino acid 67 to 125 (22% proline of 59 amino acids), and there is a region from amino acid 304 to 331 that is predicted to be α-helical and highly charged, especially over an acidic region (318-EEMEEADDEEE-328). The predicted α-helix would not be amphipathic.

One point mutation was found in each of the pag-3 mutant alleles (Fig. 2). cDNAs corresponding to each mutant allele were generated by RT-PCR. All mutations are C→T transitions, a type of mutation often induced by EMS mutagenesis (Anderson, 1995). Previous studies have shown that the pag-3(ls20) and pag-3(ls64) alleles have the genetic characteristics of null alleles (Jia et al., 1996). The mutation in pag-3(ls20) at nucleotide 829 changes a coordinating histidine in the fourth zinc finger (HIS 228) to a tyrosine. The mutation in pag-3(ls64) at nucleotide 490 introduces a stop codon 13 residues before the first zinc finger at ARG 115. Thus, the zinc finger domain, and the fourth zinc finger specifically, is essential for the function of PAG-3. The third allele, pag-3(ls65), is a hypomorph with the least severe effects on motility (Jia et al., 1996). The mutation in pag-3(ls65) introduces a stop codon (at GLN 47) upstream of several AUG codons, so translation of the mutant mRNA may initiate at a downstream AUG codon to generate a truncated protein with intact zinc fingers.

To determine where pag-3 is expressed, we made a translational fusion construct with a 2.9-kb genomic fragment from the pag-3 gene and the E. coli lacZ gene (Fig. 4). The 2.9-kb pag-3 fragment includes about 2.2 kb of 5′-flanking sequence and encodes the N-terminal 142 amino acids. The vector includes a SV40 T-antigen nuclear localization signal. We transformed this fusion construct into wild-type C. elegans and obtained stable transgenic lines carrying the fusion gene in extrachromosomal arrays. We stained transgenic animals for β-galactosidase activity or with antibodies against β-galactosidase and examined them at all stages. The pag-3lacZ fusion gene was expressed in the BDU neurons and the six touch neurons (Fig. 5). Staining was also seen in 11 of 12 VA motor neurons, all 11 VB motor neurons, two AVF interneurons and the immediate precursors of all these cells (see Figs 5 and 6). Staining was seen not only in the nuclei, but also in the cell bodies and axons of the stained cells. The axonal trajectories of stained cells were clearly seen in transgenic animals stained with anti-β-galactosidase antibody. In addition, a few cells in the retrovesicular ganglion were stained; one consistently stained well, others sporadically stained weakly. All the staining patterns observed with the pag-3lacZ construct were also observed with a pag-3gfp construct (data not shown).

The expression of pag-3lacZ in the BDU neurons and the ALM and PLM touch neurons was first seen in twofold stage embryos (data not shown) when mec-7, a touch neuron-specific gene, is first expressed. These neurons are generated in embryos at this stage (Sulston et al., 1983). The AVM and PVM touch neurons are derived post-embryonically at the L1 larval stage (Sulston and Horvitz, 1977), and were stained at this stage (Fig. 5A). Staining in all of these cells lasted until the early L2 stage.

Expression in the VA and VB motor neurons and their immediate precursors was observed in late L1 to early L2 stage transgenic animals (Fig. 5). The ventral nerve cord consists of eight distinctive classes of motor neurons (Sulston, 1976; Sulston and Horvitz, 1977; Sulston et al., 1983). Three classes (DA, DB and DD) are generated during embryogenesis. The remaining five classes (VA, VB, VC, AS and VD) are descendants of P neuroblasts and are postembryonically derived during late L1 and early L2 larval stages. The cell lineages that generate these five classes of motor neurons are depicted in Fig. 6. The positions of all the motor neurons in the ventral cord are relatively invariant, and they exhibit distinctive axonal trajectories (White et al., 1986). Cells expressing pag-3lacZ have either anteriorly or posteriorly directed axons, which resemble the axonal morphologies of the V-type motor neurons (VA, VB and VC). The stained cells correspond in number and position to the VA and VB motor neurons. The remaining classes of motor neurons have not only anteriorly or posteriorly directed processes, but also commissures that run around to the dorsal cord. Staining was not observed in neurons with such commissures.

Northern analysis of total RNA from wild-type hermaphrodites at each developmental stage reflected the same pattern of temporal regulation (data not shown). A single transcript of about 1.6 kb is present in RNA from embryos and L1, L2 and L3 larvae and is most abundant in L2 larvae. The mRNA disappears in L4 larvae and young adults and reappears in mature adult hermaphrodites bearing embryos. Northern analysis of pag-3 expression in wild-type and mutant animals showed that pag-3 transcripts are produced at about twofold higher levels in mutant animals (data not shown).

Because the northern analysis suggested that pag-3 negatively regulates its own expression, we examined the effect of pag-3 mutations on the expression of the pag-3lacZ fusion gene. We introduced pag-3 mutations into strains carrying the fusion gene. We did not detect any change of pag-3lacZ expression in embryos, L1 or L2 larvae. However, the ALM and PLM touch receptor neurons and the BDU neurons express the fusion gene at a higher frequency in L3, L4 and adult animals in pag-3 mutant backgrounds than in the wild-type background (Table 1).

To define the cell lineages that require pag-3(+) activity for coordinated locomotion, we used the method of Herman (1984, 1987, 1989) to generate mosaic animals. The locomotion of C. elegans is controlled by 75 motor neurons that innervate the ventral and dorsal body muscles (Chalfie and White, 1988), and coordination between the motor neurons is mediated by five pairs of interneurons. Motility defects could be due to muscle or neural defects. Almost all body wall muscle cells (94 of 95) descend from P ¹, whereas 74 of 75 motor neurons are descendants of AB.p (Fig. 7B). The chromosomal location of pag-3 enabled us to use a scheme very similar to that used for mosaic analysis of unc-3 and unc-7 (Herman, 1984, 1987; Starich et al., 1993). pag-3 maps to the right arm of the X chromosome, in close proximity to daf-6 and osm-1 (Fig. 7A). Both genes are required for the uptake of FITC by the amphid neurons in the head and the phasmid neurons in the tail. daf-6 or osm-1 mutant animals carrying a chromosomal duplication of this region (mnDp14) were scored for FITC-staining pattern and motility phenotypes to determine at which cell division the duplication was lost.

To determine whether the Unc phenotype of pag-3 mutants results from a muscle defect, we first identified mosaic animals that had lost the duplication in P1. The strain we used has the genotype unc-93(e1500) III; pag-3(ls20) unc-7(bx5) osm-1(p808) sup-10(mn219) X; mnDp14(X;f)[pag-3(+) unc-7(+) osm-1(+) sup-10(+)]. unc-93(e1500) animals have the ‘rubberband’ phenotype: they contract and relax when they are prodded on the head (Greenwald and Horvitz, 1980). sup-10(mn219) is a recessive suppressor of unc-93(e1500) and has no phenotype alone (Greenwald and Horvitz, 1980). unc-7(bx5) animals exhibit a forward kinker Unc phenotype. pag-3(ls20) unc-7(bx5) double mutants show forward and reverse kinker Unc phenotypes. The duplication mnDp14 contains the dominant wild-type copy of pag-3, unc-7, osm-1 and sup-10. When the duplication is lost, sup-10(+) is absent and the rubberband phenotype is suppressed. Thus, duplication-bearing animals have the rubberband phenotype, while non-duplication-bearing animals have the forward and reverse kinker Unc phenotypes. Because the germ line descends from P1, mosaic animals that have lost the duplication at P₁ produce only progeny without the duplication. Because the mutation osm-1(p808) acts cellautonomously to prevent the uptake of FITC in 16 neurons that descend from AB.a and AB.p, the presence of the duplication in the AB lineage will render the animals FITC- staining. Thus, mosaics that have lost the duplication at P₁ [P₁ (−)] are non-Unc-7, FITC-staining and non-rubberband, and they should have only forward and reverse kinker Unc, non-rubberband, FITC-nonstaining self-progeny. We screened for such animals and found 15 P₁ (−) mosaics satisfying all of the above criteria from about 6200 duplicationbearing animals. All of these mosaics were wild type in locomotion. This result suggested that the wild-type pag-3 gene product is not required in muscle cells for coordinated movement.

We then addressed whether pag-3(+) is required in cells derived from AB to promote coordinated movement. We used two strains to generate AB(−) mosaics. The first strain has the genotype: daf-6(e1377) pag-3(ls20) sup-10(n983); mnDp14. sup-10(n983) is another recessive allele of sup-10, which confers the rubberband phenotype described above for unc-93 (Greenwald and Horvitz, 1986). daf-6 acts cell-nonautonomously in the four sheath cells, AMshL/R and PHshL/R, to assist FITC uptake by the amphid and phasmid neurons, respectively. Duplication-bearing selfprogeny are wild type for movement and FITC-staining. Selfprogeny without the duplication have the rubberband phenotype and are FITC-nonstaining due to the daf-6 mutation. We first screened for pag-3 reverse kinker Unc animals from the above strain and then analyzed their FITC- staining pattern to determine the point of duplication loss in the AB lineage. From about 25,900 duplication-bearing animals, 26 reverse kinker Unc mosaic animals were found. 22 of those segregated two types of progeny: FITC-staining animals of wild-type motility and rubberband, FITC-nonstaining animals. The inheritance of the staining phenotype showed that mnDp14 had been maintained in the mosaic parent’s germ line that is derived from P ₁. The remaining four pag-3 mosaics had only rubberband, FITC-nonstaining offspring, which indicated that the mosaic parents had lost mnDp14 in the germ line. The points of duplication loss of the pag-3 mosaics are summarized in Table 2A.

The second strain that we used to screen for AB(−) mosaics has the genotype: pag-3(ls20) sup-10(n983) osm-1(p808); mnDp14. Self-progeny with the duplication are wild type and FITC-staining. Self-progeny without the duplication have the rubberband phenotype and are FITC-nonstaining. Of about 15,700 duplication-bearing animals screened, 17 reverse kinker Unc mosaic animals were found (Table 2B).

Because all 43 pag-3 reverse kinker Unc mosaics obtained from both strains showed duplication losses at AB or AB.p, the focus of pag-3 action is probably distributed among descendants of AB.p. During the course of mosaic analysis, we noted that many mosaic animals had experienced more than one duplication loss. We also found a number of pag-3 reverse kinker Unc mosaic candidates that were later shown by progeny testing to not be true mosaics. Most of these individuals seemed to contain partially deleted duplications. Both phenomena have been observed in several other mosaic analysis experiments (Greenwald and Horvitz, 1986; Herman, 1984, 1987; Starich et al., 1993; Villeneuve and Meyer, 1990).

To identify the cells that are responsible for the misexpression phenotype conferred by mutations in pag-3, we constructed a strain of genotype pag-3(ls20) sup-10(n983) osm-1(p808); uIs9; mnDp14. The integrated mec-2gfp array uIs9 was used as a marker for scoring the misexpression phenotype. 14 pag-3 reverse kinker Unc mosaics were found from about 19,500 duplicationbearing animals. These mosaics were examined for mec-2gfp expression in their BDU neurons and then for their points of duplication loss by FITC uptake (Table 3). Because duplication loss in AB was sufficient to cause the misexpression phenotype while duplication loss in AB.p did not lead to the misexpression phenotype, the focus of pag-3 action is very likely among descendants of AB.a, from which the ALM and BDU neurons are derived.

The C. elegans pag-3 gene represses touch neuron-specific genes in the BDU neurons and is required for coordinated movement (Jia et al., 1996). We have cloned and sequenced pag-3 and investigated where pag-3(+) is required and when and where pag-3(+) is expressed. pag-3 encodes a C2H2-type zinc finger protein with zinc fingers that are highly homologous to those encoded by a mammalian proto-oncogene, gfi-1. Expression analyses showed that pag-3 transcription is regulated during development and that pag-3 negatively regulates its own expression. A pag-3lacZ fusion gene was expressed in the BDU neurons, the touch neurons, VA and VB motor neurons, the AVF neurons and unidentified neurons of the retrovesicular ganglion. Mosaic analysis showed that pag-3 functions in the AB.a lineage to promote the fate of the BDU neurons, and in the AB.p lineage to promote locomotion. Together these results show that pag-3 is expressed in and is required in multiple types of neurons.

Because PAG-3 is a zinc finger protein, it probably binds DNA and acts as a transcriptional regulator. At least four genes expressed in touch neurons (mec-2gfp, mec-4lacZ, mec-7 and mec-9lacZ) are negatively regulated by pag-3 in the BDU neurons (Jia et al., 1996; G. Xie and E. Aamodt, unpublished results), and pag-3 also negatively regulates its own expression. The allele ls64, which encodes a truncated protein without the zinc finger region, and the ls20 allele, which has a missense mutation at a zinc-binding histidine in the fourth zinc finger, have the genetic characteristics of null alleles. This suggests that the zinc finger region is essential for the function of PAG-3. The N-terminal proline-rich region of PAG-3 may function in transcriptional regulation (Ma and Ptashne, 1987; Mitchell and Tjian, 1989).

The sequence similarity of the zinc fingers encoded by pag-3 and gfi-1 is striking and suggests conservation of function. The amino acid sequences are 77% identical between zinc fingers 1-5 in PAG-3 and zinc fingers 2-6 in GFI-1, and 89% identical between the last four zinc fingers in these two proteins. The sequence similarity of these zinc fingers suggest that they may bind to a similar DNA sequence, but DNA binding seems insufficient to explain this level of sequence conservation. That is, certain residues of C2H2 zinc fingers have been shown through crystallography to contact DNA (Pavletich and Pabo, 1991, 1993; Pieler and Bellefroid, 1994) or to be important for DNA site selection (Pavletich and Pabo, 1991), but other residues face away from the DNA helix and do not seem to be critical for binding. Because the region of sequence conservation between these two proteins is continuous in the zinc finger domain, we postulate that the whole region is important to the functions of these proteins, and that they will have not only similar binding sites, but other functional similarities. For example, both PAG-3 and GFI-1 may interact with other proteins through their zinc finger domains. gfi-1 (Growth Factor Independence-1), was identified from a viral insertion that rendered interleukin-2 (IL-2)-dependent T-cell lymphoma lines IL-2-independent (Gilks et al., 1993). Zornig et al. (1996) found that gfi-1 cooperates with two other oncogenes, myc and pim-1, in lymphomagenesis. GFI-1, like PAG-3, may be a transcriptional repressor (Zweidler-McKay et al., 1996).

Expression of pag-3 coincides with the generation of touch and motor neurons, and may be negatively self-regulated. A pag-3lacZ fusion gene was expressed in the touch neurons and BDU neurons from the time of their generation to the L2 larval stage, and in the VA and VB motor neurons and the AVF interneurons from late L1 to L2 larval stages (Fig. 5). This expression pattern is consistent with the results from the developmental northern analysis (not shown), which showed that pag-3 mRNA was present from the embryonic to the L3 stage and peaked during the L1 and L2 stages. Northern analysis showed higher pag-3 mRNA message levels in mixed-stage populations of pag-3 mutants than in wild type (not shown), and we found that the pag-3lacZ transgene was expressed into adulthood more frequently in pag-3 mutant animals (Table 1). These results suggest that pag-3 may negatively regulate its own expression late in development. Although Pulak and Anderson have shown that the presence of in-frame stop codons decreases the stability of a message, particularly near the 5′ end (Pulak and Anderson, 1993), we observed a twofold increase in message levels of the pag-3(ls65) mRNA, which encodes a stop in the predicted N-terminal region. Thus, either the stop codon did not destabilize the pag-3(ls65) message or the loss of autoregulation of pag-3 more than compensated for this destabilization.

Genetic mosaic analysis suggested that pag-3(+) is required in the AB.a lineage to prevent the misexpression phenotype and in the motor neurons of the AB.p lineage to promote locomotion. The BDU neurons and the ALM touch neurons are descendants of AB.a. Mosaic animals that have lost the duplication at AB ectopically express mec-2gfp in the BDU neurons, whereas AB.p(−) mosaics do not. We can conclude from this study that the activity of the gene is required in the AB.a lineage, but we are unable to determine more precisely in what sublineage pag-3 acts.

Loss of pag-3(+) in the progenitors of the motor neurons rather than in the progenitors of the body muscle cells led to motility defects. The cord motor neurons descend from AB, and 74 of 75 of the cells are descendants of AB.p. Half of the VA and VB motor neurons descend from AB.pl, and the rest descend from AB.pr. All but one body muscle cell descend from P1. All P1(−) mosaics were wild type in locomotion and all pag-3 reverse kinker mosaic animals had the duplication loss in the AB or AB.p lineages. Because the loss of duplication at either AB.pl or AB.pr resulted in pag-3 Unc animals, pag-3(+) is required in both of these lineages. We did not find any pag-3 Unc mosaic animals that had lost the duplication solely in the AB.a lineage, so it is unlikely that pag-3(+) is required in AB.a to coordinate movement.

The VA motor neurons are required for backward movement whereas the VB motor neurons are required for forward locomotion. pag-3 mutant animals exhibit both backward and forward locomotion defects (Jia et al., 1996). Wild-type C. elegans move forward and backward with a smooth sinusoidal movement, but pag-3 mutant animals are lethargic and are unable to propagate a smooth sinusoidal movement when moving backwards. Instead, they make irregular bends. This phenotype is referred to as a reverse kinker uncoordinated phenotype. The forward movement of pag-3 mutants is also affected, in that they move forward slowly when prodded. These defects probably result from the loss of pag-3(+) activity in either the VA or VB motor neurons or in both.

In an earlier report, we postulated that PAG-3 helps to differentiate the ALM and BDU neurons by repressing the touch neuron genes in the BDU neurons (Jia et al., 1996). The data reported here on the pag-3 sequence and mosaic analysis supported this model, but we were surprised to find that pag-3lacZ is expressed in the BDU neurons and in the touch neurons at comparable levels. Why does pag-3 prevent touch neuron gene expression in the BDU neurons but not in the touch neurons? Clearly another factor must differentiate the ALM and BDU neurons. One strong candidate for such a factor is MEC-3. MEC-3 and UNC-86 are thought to act together to activate the touch neuron genes (Lichtsteiner and Tjian, 1995; Xue et al., 1993). unc-86 is expressed in both the BDU neurons and in the touch neurons, but mec-3 is apparently only expressed at observable levels in the touch neurons and in four other cells: the PVD neurons and the FLP neurons (Way and Chalfie, 1989). mec-3lacZ expression is not detected in the BDU neurons (Way and Chalfie, 1989); however, we found that expression of mec-7lacZ in the BDU neurons of pag-3 mutant animals requires mec-3 (Jia et al., 1996). Perhaps, then, mec-3 is expressed at a high level in the touch neurons but at a very low level in the BDU neurons, and PAG-3 and MEC-3 act competitively by binding to the same or overlapping sites. MEC-3 predominates in the touch neurons to activate touch neuron genes, and PAG-3 represses touch neuron genes in the BDU neurons where MEC-3 levels are low. We tested this prediction with pag-3(ls20) and two hypomorphic mec-3 alleles, u298 and u312 (data not shown). In both cases, the presence of a mutated pag-3 gene did not affect the frequency at which mec-7lacZ was expressed in the touch neurons. These experiments assume that the pag-3(ls20) product has lost DNA-binding activity and that the MEC-3 protein in these strains can bind its targets. The nature of the mec-3 lesions is not known, however, so this experiment does not entirely rule out the possibility of a PAG-3:MEC-3 competition.

The role of pag-3, if any, in determination of cellular fate in asymmetric cell divisions remains to be elucidated. The ALM touch neurons and the BDU interneurons are lineal sister cells (Sulston et al., 1983) that have distinct morphologies and perform different functions: the two ALM cells are touch receptor neurons, while the two BDU neurons are interneurons of unknown function (Chalfie and Au, 1989; Chalfie and Sulston, 1981; White et al., 1986). Likewise, the VA and VB motor neurons are sister cells that have distinct morphologies and functions (Chalfie and White, 1988; White et al., 1986). The AVF interneurons derive from the W and P1a neuroblasts in lineages parallel to those of the VA motor neurons. As a putative DNA-binding protein, PAG-3 may contribute to fate determination as a transcriptional repressor since loss-offunction mutations in pag-3 caused ectopic expression of touch neuron genes in the BDU neurons, and since PAG-3 represses its own expression in the ALM and BDU neurons late in development (Table 1). PAG-3 may directly bind its own control regions and those of several touch neuron genes to prevent transcription of these genes. Alternatively, pag-3 may affect the transcription of touch neuron genes indirectly by activating downstream activators and repressors. In either case, PAG-3 must act together with another protein(s) not expressed in the touch neurons to repress touch neuron gene expression in the BDU neurons, or another protein in the touch receptor cells may inactivate PAG-3 and allow expression of the touch receptor program. The identification of potential binding sites for PAG-3 in genes expressed either in the touch neuron or the motor neuron lineages may clarify the mechanism of PAG-3 activity.

We thank M. Chalfie for the mec-2gfp array uIs9; J. Shaw for strains SP1490, SP1491 and SP1528; M. Krause for plasmid pT7-T3-18-103, R. Holmgren for the anti-β-galactosidase antiserum, and A. Coulson for cosmids and YACs. We are grateful to X. Alvarez-Hernandez and J. Glass for the use of their microscope, D. Gross for the CHEF gel apparatus, P. Good and L. Robinson for supplies and discussions, S. Aamodt for the staged RNAs and for helpful discussions and comments on this manuscript, T. Harrington and K. Simpson for technical assistance, and B. Norman for oligonucleotides. Some nematode strains were provided by the Caenorhabditis Genetics Center, which is funded by the National Institutes of Health National Center for Research Resources (NCRR). This study was supported by grants to E.A. from the LSU Medical Center-Shreveport, the Biomedical Research Foundation of Northwest Louisiana and the LSUMC Center for Excellence in Cancer Research.