Eleven residues determine the acyl chain specificity of ceramide synthases

Lipids display large structural complexity, with ∼40,000 different lipids identified to date, ∼4000 of which are sphingolipids. A critical factor determining the biological activities of the sphingolipid, ceramide, and of more complex sphingolipids is their N-acyl chain length, which in mammals is determined by a family of six ceramide synthases (CerS). Little information is available about the CerS regions that determine specificity toward different acyl-CoA substrates. We previously demonstrated that substrate specificity resides in a region of ∼150 residues in the Tram-Lag-CLN8 domain. Using site-directed mutagenesis and biochemical analyses, we now narrow specificity down to an 11-residue sequence in a loop located between the last two putative transmembrane domains (TMDs) of the CerS. The specificity of a chimeric protein, CerS5(299–309→CerS2), based on the backbone of CerS5 (which generates C16-ceramide), but containing 11 residues from CerS2 (which generates C22–C24-ceramides), was altered such that it generated C22–C24 and other ceramides. Moreover, a chimeric protein, CerS4(291–301→CerS2), based on CerS4 (which normally generates C18–C22 ceramides) displayed significant activity toward C24:1-CoA. Additional data supported the notion that substitutions of these 11 residues alter the specificities of the CerS toward their cognate acyl-CoAs. Our findings may suggest that this short loop may restrict adjacent TMDs, leading to a more open conformation in the membrane, and that the CerS acting on shorter acyl-CoAs may have a longer, more flexible loop, permitting TMD flexibility. In summary, we have identified an 11-residue region that determines the acyl-CoA specificity of CerS.

Recent advances in analytical techniques, such as MS, have revealed an unexpected complexity in lipid structures (1), with ϳ40,000 lipids identified to date (http://www.lipidmaps.org). 5 Of these, ϳ4000 individual sphingolipid (SL) 6 structures have been identified (2). SLs differ in the headgroups added at the C1 position, the length and degree of saturation of the sphingoid long-chain base, and of the acyl chain that is attached to the sphingoid backbone via an amide bond at C2 (3,4).
The acyl chain composition of ceramide, the simplest SL, is determined by the activity of a family of ceramide synthases (CerS), of which six are found in mammals (5,6), with each generating ceramides of a different acyl chain length (7). Since the identification of the first mammalian CerS in 2002 (8), some effort has been invested in attempting to determine the basis for the acyl chain specificity of the mammalian CerS; thus, it is known that the amino acids involved in acyl chain specificity are located in the Tram-Lag-CLN8 (TLC) domain (9), a region of ϳ200 residues also found in other proteins (10). The active site was suggested to reside within a region known as the Lag1p motif, containing two conserved histidine residues (11,12), which is within the TLC domain (13,14). Other regions of the CerS include the homeobox (Hox)-like domain (14) that is found in CerS2-S6 (6) and contains residues that are critical for activity (15), but they do not appear to play a role in determining acyl-CoA specificity (9).
CerS are located in the endoplasmic reticulum (ER) membrane (6), although the exact number of transmembrane domains (TMDs) and their topology have not been resolved experimentally (16). Although most prediction programs give similar suggestions for the topology of the N and C termini, there is a significant disagreement about almost every other feature of CerS topology (3). That being said, recent experimental data support an odd number of TMDs (17).
As stated above, we previously determined that the domain responsible for acyl chain specificity resides in a minimal region of 150 residues in the TLC domain (9). This was shown by generation of chimeric proteins consisting of portions of CerS2 (which generates C22:0 -C24:1-ceramides) and CerS5 (which generates C16:0-ceramide) and consequently examining which ceramide species were generated by the chimeric proteins. We now provide data in which we considerably narrow down the region involved in determining acyl chain specificity, and we identify a region of 11 residues that appears to be critically involved in determining the acyl chain specificity, not only of CerS5 and CerS2 but also of CerS4.

N terminus of CerS faces the lumen of the ER
To further analyze CerS topology in the ER (16 -18), we took advantage of a conserved N-glycosylation motif (NX(T/S)) found in CerS2 and CerS4 -CerS6. This sequence is not found in CerS3, and CerS1 does not align with the other CerS at the N terminus (Fig. 1A). Consequently, upon treatment of CerS2-CerS6 with endoH, a decrease in molecular weight consistent with deglycosylation was observed for all CerS except CerS3 (Fig. 1B). Interestingly, CerS5 activity increased upon endoH treatment (1189 Ϯ 63 pmol/min/mg before endoH treatment and 1768 Ϯ 204 pmol/min/mg after endoH treatment; n ϭ 3). Mutation of the key asparagine residue within the N-glycosyl-ation motif in CerS5 (CerS5(N26Q)) resulted in a similar reduction in molecular weight (data not shown). However, a small decrease in CerS5 activity (1887 Ϯ 325 pmol/min/mg for CerS5 and 1353 Ϯ 327 pmol/min/mg for CerS5(N26Q)) was observed, perhaps suggesting an important role for the N-glycan in protein folding. Irrespective of the effect of deglycosylation on CerS activity, these data unambiguously demonstrate that the N terminus of CerS faces the lumen of the ER (Fig. 1C), where the first step in N-glycosylation takes place (20), as suggested in some but not all previous topological predictions (9,17,18). This is of relevance to the studies below where specific sequences and motifs are modified, based to some extent on the putative topology of the CerS.

Acyl-CoA specificity is associated with one of two frameshifts in CerS2 and CerS5 sequences
Pairwise alignment of human CerS2 and CerS5 revealed two frameshifts in regions of low sequence similarity. The first is in the second putative TMD, between residues 156 and 165 of CerS5, and the other frameshift is in the putative last loop between residues 299 and 309 of CerS5 ( Fig. 2A). CerS2 and CerS5 belong to two different branches of the CerS phyloge- The N-glycosylation consensus sequence in CerS2 and CerS4 -6 is in red. B, cell lysates (2 mg of protein/ml) from HEK cells overexpressing the indicated CerS-HA were incubated with or without endoH prior to Western blotting using an anti-HA antibody. Molecular weight markers are indicated. Tubulin was used as a loading control. This experiment was repeated three times with similar results. C, revised putative topology of the CerS. Unlike previous models (9), we now suggest that the N terminus of the CerS faces the ER lumen and that the fourth TMD does not fully penetrate the ER membrane.

Acyl chain specificity of ceramide synthases
netic tree (9), which can essentially be defined based on these frameshifts.
Two sets of chimeric proteins were generated ( Fig. 2B) to determine whether the frameshifts play a role in defining the acyl-CoA specificities of Cer2 and CerS5. Thus, in CerS5 (153-1653CerS2) , all of the sequence is derived from CerS5 with the exception of residues 153-165, which are from CerS2. To analyze acyl chain specificity, each chimeric protein was assayed under optimal kinetic conditions for the parent CerS using the relevant acyl-CoA. Thus, CerS2 activity was assayed using very-long-chain acyl-CoAs and a reaction time of 20 min with 40 g of protein, whereas CerS5 activity was assayed with C16-acyl-CoA using 1 g of protein for 5 min (21,22).

Critical role for the putative last loop of CerS in determining acyl-CoA specificity
We generated CerS2 Ϫ/Ϫ HEK293T cells using CRISPR/Cas9 technology. The main motivation for doing this was to refute the possibility that apparent changes in acyl-CoA specificity could be due to more efficient dimerization (17) of CerS5 (299 -3093CerS2) with endogenous CerS2, leading to an increase in CerS2 activity rather than to altered acyl-CoA specificity of CerS5 (299 -3093CerS2) . A HEK293T clone, which contained a deletion of 10 nucleotides in both alleles of the genomic DNA sequence, was obtained, resulting in a premature stop codon after 63 residues (Fig. 4A). As a consequence, no CerS2 protein was generated (Fig. 4B), and no very-long acyl chain ceramides were synthesized (Fig. 4C) or detected in the cells (Fig. 4D). The SL profile of the CerS2 Ϫ/Ϫ HEK293T cells was similar to that of the CerS2 null mouse (23, 24), with decreased levels of very-long acyl chain ceramides and increased levels
We next characterized the activity of CerS5 (299 -3093CerS2) toward various CoAs. An obstacle in using an in vitro assay for a chimeric protein composed of different regions of different CerS is that each of the parent CerS exhibits different kinetic parameters with respect to reaction time and amount of protein (21,22). To attempt to overcome this complication, we assayed CerS5 (299 -3093CerS2) activity with increasing amounts of protein while utilizing different acyl-CoAs (Fig. 5A). CerS5 (299 -3093CerS2) generated C22:0-ceramide and C16:0-ceramide to a similar extent 7 but also generated C18:0-, C20:0-, and C24:1-ceramides (Fig. 5A), suggesting that CerS5 (299 -3093CerS2) has lost its specificity toward C16-CoA and shows specificity toward all of the CoAs tested. Importantly, upon increasing amounts of C22-CoA, CerS5 (299 -3093CerS2) generated more C22-ceramide than CerS2 itself (Fig. 5B), indicating that under assay conditions optimized for CerS2, CerS5 (299 -3093CerS2) generates bona fide amounts of very-long-chain ceramides. However, the SL profile of HEK cells upon overexpression of CerS5 (299 -3093CerS2) 7 The apparent lack of activity of CerS5 (299 -3093CerS2) toward C16-CoA in Fig. 3B was due to short assay times and the low amounts of protein that were used, corresponding to optimal assay conditions of 1 g of protein for 5-min reaction time that is normally used for CerS5, whereas activity towards C16-CoA in Fig. 5A was assayed using more protein (5-60 g) and for a longer reaction time (20 min), during which time more C16-CoA is formed.

Acyl chain specificity of ceramide synthases
did not change significantly compared with controls ( Fig. 5C) with the exception of a small increase in C14-and C18-ceramide; however, levels of C16-ceramide, which is generated by CerS5, were unaltered (25).

Discussion
Studies performed ϳ25 years ago indicated that ceramide is generated in the ER (26,27), but it was only when CerS was identified that their localization in the ER membrane could be unambiguously determined (8, 17, 18, 29). However, solubilization and reconstitution of the CerS have proved extremely dif-ficult (17,31), and therefore structure-function studies are few and far between.
In this study, we extend our earlier work and suggest a new putative topology in which the N terminus is lumenal, and the fourth TMD is unlikely to completely cross the ER membrane (Fig. 1C). This revised topology, along with the discovery of an unexpected frameshift when aligning CerS2 and CerS5, led us to focus on the loop between the 5th and 6th TMDs and to the principal finding of this study, namely that CerS specificity can be altered by modifying the sequence of only 11 amino acid residues within this lumenal loop. How the two CerS substrates gain access to this loop is currently unknown, but it should be noted that acyl-CoAs are generated in the cytosol, whereas sphingosine is presumably generated within the membrane.
Whether changes in this loop cause loss of specificity or gain of activity is unclear because in vitro assays using CerS5  indicate that this chimeric protein is able to generate ceramides with a variety of different acyl chain lengths, albeit under conditions optimized for the activity of CerS2, which requires a longer reaction time and more protein to obtain similar levels of enzymatic activity as CerS5, which uses shorter acyl chains. Under these conditions, C22:0-cer-

Acyl chain specificity of ceramide synthases
amide is the main ceramide generated; moreover, when using increasing amounts of C22:0 acyl-CoA in the reaction mixture, CerS5 (299 -3093CerS2) is more effective in generating C22:0-ceramide than the cognate CerS2. Thus, altering 11 residues in the CerS5 sequence unequivocally allows it to generate C22:0-ceramide, although CerS5 (299 -3093CerS2) is also able to generate other ceramides. The reason that C22:0-ceramide levels were not elevated in cells upon overexpression may be related to the availability of acyl-CoAs in cells or to a currently unknown level of regulation of chimeric CerS upon their overexpression.
Because the structure of the CerS is not known, any discussion of how altering these residues affects specificity is by definition somewhat speculative. All topology predictions are consistent with this region forming a loop (Fig. 7A and Fig. S7A), but we are unable to determine experimentally at this stage how the loop folds to determine acyl-CoA specificity. An obvious proposition would be that the very-long-chain acyl-CoAs, such as those required by CerS2, might fit in a more extended loop, whereas the long-chain acyl-CoAs, such as those required by CerS5, might fit in a more compact loop. However, the opposite is true, as the CerS5 loop contains more residues than the CerS2 loop and thus the adjacent TMDs might be shorter (Fig.  7). Moreover, examination of the TMDs adjacent to the loop indicate that the loop in WT CerS5 and CerS6 consists of 15 and 16 residues, CerS1 and CerS4, 21 and 20 residues, respectively, and CerS2 and CerS3, only 11 and 9 residues (Fig. 7A). This indicates that the CerS that utilize the longest acyl-CoAs have the shortest number of residues in this loop. We suggest that the short loop may restrict the positioning of the adjacent TMDs such that they have a more open conformation in the membrane, whereas the CerS that use shorter acyl-CoAs may have a longer and more flexible loop affecting the flexibility of the TMDs and thus permitting use of shorter length acyl-CoAs (Fig. 7B). TMD prediction of the chimeric proteins (Fig. 7A) is consistent with this suggestion as the TMDs are intermediate in length. Importantly, the CerS2 (291-3013CerS5) chimera has a longer loop (18 versus 11 residues) consistent with its ability to accommodate C16-CoA. The suggestions of a relationship between the number of residues in the loop and acyl-CoA specificity remain speculative but are nevertheless consistent with the altered specificities of the chimeric proteins.
In summary, we have identified a critical region that appears to determine the acyl-CoA specificity of the CerS. The precise mechanistic details of how this lumenal region gains accessibility to acyl-CoAs, which are generated in the cytosol, and the relationship to the CerS active site, await insight into the threedimensional structure of the CerS.

Acyl chain specificity of ceramide synthases
anti-CerS2 antibodies, and polyethyleneimine were from Sigma. Horseradish peroxidase was from The Jackson Laboratory (Bar Harbor, ME). An ECL detection system was from Thermo Fisher Scientific (Waltham, MA). EndoH was from New England Biolabs Inc. (Ipswich, MA). Silica gel 60 TLC plates were from Merck (Billerica, MA). All solvents were of analytical grade and were purchased from Bio-Lab (Jerusalem, Israel).
The terminology used for each construct is given in Fig. 2B. All sequences were confirmed prior to use.

Cell culture and transfection
HEK293T cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 100 IU/ml penicillin, 100 g/ml streptomycin, and 110 g/ml sodium pyruvate. Transfections were performed with the polyethyleneimine reagent using 8 g of plasmid per 10-cm culture dish. 36 -48 h after transfection, cells were removed from culture dishes and washed twice with PBS. Cell homogenates were prepared in 20 mM Hepes-KOH, pH 7.2, 25 mM KCl, 250 mM sucrose, and 2 mM MgCl 2 containing a protease inhibitor mixture. Protein was determined using the Bradford reagent (Bio-Rad).

Western blotting
Proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes. HA-tagged constructs were identified using a mouse anti-HA antibody (1:10,000) and goat anti-mouse horseradish peroxidase (1:10,000) as the secondary antibody. Equal loading was confirmed using a mouse anti-tubulin antibody. Detection was performed using the ECL detection system.

Ceramide synthase assays
Cell homogenates were incubated with 15 M NBD-Sph, 20 M defatted BSA, and 50 M fatty acyl-CoA in a 20-l reaction volume at 37°C for various times and using various amounts of tissue homogenate as indicated in the figure legends. Reactions were terminated by addition of chloroform/methanol (1:2, v/v) and lipids extracted (34). Lipids were dried under N 2 , resuspended in chloroform/methanol (9:1, v/v), and separated by TLC using chloroform/methanol, 2 M NH 4 OH (40:10:1, v/v/v) as the developing solvent. NBD-labeled lipids were visualized using a Typhoon 9410 variable mode imager and quantified by ImageQuantTL (GE Healthcare, Chalfont St Giles, UK).

Liquid chromatography electrospray ionization tandem MS
SL analysis by LC-ESI-MS/MS using an ABI 4000 quadrupole-linear ion trap mass spectrometer was performed as described previously (36).

Prediction of transmembrane domains
Transmembrane domain prediction was performed with the following programs using default parameters: DAS TMfilter