Modulation of Ceramide Synthase Activity via Dimerization*

Background: Ceramide is synthesized by a family of six ceramide synthases (CerS), which use different acyl-CoAs for N-acylation of the sphingoid base. Results: CerS form homo- and heterodimers, which regulate ceramide synthesis. Conclusion: CerS activity is modulated by dimer formation. Significance: The acyl chain composition of ceramide in different tissues may depend on interaction between different CerS. Ceramide, the backbone of all sphingolipids, is synthesized by a family of ceramide synthases (CerS) that each use acyl-CoAs of defined chain length for N-acylation of the sphingoid long chain base. CerS mRNA expression and enzymatic activity do not always correlate with the sphingolipid acyl chain composition of a particular tissue, suggesting post-translational mechanism(s) of regulation of CerS activity. We now demonstrate that CerS activity can be modulated by dimer formation. Under suitable conditions, high Mr CerS complexes can be detected by Western blotting, and various CerS co-immunoprecipitate. CerS5 activity is inhibited in a dominant-negative fashion by co-expression with catalytically inactive CerS5, and CerS2 activity is enhanced by co-expression with a catalytically active form of CerS5 or CerS6. In a constitutive heterodimer comprising CerS5 and CerS2, the activity of CerS2 depends on the catalytic activity of CerS5. Finally, CerS dimers are formed upon rapid stimulation of ceramide synthesis by curcumin. Together, these data demonstrate that ceramide synthesis can be regulated by the formation of CerS dimers and suggest a novel way to generate the acyl chain composition of ceramide (and downstream sphingolipids), which may depend on the interaction of CerS with each other.

The CerS are multispanning membrane proteins (7)(8)(9), although the exact number of transmembrane domains is still uncertain, with estimates ranging from 5-8 (discussed in Ref. 9). Recently, the active site of the CerS was localized to a region of 150 residues in the Tram-Lag-CLN8 domain (9). However, the active site residues have not been experimentally determined, although site-directed mutagenesis of a number of residues (10,11) abrogates CerS activity, including two conserved histidines that are presumed to be involved in the N-acyltransferase reaction, based on homology to other acyl-CoA transferases (12,13).
Little is known about how CerS activity is regulated. CerS mRNA is differentially expressed in different tissues (5)(6)(7), but the sphingolipid acyl chain composition does not always correlate with CerS expression (5,14), prompting suggestions of cross-regulation of CerS expression and/or activity (15). In addition, post-translational mechanisms of CerS regulation likely exist, which might facilitate the rapid changes in CerS activity observed upon various stimuli and under various conditions (16 -26).
CerS can be phosphorylated (27) and glycosylated (28), and it was suggested recently that CerS2, -5, and -6 exist as heterocomplexes in HeLa cells (22). We now explore the possibility that CerS activity is regulated by dimer formation and demonstrate that CerS can form both homo-and heterodimers. Moreover, we have generated constitutive dimers and demonstrate that the activity of one member of a heterodimer depends upon and can be modulated by the activity of the other member. Together, our results suggest a rapid and reversible mechanism of regulating CerS activity involving dimer formation.
Cloning-Human CerS2 with an HA tag at the C terminus (CerS2-HA) was cloned as described (5). Primers for subcloning are given in Table 1. A CerS5-HA fragment was amplified by PCR using primers A and B. The PCR product was digested with KpnI and EcoRI followed by ligation into a pcDNA3 vector. A CerS5 constitutive dimer was synthesized by Genscript (Piscataway, NJ) after modifying and optimizing the coding sequence for use with a mammalian expression system (supplemental Fig. 1). The sequence was subcloned into a pcDNA3 vector and an HA tag added at the 3Ј end (CerS5:CerS5-HA) using primers C and D ( Table 1). Insertion of a transmembrane (TM) domain 3 between the two monomers of the dimer (CerS5:TM:CerS5-HA) was performed using primers E and F (supplemental Fig. 2). Mutagenesis of two His residues in CerS5 (H220A and H221A) in the N-terminal monomer (CerS5 HH : TM:CerS5-HA) was performed using primers G and H, and mutagenesis of H220A and H221A in the C-terminal monomer (CerS5:TM:CerS5 HH -HA) using primers I and J ( Table 1). Cloning of a CerS5:CerS2 heterodimer (CerS5:TM:CerS2-HA) was performed using primers K and L to amplify CerS2; a second PCR reaction was performed in which the CerS2 fragment replaced the 3Ј CerS5 fragment by restriction-free cloning (30). Mutagenesis of CerS5:TM:CerS2-HA was performed using primers E and F (supplemental Fig. 3). A truncated CerS5 chimera was cloned using primers M and N, followed by restriction cleavage using BamHI and ECoRI and ligation into a PCMV-2B vector (CerS5 ⌬C332-392 ). The sequences of all constructs were confirmed prior to use.
Cell Culture and Transfection-Human embryonic kidney (HEK) 293 and HepG2 cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 100 IU/ml penicillin and 100 g/ml streptomycin. Transfection with the various CerS constructs was performed using the PEI transfection reagent (Sigma). Twenty four hours after transfection, cells were collected by trypsinization and stored at Ϫ80°C. Expression levels after transfection were confirmed by Western blotting.
Cross-linking with Formaldehyde-HEK cells were transfected with FLAG-CerS (31). Twenty four hours after transfection, live cells were incubated with 1% formaldehyde for 15 min. Cross-linking was terminated by addition of 125 mM glycine for 5 min. Cells were collected and CerS dimer formation analyzed by Western blotting.
Co-immunoprecipitation of CerS-Cells were co-transfected with CerS5-HA and FLAG-CerS1-6. Twenty four hours after transfection, cells were collected, homogenized in lysis buffer (150 mM NaCl, 50 mM Tris-HCl, pH 7.4, 1% Nonidet P-40) containing a protease inhibitor mixture, and centrifuged (2,700 ϫ g, 4°C, 10 min). Pellets (containing the non-soluble material) were discarded, and the supernatant was incubated for 5 h with 2 g of an anti-FLAG antibody in an orbital shaker at 4°C. Following incubation, 30 l of agarose beads conjugated to protein A (Bio-vision, Milpitas, CA) was added overnight. Beads were collected subsequently by centrifugation and washed three times with lysis buffer. Proteins were removed from the beads by boiling in sample buffer (30 mM Tris-HCl, pH 6.8, 0.7 M glycerol, 35 mM sodium dodecyl sulfate, 0.1 M dithiothreitol, 0.14 mM bromphenol blue). Eluates were blotted with either anti-FLAG or anti-HA antibodies (1:10,000 dilution). A similar procedure was performed for endogenous CerS2 and CerS6 in HepG2 cells using anti-CerS2 and anti-CerS6 antibodies (both at a dilution of 1:1,000).
Western Blotting-Western blotting was performed as described (5,9). When cross-linking was performed using formaldehyde prior to Western blotting, homogenates were not boiled in the sample buffer. When non-denaturing conditions were used, proteins were treated with a non-reducing sample buffer (30 mM Tris-HCl, pH 6.8, 0.7 M glycerol, 35 mM sodium dodecyl sulfate, 0.14 mM bromphenol blue) and heated to 50°C for 5 min. Protein levels were quantified by densitometry.
CerS Reconstitution-CerS5-HA and CerS5:TM:CerS5-HA were reconstituted in liposomes by a similar method used previously for CerS5 (39). Briefly, following expression of either of the two constructs in HEK cells, protein was solubilized using 20 mM HEPES KOH, pH 7.4, 25 mM KCl, 250 mM sucrose, 2 mM MgCl 2 , and 1% digitonin containing a protease inhibitor mixture. The proteins were incubated for 3 h with an anti-HA antibody followed by overnight incubation with agarose beads conjugated to protein A. The beads were washed three times with the solubilization buffer, and protein was eluted using 100 mM glycine, pH 2.5; 1 M Tris buffer, pH 11, was added immediately after elution. Eluted proteins were concentrated using centrifugal filter units (Millipore) and were reconstituted with 1,2dioleoyl-sn-glycero-3-phosphocholine liposomes.
Immunofluorescence-The intracellular localization of various HA-tagged CerS constructs was determined by confocal laser scanning microscopy (6) using Bip as an endoplasmic reticulum marker.

CerS Form High Molecular Weight Complexes and Co-
Immunoprecipitate-To determine whether CerS can oligomerize, HEK cells overexpressing each of the FLAG-tagged CerS were treated with formaldehyde for 15 min. Upon analysis by Western blotting, a high M r band was detected, which was approximately twice the M r of monomeric FLAG-CerS (Fig.  1B). Bands with a similar M r could be detected when samples were treated with a non-reducing sample buffer and heated to 50°C (Fig. 1C). No higher M r bands could be detected in Western blots under denaturing conditions (Fig. 1A), consistent with earlier studies examining CerS expression under denaturing conditions (7,10,39). These results demonstrate that higher M r CerS complexes can be detected under suitable conditions. HEK cells were next co-transfected with CerS5-HA and each of FLAG-CerS1-6. The CerS were then immunoprecipitated using an anti-FLAG antibody and HA-CerS5 detected in the immunoprecipitates by Western blotting using an anti-HA antibody. HA-CerS5 was co-immunoprecipitated by each of the six FLAG-CerS ( Fig. 2A) albeit in varying amounts; similar results were obtained using the anti-HA antibody for the immunoprecipitation step (data not shown). To confirm that the co-immunoprecipitation was not due to overexpression, endogenous CerS2 and CerS6 were immunoprecipitated using anti-CerS2 and anti-CerS6 antibodies, respectively. CerS2 was able to immunoprecipitate CerS6 and vice versa (Fig. 2B).
CerS Interact in Vivo-To determine whether dimerization modulates CerS activity, HEK cells were co-transfected with various combinations of CerS. In the first experiment, cells were transfected with a catalytically active full-length CerS5-HA together with increasing amounts of a catalytically inactive FLAG-CerS5 (FLAG-CerS5 ⌬C332-392 ) lacking the last putative transmembrane domain. The two proteins were coimmunoprecipitated (data not shown) using both anti-FLAG and anti-HA antibodies for immunoprecipitation. CerS5-HA activity decreased upon transfection with increasing amounts of FLAG-CerS5 ⌬C332-392 (Fig. 3), demonstrating that FLAG-CerS5 ⌬C332-392 acts in a dominant-negative fashion to inhibit the activity of full-length CerS5-HA.
Next, HEK cells were transfected with FLAG-CerS2 and CerS5-HA or CerS6-HA and assayed for CerS2 activity using C 22 -CoA. Of all of the CerS, CerS2 displays the lowest activity in in vitro assays, and there is often a lack of apparent correspondence between levels of CerS2 expression and levels of enzyme activity (5,31,40). When FLAG-CerS2 was expressed by itself, C 22 -ceramide synthesis was elevated by ϳ50% (Fig. 4) (5). However, upon co-expression of FLAG-CerS2 with either CerS5-HA or CerS6-HA, a significant increase in CerS2 activity was observed (Fig. 4), consistent with the notion that CerS interact with each other and that some CerS can modulate the activity of other CerS.
CerS Activity Is Modified in CerS-constitutive Dimers-To directly determine the regulation of CerS activity by dimer formation, a series of constitutive CerS dimers were generated. The first dimer that was generated, consisting of two monomers of CerS5 directly attached to each other via the N terminus of one monomer and the C terminus of the other monomer (CerS5:CerS5-HA), did not display catalytic activity (Fig. 5A). However, insertion of a transmembrane domain between the two monomers (CerS5:TM:CerS5-HA) generated a dimer with somewhat higher levels of activity than the CerS5-HA monomer (Fig. 5A). After immunoprecipitation and reconstitution in 1,2-dioleoyl-sn-glycero-3-phosphocholine liposomes, CerS5:TM:CerS5-HA displayed ϳ6-fold higher activity than CerS5-HA (Fig. 5B). As expected, mutation of two residues (H220A and H221A) involved in catalytic activity (CerS5 HH :TM:CerS5 HH -HA) completely abrogated CerS5 activity in the constitutive dimer (Fig. 5A).

FIGURE 2. Detection of CerS heterodimers by co-immunoprecipitation.
A, HEK cells were co-transfected with CerS5-HA and FLAG-CerS1-6. CerS were subsequently immunoprecipitated using an anti-FLAG antibody and detected by Western blotting using either an anti-FLAG antibody (upper panel) or an anti-HA antibody (lower panel). B, CerS2 and CerS6 were co-immunoprecipitated from HepG2 cells using antibodies to endogenous CerS. Each experiment was repeated three times with similar results.  CerS5:CerS2 heterodimers were also generated and assayed for activity using C 16 -CoA (for CerS5) and C 22 -CoA (for CerS2). CerS5:TM:CerS2-HA displayed slightly more activity using C 16 -CoA as substrate than CerS5, but remarkably, CerS2 activity measured using C 22 -CoA was elevated by ϳ3-fold (Fig.  6). This increase in CerS2 activity was abolished using a noncatalytically active form of CerS5 in the constitutive dimer (CerS5 HH :TM:CerS2-HA), demonstrating that optimal CerS2 activity depends on an interaction with a catalytically active form of CerS5.
To exclude the possibility that the modulation of CerS2 activity by CerS5 is due to altered subcellular localization, the intracellular localization of CerS5:TM:CerS5-HA and CerS5: TM:CerS2-HA were compared with that of CerS5. No differences in intracellular localization were detected (Fig. 7).
CerS Dimers Are Formed Rapidly upon Stimulation of Ceramide Synthesis-Finally, we examined the relevance of CerS dimer formation for ceramide synthesis in vivo. HEK cells overexpressing FLAG-CerS2 or FLAG-CerS5 were incubated with curcumin, which rapidly activates ceramide synthesis via the de novo pathway (19,23). Within 5 min after curcumin treatment, high M r bands corresponding to FLAG-CerS2 and FLAG-CerS5 dimers could be detected (data not shown), which became more pronounced after 30 min and remained elevated for up to 180 min (Fig. 8A). Next, ceramide synthesis was measured in non-transfected curcumin-treated cells, in which a significant increase of [4, H]sphinganine incorporation into [ 3 H]ceramide (Fig. 8B) was observed. Furthermore, levels of various ceramide species (particularly C20-, C18-, C26:1) were elevated after curcumin treatment) (Fig. 8) consistent with the idea that curcumin treatment induces formation of both homo-and  heterodimers. Together with the data showing dimer formation in overexpressing cells, these results imply that curcumin elevates ceramide synthesis by the rapid formation of CerS dimers.

DISCUSSION
In the current study, a variety of techniques were used to demonstrate that CerS activity is regulated by dimer formation. This suggestion is consistent with earlier studies demonstrating co-immunoprecipitation of some CerS in both mammalian cells (22) and in yeast (41) and with a kinetic study examining the mode of inhibition of CerS by FTY720 (42). In the latter, the uncompetitive inhibition of sphinganine binding to CerS by FTY720 suggested that there may be two sphinganine-binding sites that act allosterically with respect to one another, or that CerS may form dimers that interact allosterically.
The current study is consistent with a model in which the binding of a substrate to one monomer of the dimer allosteri-  cally affects binding to the other monomer. By way of example, CerS2 has a K m toward sphinganine of ϳ5 M, whereas the other CerS have a K m of ϳ2 M (31). Thus, CerS2 activity is relatively low by itself but increases significantly either when co-expressed with another CerS or expressed as a constitutive dimer. Thus, the binding of either sphinganine and/or C 16 -CoA to CerS5 (Fig. 9) would cause a conformational change in CerS2 that increases the affinity of CerS2 toward sphinganine and/or C 22 -CoA, and thus enhances its activity. Alternatively, one substrate binding site might be on one monomer and the other on the other monomer (a trans interaction) (Fig. 9).
Irrespective of the precise details of the mode of interaction of the CerS in the dimer, or of the order and location of the substrate binding sites, the finding that this interaction modulates CerS activity is of great importance for understanding ceramide and CerS biology. Previously, it had been assumed that the acyl chain composition of ceramide (and consequently of downstream sphingolipids) is likely to simply reflect the CerS expression pattern in a particular tissue (see Ref. 5 for details). However, this is now unlikely to be the major determinant of acyl chain composition, which will rather depend on the com-binatorial expression of the different CerS and their ability to interact. Moreover, CerS splice variants, many of which are unlikely to have activity by themselves (2), could act in a dominant-negative manner to regulate CerS activity. Specifically, we suggest that maximal CerS2 activity depends on its interaction with other CerS; of all of the CerS, CerS2 has the lowest activity in vitro (43) but has the widest tissue distribution and is the main CerS responsible for synthesis of very long acyl chain (C 22-24 ) ceramides (5).
The yeast CerS, Lac1 and Lag1, form a high M r complex, but the role of possible homo-or heterodimeric Lag1-Lac1 complexes in regulating ceramide synthesis has not been evaluated. Moreover, it is unlikely that yeast require the same finesse of CerS regulation because yeast ceramides only contain one kind of fatty acid (namely C 26 ) compared with the wide variety of fatty acids found in mammalian cells. However, levels of ceramide synthesis, in both yeast and mammals, are likely to be regulated by dimer formation since the activity of a CerS5 homodimer is somewhat higher than that of the monomer.
Molecular and structural details of how the CerS interact as either homo-or heterodimers in the endoplasmic reticulum membrane are lacking. The constitutive dimers generated in this study do not help in this regard but may give clues about the topology of the CerS because an additional transmembrane domain was required for activity of the constitutive dimers; this may suggest that the CerS contain an odd number of transmembrane domains (i.e. Ref. 5, and see Ref. 7) in which the N and C termini are located on opposite sides of the endoplasmic reticulum. Indeed, in the endoplasmic reticulum, CerS may exist in equilibrium between monomers and dimers, and the formation/dissociation of the dimers might be a major way of regulating their activity. Dimers are formed relatively quickly, and this mechanism provides a rapid way to increase levels of ceramide synthesis under various physiological conditions. This could be particularly important in such cases where de novo ceramide synthesis plays a role in one or other signaling pathway and thus needs to be activated by a rapid, post-translational mechanism.
In summary, we suggest a novel means of modulating CerS activity by dimer formation and dissociation. Whether other enzymes of sphingolipid metabolism are also regulated by such means is not known (with the exception of sphingosine palmitoyl transferase, where heterodimer formation is required for activity (32)). CerS dimer formation adds a new element to our understanding of how ceramide synthesis is regulated. The left-hand panel shows how C 16 -CoA synthesis by CerS5 regulates C 22 -CoA synthesis by CerS2 in a cis manner, i.e. the sphinganine and CoA binding sites are provided by the same CerS monomer. The right-hand panel shows the same mode of regulation but in a trans manner, in which the sphinganine binding site is on one monomer and the acyl-CoA binding site is on the other monomer. In both models, C 16 -CoA first binds to CerS5 (i). As a result, the sphinganine binding site is allosterically modified (ii), followed by binding of C 22 -CoA to CerS2. The enzymatic reaction (iv) then proceeds in either a cis or trans manner.