During Caenorhabditis elegans development, the patterns of cell divisions, cell fates and programmed cell deaths are reproducible from animal to animal. In a search for mutants with abnormal patterns of programmed cell deaths in the ventral nerve cord, we identified mutations in the gene pag-3, which encodes a zinc-finger transcription factor similar to the mammalian Gfi-1 and Drosophila Senseless proteins. In pag-3 mutants, specific neuroblasts express the pattern of divisions normally associated with their mother cells, producing with each reiteration an abnormal anterior daughter neuroblast and an extra posterior daughter cell that either terminally differentiates or undergoes programmed cell death, which accounts for the extra cell corpses seen in pag-3 mutants. In addition, some neurons do not adopt their normal fates in pag-3 mutants. The phenotype of pag-3 mutants and the expression pattern of the PAG-3 protein suggest that in some lineages pag-3 couples the determination of neuroblast cell fate to subsequent neuronal differentiation. We propose that pag-3 counterparts in other organisms determine blast cell identity and for this reason may lead to cell lineage defects and cell proliferation when mutated.
The known and essentially invariant cell lineage of the nematode Caenorhabditis elegans has provided a basis for analyzing at single-cell resolution the mechanisms that establish cell lineages and cell fates during metazoan development (Sternberg and Horvitz, 1984). Programmed cell death is a major aspect of animal development, and abnormalities in programmed cell death contribute to a variety of human diseases (reviewed by Lockshin et al., 1998). Genetic and molecular analyses of C. elegans have defined a pathway for programmed cell death (Metzstein et al., 1998). This pathway is extensively conserved between animals (Horvitz, 1999; Ranger et al., 2001). How the process of programmed cell death is controlled during development is less well understood. Two genes, ces-1 and ces-2, regulate programmed cell death in C. elegans and have mammalian homologs with conserved functions (Ellis and Horvitz, 1991; Inaba et al., 1996; Inukai et al., 1999; Metzstein et al., 1996; Metzstein and Horvitz, 1999). Mutations that affect the mammalian homolog of ces-2 can cause leukemia in children, perhaps in part by rendering malignant lymphoblasts resistant to programmed cell death (Inaba et al., 1996).
To define additional genes that control the programmed deaths of specific cells in C. elegans we have sought mutations that alter the pattern of cell deaths in the ventral nerve cord. A newly hatched C. elegans larva contains 22 motoneurons in its ventral cord and associated ganglia (Sulston and Horvitz, 1977; White et al., 1986). During postembryonic development, 12 ventrolateral P blast cells migrate into the ventral cord and undergo a series of divisions. The divisions of the P cells, together with those of a 13th blast cell, W, produce 65 neural cells, 10 of which undergo programmed cell death (Sulston and Horvitz, 1977) (Fig. 1). Seven of the 10 cell deaths are of cells at equivalent positions in the W and P cell lineages. Specifically, the six Pn.aap (Pn.aap, the posterior daughter of the anterior daughter of the anterior daughter of any P cell) cells derived from the P lineages in the anterior and posterior ventral cord (P1, P2 and P9-P12) and W.ap, a cell lineally equivalent to the Pn.aap cells, undergo programmed cell death; the six Pn.aap cells in the midbody (from P3-P8) survive and differentiate to become VC motoneurons.
In our search for mutants abnormal in ventral cord cell deaths, we have identified alleles of the gene pag-3. We describe the effects of mutations in pag-3 on programmed cell deaths in the ventral nerve cord and the role of pag-3 in the determination of neuroblast cell fate and in neuronal differentiation.
MATERIALS AND METHODS
Isolation of pag-3(n3098)
ced-1 hermaphrodites were mutagenized with ethylmethane sulfonate (Brenner, 1974). A synchronized population of F2 progeny was examined as late L2/early L3 larvae using Nomarski microscopy (Sulston and Horvitz, 1977). Animals with abnormal numbers or patterns of cell corpses in the ventral cord were recovered from the slide for analysis. A single allele of pag-3, n3098, was isolated after screening approximately 9000 animals, or 4500 mutagenized haploid genomes.
Strains, alleles and cell lineage analysis
All strains were maintained at 20°C as described by Brenner (Brenner, 1974). Cell nomenclature and cell lineage analysis were as described by Sulston and Horvitz (Sulston and Horvitz, 1977), except that ‘P0.a’ is now known as W (Sulston et al., 1983). Unless otherwise indicated, mutations used are cited by Riddle et al. (Riddle et al., 1997). Mutations used were:
LGII, wdIs4 (the integrated Punc-4gfp plasmid pNE-1, generously provided by David Miller);
LGIII, wdIs1 (the integrated Punc-4lacZ plasmid pNC4-22Lz) (Miller et al., 1992), lin-39(n1760), mab-5(e1239) and egl-5(n945);
LGV, oyIs14 (Hao et al., 2001) kindly provided by T. Sarafi-Reinach and P. Sengupta;
Construction and analysis of pag-3/Df animals
ced-1(e1735) males were mated with mnDp1/+; mnDf19 hermaphrodites, and non-Unc (mnDp1-carrying) cross progeny males were mated with ced-1(e1735); dpy-17(e164) unc-32(e189); pag-3 hermaphrodites. Non-Dpy non-Unc-32 Pag-3 hermaphrodites were picked as late L2 Unc larvae, and the patterns of cell corpses in the ventral cord of Ced animals examined using Nomarski optics. Similar results were obtained with a second deficiency, mnDf43, which also deletes the pag-3 locus (Meneely and Herman, 1979). To construct pag-3/pag-3/+ animals, crosses were performed as above, except non-Dpy non-Unc cross progeny males were mated with ced-1; dpy-17 unc-32; pag-3 hermaphrodites, and non-Dpy non-Unc Ced animals were scored.
Antibody preparation and staining of fixed animals
A DNA fragment encoding the Zn-finger domains of PAG-3 (amino acids 124-333, PAG-3ZC) was amplified from the pag-3 cDNA (Jia et al., 1997) and cloned into pET21b(+) (Novagen) to construct a HIS-tagged fusion protein. The expressed protein was partially purified using the method described by the manufacturer.
PAG-3 antibodies were purified from rabbit serum as described (Koelle and Horvitz, 1996). The specificity of the antiserum was demonstrated by recognition on western blots of an approximately 33 kDa band in E. coli expressing PAG-3ZC and a 47 kDa band in extracts of wild-type C. elegans. Fixed pag-3(n3098) animals did not stain with the antiserum, but dpy-17(e164) mutants fixed in the same tube did stain. For whole-mount staining, animals were fixed with Bouin’s fixative as described (Nonet et al., 1997). Primary antiserum was diluted 1:50 for staining and detected with FITC- or Texas Red-conjugated goat anti-rabbit secondary antibody. For experiments with PAG-3 antiserum and the Punc-4lacZ construct wdIs1 co-staining, β-galactosidase was detected in fixed animals using a 1:500 dilution of a α-lacZ monoclonal antibody (Promega). FMRFamide staining was done as described (Li and Chalfie, 1990), using collagenase for permeabilization. Antiserum against FMRFamide was kindly provided by Chris Li.
Mutations in pag-3 affect the pattern of cell deaths in the ventral cord
To visualize the pattern of programmed cell deaths in the C. elegans ventral cord, we used a mutation in the gene ced-1, which encodes a receptor that allows engulfing cells to recognize dying cells (Zhou et al., 2001). In ced-1 mutants cell corpses persist rather than being engulfed and degraded and are easily detected using Nomarski differential interference contrast microscopy (Hedgecock et al., 1983). We mutagenized ced-1 hermaphrodites with ethylmethane sulfonate (Brenner, 1974) and examined the patterns of cell corpses in the ventral nerve cords of F2 progeny.
In our screen, we identified a mutation, n3098, that recessively conferred an increased number of cell corpses in the anterior ventral cords of many animals (Fig. 2). In the posterior ventral cord, some individual n3098 mutants had an increased number of cell corpses, while others had fewer than the normal number of corpses. Not all postembryonic cell deaths were affected by n3098, as the number of corpses generated by the Q neuroblasts (Sulston and Horvitz, 1977) was normal (1.6±0.5 versus 1.7±0.5 and 1.3±0.1 versus 1.4±0.1 for ced-1 versus ced-1; n3098 mutants for QL and QR, respectively).
The additional corpses seen in the ventral cords of n3098 mutants resulted from programmed cell deaths, as mutations in the ced-3 gene, which encodes a caspase required for programmed cell death in C. elegans (Yuan et al., 1993), prevented their appearance (data not shown). n3098 mutants had sluggish and uncoordinated movement. ced-3; n3098 mutants (but not ced-3 mutants) were also sluggish and uncoordinated, indicating that the movement defect was not caused by the deaths of ventral cord motoneurons required for normal movement. The n3098 mutation mapped to the right arm of LG X. The movement defect of n3098 mutants and the map location of n3098 suggested it might be an allele of the gene pag-3, originally identified in mutants that overexpressed a Pmec-7lacZ transgene in the BDU neurons (Jia et al., 1996). pag-3 is predicted to encode a 336 amino acid C2H2 zinc-finger protein with greater than 80% amino acid identity in its five Zn-finger domains to the mammalian Gfi-1 and Gfi-1B proteins (Jia et al., 1997; Tong et al., 1998). Genetic mosaic analysis suggests that pag-3 is required in the nervous system for normal movement (Jia et al., 1996; Jia et al., 1997).
We found that n3098 mapped to the region of pag-3 and failed to complement the movement defect of pag-3(ls20) mutants. In addition, in a ced-1 background, pag-3(ls20) mutants, like n3098 mutants, had an abnormal pattern of cell corpses in the ventral cord (Fig. 2). ced-1; pag-3(ls20) mutants had more corpses in the ventral cord than did ced-1; pag-3(n3098) mutants, suggesting that at least one of these alleles is not null. Analysis of the phenotype of mutants heterozygous for the pag-3(n3098) and pag-3(ls20) alleles in trans to chromosomal deficiencies that delete the pag-3 locus suggested that n3098 is a strong loss-of-function or null allele (Fig. 2). We determined the sequence of pag-3 genomic DNA from pag-3(n3098) mutants and identified a G-to-A mutation changing the codon for tryptophan at amino acid 113 to an opal stop codon, which is located before the sequences encoding the Zn-finger domains of PAG-3. pag-3(ls20) homozygotes had more corpses than did pag-3(ls20)/Df mutants, suggesting that pag-3(ls20), while generally acting as a strong loss-of-function allele, may cause abnormalities in addition to those resulting from a complete loss of pag-3 function (e.g. by a chromosomal deficiency that spans the pag-3 locus). pag-3(ls20)/pag-3(+) and pag-3(ls20)/pag-3(ls20)/pag-3(+) mutants (the latter wild-type allele was provided on an attached chromosomal duplication – see Materials and Methods) had a wild-type pattern of cell corpses. The ls20 allele is an H228Y mis-sense mutation affecting the fourth Zn finger (Jia et al., 1997). The corresponding Zn finger in Gfi-1 is essential for DNA binding (Zweidler-Mckay et al., 1996), consistent with our hypothesis that the PAG-3(ls20) protein interacts abnormally with DNA or other proteins and interferes with the function of a multiprotein complex in a way that is recessive to a single wild-type allele of pag-3. Taken together, our data indicate that mutations in pag-3 can affect the pattern of cell corpses in the ventral nerve cord of C. elegans and that pag-3(n3098) is a strong loss-of-function or null allele.
Differentiation of the BDU interneurons in pag-3(n3098) mutants is probably abnormal
Previously described mutations in pag-3 lead to misexpression of touch neuron genes in the BDU interneurons (Jia et al., 1996). We observed that pag-3(n3098) mutants similarly expressed an integrated Pmec-7lacZ reporter in the BDU interneurons, which do not normally express mec-7 (Jia et al., 1996), and we quantified this effect. In 50 wild-type animals examined carrying the Pmec-7lacZ reporter, 0/100 BDU interneurons expressed β-galactosidase, whereas 92/100 ALM mechanosensory neurons did so. By contrast, in 50 pag-3(n3098) mutants 39/100 BDU and 83/100 ALM neurons expressed β-galactosidase, similar to the previously reported phenotypes of mutants with other mutations in pag-3 (Jia et al., 1996) (data not shown). The ALM and BDU neurons of wild-type animals and pag-3(n3098) mutants were easily identified by their axonal morphologies and nuclear positions, and no animal had more than two ALM or BDU neurons. These data suggest that the misexpression of mechanosensory genes in the BDU interneurons results from a defect in neuronal differentiation, as previously proposed (Jia et al., 1996).
pag-3 is required for cell-fate determination by the Pn.aa neuroblasts
To determine the cause of the abnormalities in the patterns of cell deaths in pag-3 mutants, we directly observed the postembryonic cell lineages in the ventral cords of developing pag-3(n3098) and pag-3(ls20) larvae. As shown in Fig. 3, P-cell lineages in pag-3 mutants had a reiterative cell lineage defect. Specifically, in the P-cell lineages of wild-type animals, the neuroblast Pn.aa divides to produce an anterior daughter Pn.aaa that is itself a neuroblast and a posterior daughter Pn.aap that terminally differentiates or undergoes programmed cell death. Each Pn.aaa neuroblast then divides to produce two terminally differentiated neurons. In pag-3 mutants the Pn.aaa neuroblasts, rather than producing two terminally differentiated neurons, instead often produced an anterior daughter that was a neuroblast and a posterior daughter that underwent programmed cell death or terminally differentiated with a morphology typical for a neuron.
For example, seven of the nine anterior-most daughter cells (the W.aaa and Pn.aaaa cells) we observed in the W, P1 and P2 lineages of three pag-3(n3098) mutants underwent one or two extra rounds of division (Fig. 3B). The anterior-most daughter cells ultimately exited the cell cycle and adopted a nuclear morphology typical of a neuron. The Pn.aaap cells, the posterior daughters resulting from the Pn.aaa divisions, in the W, P1 and P2 lineages of wild-type animals survive to become motoneurons, but in pag-3(n3098) mutants instead often underwent programmed cell death, as do the Pn.aap cells in the wild type (Fig. 3B).
We also observed in the midbody and posterior ventral cord of pag-3(n3098) mutants extra cell divisions by Pn.aaaa cells and the programmed cell deaths of the P9- and P10.aaap cells, which survive in wild-type animals (Fig. 3B). Compared with the anterior ventral cord cell lineages, extra divisions by Pn.aaaa cells were rare in the more posterior lineages of pag-3(n3098) mutants. We observed only a single extra division in the P9-12 lineages of six pag-3(n3098) mutants (Fig. 3B). However, in three pag-3(ls20) mutants, five of 12 P9-12.aaaa cells underwent one extra round of division (Fig. 3C). The more frequent extra divisions by Pn.aaaa neuroblasts in pag-3(ls20) mutants and the increased number of posterior daughter cells generated by these divisions (some of which undergo programmed cell death) can account at least in part for the greater number of cell corpses in the ventral cords of pag-3(ls20) compared with pag-3(n3098) mutants. The greater likelihood of extra divisions by the Pn.aaaa neuroblasts in the anterior compared with the posterior ventral cord suggests that positional information influences the pag-3 mutant phenotype.
The patterns of cell divisions and programmed cell deaths we observed are most easily explained by proposing that in pag-3 mutants, the Pn.aaa neuroblasts adopt the fate of their mothers, the Pn.aa neuroblasts (Fig. 3D). The predicted consequences of this reiterative lineage defect are that the neurons normally generated by the Pn.aaa neuroblasts would be abnormal or absent and that extra Pn.aap-like cells would be produced.
The VA and VB motoneurons are abnormal or absent in pag-3 mutants
The normal descendants (Pn.aaaa and Pn.aaap) of the Pn.aaa neuroblasts are the VA and VB motoneurons (Sulston and Horvitz, 1977). pag-3 mutants are defective in both forwards and backwards movement, suggesting that the VA and VB motoneurons, which control backwards and forwards movement, respectively (Chalfie et al., 1985), are defective. We used reporter constructs expressed in the VA and VB motoneurons to test whether these neurons were present in pag-3 mutants. At hatching and in young L1 larvae, the ventral nerve cord and associated ganglia contain nine DA, seven DB and six DD juvenile motoneurons (White et al., 1986). The postembryonic P cell lineages, which commence during the late L1 stage, add 12 VA, 11 VB and 13 VD motoneurons to the juvenile neurons, as well as additional cells (Sulston and Horvitz, 1977). We used an integrated Punc-4gfp transgene (generously provided by David Miller), which is expressed in the DA and VA motoneurons (Pflugrad et al., 1997), to determine the number of VA motoneurons in pag-3(n3098) mutants. We counted the number of fluorescent nuclei in L3 larvae carrying the Punc-4gfp transgene. From this number, which represented expression in the DA and VA motoneurons, we subtracted the number of fluorescent nuclei in L1 larvae, which represented expression in the juvenile DA motoneurons. The number of VA motoneurons in Punc-4gfp; pag-3(n3098) mutants was significantly reduced compared with that in otherwise wild-type animals carrying the Punc-4gfp transgene (Table 1). As noted above, the anterior-most daughters generated by the P2-P12 cell lineages of wild-type animals adopt a VA motoneuron fate. Our observations suggest that as assessed by expression of unc-4, a characteristic of normal VA motoneurons (Miller and Niemeyer, 1995), the corresponding cells of pag-3 mutants differentiate abnormally.
More strikingly, the VB motoneurons were essentially entirely absent from the ventral cord of pag-3(n3098) mutants, as determined by expression of the Pacr-5gfp reporter, which is expressed in the DB and VB motoneurons (Winnier et al., 1999). By contrast, the number of VD neurons in pag-3(n3098) mutants was normal, as determined by staining for the neurotransmitter GABA, which is expressed by the DD and VD motoneurons (McIntire et al., 1993). This observation suggests that the development of the VD motoneurons, which are generated as the posterior daughters of the Pn.ap neuroblasts, was not affected by loss of pag-3 function. This conclusion was consistent with our observation that the cell divisions of the Pn.ap neuroblasts were normal in pag-3 mutants. The defects in P cell lineages and the abnormalities in cell fates in pag-3 mutants are summarized in Fig. 3E.
The absence of VB motoneurons and the abnormality in VA motoneurons of pag-3 mutants accounts for the uncoordinated movement of these animals.
pag-3 mutants have extra VC-like motoneurons
In the P3-P8 lineages of the midbody, the Pn.aap cells, the normal posterior daughters of the Pn.aa neuroblasts, become VC motoneurons, while in the W, P1, P2 and P9-12 lineages of the anterior and posterior the Pn.aap (and W.ap) cells undergo programmed cell death (Sulston and Horvitz, 1977) (Fig. 1). The cell lineage reiteration defect of pag-3 mutants completely accounts for the extra programmed cell deaths in the anterior and posterior ventral cord (extra Pn.aap-like daughter cells) and predicts that pag-3 mutants should also contain extra VC-like motoneurons along the midbody.
To assess the number of VC motoneurons we used a Plin-11gfp construct. lin-11 encodes a LIM homeodomain protein (Freyd et al., 1990) required for vulval morphogenesis (Ferguson et al., 1987) and is expressed in the VC motoneurons, vulval cells, gonadal and uterine cells, and several head neurons (Newman et al., 1999; Freyd, 1991) (this work). We generated an integrated reporter construct containing a portion of the lin-11 promoter fused to gfp (Reddien et al., 2001) and used it to detect VC and VC-like cells in pag-3 mutants. pag-3 mutants had about three times as many Plin-11gfp-expressing neurons as did wild-type animals (Fig. 4). To test whether the Pn.aap-like cells (i.e. Pn.aaap, Pn.aaaap, etc.) in pag-3 mutants showed other characteristics of VC motoneurons, we stained pag-3 mutants with antiserum directed against the FMRF-amide neuropeptide expressed by the VC motoneurons (Schinkmann and Li, 1992) and observed many extra FMRF-amide-positive neurons in the ventral cords of pag-3 mutants (Fig. 4).
Most of the extra VC-like neurons were in the midbody, as visualized by expression of the Plin-11gfp construct (Fig. 4). However, pag-3 mutants also had some fluorescent nuclei in the anterior and posterior cord, where the Pn.aap cells in wild-type animals normally undergo programmed cell death. As seen in the cell lineages diagrammed in Fig. 3, not all Pn.aap and ‘Pn.aap-like’ cells generated by the anterior and posterior P lineages died in pag-3 mutants. For example, eight of 24 P9-12.aap survived in the six pag-3(n3098) mutants we observed. Some of the surviving Pn.aap cells may differentiate to become VC-like cells and express Plin-11gfp [as has been directly confirmed in ced-3 mutants or mutants defective in engulfment of cell corpses (Reddien et al., 2001)] accounting for the fluorescent nuclei in these regions of the ventral cords of pag-3 mutants.
We conclude that the number of corpses in the ventral cords of pag-3 mutants was determined by two factors: generation by reiterated cell lineages of extra posterior daughter cells (Pn.aap-like cells) that can undergo programmed cell death and a failure of some of these cells to die in pag-3 mutants.
The HOM-C genes do not encode the positional information that influences the pag-3 phenotype
As noted above, the likelihood that a W.aaa or Pn.aaaa cell would divide and thereby generate extra posterior daughter cells (which could undergo programmed cell death) varied with position along the anteroposterior body axis and differed between the two pag-3 alleles. We tested the hypothesis that genes of the C. elegans homeotic gene cluster either promote extra divisions in the anterior ventral cord or prevent extra divisions in the posterior ventral cord. These genes – ceh-13, lin-39, mab-5 and egl-5 – specify spatial patterning along the anteroposterior body axis (Brunschwig et al., 1999; Burglin and Ruvkun, 1993; Chisholm, 1991; Clark et al., 1993; Kenyon, 1986; Salser et al., 1993). We determined the numbers of cell corpses in the ventral cords of ced-1; pag-3(n3098) mutants defective in lin-39, mab-5, or egl-5 [ceh-13 mutants die during embryogenesis; (Brunschwig et al., 1999)]. We found that loss of the spatial information controlled by lin-39, mab-5 and egl-5 did not affect the likelihood of extra divisions (Table 2). How positional information influences the pag-3 phenotype remains unknown.
PAG-3 is expressed in the Pn.aa neuroblasts and their descendants
To characterize the PAG-3 expression pattern, we raised a rabbit polyclonal antiserum that specifically recognized PAG-3 and used it to stain wild-type animals of various stages (see Materials and Methods). Previous studies have indicated that a transgene reporter construct with β-galactosidase under the control of pag-3 promoter sequences is expressed in the BDU neurons, the mechanosensory neurons, the two AVF interneurons, and the VA and VB motoneurons in the ventral cord (Jia et al., 1997). We found that PAG-3 protein was expressed more widely in the nervous system than had been observed using the Ppag-3lacZ reporter. We detected PAG-3 during embryonic development in many nuclei ∼280 minutes after fertilization (data not shown). At this developmental stage, the cells derived from the AB cell lineages, which generate most of the neuronal cells, are beginning to terminally differentiate (Sulston et al., 1983). In late L2 larvae anti-PAG-3 staining was seen in approximately two dozen cells in the head, all six mechanosensory neurons, the BDU neurons, approximately ten cells in the tail as well as in the ventral cord. PAG-3 staining of many cells in the head and tail remained detectable in adult animals (data not shown).
In the ventral cord, PAG-3 was first detected in the Pn.aa neuroblasts (Fig. 5), the daughters of which are abnormal in pag-3 mutants. PAG-3 was not detected in the Pn.ap neuroblasts or their descendants, consistent with our observation that these cells appear to be unaffected by loss of pag-3 function. PAG-3 was present equally in each of the descendant cells of Pn.aa after subsequent rounds of division (i.e. the Pn.aaa, Pn.aap, Pn.aaaa and Pn.aaap cells), including the three differentiating neurons generated by each Pn.aa neuroblast. In most cells, PAG-3 protein became undetectable shortly after the cells had been generated in the L1, but PAG-3 was present in six cells in the ventral cords of adults (see below). Pn.aap undergoes programmed cell death in the W, P1, P2 and P9-12 lineages (Fig. 1). PAG-3 protein was observed in the P2.aap and P9-P12.aap cells that died, as well as in the homologous P3-P8.aap cells that survived (Fig. 5 and data not shown), indicating that the simple presence or absence of PAG-3 as these cells are being generated does not determine the life versus death decision. The precise expression pattern of PAG-3 in the W and P1 descendants was difficult to define unambiguously, as there are many nuclei in the retrovesicular ganglion at the anterior end of the ventral cord (in which W and P1.a divide) and as two embryonic cells in this region expressed PAG-3. Taken together these data suggest that in the ventral cord PAG-3 protein is first produced precisely in those neuroblasts abnormal in pag-3 mutants, persists through the divisions of the Pn.aaa cells, and is present at least transiently in the three differentiating cells generated by each Pn.aa neuroblast.
PAG-3 expression persisted throughout the life of the animal in four cells in the retrovesicular ganglion at the anterior end of the ventral cord and in two cells in the posterior ventral cord. In newly hatched L1-stage larvae, before the initiation of the postembryonic W and P cell lineages, two cells in the retrovesicular ganglion expressed PAG-3. Based on position, these cells were most likely the RIG interneurons (Fig. 6 and data not shown). After completion of the W and P cell lineages, two additional cells in the retrovesicular ganglion and two cells in the posterior ventral cord contained detectable PAG-3 protein. Based on the locations of the nuclei and their appearance only after completion of the W and P cell lineages, we thought these nuclei might be the two AVF and the VA11 and VA12 neurons, respectively. We confirmed this hypothesis by staining animals carrying an integrated Punc-4lacZ reporter, which is expressed in the AVF and VA as well as other neurons (Miller and Niemeyer, 1995), with PAG-3 antiserum and monoclonal antibody against β-galactosidase (Fig. 6). The four postembryonically generated neurons that expressed PAG-3 in adults – AVFR/L, VA11 and VA12 – are lineally related as the anterior-most descendants of W, P1, P11 and P12, respectively.
pag-3 probably affects differentiation of the PVQ neurons
pag-3 was initially identified in mutants that misexpressed mechanosensory genes in the BDU interneurons and was proposed to affect specification or differentiation of the BDU cell fate (Jia et al., 1996). We found that pag-3 similarly affects the PVQ interneurons. The PVQs are a pair of interneurons (PVQL and PVQR) located in the tail lumbar ganglia and can be visualized using an Psra-6gfp reporter (Troemel et al., 1995). In 30 adult wild-type animals, a total of 60 fluorescent nuclei were observed, whereas pag-3(n3098) mutants failed to express the reporter (one fluorescent nucleus was observed in 30 adult animals). In those rare pag-3 mutants in which the PVQ neurons expressed Psra-6gfp, the axons appeared normal (data not shown). Expression of Psra-6gfp in the ASH and ASI neurons, which also express this reporter (Troemel et al., 1995), was normal in pag-3 mutants, suggesting that the PVQ neurons were missing or failed to express the Psra-6gfp reporter. Using Nomarski optics, we examined the tail region of wild-type animals carrying the Psra-6gfp reporter and in blinded assays identified the PVQ neurons, which have slightly larger nuclei than adjacent neurons and occupy characteristic positions at the anterior margins of the lumbar ganglia adjacent to the rectum (White et al., 1986). In similar assays of Psra-6gfp; pag-3(n3098) mutants, we identified nuclei with the positions and morphology expected of PVQ neurons, but, unlike in wild-type animals, these nuclei did not express the reporter. The abnormalities in the PVQ neurons could reflect a defect in the cell lineages generating these neurons. The PVQ neurons are generated during embryogenesis, making direct analysis of cell lineages difficult. The sister cells of the PVQR and PVQL neurons undergo programmed cell death (Sulston et al., 1983). A defect in the cell lineages that generate the PVQ neurons might generate an abnormal number of cell corpses, as we have observed in the ventral nerve cord lineages of pag-3 mutants. We compared the number of cell corpses in the lumbar ganglia of ced-1; pag-3(n3098) and ced-1 mutants, and found them to be similar (0.7±0.1 and 0.6±0.1, respectively, n=70 animals). PAG-3 was expressed in the PVQ neurons throughout the life of wild-type animals, as demonstrated by co-staining with PAG-3 antiserum of animals carrying the Psra-6gfp reporter (data not shown). These data suggest that pag-3 is required for differentiation of the PVQ interneurons.
pag-3 determines neuroblast fate
Prior studies of the pag-3 gene established that loss of pag-3 function results in abnormal gene expression in the BDU interneurons and a behavioral defect in locomotion (Jia et al., 1996). Mosaic analysis suggests that pag-3 is required within the nervous system to prevent the locomotory defects, although how pag-3 acts has remained unclear (Jia et al., 1997). Through an analysis of cell lineages in the ventral nerve cord and of PAG-3 protein expression in individual cells of the developing nervous system, we have established that pag-3 determines the fate of specific anterior neuroblasts in the ventral cord of C. elegans. Our data suggest that pag-3 may also act in neuronal differentiation, as discussed below.
We observed that in pag-3 mutants the Pn.aaa neuroblasts of the ventral nerve cord reiterated the fate of their Pn.aa mothers. The posterior daughter cells produced by the reiterated divisions (e.g. Pn.aap cells) generally adopted fates appropriate for the corresponding wild-type lineages. However, some of the Pn.aap cells in pag-3 mutants, none of which divided, were variably abnormal. Some Pn.aap and Pn.aap-like cells derived from P cell lineages in the anterior and posterior ventral cord, all of which die in wild-type lineages, instead survived and differentiated. These observations suggest that both daughter cells of the Pn.aa neuroblasts were abnormal: the Pn.aaa cells failed to determine their proper fate as neuroblasts and the Pn.aap cells did not always undergo programmed cell death. These abnormalities presumably reflect a requirement for pag-3 in the Pn.aa mother cells to establish the Pn.aaa and Pn.aap cell fates prior to cell division, separate functions for pag-3 in the two daughter cells after this division has been completed, or both. The expression pattern of PAG-3 protein was consistent with all three of these possibilities, as protein was present in the parental neuroblast Pn.aa and in both daughter cells.
pag-3 may also affect neuronal differentiation
The PVQ and BDU neurons in pag-3 mutants are abnormal, despite being generated by apparently normal cell lineages, suggesting that these neurons fail to differentiate properly in pag-3 mutants. The VA motoneurons also differentiate abnormally in pag-3 mutants, as demonstrated by the defective backwards movement of these animals and by abnormal expression of an unc-4 reporter gene. A role for pag-3 in differentiating and differentiated neurons is supported by the expression pattern of the protein. In the Q neuroblast lineages PAG-3 protein was detected only after the generation of the terminally differentiating AVM and PVM mechanosensory neurons (data not shown). In many neurons, including the PVQ, BDU and touch neurons, PAG-3 protein was present throughout the life of the animal. Some of these cells are abnormal in pag-3 mutants (the BDU and PVQ neurons), while others have no obvious defect [the touch neurons (Jia et al., 1996)]. These data suggest that pag-3 functions in diverse contexts within the developing nervous system, with an essential role in the development of some neuroblasts and neurons and a currently unapparent role in other neurons.
PAG-3 does not appear to be associated with any obvious single characteristic of differentiated neurons. In the ventral nerve cord, PAG-3 was expressed in cells that live and in cells that undergo programmed cell death; in cells that express lin-11 and in cells that do not; and transiently in the VA2-10 motoneurons but continuously in VA11 and VA12. PAG-3 is required for cell-fate determination by the Pn.aa neuroblasts but not by the Pn.ap neuroblasts. PAG-3 was expressed in adult animals in neurons of several different types, including motoneurons (VA11, VA12), sensory neurons (ALM, PLM, AVM and PVM) and interneurons (BDU, RIG and AVF). Given its extensive similarity to mammalian Gfi-1 transcription factors, PAG-3 most probably affects cell fates by regulating transcription. Rather than affecting expression of a single common set of genes in all neuronal lineages, our finding that pag-3 is expressed in many neuronal subtypes at different points in neuronal development suggests that pag-3 cooperates with other factors to regulate the expression of cell type- and developmental stage-specific sets of genes to generate the complex pattern of neuronal subtypes seen in C. elegans.
pag-3 expression was specifically activated in the Pn.aa neuroblasts, the descendants of which were abnormal in pag-3 mutants. By contrast, expression was not activated to a detectable level in the sisters of these cells, the Pn.ap neuroblasts, the descendants of which appeared normal in pag-3 mutants. These observations suggest that pag-3 acts specifically in the Pn.aa neuroblasts and their descendants. PAG-3 protein was present at the time of generation of neurons descended from each Pn.aa cell. It is shortly after the generation of the terminal cells in these lineages that these neurons begin to adopt identifying characteristics, such as class-specific patterns of axonal projections or the morphological characteristics of a dying cell (Sulston, 1976; Sulston and Horvitz, 1977; White et al., 1992; Knobel et al., 1999). PAG-3 thus may well be expressed in response to Pn.aa lineage-specific signals to determine blast cell fates and then act later to induce distinctive differentiated features characteristic of the neurons produced by those lineages.
The function of PAG-3 may be similar to that of UNC-86, a POU-homeodomain protein that couples cell lineage cues to aspects of terminal differentiation (Finney and Ruvkun, 1990; Finney et al., 1988). Like mutations in pag-3, mutations in unc-86 result in the reiteration of some neuroblast lineages (Chalfie et al., 1981). unc-86 is the only other C. elegans gene known to be able to mutate to cause reiterative cell lineage defects that specifically affect development of the nervous system. In addition to being required for neuroblast determination, unc-86 also specifies characteristics of the mechanosensory neurons generated by those neuroblasts (Chalfie et al., 1981; Duggan et al., 1998; Finney and Ruvkun, 1990; Sze et al., 1997). pag-3 may function similarly in the ventral nerve cord lineages. In these lineages, pag-3 determines Pn.aa neuroblast fate and may also establish the fate of the VA and VB motoneurons generated by those neuroblasts.
Does pag-3 have a direct role in regulating programmed cell death?
We identified pag-3 in a screen for mutations in genes that affect whether individual cells in the ventral nerve cord survive or undergo programmed cell death. Does pag-3 function during cell-death specification? Two observations intrigued us. First, the protein most similar to PAG-3 encoded by the C. elegans genome is CES-1, which specifies the fate of programmed cell death for particular neurons (Ellis and Horvitz, 1991; Metzstein and Horvitz, 1999). Second, Gfi-1, a mammalian protein structurally very similar to PAG-3, can directly repress transcription of the proapoptotic gene Bax in cultured lymphocytes and prevent apoptosis (Grimes et al., 1996). These observations suggested that in C. elegans, pag-3 might act to prevent programmed cell death and were consistent with our observation that pag-3 mutants contained extra cell deaths. However, our studies established that the extra cell deaths in the ventral cord of pag-3 mutants are instead explained by a defect in the cell-fate determination of the Pn.aa neuroblasts.
Nonetheless, it remains possible that PAG-3 directly regulates the programmed cell death pathway in ventral cord neurons or in other cells in which PAG-3 is expressed. Specifically, loss-of-function mutations in ces-1 have normal patterns of cell deaths; the role of ces-1 in regulating programmed cell deaths of individual cells was identified by a gain-of-function mutation (Ellis and Horvitz, 1991). The current model for ces-1 function is that ces-1 expression must be repressed in cells that are to undergo programmed cell death (Metzstein and Horvitz, 1999). No similar gain-of-function mutations in pag-3 are currently available. We tested whether overexpression of pag-3 using the broadly expressed heat-shock promoter could prevent programmed cell deaths, but found pag-3 overexpression to be lethal (data not shown). Further work will be required to assess whether PAG-3 directly regulates the transcription of genes that control programmed cell death.
pag-3 counterparts may determine blast cell identity in other organisms
The Drosophila protein Senseless is very similar in sequence to PAG-3 and is necessary for the development of sensory organs in the peripheral nervous system (Nolo et al., 2000). In senseless mutant flies, there is a severe loss of neurons in the peripheral nervous system and an excess of programmed cell deaths (Salzberg et al., 1997). It is unclear at present whether the defects in senseless mutants result from abnormal cell-fate determination by peripheral nervous system neuroblasts, abnormalities in neuronal survival and differentiation, or both. pag-3 mutants differ from senseless mutants in that neuronal loss was not a prominent aspect of the pag-3 mutant phenotype and was noted in the ventral cord only after analysis of the cell lineages. However, this difference may be only quantitative. There may well be equivalent underlying effects on neuroblast determination. The excess of programmed cell deaths in senseless mutants could result from cell-lineage reiterations that generate extra dying cells, as in pag-3 mutants. Alternatively, if senseless acts in the neuroblasts of the peripheral nervous system of Drosophila to determine neuroblast fate in a manner analogous to the way pag-3 acts in the ventral nerve cord of C. elegans, the abnormal neuronal cells generated may undergo programmed cell death, as do abnormal neurons in other Drosophila mutants defective in development (Steller and Grether, 1994).
Two mammalian genes closely related to pag-3, Gfi-1 and Gfi-1B, are expressed in hematopoietic cells of adult rats and mice (Gilks et al., 1993; Tong et al., 1998). Proviral activation of Gfi-1 expression by Moloney murine leukemia virus in lymphocytes promotes the development of lymphomas (Gilks et al., 1993; Scheijen et al., 1997; van Lohuizen et al., 1991). The ectopic expression of Gfi-1 using the Lck promoter disturbs early T lymphocyte development (Schmidt et al., 1998). We report here that in the absence of pag-3 function, specific blast cells, the Pn.aa neuroblasts, adopt a stem cell-like pattern of division. Our finding of a required role for pag-3 in this determination of blast cell fates leads us to suggest that in mammals mutations in the Gfi-1 genes may similarly affect cell lineage determination and lead to cellular proliferation. Perhaps, independently or together with an effect of Gfi-1 on apoptosis, such a defect may lead to a loss of particular cell types or a pool of abnormal cells prone to the development of malignancy.
We thank Craig Ceol, Peter Reddien, Brad Hersh, Mark Alkema and Hillel Schwartz for their suggestions concerning this manuscript, and other members of the Horvitz laboratory for sharing expertise and ideas that contributed to this work. We thank Na An for maintenance of strains, Beth Castor for determining DNA sequences and the Caenorhabditis Genetics Center for strains. S. C. is especially thankful for the devoted support of his family and his wife Doreen. S. C. was financially supported by postdoctoral fellowships from the Damon Runyon-Walter Winchell Cancer Research Foundation, the Howard Hughes Medical Institute, the Merck-M.I.T. collaboration and NIH training grant HL07574. This research was supported in part by NIH research grant GM24663. H. R. H. is an Investigator of the Howard Hughes Medical Institute.