DAF-5 is a Ski oncoprotein homolog that functions in a neuronal TGFβpathway to regulate C. elegans dauer development

Like many organisms, free-living nematodes can alter development in response to environmental conditions. When food is scarce, population density is high and temperature is high, animals arrest development after the second molt as third larval stage (L3) dauers(Golden and Riddle, 1984a;Golden and Riddle, 1984b). Dauers are resistant to environmental insults and do not age (Riddle and Albert,1997), which allows survival and dispersal from environments with poor resources. If conditions improve, the dauer molts and resumes reproductive growth.

A transforming growth factor β superfamily (TGFβ) pathway is required for a normal dauer decision, and is thought to act as a step in a neuroendocrine pathway that couples external cues to dauer development(Riddle and Albert, 1997; Patterson and Padgett, 2000). Food availability and population density are sensed as chemosensory cues. These inputs, along with temperature, are sensed during early larval stages(Golden and Riddle, 1984b). High food and low pheromone stimulate transcription of daf-7, the gene for the ligand in the TGFβpathway (Ren et al., 1996; Schackwitz et al., 1996). DAF-7 binds to the receptors DAF-1 and DAF-4, which probably function in neurons (Inoue and Thomas,2000; Gunther et al.,2000). DAF-8 and DAF-14 are Smads that appear to be the direct targets of the receptors in the dauer TGFβ pathway (A. O. Z. Estevez, PhD thesis, University of Missouri, 1997)(Riddle and Albert, 1997; Inoue and Thomas, 2000). These Smads antagonize the function of another Smad called DAF-3(Patterson et al., 1997). The daf-5 gene has similar genetic properties to daf-3(Thomas et al., 1993), and so may be acting as a co-factor to DAF-3.

Little is known about events controlled by the receptors and Smads in this pathway, or the mechanism by which the pathway controls the dauer decision. Genetic analysis places the TGFβ pathway upstream of daf-12 (a gene encoding a nuclear hormone receptor) and daf-9 (a gene encoding a putative biosynthetic enzyme for a hormone that regulates DAF-12)(Thomas et al., 1993; Antebi et al., 1998; Gerisch et al., 2001; Jia et al., 2002). The daf-9 gene appears to be expressed in the XXX cell, which is little studied, but has neuronal properties and is located in a head ganglion(Ohkura et al., 2003). These facts, as well as the expression of DAF-7 in neurons and the suggested function of DAF-4 in neurons (Inoue and Thomas, 2000), suggest a model in which DAF-3 and DAF-5 function in the nervous system. However, some evidence can be interpreted as suggesting a function for DAF-3 and DAF-5 outside the nervous system. For example, a small sequence element derived from a pharynx specific promoter binds DAF-3 in vitro, and mediates daf-3-dependent repression of a reporter gene in the pharynx (Thatcher et al.,1999).

The dauer TGFβ pathway is unconventional in that upstream components of the pathway directly or indirectly antagonize DAF-3, a Smad protein(Patterson et al., 1997). In other pathways, Smad transcription factors are activated, not repressed, by the upstream components of the pathway. Anti-Smads antagonize receptor or Smad function (Shi and Massague,2003), but DAF-3 is antagonized by receptor signaling, and functions in the absence of receptor signaling to regulate genes that control dauer formation (Patterson et al.,1997).

Genetic analysis is consistent with DAF-5 acting as a cofactor of the DAF-3 Smad. The most extensively studied roles of TGFβ superfamily pathways are in the control of cell fate and in control of the cell cycle. The neuroendocrine role of the dauer pathway is very different. The dauer TGFβ pathway has evolved a unique mode of signaling, in which the receptors and receptor-activated Smads antagonize another Smad protein, DAF-3. Studies that reveal mechanisms of daf-5 function will help us understand how signaling pathways evolve new functions.

We show that DAF-5 is a diverged homolog of human Sno and Ski, which antagonize TGFβ signaling in cell culture by binding Smads, preventing their interaction with each other and with co-activators, and by recruiting co-repressors (Akiyoshi et al.,1999; Luo et al.,1999; Stroschein et al.,1999; Sun et al.,1999; Xu et al.,2000; Frederick and Wang,2002). DAF-5, Sno and Ski share a domain we call the SDS box; this domain in Ski mediates its interaction with Smad4. Yeast two-hybrid experiments demonstrate that DAF-5 interacts with DAF-3, and deletion studies are consistent with this interaction being mediated by the same domains as in the vertebrate Ski/Smad interaction. This functional similarity, combined with our phylogenetic analysis of DAF-5/Sno/Ski homologs suggests that, despite a highly divergent sequence, DAF-5 is an ortholog of Sno and Ski. DAF-5 is the first example of a Sno/Ski with a genetically defined function in a TGFβpathway. We identified a mutation hotspot in the conserved Dachbox domain. These mutants are the first direct evidence for a function for the Dachbox in TGFβ signaling. Our analysis of regulation of a pharynx-specific promoter does not support a previously hypothesized role for daf-3 and daf-5 in the pharynx. Furthermore, we show that daf-5 is expressed and functions in the nervous system. Therefore, we propose a model in which daf-3 and daf-5 have evolved novel functions that allow them to act in a neurosecretory pathway to control C. elegansdauer developmental arrest.

Details of strains used can be found in supplemental data. For egg-laying assays, worms were well fed and grown to L4 or young adult at 15°C, and were transferred to continue development and lay eggs at 25°C. Animals were scored 6 or 26 hours after the first egg was laid; for all genotypes, the data were similar for scoring at any time in this interval. The total number of eggs and the stage of the four oldest eggs inside of a worm were scored with DIC optics. For dauer formation, synchronized worms were incubated at 25-25.8°C for 3-4 days, and scored for dauer formation. High temperature Daf-c assays were very sensitive to temperature change; therefore, two printing thermometers were used to record temperature every 2 hours. Reported temperatures are an average of all readings. For scoring rnr-1 and cki-1 expression, animals were grown at 15°C. Gravid adults were transferred to fresh plates and allowed to lay eggs for 2-3 hours at room temperature. The progeny were incubated at 23°C. Larvae were staged by scoring the lethargus at the end of L2, and by body size and gonad size.

DNA polymorphism and mapping data were submitted to Wormbase(http://www.wormbase.org). Clones were injected as previously described(Mello et al., 1991) with pRF4 used as transformation marker.

We use the terms ortholog and paralog as defined in(Fitch, 1970) and(Jensen, 2001). Simply put,orthologs are duplicates of a character (e.g. a gene sequence) that arise from speciation events, whereas paralogs arise from a duplication within a single genome. Orthology and paralogy are typically hypotheses that are created based on sequence comparisons and other data. These definitions make no suggestion of conserved function.

Yeast two hybrid assays were performed as described(Walhout and Vidal, 2001). His/3-AT growth assay was scored on scale 0-3, and β-gal activity assay was scored on scale 0-5. All protein-coding sequences of interest were fused to the activation domain and the DNA-binding domain, and all assays were performed with both AD/DB combinations. We assigned `+' scores based on the total of His/3-AT and β-gal scores from the two combinations: ±,1-3; +, 4-7; ++, 8 or 9; +++, 10-13; ++++, 14-16.

We wished to evaluate the requirement for daf-5 in inducing dauer formation when TGFβ pathway genes are mutant. Therefore, we carried out an epistasis study of daf-5 using two alleles that are the most likely nulls, based on gene structure (see below). Smads and receptors in this pathway have redundant functions, and some single mutants do not completely disable the pathway (Gunther et al.,2000; Inoue and Thomas,2000). In addition, the dauer TGFβ pathway is partially redundant with a pathway that controls body size(Krishna et al., 1999; Morita et al., 1999). Therefore, we tested for the ability of daf-5 mutants to suppress when multiple pathway components are mutated. daf-8(sa343) and daf-14(m77) have early stop codons, and each gives the strongest known Daf-c (dauer formation constitutive) phenotype for the respective genes(A. O. Z. Estevez, PhD thesis, University of Missouri, 1997)(Inoue and Thomas, 2000). At 25°C, we see complete suppression of the Daf-c phenotype of daf-8;daf-14 double mutants and daf-8; daf-14; sma-2 triple mutants(Table 1). Therefore, at this temperature, daf-5 is essential for dauer formation induced by TGFβ pathway mutants. At slightly higher temperatures, suppression is incomplete, and continues to diminish as the temperature is raised from 25.4°C to 25.8°C. Our results are consistent with previous results(Thomas et al., 1993; Ailion and Thomas, 2000), but we show that failure to completely suppress daf-c mutants at temperatures higher than 25°C is not caused by incomplete loss of daf-5 function (previous studies used alleles of daf-5 that may not be null).

Animals mutant for daf-8 or daf-14 Smad genes also have an Egl-d (egg laying defective) phenotype. These animals have a structurally normal reproductive system, but lay eggs less frequently than wild type, and accumulate eggs inside. This phenotype is suppressed by daf-5 and daf-3 mutations (Trent et al.,1983; Thomas et al.,1993). daf-8, daf-14 and daf-8; daf-14 mutants contained more eggs (Table 2)and older eggs (see Table S1 and Fig. S1),than wild type. Introducing daf-5 mutants into daf-8 or daf-14 mutants or daf-8; daf-14 double mutants fully suppressed the Egl-d phenotype, except in the case of daf-8; daf-5(mg89);daf-14, which was intermediate between wild type and the Daf-c mutants. Interestingly, daf-5 single mutants have fewer eggs in the uterus than the other genotypes, and have younger eggs inside (for example, 50% of the oldest eggs in daf-5(sa310) have fewer than eight cells, versus 15% in wild type). This Egl-c (egg-laying constitutive) phenotype has not been previously reported for daf-5 mutants.

We identified the daf-5 coding region using DNA polymorphism mapping, and transgenic rescue of mutants. Our three factor mapping (data submitted to WormBase) placed daf-5 in a ∼120 kb interval(Fig. 1A). Cosmid rescue of daf-5 mutants suggested that the daf-5 gene was contained on the cosmid W01G7. Only one predicted gene was within the interval to which daf-5 was mapped (Fig. 1A). We sequenced three cDNAs provided by Y. Kohara. The longest clone, yk130g8, is identical to the structure inferred from a concatenation of EST sequences shown by WormBase(http://www.wormbase.org)and GenBank (NM_064540).

A blast of the predicted DAF-5 protein sequence against databases at NCBI reveals similarity to the oncoproteins Ski (for Sloan Kettering Virus) and Sno(for Ski-related novel sequence). A careful examination of many Sno/Ski homologs demonstrates that the homology of DAF-5 is significant, and we suggest that DAF-5 is the C. elegans ortholog of Sno/Ski. We show two domains from the Sno/Ski/Dachshund superfamily. Fig. 1C shows the SDS box (for Sno, Daf-5 and Ski); in human Ski, the SDS box and about 20 amino acids on either side constitute the minimal region required for binding to Smad4. Fig. 1D shows the Dachbox-N, a domain shared by the Sno/Ski family and Dachshund, which is a transcription regulator conserved throughout bilateria. DAF-5 has a predicted coiled-coil at the C terminus; Sno and Ski also have a coiled coil, but have a pattern of charged residues not shared by DAF-5.

This alignment is very informative regarding the relationship of DAF-5 to other members of the family. Sno and Ski are found in humans and all major groups of vertebrates, but in insects have only one ortholog of these two proteins (see Materials and methods for definitions of ortholog and paralog). Sno and Ski are more similar to each other than either is to the single Drosophila or mosquito ortholog; therefore Sno and Ski are probably paralogs that were duplicated after the divergence of the protostome and deuterostome lineages. We have named the insect genes Snowski(Snk) to reflect the orthology to both Sno and Ski.

A new family of proteins closely related to the Snowski group is shown in Fig. 1C,D. Humans have two genes in this group. We have named these genes Skate (for Ski-related gene) and Icy (for Ski sequence family). Drosophila and mosquito each have a gene that is much more similar to human Icy and Skate than to Drosophila Snk. We have named the single Drosophila and mosquito genes iceskate (isk) to reflect their orthology to both Icy and Skate.

We suggest that DAF-5 is an ortholog of Snowski or Iceskate. First, DAF-5 clearly has an SDS box, which is not found in any other protein in C. elegans or C. briggsae. This SDS box is more similar to the Snowski group than to the Iceskate group, including amino acids that are important for the ability of Ski to bind Smad proteins. Second, DAF-5 binds the DAF-3 Smad (see below). This binding is mediated by the SDS box in Ski,and may thus be a conserved function of the SDS box. Third, the rate of divergence in the Snowski/Iceskate family is so high that relatively modest sequence conservation is not surprising. This rapid change can be seen when examining the SDS box of B. malayi and potato cyst nematode Snowski. These two nematode proteins are more different from each other than insect Snowski is from Human Sno and Ski. The DAF-5 gene is even more rapidly diverging in the Caenorhabditis genus. C. briggsae and C. elegans proteins average more than 70% amino acid identity. The DAF-5 sequence is only 40% identical overall between C. briggsae and C. elegans. In fact, in the Dachbox and SDS box, the difference between C. elegans and C. briggsae DAF-5 is greater than the difference between insect and human Snowski.

Sequencing of daf-5 mutants identified a mutation hotspot. We identified mutations in 15 daf-5 alleles(Fig. 1B; see Table S2). All five missense mutants were found in a 16 amino acid stretch of the 627 amino acid protein. This hotspot is in the region of the Dachbox where DAF-5 is most similar to Snowski, Iceskate and Dachshund. Three of the mutants affect two very strongly conserved residues (two of the mutants are identical but independently isolated). One mutant has an in frame deletion of this region, and the final mutant is in a glycine that is unique to DAF-5. This region is critically important for DAF-5 function; the sa310 mutation(E162K) has a phenotype as severe as putative null alleles(Table 1). In vitro analysis showed that an insertion of four amino acids next to residue 168, which is homologous to the residue that is mutated in daf-5(sa310), eliminates the transforming and myogenic activity of vSki(Zheng et al., 1997). Thus,the Dachbox is required for a gain-of-function phenotype of vSki, but how this function relates to wild-type function of cSki or to TGFβ signaling has not been experimentally determined. A point mutation in the Dachbox disrupts the interaction of Ski with NCoR (Ueki and Hayman, 2003). However, this mutation affected Vitamin D receptor-dependent gene expression, but not TGFβ-dependent gene expression. These in vitro experiments suggest possible functions for the Dachbox, but our daf-5 mutants are the first missense mutants identified in any Sno/Ski gene, and thus the first in vivo evidence for the wild-type function of a particular domain of Sno/Ski.

A yeast two-hybrid screen for proteins that interact with DAF-3 identified the predicted protein W01G7.1 (M. Tewari, P. J. Hu, G. B. Ruvkun and M. Vidal,personal communication), which we show is DAF-5. We used yeast two-hybrid assays to identify regions of DAF-3 and DAF-5 required for interaction. We find that the DAF-3 MH2 domain strongly interacts with a DAF-5 fragment that is truncated after the SDS domain (Fig. 2, see Table S3). Thus, as in vertebrates, the region downstream of the SDS box is dispensable for binding to Smads and the Smad domain that binds Sno/Ski is the MH2 domain(Akiyoshi et al., 1999; Luo et al., 1999; Stroschein et al., 1999; Sun et al., 1999; Frederick and Wang, 2002; Wu et al., 2002). These interaction results suggest that DAF-3 and DAF-5 function as part of a transcriptional regulatory complex.

We made reporter gene GFP fusions to identify the cells in which daf-5 might function. Our full-length construct rescues a daf-5 mutant as efficiently as a wild-type genomic clone(Table 4). We saw relatively strong expression in ganglia in the head and tail and in the anterior pharynx(Fig. 3). Cell ablation and other experiments have shown that several cells in the nervous system (ASI,ADF, ASG, ASJ and XXX) are required for normal regulation of dauer formation(Bargmann and Horvitz, 1991; Schackwitz et al., 1996; Jia et al., 2002; Gerisch et al., 2001). We examined these cells for DAF-5::GFP expression. GFP was not seen in ASI, ASJ,ADF or ASG (0 of 86 animals). Fluorescence in the anterior ganglion, where XXX is found, is at least an order of magnitude less than in the bright cells of the ganglia posterior to the nerve ring. The weakness of fluorescence has precluded identification of specific cells, but we sometimes see weak fluorescence in the ventral, anterior part of the ganglion, where XXX resides(seven out of 53 animals). A small number of animals show weak expression in the hypodermis, muscles, intestine and distal tip cells (see Table S4). We also constructed a transcriptional GFP fusion, which consists of 6.5 kb upstream of the first ATG and also the first 57 codons in the first exon. The non-rescuing GFP construct is more strongly expressed, and shows more consistent expression in the hypodermis, muscles, intestine and distal tip cells; still, its expression is strongest in the head and tail ganglia (see Table S4). DAF-5::GFP from the rescuing construct mostly localizes to nuclei(Fig. 3, Table 3), which is consistent with the idea that DAF-5 is a transcription factor and functions in the nucleus. We examined the intensity of fluorescence and nuclear localization of DAF-5::GFP from the rescuing construct in wild-type and in TGFβ pathway mutants, including dauers, and saw no obvious differences.

We used tissue-specific expression of daf-5 to identify cells in which it functions. We expressed daf-5cDNA::GFP with various tissue-specific promoters (Aamodt et al.,1991; Okkema et al.,1993; Hsu et al.,1995; Maduro and Pilgrim,1995; Gilleard et al.,1997; Ogura et al.,1997); these constructs had GFP inserted at the same site as a genomic construct that rescued a daf-5 mutant(Table 4). Rescue was assayed in daf-7; daf-5 double mutants. Rescued animals would be expected to have the Daf-c phenotype of a daf-7 mutant. pF25B3.3strongly expressed daf-5::GFP exclusively in nervous system and rescued daf-5 mutants as well as two positive controls. Similarly, punc-14, which expressed daf-5::GFP in nervous system at high level in addition to some non-neuronal expression, also showed strong rescue. Weak ubiquitous expression of daf-5::GFP from the pdpy-30 promoter gave partial rescue. unc-119::daf-5::GFPwas weakly expressed in the nervous system but did not rescue, perhaps owing to the low level of expression. Strong expression of daf-5::GFP from the muscle promoter pmyo-3 did not rescue at all. Initial tests of two strains of pdpy-7::daf-5::GFP gave very weak expression and no rescue. Therefore, we isolated additional lines that had strong hypodermal expression of daf-5::GFP, and these did not rescue either. For unknown reasons, the expression of pges-1::daf-5::GFP was undetectable. Overall, our results show neuronal expression of daf-5is sufficient to rescue daf-5 dauer formation defect, while muscle or hypodermal expression is not.

In vertebrate cells, Smad proteins control the cell cycle by regulating transcription of cyclin kinase inhibitor genes(Moustakas and Kardassis,1998). The division of hypodermal seam cells is arrested in dauers because CKI-1, a cyclin kinase inhibitor, is transcriptionally upregulated,and this upregulation is inhibited by wild-type daf-7(Hong et al., 1998),suggesting that daf-5 may directly or indirectly upregulate cki-1. We tested whether cell cycle arrest in dauers requires daf-5. The reporter, rnr-1::GFP, drives the expression of GFP in the seam cells as they enter S phase and through the division at the beginning of L3. daf-7 (e1372) mutants do not express rnr-1::GFP at the corresponding time, which reflects seam cell cycle arrest in dauers (Hong et al.,1998). We observed that 47% of N2 show rnr-1::GFPexpression (Fig. 4A, Table 5), whereas none of the daf-1(sa184)-induced dauers does. Thus, this daf-1mutation causes a complete arrest of the cell cycle. Interestingly, in daf-5(sa310); daf-1(sa184) double mutants, the percentage of animals with green seam cells is 31%, which is similar to wild type (Fig. 4B; Table 5). The reason that we did not see rnr-1 expression in some N2 or daf-5; daf-7double mutants even after careful synchronization was probably due to the transient expression of the rnr-1 gene and variation in the timing of cell cycle among individual animals. We conclude that a daf-5mutation suppresses the cell cycle arrest caused by daf-1.

We wanted to test whether daf-5 controls seam cell cycle arrest via the cyclin kinase inhibitor, cki-1. In wild type, after cell division is complete, cki-1::GFP expression is seen, reflecting the role of cki-1 in restricting seam cells to one and only one division(Hong et al., 1998). In dauer larvae, cki-1 is expressed during the period where seam cells would divide in wild type; as a result, the seam cells do not divide. In daf-1(sa184) induced dauers, 100% of the animals with the cki-1::GFP reporter had a continuous line of green seam cells on both sides (Fig. 4C, Table 5), confirming the regulation of cki-1 by TGFβ signaling(Hong et al., 1998). Eighty-one percent of wild-type animals had either no green seam cells (53%)or a few faintly green seam cells (28%) expressing GFP. Similar to the N2 animals, 66% of daf-1; daf-5 double mutants had either no green seam cells (37%) (Fig. 4D, top animal, Table 5) or a few faintly green seam cells (29%). Nineteen percent of N2 and 34% of daf-1;daf-5 double mutants have many green cells(Fig. 4D bottom animal, Table 5). Expression in a subset of animals is expected because cki-1 is transiently expressed at the beginning of each larval stage to limit seam cells to a single division(Hong et al., 1998). Therefore, daf-5 mediates cell cycle arrest in dauers by controlling cki-1 expression. However, unlike in vertebrate TGFβ signaling,this control is likely to be indirect, because daf-5 expression in the nervous system is sufficient to cause dauer arrest.

Previous work is consistent with DAF-5 and DAF-3 having a function in the adult pharynx, but we found that expression of DAF-5 solely in the nervous system can cause dauer arrest. Therefore, we re-examined the role of DAF-3 and DAF-5 in adult pharynx. DAF-3 was shown to bind to a 5 bp sequence within a 32 base pair regulatory element isolated from the C. elegans myo-2promoter (the `C subelement'). This regulatory element drives pharynx-specific gene expression when placed upstream of a minimal promoter(Thatcher et al., 1999). Expression from this construct in adults is strongly repressed in mutants of daf-7 and other Daf-c mutants in the TGFβ pathway, but a mutation of daf-3 or daf-5 relieves this repression. However, all of this regulation was observed with the C subelement removed from its normal context and concatamerized upstream of a minimal promoter, and we wished to see if this regulation occurs within the normal context of the full myo-2 promoter.

We used a GFP reporter fused to a full-length myo-2 promoter (with 1.6 kb upstream of the translation start) to examine regulation by genes in the TGFβ pathway. If the full-length promoter is regulated similarly to the C subelement reporter, we expected that expression would be repressed in daf-7 mutant adults relative to wild-type adults, and that mutations in daf-5 or daf-3 would alleviate that repression. However,expression in adults was indistinguishable in wild type and in daf-7mutants (Table 6). Therefore,the regulation of the full-length promoter is unlike the regulation of the C subelement reporter. Similarly, in wild-type L2 stage larvae and daf-7 larvae in the corresponding L2d stage, expression was indistinguishable between N2 and daf-7, and expression in the daf-7;daf-5 and daf-7;daf-3 double mutants was modestly reduced. Finally, we see that expression in the daf-7 dauer is less than in wild type or the daf-7;daf-3 and daf-7;daf-5 double mutants. The C subelement reporter also shows a reduction in expression in dauers, comparable with what we see. However, unlike in adults, the reduction of expression of the C subelement reporter in dauers is not dependent on daf-3 (Thatcher et al.,1999), and is therefore mediated by some other pathway. In summary, we see no evidence that the full-length myo-2 promoter is regulated by daf-3 or daf-5.

We show that DAF-5 is homologous to the Sno/Ski family of transcriptional co-repressors, and we argue that DAF-5 is likely to be the C. elegansortholog of human Sno/Ski and Drosophila Snowski. In vertebrates, Sno and Ski antagonize TGFβ signaling by binding to Smads, but our analysis clearly demonstrates that DAF-5 is not an antagonist of the DAF-3 Smad. We show that DAF-5 is localized to the nucleus, and that it binds DAF-3, which strongly suggests that DAF-5 is a co-factor of DAF-3, and that DAF-5 assists DAF-3 in regulating gene expression. Our findings are novel in that we have identified a wild-type function for a Ski/Sno protein in a TGFβ pathway in vivo.

Recent work in cell culture has showed that Sno and Ski function in vitro as key factors in TGFβ regulation of gene expression and cell division(Akiyoshi et al., 1999; Luo et al., 1999; Stroschein et al., 1999; Sun et al., 1999; Wang et al., 2000; Xu et al., 2000; Frederick and Wang, 2002; Miyazono et al., 2003; Ueki and Hayman, 2003). These studies suggest numerous possible functions for Sno and Ski in TGFβsignaling, but understanding the specific contexts in which Sno and Ski function in vivo will require further work. Some studies of Sno/Ski function in vivo have used gain-of-function phenotypes in Xenopus development(Wang et al., 2000) or in cancer (Frederick and Wang,2002; Miyazono et al.,2003). These studies have identified mechanisms of Sno/Ski action and suggest possible roles for the proteins in vivo, but are accompanied by a caveat. Overexpression and gain-of-function mutants might cause the genes to act in an event that is not part of the normal function of the gene. Mouse Ski and Sno mutants have interesting phenotypes, including skeletal and muscular developmental defects and cancer susceptibility(Berk et al., 1997; Shinagawa et al., 2000; Shinagawa et al., 2001). These mutants will be useful in dissecting the wild-type functions of Sno and Ski in TGFβ signaling; however, the phenotypes of these mutants are complex, and Sno and Ski function in multiple signaling pathways(Dahl et al., 1998; Nomura et al., 1999), so tying a specific phenotype to the effect of Sno or Ski on TGFβ signaling will require careful analysis. Whereas Sno and Ski function in many pathways, daf-5 mutants exclusively affect phenotypes controlled by the dauer TGFβ pathway; this fact makes the assignment of function much more straightforward in C. elegans.

From analysis of the Snowski family, we identified several interesting new genes and relationships among them. First, we show that Sno and Ski are likely to be paralogs that arose after the divergence of protostomes and deuterostomes, perhaps as late as the divergence of urochordates and cephalochordates (because sea squirt has only one ortholog). Second, we identify the Iceskate family: an uncharacterized, highly conserved Snowski-related gene family. Genes in the Iceskatefamily have a striking difference from Snowski. Iceskate shows virtually no similarity to a set of amino acids in the SDS box of Ski that provide critical hydrogen bonds and van der Waals contacts in the Ski/Smad4 structure (Wu et al., 2002). To our knowledge, Iceskate proteins have not been studied experimentally; the functional relationship of Iceskate to Sno/Ski is an interesting issue for future research. Third, we find that Snowski is changing rapidly in nematode evolution. In Caenorhabditis, we see that daf-5 is changing rapidly, much more rapidly than the typical C. elegans gene, which may explain the relatively low primary sequence conservation between DAF-5 and other Snowski proteins.

Despite the rapid evolution of DAF-5, it binds DAF-3, a Smad. We find that the region downstream of the SDS box is entirely dispensable for DAF-5 binding to Smads, and the MH2 domain of DAF-3 is sufficient for interaction with DAF-5. The region of DAF-5 homologous to the region of Ski that contacts Smad4 is shown in Fig. 1C. Sno, Ski,DAF-5 and all other members of this family have a conserved zinc finger in the SDS domain. This zinc finger is an important structural component of Ski, and presumably all members of this family, because it is absolutely conserved. Only a subset of the residues of Ski that contact Smad4 are conserved in DAF-5. The binding of DAF-5 to DAF-3 may be somewhat different from the binding of Ski to Smad4, because DAF-3 is also greatly diverged from Smad4.

The rapid change in daf-5 in the Caenorhabditis genus prompts the question of to what extent have daf-5 and other members of the TGFβ pathway evolved new functions? The model for Sno/Ski function in vertebrate TGFβ signaling is shown in Fig. 5A. Briefly, activation of TGFβ receptors causes phosphorylation of Smads. This activation allows the Smads to bind Sno/Ski proteins and cause their degradation. With Sno/Ski gone, the Smads are free to regulate gene expression. However, Snoand Ski transcription is activated by Smads, leading to accumulation of Sno/Ski protein after receptor activation. Sno and Ski bind to Smads and recruit a variety of co-repressors, which prevent Smads from activating gene expression.

Two major functional changes have occurred within the dauer TGFβpathway (Fig. 5B). First, DAF-3 is unique among Smads in that its function is antagonized by receptor signaling, and that it acts in the absence of the receptors and R-Smads. DAF-3 might recruit factors that in other systems are recruited by R-Smads, or the pathway may have evolved to use other factors. Second, DAF-5 is the only known example of a Snowski protein acting in a TGFβ pathway as a co-factor rather than an antagonist. In principle, Sno or Ski could act as cofactors rather than antagonists on promoters that are negatively regulated by TGFβ, but this function has not yet been observed. One function that DAF-5 probably does not retain is a function in a variety of signal transduction pathways. Ski acts as a co-repressor for Mad, multiple nuclear hormone receptors, Rb, and other transcriptional regulators(Nomura et al., 1999; Dahl et al., 1998); Sno or Ski mutant mice have severe developmental defects,consistent with their role in multiple pathways(Berk et al., 1997; Shinagawa et al., 2001). daf-5 mutants, however, have much more modest defects, and daf-5 has no known phenotypes that are not shared by daf-3,suggesting that DAF-5 acts only as a co-factor to DAF-3.

The question of where the TGFβ pathway is acting is important, and evidence is not conclusive. Many of the genes in the pathway have broad patterns of expression, including many non-neuronal cells(Patterson et al., 1997; Gunther et al., 2000; Inoue and Thomas, 2000). We found that expression from a neuron-specific promoter efficiently rescued the daf-5 mutant, and that expression in muscle or hypodermis did not rescue. This result is consistent with daf-5 acting in neurons to regulate a hormonal cue for dauer.

Experiments to directly address the site of action of DAF-4 have been reported (Inoue and Thomas,2000), and suggest that this gene also has a neuronal focus of action. In these experiments, a daf-4 cDNA was fused to several tissue-specific reporters. A promoter expected to give neuronal and intestinal expression fused to a daf-4 cDNA rescued a daf-4 mutant, but putative intestine-specific, muscle-specific and other promoters did not. However, the authors could not monitor the expression of the constructs directly, and they were appropriately cautious in interpreting their results. Our results suggest that unexpected expression is not just a formal possibility. Rather, misexpression of `tissue specific' reporter constructs may be common. Several of our constructs gave unexpected expression patterns(Table 4). The unc-14promoter has been reported to be neuron-specific, but we saw expression elsewhere. The unc-119, dpy-7, dpy-30 and ges-1 promoters gave unexpectedly weak or undetectable expression, although in the case of dpy-7, we were able to correct the problem by isolating transgenic strains with stronger expression.

Two reports suggest functions for daf-3 and daf-5 outside the nervous system, which is inconsistent with our conclusion. One report suggested that daf-3 and daf-5 function in the pharynx to directly regulate gene expression in adults(Thatcher et al., 1999). This work used a reporter with a small element derived from the myo-2reporter (the `C subelement'). However, our examination of the full-length myo-2 promoter indicates that daf-3 and daf-5 do not regulate myo-2 gene expression in adults. In dauers, the full-length promoter does show regulation similar to that of a reporter containing only the C subelement, but regulation of the C subelement reporter in dauers is daf-3-independent. This daf-3-independent downregulation may be caused by a general reduction in expression of housekeeping genes that is seen in dauers (T. Liu and G.I.P., unpublished). We suggest that daf-3 and daf-5 do not actually regulate the expression of myo-2 in a normal context. daf-3 and daf-5 may function in the nervous system to regulate a secondary signal that in turn regulates the reporter with the C subelement.

Cell division arrest in hypodermis and other non-neuronal cells of dauers is dependent on a cyclin kinase inhibitor, cki-1(Hong et al., 1998). This gene is similar to a gene that is directly regulated by Smads in vertebrates(Moustakas and Kardassis,1998). In C. elegans, this gene is repressed in a daf-7 mutant, and we found that this regulation was daf-5dependent, which suggests that daf-3 and daf-5 could act outside the nervous system to directly regulate this gene. However, our demonstration that daf-5 expression in the nervous system is sufficient for dauer arrest suggests that regulation of cki-1 is indirect.

We observed a phenotype of daf-5 that has not been previously reported; daf-5 mutants are Egl-c, meaning that compared with wild type, daf-5 mutants have fewer eggs retained in the uterus and lay eggs with younger embryos. This phenotype suggests that the daf-5mutants have hyperactive egg-laying behavior. We were concerned that a low rate of egg production in the spermatheca might cause this phenotype, but we found that the rate of egg production was not correlated with the Egl phenotype. We found that all of the mutant genotypes in our experiment produce eggs at a rate of 1/2 to 2/3 that of wild type (data not shown), but have dramatically different Egl phenotypes. For example, the daf-5 single mutants make eggs at the same rate as the daf-8; daf-5; daf-14 triple mutants, which are not Egl-c. This new result suggests that daf-5functions to restrain egg laying even when the TGFβ pathway is active. The role of daf-5 in egg laying is not TGFβ pathway independent,as daf-c; daf-5 double mutants are not significantly Egl-c.

In addition to dauer and egg laying, the dauer TGFβ pathway also regulates social feeding behavior and fat accumulation(Trent et al., 1983; Thomas et al., 1993). In addition, daf-3 and daf-5, but not the other Smads or receptors, have a role in chemotaxis to both olfactory and gustatory cues(Daniels et al., 2000). All of these behaviors and developmental events are also affected by chemosensory input. Thus, the role of the TGFβ pathway can be said to be a general role in coupling chemosensory events to developmental and behavioral output. How might this pathway be coupled to these events? One possibility is that all of these events are affected, directly or indirectly, by a hormonal cue or cues, and that the TGFβ pathway regulates the expression of genes needed for production of the hormone(s). Orientation to a chemical gradient is quick,occurring within minutes. This is perhaps too brief an interval for the hormone to be acting during the chemotactic process, but a hormone might cause structural changes in the nervous system that alter chemotactic behavior.

A second interesting possibility is suggested by recent work that has identified new functions for TGFβ superfamily signaling in the nervous system. Retrograde signals between Drosophila neurons and their postsynaptic partner cells can affect the nature of the synaptic connection(Aberle et al., 2002; Marques et al., 2002) as well as the neurotransmitters produced by the presynaptic cells(Allan et al., 2003), and genetic analysis suggests that these retrograde signals are mediated by Gbb, a TGFβ superfamily ligand, and Wit, a TGFβ superfamily receptor. Thus,the dauer TGFβ pathway might act to affect the synaptic connections or other properties of neurons to alter signaling in the nervous system. This model is not mutually exclusive with the hormone regulation model above, as changes in connectivity or neurotransmitter production might affect synaptic signaling to a neurosecretory cell that makes hormone.

Study of the Ski and Sno proteins has provided an opportunity to learn how a co-repressor can play a variety of roles in different regulatory events. Perhaps the next important task will be to tie the biochemical and cell biological mechanisms identified in cell culture systems to important functions of Sno and Ski in vivo. Cancer biology is one field that may provide this connection, as the role of Ski and Sno in cancerous cells suggests testable hypotheses about the role of these proteins in the normal context. Sno and Ski function in multiple pathways, which makes understanding the role of the proteins in any one pathway difficult. In C. elegans, DAF-5 appears to function predominantly or exclusively in a single TGFβpathway. This simplicity has allowed us the unique discovery of an unambiguous connection of a Sno/Ski family member to a specific regulatory event in vivo,and this discovery provides a genetic system in which to pursue further understanding of Sno/Ski function.

We thank Dr Cole Zimmerman and Dr Richard W. Padgett for rnr-1::GFP construct; Dr Victor Ambros for strain VT825; J. Babiarz for microinjection; L. Tarantino for construction of daf-5 mutant strains; Y. Kohara for sharing cDNA clones; and G. Thio for providing genomic DNA from daf-5 mutants. We especially thank Dr M. Tewari and J. S. Ahn for generously sharing their time to perform yeast two-hybrid assays. Some strains in this work were provided by the C. elegans Genetics Center, which is supported by the National Center for Research Resources. This work was supported by NIH grant GM60994.