Expression of ceramide glucosyltransferases, which are essential for glycosphingolipid synthesis, is only required in a small subset of C. elegans cells

Glycosphingolipids (GSLs) are complex glycolipids derived from ceramide. They consist of a ceramide lipid moiety (a fatty acid chain linked to a sphingoid base) embedded in the cell membrane is attached to a wide variety of oligosaccharide structures that extend from the non-cytosolic leaflet of the lipid bilayer (Degroote et al., 2004; Lahiri and Futerman, 2007). The ceramide is made in the endoplasmic reticulum (ER), and the oligosaccharide is synthesized in the Golgi complex by sequential addition of sugars to ceramide in the lumenal leaflet of the membrane. The precursor of most GSLs, glucosylceramide (GlcCer), is synthesized by ceramide glucosyltransferase (CGT) (Fig. 1), which catalyzes the addition of glucose to ceramide. More complex glycosphingolipids are generated by dedicated glycosyltransferases, which add further sugar molecules to GlcCer (Fig. 1) (Varki et al., 1999).

GSLs are proposed to be important structural components of cell membranes and are particularly enriched in membrane microdomains known as lipid rafts. These domains are thought to help assemble signalling molecules and glycosylphosphatidyl-inositol-anchored proteins in specific membrane compartments, thus regulating signal transduction (Simons and Ikonen, 1997), although the existence and function of lipid rafts remains controversial (Munro, 2003).

Experiments in cell culture suggest that complex GSLs are involved in intracellular protein sorting (Sprong et al., 2001; Groux-Degroote et al., 2008), neuronal development (Futerman et al., 1999), cell growth and proliferation (Deng et al., 2000; Li et al., 2000), apoptosis (Kolesnick and Kronke, 1998), and cell adhesion and migration (Iwabuchi et al., 1998; Hakomori and Handa, 2002). However, analysis of mutant animals lacking enzymes important for GSL biosynthesis has failed to clarify the functions of GSLs in vivo. CGT-knockout mice, for example, which are unable to synthesize GSLs, die early during embryogenesis (E7.5-E9.5), with major defects in cell differentiation and a massive increase in apoptosis (Yamashita et al., 1999). It is not clear whether the increased apoptosis is caused by the accumulation of ceramide and its metabolites, which are known to promote apoptosis (Kolesnick and Kronke, 1998; Lahiri and Futerman, 2007), or by a lack of complex GSLs. However, apoptosis is apparently not increased in mice with CGT selectively knocked out in the nervous system, possibly because ceramide levels during development are not increased much in these animals, which display abnormal post-embryonic brain maturation (Jennemann et al., 2005). Surprisingly, several of the defects caused by CGT knockout in vivo are not observed in ES cells from CGT-deficient mice (Yamashita et al., 1999) or melanoma GM95 cells, which are devoid of GSLs (Ichikawa et al., 1994), suggesting that these lipids are particularly important for cell organization and differentiation in vivo. Knockout mice unable to synthesize a subset of complex GSLs, such as derivatives of gangliosides GM2, GD2 or GD3, display a variety of phenotypes, including disrupted male fertility, attenuated interleukin-mediated proliferative response of T cells, reduced regenerative capacity, motor deficits and neuronal degeneration caused by defects in myelination (Furukawa et al., 2004).

Because GSL-knockout studies in mammals have proved difficult to interpret, simpler model organisms have been used to investigate GSL function (Haltiwanger and Lowe, 2004). Drosophila CGT mutants are not available, but when CGT is knocked down by RNA interference (RNAi) in the fly embryo, apoptosis is increased (Kohyama-Koganeya et al., 2004). Moreover, deficiencies in the enzymes that add sugars to GlcCer lead to defects in several signalling pathways, including those activated by EGF receptors and Notch (Goode et al., 1992; Goode et al., 1996; Chen et al., 2007). Studies in Caenorhabditis elegans have also addressed important aspects of the physiological role of complex GSLs (GlcCer with additional sugars). Analysis of mutants resistant to the Bacillus thuringiensis toxin show that complex GSLs act in the intestine as toxin receptors and that the oligosaccharide chains are essential for this function (Griffitts et al., 2003; Griffitts et al., 2005). Additional studies show that mutations in genes involved in oligosaccharide addition to GSLs suppress gain-of-function mutations in lin-12/Notch, implying that complex GSLs act as positive regulators of Notch signalling (Katic et al., 2005). Recent work shows that mutations in genes required for the synthesis of C17 branched-chain fatty acids – which are normally incorporated into the GSL backbone – cause arrest at the first larval stage (L1) (Chitwood et al., 1995; Kniazeva et al., 2004; Entchev et al., 2008; Kniazeva et al., 2008). Mutant animals carrying these mutations have altered levels of C17 ceramides, as well as unusual ceramides (Entchev et al., 2008). It is possible that these alterations lead to reduced GSL function and L1 arrest. Interestingly, all the C. elegans mutants unable to synthesize complex GSLs are viable, suggesting that these lipids are dispensable in this organism, perhaps because the simple GSL GlcCer can functionally compensate for defects caused by complex GSL depletion or, alternatively, GSLs do not have essential functions. However, all these in vivo and in vitro studies have only partially clarified the physiological functions of simple GSLs, and it is still unclear whether all cells in C. elegans require these lipids, and, if so, for which physiological functions.

To answer these questions more definitively, we made C. elegans mutant strains that were unable to synthesize the GSL precursor GlcCer. In contrast to mammals and flies, which synthesize a single CGT protein, the C. elegans genome encodes three CGT proteins (Fig. 2), and we generated strains deficient in all of them. These mutants arrest at L1, become increasingly uncoordinated, and eventually die as L1 larvae. We characterize these mutants in detail and show that they are rescued by expressing CGT in a small number of cells in the intestinal tract, suggesting that CGT function is essential only in these cells.

CGT transfers glucose from UDP-glucose to ceramide to form glucosylceramide, the precursor of all C. elegans GSLs (Fig. 1) (Ichikawa and Hirabayashi, 1998; Haltiwanger and Lowe, 2004). CGT is a transmembrane protein anchored to Golgi membranes through its N-terminus (Fig. 2A), with the active site facing the cytosol (Ichikawa et al., 1996; Marks et al., 1999). The amino acid sequence of CGT does not have immediately identifiable functional domains. However, site-directed mutagenesis associated with biochemical analysis in rat CGT showed that mutation of a conserved histidine (Fig. 2A, asterisk) in the UDP-glucose binding site of the protein causes a strong reduction of enzymatic activity. In addition, a detailed functional analysis has revealed that several conserved amino acids in the C-terminal two thirds of CGT (such as K124, D144, D236, R195, R272, R275, G224, G225 and W276) (Fig. 2A,B, green dots) are essential for enzymatic activity in vitro (Wu et al., 1999; Marks et al., 2001).

Blast analysis of the C. elegans genomic sequence revealed the presence of three genes highly homologous to the single mammalian CGT (UGT8) gene: cgt-1, cgt-2 and cgt-3 (Fig. 2A). The deduced protein sequences show a high level of identity with each other (at least 51%) and with mammalian CGT (37% identity for CGT-3 or CGT-2 and 34% for CGT-1) (Fig. 2C). Each amino acid essential for enzymatic activity is conserved in all three C. elegans CGTs, suggesting that they are functional proteins. Indeed, CGT-2 and CGT-3 have CGT activity when heterologously expressed in GM95 cells or in yeast cells (Ichikawa and Hirabayashi, 1998; Leipelt et al., 2001). To our knowledge, CGT-1 has not been tested for enzymatic activity. Based on the known expressed sequence tags and on serial analysis of gene expression tags, cgt-3 is predicted to have four alternatively spliced mRNAs, all containing the amino acids essential for enzymatic activity (http://wormbase.org/db/gene/gene?name=WBGene00019127).

To investigate the role of GSLs in C. elegans and to determine the contribution of the three CGT genes, we used several CGT mutants. We generated the deletion cgt-1(qa1809) using PCR to screen chemically mutagenised C. elegans libraries with cgt-1-specific primers (Lesa, 2006). In addition, the deletion alleles cgt-1(tm1027), cgt-2(tm1097), cgt-2(tm1192) and cgt-3(tm504) were a generous gift from Shoehi Mitani (Tokyo Women's Medical University, Japan). All the cgt-1 and cgt-3 deletions (Fig. 2B) are likely to strongly reduce or completely eliminate CGT activity, because they lack several amino acids required for enzymatic activity (Marks et al., 2001). The cgt-1(qa1809) mutant lacks 908 bp that include exon 4 and exon 5 and part of exon 3 and exon 6, and therefore results in a protein with none of the required amino acids. The cgt-1(tm1027) mutant lacks 936 bp that include part of the cgt-1 promoter, the start codon, the first three exons and part of exon 4, including K180 and D200. The cgt-3(tm504) mutant has a deletion of 1050 bp, which includes part of exon 2, the whole of exon 3 and exon 4, and the majority of exon 5, and inserts 57 new nucleotides, resulting in a protein that lacks the entire CGT gene active site. The cgt-2(tm1097) and cgt-2(tm1192) mutants contain a deletion of a small portion of exon 4 and part of the following intron (Fig. 2B). To test whether these deletions are likely to cause loss of CGT-2 protein function, we analyzed cgt-2 mRNA transcripts by performing RT-PCR. Both cgt-2(tm1097) and cgt-2(tm1192) mutants produce a major and a minor mRNA transcript (Fig. 3). We sequenced the cDNAs derived from these transcripts and found that the protein corresponding to the major tm1097 transcript is truncated at W325, with the following seven new amino acids added: IASLLHS. Since tm1097 does not remove amino acids known to be essential for CGT activity, it is not clear whether it causes loss of CGT-2 function. The protein corresponding to the major tm1192 mRNA has a 31-amino acid deletion (313-343), which includes the required amino acids R321, R324 and W325 (Marks et al., 2001). It is therefore likely that tm1192 results in complete loss of CGT-2 function. The minor tm1097 and tm1192 transcripts result in identical proteins lacking all amino acids encoded by exon 4 (Fig. 3) and are expected to produce mutant proteins with no enzymatic activity. All the cgt deletions we have used in this work, except cgt-2(tm1097), are likely to completely eliminate CGT enzymatic activity.

All single CGT mutants showed no observable phenotype, suggesting that GSL levels might not be altered significantly in these mutants. Indeed, we found that levels of GlcCer derivatives in the putative null alleles cgt-1(qa1809) and cgt-3(tm504) were not affected in comparison with those of wild-type animals (Fig. 4, lanes 1, 2 and 7). Mutants carrying inactivated cgt-2 in combination with inactivated cgt-1 or cgt-3 also did not display any observable phenotype. However, cgt-3;cgt-1 double mutant animals arrested growth at an early larval stage and became increasingly uncoordinated and died after a few days (Fig. 5A,B). In addition, the growth-arrest phenotype caused by depletion of cgt-1 and cgt-3 activities by either gene deletion or by RNAi was not exacerbated by the presence of cgt-2(tm1097) or cgt-2(tm1192) or by RNAi for cgt-2 (data not shown), suggesting that cgt-2 does not influence the growth-arrest phenotype.

The cgt-3;cgt-1 double mutants contained only ∼25% of the GlcCer present in wild-type animals (Fig. 4, lane 6). Moreover, GlcCer levels were further reduced when all three cgt genes were knocked down (Fig. 4, lane 3), suggesting that cgt-2 encodes a functional CGT protein. The level of GlcCer (7.5±1.3% of wild type) in the triple mutant might reflect an incomplete RNAi effect; it could represent GlcCer inherited from the mothers, which carry wild-type copies of cgt-3 and have significant GSL levels, or it could represent maternal GlcCer, because some mothers could not be removed from the sample we analyzed. It is unlikely to reflect a dietary contribution, because, to our knowledge, E. coli (worm food) do not produce GlcCer.

Inspection under a dissecting microscope suggested that most cgt-3;cgt-1 mutants arrested in L1 (Fig. 5A,B). Indeed, electron microscopy revealed that the arrested animals had alae, which are cuticular ridges only found in L1 larvae or adults (Fig. 5C). To determine how far development proceeded in L1, we focused on the M-cell lineage, which gives rise to sexually dimorphic muscles. L1 larvae hatch with one M cell (Fig. 5D), which divides after 5-6 hours at 20°C. The daughter cells then divide sequentially after ∼3 hours, ∼2 hours and ∼3 hours, so that by the end of L1 (∼15 hours after hatching) 16 M-lineage cells are present (Fig. 5E) (Sulston and Horvitz, 1977). We used the integrated hlh-8::gfp reporter transgene to visualize the M cells, because it is specifically expressed in the M cell lineage (Harfe et al., 1998). Three days after hatching, more than 84% of the progeny of cgt-3(RNAi);cgt-1 mutants had one or two GFP-positive cells, whereas all control animals had 16 GFP-positive cells (Fig. 5F), suggesting that most of the mutants arrested in early L1.

To verify that the growth-arrest phenotype is caused by inactivation of cgt-1 and cgt-3, we performed transformation rescue experiments in animals with inactivated cgt-1 and cgt-3. As shown in Table 1, expression of extrachromosomal copies of either cgt-1 or cgt-3 bypassed the growth arrest at L1 (Table 1). Moreover, a PCR fragment containing the promoter, coding region and 3′-untranslated regions of wild-type cgt-1, rescued the cgt-3(RNAi);cgt-1(qa1809) mutants as did wild-type cgt-3 in cgt-3(tm504);cgt-1(RNAi) animals. These results show that the growth-arrest phenotype is caused by simultaneous inactivation of cgt-1 and cgt-3.

To test when genes encoding CGTs are normally required to allow proper growth, we inactivated them at different developmental stages. First, we used cgt-3 siRNA in cgt-1 mutant mothers (n=12) and found that the progeny arrested at the L1 stage. However, when we exposed cgt-1 mutant animals to cgt-3 siRNA immediately after hatching (n=32), all the animals reached adulthood. These results suggest that cgt-1 activity is either required embryonically or during the L1 stage, since cgt-3 RNAi takes approximately 12-14 hours to produce the growth-arrest phenotype in cgt-1(qa1809) mutant animals at 20°C (G.M.L., unpublished results), which is roughly the duration of the L1 stage (Byerly et al., 1976). It is unlikely that CGT genes are required for general embryonic development, because, upon hatching, the cgt-3;cgt-1 L1 larvae did not show gross morphological defects and we could rescue the cgt-3;cgt-1 phenotype by tissue-specific expression of either cgt-1 or cgt-3 (see below). It is also unlikely that the L1 arrest is a maternal effect phenotype, because we observed an identical L1 arrest in progeny of cgt-3;cgt-1(RNAi);cgt-2 mothers [RNAi often reduces stores of maternal message that would normally be inherited by the oocyte (Maine, 2001)] and in cgt-3;cgt-1;cgt-2 progeny of cgt-3;cgt-1/+;cgt-2 mothers (data not shown).

Apoptosis is dramatically increased in mice and Drosophila when CGT is knocked out, possibly because loss of CGT function causes accumulation of ceramide (Yamashita et al., 1999; Kohyama-Koganeya et al., 2004), a known inducer of apoptosis (Pettus et al., 2002). Since our lipid analysis revealed a modest increase in ceramide levels in mutants undergoing growth arrest (Fig. 4), we tested whether apoptosis has a role in the growth-arrest phenotype caused by simultaneous inactivation of cgt-1 and cgt-3. In mammalian cells, overexpression of BCL2, an anti-apoptotic gene, suppresses ceramide-induced apoptosis (Zhang et al., 1996). Likewise, gain-of-function mutations of the C. elegans homolog ced-9, and loss-of-function mutations of two pro-apoptotic genes, ced-3 (a caspase) and ced-4 (the caspase activator apaf-1), suppress the death of all the cells that normally undergo apoptosis (Hengartner and Horvitz, 1994). Thus, if the growth arrest of cgt-3;cgt-1 mutant animals is caused by excessive apoptosis, mutants carrying any of the CED protein mutations should grow normally. To test this, we inactivated cgt-1 and cgt-3 by RNAi in ced-3(n717) (n=7), ced-4(n1162) (n=4) or ced-9(n1950) (n=6) mutant worms. All of the 17 animals exposed to RNAi produced progeny that arrested at the L1 larval stage, as did wild-type animals exposed to the same siRNAs (n=5). Moreover, we used the nucleic acid fluorescent dyes SYTO11 and acridine orange to visualize apoptotic nuclei. The nuclear staining pattern was similar in cgt-3(RNAi);cgt-1(qa1809);cgt-2(tm1192) arrested mutants and in wild-type animals (data not shown), suggesting that no additional apoptotic events were occurring in the mutants. In agreement with these results, we could not detect an increase in button-like cells (a typical apoptotic phenotype) under Nomarski optics in arrested mutants. It is therefore unlikely that the growth-arrest phenotype observed in cgt-3;cgt-1 worms is caused by increased apoptosis.

To gain insight into the function(s) of cgt-1 and cgt-3 we looked at where cgt-1 and cgt-3 are normally expressed. We generated constructs of cgt-1 and cgt-3 containing the endogenous promoter, the coding sequence fused in-frame with GFP, and the 3′ untranslated region. These constructs encode fully functional proteins, because they rescued the growth-arrest phenotype of cgt-3;cgt-1 mutant animals. Their expression pattern is therefore likely to reflect that of the corresponding endogenous genes.

Both cgt-1::gfp and cgt-3::gfp were strongly expressed in pharyngeal muscles and in other pharyngeal cells, although their expression did not completely overlap (Fig. 6). In addition, they showed strong expression in the pharyngeal intestinal valve (PIV), a set of six epithelial cells connecting the posterior bulb of the pharynx to the anterior intestine (Hall and Altun, 2008), in the intestinal cells (particularly the most anterior and the most posterior), in the intestinal rectal valve (IRV), two epithelial cells that occlude the lumen of the posterior intestine, and in the three rectal gland cells (RGCs) (Hall and Altun, 2008). We detected cgt-1::gfp expression in the excretory cell, excretory canals, duct cell and pore cell; these cells form the excretory system, which helps to maintain osmotic or ionic balance and secretes material of unknown nature and function (Hall and Altun, 2008). Some cgt-3::gfp expression was also visible in head mesodermal cells. Both cgt-1 and cgt-3 showed additional, non-overlapping, expression in some non-identified head cells. Interestingly, with the possible exception of a few amphid neurons, we could not detect any expression of cgt-1 or cgt-3 in the nervous system.

To determine where the genes encoding CGT are essential, we tested whether a single wild-type copy of cgt-1 is sufficient to rescue the growth-arrest phenotype in worms with no active genes encoding CGT. We produced the strain cgt-3(tm504);cgt-1(qa1809)/DnT1;cgt-2(tm1192). DnT1 is a reciprocal translocation between chromosomes IV and V, which contains wild-type cgt-1; therefore, the strain we produced contains cgt-1 as the only cgt gene. We found that this strain did not arrest growth, indicating that the essential CGT function resides within the tissues where cgt-1 is normally expressed (Fig. 6). Since we observed the growth-arrest phenotype only when both cgt-1 and cgt-3 were inactivated, it is likely that the essential CGT gene function is exerted in a tissue where both genes are normally expressed, that is the PIV, the IRV, the intestine, the RGCs and the pharyngeal cells. We therefore generated constructs that drive gene expression in specific tissues and tested their ability to rescue the growth arrest phenotype associated with loss of CGT activity. As expected, constructs containing the endogenous promoter, the coding sequence and the 3′ untranslated region of cgt-1 or cgt-3 restored normal growth of cgt-3;cgt-1 or cgt-3;cgt-1;cgt-2 mutant animals (Table 1).

To test whether GSLs can rescue the defect cell non-autonomously, we expressed CGT genes in muscles and in the nervous system, where no CGT expression is normally detected, and examined whether the growth-arrest phenotype of cgt-3;cgt-1 mutant animals was rescued. We placed cgt-3 under the control of the myo-3 promoter, which drives expression in most muscles, and cgt-1 and cgt-3 under the control of the neuron-specific unc-119 promoter. As expected, these chimeric constructs could not rescue the growth-arrest phenotype of cgt-3;cgt-1 double mutant animals, even when coexpressed (Table 1), suggesting that GSLs or their products do not effectively cross cell boundaries.

We then investigated whether a construct selectively driving cgt-1 expression in the pharyngeal muscles (myo-2p::cgt-1) (Okkema et al., 1993) could rescue the growth arrest of cgt-3;cgt-1 mutant animals. We generated several stable lines and found that this chimeric construct could not rescue growth arrest (Table 1), suggesting that CGT function is required in non-pharyngeal tissues to overcome the growth-arrest phenotype of cgt-3;cgt-1 mutant animals.

We then tested whether cgt-1 or cgt-3 expressed in the PIV, IRV and a subset of intestinal cells could rescue the growth arrest of cgt-3;cgt-1 or cgt-3;cgt-1;cgt-2 mutant animals. We placed cgt-1 and cgt-3 under the control of the F54D5.1 promoter, which drives expression specifically in PIV, IRV, the RGCs, and the most anterior and posterior intestinal cells from late embryonic stages until adulthood (Hope et al., 2004) (Fig. 7A-C and data not shown). Transgenic cgt-3;cgt-1 double mutants, and even triple mutants expressing these constructs, grew and reproduced similarly to wild-type animals. In addition, they produced levels of GSLs close to (∼70%) those of wild-type animals (Fig. 7D). These findings suggest that CGT function is required in the IRV, PIV, RGCs, a subset of intestinal cells, or in a combination of these cells.

To test whether intestine-specific expression of CGT was sufficient to rescue the cgt-3;cgt-1 mutant phenotype, we placed cgt-1 under the control of the intestine-specific promoter elt-2 (Fukushige et al., 1998), which is expressed in the intestine and its precursor cells from early embryonic stages. This construct rescued the phenotype of cgt-3;cgt-1 mutants. Since we also rescued the mutant phenotype by expressing CGT in PIV, IRV and in the most anterior and posterior intestinal cells, it is likely that CGT function is only required in the most anterior and posterior intestinal cells.

When C. elegans larvae hatch, they are developmentally arrested and require food to start and maintain normal post-embryonic development (Johnson et al., 1984; Hong et al., 1998). We therefore tested the possibility that cgt-3;cgt-1 mutants arrest growth because they are unable to feed normally. We mixed small fluorescent beads (with a diameter of ∼20% of that of the bacteria that the worms feed) with food and examined whether the beads appeared in the digestive tract of 1-day old, L1-arrested CGT mutants (Fig. 8). The vast majority of control cgt-1;cgt-2 L1 larvae (85.1±3.3%) had multiple fluorescent beads along their digestive tract, and very few had no beads at all in the digestive tract (9.6±4.0%) or beads that had not passed the pharynx (5.3±1.8%) (Fig. 8A-C). By contrast, most of the L1-arrested cgt-3(RNAi);cgt-1;cgt-2 mutants either had no fluorescent beads at all in their digestive tract (42.7±6.6%, P<0.0001) or had beads stuck (29.7±4.9%, P<0.0001) just before (Fig. 8D-F) or just after (Fig. 8G-I) the PIV. Only 27.6±5.4% (P<0.0001) of the mutants had beads inside the intestine. To confirm that CGT mutants were starved, we examined DAF-16::GFP (Fig. 8K-N), which is normally localized in the cytoplasm (Fig. 8K-M) but displays mainly nuclear localization under starvation and other stresses. DAF-16::GFP was mainly localized in the nucleus of arrested mutants (Fig. 8N), consistent with the animals being starved.

To test whether the PIV or the most anterior intestinal cells are defective in CGT mutants (beads tend to accumulate around the PIV) we used Nomarski and fluorescence microscopy and F54D5.1::gfp to visualize PIV and IRV. As shown in Fig. 9, these cells appeared normal in newly hatched cgt-3(RNAi);cgt-1;cgt-2 mutants, and remained so 1 day after hatching, when the mutants already show defects in feeding. To check whether the canal formed by the PIV and the IRV cells was also normal in 1-day-old cgt-3(RNAi);cgt-1;cgt-2 mutants, we used confocal microscopy in animals expressing F54D5.1::gfp, and performed three-dimensional reconstructions (supplementary material Fig. S2, Movies 1-3). The canal formed by the PIV cells had a similar area in control (not-arrested) L1 cgt-1;cgt-2 larvae (3.27±1.06 μm², n=3) and in L1-arrested cgt-3(RNAi);cgt-1;cgt-2 larvae (3.62±0.57 μm², n=10), suggesting that the feeding defect is not caused by a physical obstruction within the PIV. Similarly, no difference was detectable in the IRV canal (cgt-1;cgt-2 mutant: 2.02±0.45 μm², n=3; cgt-3(RNAi);cgt-1;cgt-2 mutant: 1.89±0.46 μm², n=15). However, after the first day of L1 arrest, most of the anterior intestinal cells appeared swollen and vacuolated: an appearance consistent with necrotic cell death. In addition, the intestinal tract appeared obstructed, and in some cases the lumen seemed discontinuous. Later, the PIV and IRV also became swollen (Fig. 9). The swollen and vacuolated cells displayed dispersed, faint or no F54D5.1::gfp expression in CGT mutants, which suggests that they are necrotic, whereas they looked normal and were normally fluorescent in starved control L1 larvae (Fig. 9), suggesting that the changes were not caused by starvation. Since the PIV and IRV cells became swollen after the intestinal cells appeared to degenerate, it is possible that the PIV and IRV abnormalities were secondary to the necrosis of the intestinal cells.

Glycosphingolipids are thought to be essential membrane components of mammalian cells. However, the generation of mice defective in GSL production has not clarified GSL functions in vivo. By inactivating the genes required for GSL synthesis in C. elegans, we have generated GSL-depleted worms and show that they cannot develop beyond the first larval stage. Unexpectedly, we can rescue this growth defect by restoring CGT function in only a few cells of the digestive tract, suggesting that GSLs are dispensable in most C. elegans cells.

All three C. elegans genes encoding CGT-1, CTG-2 and CTG-3 are likely to have CGT activity. CGT-2 and CGT-3 catalyze the formation of GlcCer in cells that normally do not produce GSLs (Ichikawa and Hirabayashi, 1998; Leipelt et al., 2001), and deletion of the cgt-1 gene reduces GlcCer levels in mutant worms (Fig. 4, compare lanes 4 and 6). It is unlikely that C. elegans has other genes with CGT activity, because BLAST analyses with human CGT against the entire worm genome only identify CGT-1, CGT-2 and CGT-3.

Both cgt-1 and cgt-3 are expressed in the PIV, IRV, RGCs, intestinal cells (particularly in the most anterior and the most posterior ones) and pharyngeal muscles. Inactivation of both these genes together causes growth arrest at the L1 stage. cgt-1 can compensate for the lack of cgt-3, and vice versa, indicating that the two genes have partially overlapping functions, at least in some tissues. It is unclear why C. elegans has two genes that apparently catalyze the same enzymatic reaction (humans, mice and Drosophila have just one CGT gene) and are both expressed in some tissues. Although it is possible that they are expressed in a temporally different manner, we find that CGT-1::GFP and CGT-3::GFP are both expressed around the time when they are required: approximately at the L1 stage. Alternatively, CGT-1 and CGT-3 might catalyze glucose addition to slightly different ceramide species, with each species compensating for the absence of the other.

Expression of cgt-1 or cgt-3 in PIV, IRV, and the intestinal cells adjacent to them or expression of cgt-1 exclusively in the intestine, is sufficient to rescue the growth-arrest phenotype. The rescued animals do not show any obvious phenotype when viewed in a dissecting microscope, implying that most cells can develop and function with virtually no cgt activity. Such cells are likely to be essentially devoid of GSLs, because the C. elegans genome does not encode other CGTs, and GSLs are not likely to move from cell to cell. Our inability to rescue the CGT mutant phenotype by expressing the genes in certain tissues or by providing a mixture of exogenous GSLs (data not shown) confirms that CGTs and GSLs do not move from one cell to another.

On the one hand, CGT-knockout mice die at around E7.5, probably because many of their cells undergo apoptosis; on the other hand, cell proliferation is unaffected (Yamashita et al., 1999). It is unclear whether the increased apoptosis is caused by increased levels of ceramide, a known pro-apoptotic factor, or by GSL depletion, because it was not possible to quantify ceramide in mutant embryos. Remarkably, ES cells from CGT-knockout mice survive in vitro. These findings in mice led to the hypothesis that GSLs might be required for cell-cell or cell-matrix interactions (Yamashita et al., 1999). We find that enzymes required for GSL synthesis are only required in C. elegans in a subset of cells in the digestive tract and are therefore dispensable in most tissues, including the nervous system. This is surprising, because in mammals, GSLs have been implicated in a wide variety of cell functions that are expected to be essential in vivo, including intracellular protein sorting (Sprong et al., 2001), neuronal development (Futerman et al., 1999), and cell adhesion and migration (Iwabuchi et al., 1998; Hakomori and Handa, 2002). It seems that GSLs have extended their essential function from a few cell types to many cell types during evolution.

It is not clear what essential functions the GSLs serve in the intestinal cells in C. elegans, where normal GSL levels are required. One day after growth has arrested, intestinal cells become swollen and vacuolated, and soon after, the PIV and IRV cells also become swollen and vacuolated. These cell changes do not depend on starvation, because starved controls do not display such changes. The finding that PIV, IRV and intestinal cells appear normal at hatching in CGT mutants suggests that having essentially no GSLs is compatible with correct development of these cells. Instead, normal GSL levels appear to be essential to keep these cells alive and normal. Because we could rescue the growth-arrest phenotype by expressing CGT exclusively in the intestine or in the PIV, the IRV and in their adjacent intestinal cells, it is likely that GSLs are only essential in the most anterior and most posterior intestinal cells, which make contact with the PIV and the IRV. Our results suggest that the death or dysfunction of these cells impairs feeding, causing the worms to starve and arrest growth at the early L1 stage. Further studies will be required to understand how GSLs promote the survival and normal function of these intestinal cells and whether their survival-promoting effects are direct or indirect.

Why can cells with strongly reduced GSL levels survive in CGT mutants? It is possible that these cells adapt to the depletion of GSLs and use alternative molecules to serve the functions of the missing lipids. Such compensation happens in mammals. For example, mice that cannot synthesize galactosphingolipids, which are particularly enriched in myelin, substitute them with glucosphingolipids, which are normally absent from myelin; although this substitution allows myelin to be made, myelin function is abnormal (Coetzee et al., 1996). It is likely that compensatory lipids are also synthesized by CGT-impoverished worms. Consistent with this possibility, we identified an enzyme involved in the modification of ceramide fatty acids in a screen for mutations synthetically lethal with cgt-3(qa1809) (G.M.L., unpublished results). Perhaps, because they are relatively simple organisms, worms can change their lipid composition without compromising the survival or function of most of their cells.

C. elegans strains were cultured under standard conditions (Brenner, 1974). We used the following strains: LZ64, dpy-20(e1282);cgt-1(qa1809);gxEx64[cgt-1::gfp;pMH86;pBluescript]; LZ65, cgt-2(tm1097); LZ66, cgt-3(tm504); LZ68, cgt-1(tm1027); LZ69, cgt-2(tm1192); LZ83, cgt-3(tm504);+/DnT1;cgt-1(tm1027)/DnT1; LZ96, cgt-3(tm504);dpy-20(e1282);gxEx96[cgt-3::gfp;pMH86;pBluescript]; LZ113, cgt-3(tm504);+/DnT1;cgt-1(qa1809)/DnT1;cgt-2(tm1097); LZ114, cgt-3(tm504);+/DnT1;cgt-1(qa1809)/DnT1;cgt-2(tm1192); LZ163, dpy-20(e1282);gxEx163-[F54D5.1p::gfp;pMH86;pBluescript]; LZ196, dpy-20(e1282);cgt-1(qa1809);cgt-2(tm1192);gxEx163; LZ255, cgt-3(tm504);+/DnT1;cgt-1(qa1809)/DnT1; LZ269, cgt-1(qa1809);zIs356[daf-16::gfp;pRF4]; LZ278, cgt-3(tm504);cgt-1(tm1027);cgt-2(tm1192);gxEx145[F54D5.1p::cgt-1, pPD118.33, pBluescript]; LZ279, cgt-3(tm504);cgt-1(qa1809);cgt-2(tm1192);gxEx145; MT1522, ced-3(n717); MT2547, ced-4(n1162); MT4770, ced-9(n1950); PD4667, dpy-20(e1282);ayIs7(pBH47.70+ pMH86)]; TJ356, zIs356[daf-16::gfp;pRF4]; XA1809, cgt-1(qa1809).

We also used various stable lines of the following transgenic strains: dpy-20(e1282);cgt-1(qa1809);gxEx[myo-2p::cgt-1, pMH86, pBluescript], dpy-20(e1282); cgt-1(qa1809);gxEx[elt-2p::cgt-1, pMH86, pBluescript], cgt-3(tm504);+/DnT1;cgt-1(tm1027)/DnT1;gxEx[unc-119p::cgt-3, pPD118.33, pBluescript], cgt-3(tm504); +/DnT1;cgt-1(tm1027)/DnT1;gxEx[unc-119p::cgt-1, pPD118.33, pBluescript], cgt-3(tm504);+/DnT1;cgt-1(tm1027)/DnT1;gxEx[myo-3p::cgt-3, pPD118.33, pBluescript], cgt-3(tm504);+/DnT1;cgt-1(tm1027)/DnT1;gxEx[F54D5.1p::cgt-1, pPD118.33, pBluescript], cgt-3(tm504);+/DnT1;cgt-1(tm1027)/DnT1;gxEx[F54D5.1p::cgt-3, pPD118.33, pBluescript].

When not indicated otherwise, cgt-3;cgt-1 animals were cgt-3(tm504);cgt-1(tm1027) from LZ83 mothers, cgt-3(tm504);cgt-1(qa1809) from LZ255 mothers, cgt-3(RNAi);cgt-1(1027), or cgt-3(RNAi);cgt-1(qa1809).

To isolate deletions of the genes encoding CGT, we constructed a DNA library of approximately 1,400,000 haploid genomes from wild-type animals mutagenized with ethylmethane sulfonate (Lesa, 2006). Using PCR with cgt-1 primers (Jansen et al., 1997), we isolated the deletion cgt-1(qa1809). The deletion alleles cgt-1(tm1027), cgt-2(tm1097), cgt-2(tm1192) and cgt-3(tm504) were a generous gift from Shoehi Mitani (National Bioresource Project for the Experimental Animal `Nematode C. elegans', Tokyo Women's Medical University, Tokyo, Japan) and their molecular details are available at http://www.shigen.nig.ac.jp/c.elegans/mutants/index.jsp?lang=english. cgt-1(qa1809) lacks the fragment comprising base pairs 20,621-21,528 of cosmid T06C12. We used PCR-based genotyping on all deleted mutants to detect the deletion and to verify that no duplications of the intact CGT genes were present (not shown). We used the following primers:: GL147, GL184, GL185 for cgt-1(qa1809), GL421, GL422b, GL423b for cgt-1(tm1027): GL424, GL425b, GL426b for cgt-2(tm1097) and cgt-2(tm1192), and GL415, GL416, GL417 for cgt-3(tm504). Primer sequences are available in supplementary material Table S1.

Mothers of cgt-3;cgt-1 or cgt-3;cgt-1;cgt-2 progeny or mothers carrying one or more CGT mutations exposed to siRNA for a different CGT gene, were transferred to OP-50 plates and their progeny was observed for 5 days. Growth arrest was defined as the inability of ∼100% of the progeny to pass the L2 stage by day 4-5.

Total RNA was extracted from mixed stage populations of cgt-1(qa1809);cgt-2(tm1097) or cgt-1(qa1809);cgt-2(tm1192) worms using the RNeasy kit (Qiagen, Crawley, UK). 1.5 μg total RNA per sample was reverse transcribed with Superscript II Reverse Transcriptase (Invitrogen) and hexamer random primers (Promega, Southampton, UK) to a final volume of 30 μl. 5 μl cDNA solution was PCR amplified with primers GL599 and GL600. The bands obtained were extracted with the Qiaquik Gel Extraction kit (Qiagen) and sequenced.

Full-length cgt-1::gfp and cgt-3::gfp under the control of their respective endogenous promoter were generated by fusion PCR (Hobert, 2002) using the following primers and pPD95.75 as template for gfp::3′UTR-unc-54 sequences. For cgt-1: GL171, GL410, GL411 and GL412. For cgt-3: GL410, GL459, GL460 and GL461. myo-2p::cgt-1 was obtained by inserting an NheI-SacI fragment containing the cgt-1 coding sequence between myo-2 promoter and unc-54 3′ UTR. elt-2p::cgt-1 was made by PCR amplifying the elt-2 promoter with primers GL664 and GL665 and cgt-1::unc-54 3′UTR with primers GL666 and GL409. The two fragments were fused by PCR with primers GL667 and GL410. A similar procedure was used for unc-119p::cgt-1 and unc-119::cgt-3, using primers GL545 and GL549 (to amplify the unc-119 promoter for cgt-1) or GL545 and GL548 (unc-119 promoter for cgt-3), GL552 and GL409 (cgt-1 coding region) or GL551 and GL409 (cgt-3 coding region) and GL546 andGL410 to fuse the fragments. Similarly, myo-3p::cgt-3 was obtained with primers GL540 and GL543 (myo-3 promoter for cgt-3), GL551 and GL409 (cgt-3 coding region) and GL541 and GL410 (to fuse fragments). A similar procedure was used to obtain F54D5.1p::cgt-1 and F54D5.1p::cgt-3 using primers GL565 and GL569 (F54D5.1 promoter for cgt-1) or GL565 and GL568 (F54D5.1 promoter for cgt-3), GL551 and GL409 (cgt-3 coding region) or GL552 and GL409 (cgt-1 coding region) and GL566 and GL410 (to fuse fragments).

To test directly where F54D5.1p drives expression, we used the same strategy described above to make F54D5.1p::gfp. Primers GL565 and GL582 were used to amplify the F54D5.1 promoter. gfp was amplified from plasmid pPD95.75 with primers GL408 and GL409. The two fragments were fused by PCR with primers GL566 and GL410. We observed GFP expression in the PIV, IRV, the RGCs, often in the most anterior and posterior intestinal cells and, rarely, in the cuticle basal membrane. Constructs were injected (Mello et al., 1991), along with the appropriate markers, in the gonad of various cgt mutant or N2 mothers at 0.5-50 ng/μl. Their presence was verified by PCR. Rescue was determined by the ability of the transgenic lines to grow (see above).

For double-stranded RNA generation in vitro, an 850 bp fragment of cosmid F59G1 (oligos GL134 and GL135) and an 800 bp fragment of cosmid T06C12 (oligos GL147 and GL148) were PCR amplified and cloned into pBluescript in sense (pGL179 and pGL192) and antisense (pGL180 and pGL193) orientation. Single-stranded RNA was produced in vitro from each plasmid, using T3 and T7 RiboMax kits (final volume 20 μl). Sense and antisense RNA strands were mixed, extracted with phenol-chloroform and precipitated. They were then denatured at 68°C for 10 minutes and allowed to anneal at 37°C for 30 minutes. Double-stranded RNA was resuspended in 10 μl TE, loaded into a needle and injected in the gonad.

For RNAi by feeding, a NotI-XhoI fragment from pGL179 or from pGL192 was ligated with plasmid L4440 (a gift from Andrew Fire, Stanford University, CA). L4440 carries two promoters for T7 RNA polymerase facing each other (Timmons and Fire, 1998). To visualize the M cells, the integrated hlh-8::gfp reporter transgene ayIs7 from strain PD4667 was used (Harfe et al., 1998). Larvae hatch with one M cell that divides after 5-6 hours at 20°C resulting in two GFP-positive M cells. These divide again 8-9 hours after hatching (4 GFP-positive M cells), 10-11 hours after hatching (8 GFP-positive M cells) and again 13-14 hours after hatching (16 GFP-positive M cells). Then two of these 16 M cells divide again three times in L3 and L4 (Sulston and Horvitz, 1977).

Progeny of mothers exposed to siRNA for 3 days (Ahringer, 2006) was used for analysis. Red fluorescent microspheres (0.16 μm, Duke Scientific Corporation, Palo Alto, CA) were mixed with food and spread on NGM plates (Kormish and McGhee, 2005). The diameter of these beads is approximately five times smaller than that of E. coli, the bacteria that the worms feed on. 1-day-old L1 animals from strains LZ163 and LZ196 previously exposed or not to cgt-3(RNAi) were transferred on these plates and allowed to feed for 2 hours before observation.

Live animals immobilized in 10 mM levamisole (Sigma, Poole, UK) or directly transferred on 5% noble agar pads on ice (daf-16::gfp worms) were observed on an Axioplan 2 epifluorescence microscope (Zeiss, Oberkoken, Germany) and images were captured using Openlab software (Improvision, Coventry, UK). Confocal microscopy was performed on a Leica TCS SP5 microscope (Leica Microsystems, Milton Keynes, UK) with a ×63 oil objective. Metamorph (Molecular Devices Ltd, Wokingham, UK), and ImageJ 1.40 software (National Institutes of Health, Bethesda, MD) were used for three-dimensional reconstructions and measurements.

Wild-type L1 and arrested nematodes were prepared in parallel, fixed by immersion in buffered aldehydes and stained in osmium tetroxide (Hall, 1995). Fixed specimens were washed in water, stained in 1% uranyl acetate, dehydrated through an ethanol series, washed in propylene oxide, and embedded in epoxy resin. Ribbons of ultrathin (40 nm) serial sections were collected. Images were obtained on a Tecnai G2 Spirit electron microscope (FEI UK Ltd, Cambridge, UK) using an Olympus Morada digital camera (Olympus Soft Imaging Solutions, Münster, Germany).

Lipids were extracted from L1-arrested animals or from just starved mixed populations of worms containing mostly L1 animals and analysed by thin-layer chromatography, as previously described (Doering et al., 1999). Lipids corresponding to 100 μg worm protein were applied in each lane and chromatograms were developed in chloroform:methanol:water 70:30:5. After development, plates were air-dried, sprayed with 8% (w/v) phosphoric acid containing 10% (w/v) copper sulphate, and charred at 180°C for 10 minutes. Relative intensities of GlcCer bands were deduced by quantifying their total number of pixels using Adobe Photoshop (Adobe Systems, Uxbridge, UK). The bands used for quantification (Fig. 4A, arrowhead) were those migrating above the major bacterial band. Lipids from E. coli HT115 (DE3) bacteria were included because this strain was used for RNAi by feeding. The two bands appearing in the bacteria lane are unlikely to be GSLs or ceramides because, to our knowledge, E. coli do not produce these lipids.

Proteins were aligned using ClustalX (http://www.clustal.org) and shaded with MacBoxshade (written by Kay Hoffman, Miltenyi Biotec GmbH, Bergisch Gladbach, Germany, and Michael Baron, Institute for Animal Health, Pirbright, UK). Significance was tested with Student's t-test or Chi-Square test, depending on the experiment, using Prism (GraphPad Software, San Diego, CA).