The mammalian Golgi apparatus is composed of multiple stacks of cisternal membranes organized laterally into a polarized ribbon. Furthermore, trans-Golgi membranes come in close apposition with ER (endoplasmic reticulum) membranes to form ER–trans-Golgi contact sites, which may facilitate transfer of newly synthesized ceramide from the ER to SM (sphingomyelin) synthase at the trans-Golgi via CERT (ceramide transfer protein). CERT interacts with both ER and Golgi membranes, and together with Golgi morphology contributes to efficient SM synthesis. In the present study, we show that Golgi disassembly during pro-apoptotic stress induced by TNFα (tumour necrosis factor α) and anisomycin results in decreased levels of CERT at the Golgi region. This is accompanied by a caspase-dependent loss of full-length CERT and reduction in de novo SM synthesis. In vitro, CERT is cleaved by caspases 2, 3 and 9. Truncated versions of CERT corresponding to fragments generated by caspase 2 cleavage at Asp213 were mislocalized and did not promote efficient de novo SM synthesis. Thus it is likely that during cellular stress, disassembly of Golgi structure together with inactivation of CERT by caspases causes a reduction in ceramide trafficking and SM synthesis, and could contribute to the cellular response to pro-apoptotic stress.
Apoptosis is a highly regulated form of cell death that serves to eliminate damaged cells without causing inflammation, and is important for maintaining normal growth and development . Signals that induce cell death can come from different organelles within the cell such as the nucleus after DNA damage, or extrinsically after death receptor ligation . During this evolutionarily conserved process, damaged cells are disassembled and packaged into membrane-bound blebs, which are then phagocytosed by adjacent cells.
Some of the apoptotic machinery that mediates cleavage of cellular targets is found at the Golgi complex, suggesting that this organelle may be able to sense specific stress signals and regulate downstream signalling pathways [1,2]. Although most procaspases are localized in the cytoplasm, procaspase 2 is also localized in the nucleus and on the cytoplasmic face of the Golgi apparatus [3,4]. The localization pattern of procaspase 2 suggests that the nuclear pool of caspase 2 mediates apoptosis after DNA damage, whereas the Golgi pool mediates apoptosis in response to stress signals sensed by the secretory pathway, particularly the Golgi apparatus itself . Several known substrates of caspase 2 are localized at the Golgi, including Golgin-160  and giantin . Both of these proteins are also substrates for effector caspases 3 and 7 . Cleavage of Golgin-160 by caspase 2 is rapid and precedes cleavage by caspase 3, suggesting that caspase 2 activation at the Golgi is an early event . Many Golgi structural proteins, including GM130 (cis-Golgi matrix protein of 130 kDa) and p115, are cleaved by other caspases such as caspase 3 [6,7]. Cleavage of Golgi structural proteins promotes Golgi disassembly. First, the ribbon structure that is characteristic of mammalian cells is lost, followed by further disassembly. With disassembly of the Golgi ribbon structure, it is expected that Golgi function is also affected [8–11].
Stress signals sensed by the secretory pathway include not only disassembly of Golgi structure, but also lipid perturbations . Ceramide, which serves as the backbone for sphingolipid biosynthesis, is also an important second messenger in cells . Therefore ceramide synthesis and distribution within the cell must be tightly regulated. Ceramide is synthesized de novo in the ER (endoplasmic reticulum) and transported to various cellular locations by either vesicular or non-vesicular routes, where it is converted into sphingolipids, including SM (sphingomyelin) by SM synthase I at trans-Golgi membranes [14,15]. Ceramide can also be generated at various locations in the cell by sphingomyelinase-mediated hydrolysis of SM [16,17]. Ceramide and its derivatives including sphingosine, sphingosine 1-phosphate and ceramide 1-phosphate are important signalling lipids and are involved in regulating various aspects of cell growth, differentiation, proliferation, necrosis and apoptosis [13,16,18–21].
In the present study, we analysed how cellular stress affects de novo SM synthesis. Transfer of ceramide from the ER to the trans-Golgi mainly occurs by the activity of a soluble protein called CERT (ceramide transport protein) . CERT interacts with PI4P (phosphatidylinositol 4-phosphate)-enriched Golgi membranes through its N-terminal PH (pleckstrin homology) domain, and with the ER through a sequence in its middle region that specifically interacts with the ER-resident vesicle-associated membrane protein-associated protein . The C-terminal domain of CERT, called the START (steroidogenic acute response protein-related lipid transfer) domain, is responsible for extracting ceramide from the ER . It has been previously suggested that both the Golgi- and ER-interacting domains of CERT are required for its function [22–24]. Since CERT localizes mainly at the Golgi, it may act at ER–trans-Golgi contact sites [25,26]. Proper localization of CERT and its function depends on the morphology of the Golgi apparatus, with certain structural perturbations disrupting SM synthesis [23,27]. In the present study, we determined how cellular stress affects Golgi morphology and CERT function. Our results suggest a link between apoptotic signalling pathways and the Golgi as a platform for sensing and mediating cellular stress through CERT.
Cells and plasmids
HeLa cells were grown and maintained as described previously . Plasmids encoding Myc-tagged CERT and galT–DsRed (galactosyltransferase fused to Discosoma striata red fluorescent protein) were described previously . The plasmid encoding CERT with an N-terminal V5 tag was constructed by inserting synthetic oligonucleotides encoding the tag upstream of the CERT sequence in pcDNA3.1 between the HindIII and EcoRI sites. Myc-tagged CERT FFAT-mut (CERT lacking its ER-interacting motif) was constructed as described previously . Myc-tagged D197A and D213A CERT mutants were generated by site-directed mutagenesis using QuikChange® (Stratagene). The Myc-tagged N-terminal fragment of CERT was generated by amplifying the sequence corresponding to amino acids 1–213 of full-length CERT by PCR and inserting into pcDNA 3.1/Myc-His (Invitrogen) at the EcoRI and NotI restriction sites, resulting in a C-terminal Myc tag. The sequence was confirmed by dideoxy sequencing. Similarly, the Myc-tagged C-terminal fragment of CERT was generated by amplifying the sequence corresponding to amino acids 214–598 of the full-length CERT (with an N-terminal methionine residue preceding amino acid 214) and inserted into pcDNA 3.1/Myc-His.
Affinity-purified anti-Golgin-160 antibodies recognizing residues 60–139 and 140–311 (described in ) were used at a ratio of 1:1. Mouse anti-GM130 was obtained from BD Transduction Laboratories, monoclonal anti-Myc antibody (clone 9E10) was from Roche Molecular Biochemicals and mouse anti-V5 was from AbD Serotec. Rabbit anti-CERT IgG (recognizing an epitope between amino acids 300 and 350) was from Bethyl Laboratories. Alexa Fluor® 488-conjugated goat anti-(rabbit IgG), Alexa Fluor® 488-conjugated donkey anti-(mouse IgG), Alexa Fluor® 568-conjugated goat anti-(rabbit IgG) and Alexa Fluor® 568-conjugated donkey anti-(mouse IgG) were from Molecular Probes. HRP (horseradish peroxidase)-conjugated donkey anti-(mouse IgG) and HRP-conjugated donkey anti-(rabbit IgG) were obtained from GE Healthcare Bio-Sciences Corporation.
Labelling of endogenous sphingolipids with [3H]serine
HeLa cells were grown on 6-cm-diameter dishes as described previously . The cells were treated with 10 ng/ml TNFα (tumour necrosis factor α; Sigma) in the presence of 10 μg/ml CHX (cycloheximide), 5 μg/ml anisomycin (Sigma) or water or DMSO (Burdick and Jackson) vehicle controls for 1 h or 4 h at 37°C. During the last 1 h of drug treatment, cells were labelled with [3H]serine in the presence of CHX, as described previously . When the caspase inhibitor was used in the assay, cells were pre-incubated with 50 μM Q-VD-OPh (quinolyl-valyl-O-methylaspartyl-[2,6-difluorophenoxy]methyl ketone) (R&D Systems) for 1 h and then TNFα, anisomycin or vehicle control was added for the subsequent 4 h in presence of 50 μM Q-VD-OPh. Lipids were extracted using the standard Bligh and Dyer  method with modifications and run on HP-TLC (high-performance TLC) silica gel plates and exposed to phosphorimaging screens, as described previously . The bands were subjected to analysis using a Molecular Imager FX system (Bio-Rad Laboratories) and Quantity One software (Bio-Rad Laboratories). The amount of each lipid measured was normalized to the amount of protein in each sample.
Indirect immunofluorescence and confocal microscopy
HeLa cells were transiently transfected for approximately 24 h at 37°C with 0.5–1 μg of DNA per 3.5-cm-diameter dish with FuGENE™ 6 transfection reagent (Roche Diagnostics) according to the manufacturer's instructions. Cells were then treated with TNFα (10 ng/ml) in the presence of CHX (10 μg/ml), anisomycin (5 μg/ml) or water or DMSO vehicle for 1 h or 4 h at 37°C. During the last 1 h of drug treatment, cells were incubated in serumfree medium containing CHX to mimic the lipid-labelling experiments. Immunofluorescence studies were carried out to assess the localization of proteins of interest as described previously . For staining endogenous CERT, cells were permeabilized with 0.05% saponin (Sigma) instead of Triton X-100. For Figure 2, DNA was also stained using 0.1 μg/ml Hoechst 33342 (Sigma). Cells were visualized with an Axioskop microscope (Zeiss) equipped for epifluorescence using an ORCA-03G CCD (charge-coupled device) camera (Hamamatsu). iVision imaging software (BioVision Technologies) was used to collect and analyse the images. Cells expressing low levels of protein were selected for imaging.
Confocal imaging was performed with a single-point laser-scanning confocal microscope (Zeiss Axiovert 200 microscope with a 510-Meta confocal module) as described previously .
SM synthase assay
SM synthase activity in cell lysates was measured as described previously , except that TNFα (10 ng/ml), anisomycin (5 μg/ml) or DMSO was added along with 10 nmol of NBD-C6-Cer [NBD (7-nitrobenz-2-oxa-1,3-diazole)-C6-ceramide; Molecular Probes] complexed to BSA to each reaction before incubation. Alternatively, HeLa cells grown on 6 cm dishes were incubated with TNFα (10 ng/ml) in the presence of CHX (10 μg/ml), anisomycin (5 μg/ml) or water or DMSO vehicle for 4 h. During the last 1 h of drug treatment, cells were incubated in serum-free medium containing CHX to mimic the lipid-labelling experiments. Fluorescent NBD-SM made from the exogenously added NBD-C6-Cer substrate was analysed using a Molecular Imager FX system and quantified by Quantity One software (Bio-Rad Laboratories). The amount of each lipid measured was normalized to the amount of protein in each sample.
HeLa cells were grown to confluence on 6-cm-diameter dishes at 37°C. The cells were treated with 10 ng/ml TNFα in the presence of 10 μg/ml CHX, 5 μg/ml anisomycin or water or DMSO vehicle controls for 1 h or 4 h at 37°C. During the last 1 h of drug treatment, cells were grown in MEM (minimum essential medium; Gibco Invitrogen) in the presence of 10 μg/ml CHX to mimic lipid-labelling experiments. When the caspase inhibitor was used in the assay, cells were pre-incubated with 50 μM Q-VD-OPh for 1 h and then TNFα, anisomycin or vehicle control was added for the subsequent 4 h in the presence of 50 μM Q-VD-OPh. For analysis of CERT cleavage in cells expressing V5–CERT or V5–CERT-D213A, the transfection medium was removed after 16 h, and cells were allowed to recover for 24 h before treating with pro-apoptotic drugs. Cells were resuspended in lysis buffer [50 mM Tris/HCl, pH 8.0, 62.5 mM EDTA, 0.4% deoxycholate, 1% Nonidet P40 and protease inhibitor cocktail (Sigma)] and analysed by Western blotting as described previously .
HeLa cells were depleted of endogenous CERT by RNAi (RNA interference). Briefly, cells were transiently transfected with a pool of four duplexes (siGENOME SMARTpool M-012101-00-0005, Thermo Scientific) using Oligofectamine transfection reagent (Gibco Life Technologies) for 18 h, after which the medium was replaced. Cells from mock-transfected and CERT RNAi plates were trypsinized after 48 h and replated on 6-cm-diameter dishes. At 72 h post-RNAi transfection, the cells were transfected for 12 h with cDNAs encoding either full-length wild-type Myc-tagged CERT or the N-terminal fragment of CERT. We could not successfully express the epitope-tagged C-terminal fragment of CERT, even when different tags (Myc, FLAG, V5) were used at the N-terminus or C-terminus. The protein appeared unstable and was most probably subjected to rapid degradation. At 84 h post-RNAi transfection, cells were labelled with [3H]serine for 1 h and newly synthesized SM was quantified as described above. The experiment was carried out in parallel on 3.5-cm-diameter dishes to assess depletion of endogenous CERT and expression of full-length Myc-tagged CERT or the N-terminal fragment of CERT by immunoblotting and immunofluorescence microscopy.
In vitro transcription and translation
35S-labelled methionine substrates were generated using the TNT® T7-coupled transcription and translation kit (Promega). Templates were 1 μg of plasmid DNA (Myc-tagged CERT wild-type, Myc-tagged CERT mutants D197A and D213A, Myc-tagged caspase 2 long isoform, or Myc-tagged Golgin-160 N-terminal head (amino acids 1–393) in each reaction with 10 μl of reticulocyte lysate and 10 μCi of [35S]methionine and incubated at 30°C for 90 min.
Caspase cleavage assay
35S-labelled in vitro transcribed and translated substrates (described above) were incubated with recombinant caspases 2 (gift from Antony Rosen, Johns Hopkins University, originally obtained from Donald Nicholson at Merck-Frosst), 3, 6, 7, 8, 9 and 10 (Enzo) in their respective buffers (listed below) at 37°C for 1 h, and the reaction was stopped by adding sample buffer (50 mM Tris/HCl, pH 6.8, 2% SDS, 20% glycerol and 0.025% Bromophenol Blue) containing 3.75% 2-mercaptoethanol. Samples were run on an SDS/PAGE gel, followed by phosphorimaging. Caspase-2 cleavage reactions were carried out in buffer containing 100 mM Mes/NaOH, pH 5.5, 10% PEG [poly(ethylene glycol)], 0.1% CHAPS and 5 mM DTT (dithiothreitol) [4,29]. Caspases 3, 8 and 10 cleavage assays were carried out in buffer containing 100 mM Hepes/NaOH, pH 7.0, 10% PEG, 0.1% CHAPS and 10 mM DTT . The caspase 7 reaction was carried out in buffer containing 50 mM Hepes/NaOH, pH 7.4, 100 mM NaCl, 0.1% CHAPS, 1 mM EDTA, 10% glycerol and 10 mM DTT as per the manufacturer's directions. The caspase 9 assay was carried out in buffer containing 100 mM Mes/NaOH, pH 6.5, 10% PEG, 0.1% CHAPS and 10 mM DTT . Different concentrations of recombinant caspase 2, including 4.5 nM, 17.9 nM, 71.5 nM and 143 nM were used to determine kcat/Km values for caspase 2 cleavage of Myc-tagged wild-type CERT, Myc-tagged CERT D213A and Myc-tagged Golgin-160 N-terminal head containing amino acids 1–393. kcat/Km values were determined using the equation: substrate cleaved (%)=100×(1−e−kcat×[E]/Km×time) as described previously .
Golgi disassembly alters CERT localization after pro-apoptotic stimuli
We monitored Golgi structural perturbations in HeLa cells after 1 h and 4 h of treatment with TNFα or anisomycin. TNFα induces the apoptotic pathway extrinsically through death receptor ligation, whereas anisomycin induces the intrinsic pathway by inhibition of protein synthesis and activation of mitogen-activated and stress kinase pathways. We assessed Golgi structure by confocal microscopy after staining for the two resident Golgi proteins Golgin-160 and Golgin-97. The morphological changes to the Golgi apparatus were subtle after 1 h of treatment, but elongation of the Golgi structure occurred, and was dramatic by 4 h (Figure 1). Further disassembly at the later time was observed as reduced overlap of the two Golgi markers. Thus treatment with either TNFα or anisomycin induced Golgi disassembly, confirming previous studies [30,31].
Golgi morphology is altered after treatment with pro-apoptotic stimuli
CERT is predominantly confined to the Golgi region (Golgi membranes and nearby punctae), and its localization is dependent on Golgi morphology . Thus we determined whether localization of CERT at the Golgi region was affected by Golgi disassembly induced by pro-apoptotic stress. We incubated HeLa cells in the absence or presence of TNFα or anisomycin for 4 h, and stained for endogenous CERT and GM130, a peripheral Golgi membrane protein. As shown in Figure 2, CERT was predominantly present at the Golgi region in control cells. However, after 4 h of TNFα and anisomycin treatments, localization of CERT at the Golgi region was markedly decreased. Given that both TNFα and anisomycin induce pro-apoptotic pathways [32,33], we suspected that caspase cleavage might contribute to the decrease of CERT at the Golgi region. Pre-treating cells with the pan-caspase inhibitor Q-VD-OPh prevented the loss of CERT from the Golgi region (Figure 2). Golgi morphology was also protected in cells pre-treated with the caspase inhibitor (GM130 panels in Figure 2). These results suggest that activated caspases contribute to a reduction of CERT localization at the Golgi. To test whether caspase activation affected the Golgi pool of PI4P to which CERT binds, we assessed binding of a CERT mutant that cannot interact with the ER. Overlap of CERT lacking its ER-interacting motif with a Golgi marker (galT–DsRed) was determined as described in the Experimental section. There was no significant difference in binding to Golgi membranes in treated cells compared with control cells even though the Golgi structure was perturbed, suggesting that PI4P levels and thus phosphoinositide 4-kinase IIIβ at the Golgi  were not affected by TNFα or anisomycin treatment (Supplementary Figure S1 available at http://www.BiochemJ.org/bj/442/bj4420391add.htm).
Disruption of CERT localization by pro-apoptotic drugs is caspase-dependent
Full-length CERT decreases after treatment with pro-apoptotic drugs
The result that PI4P levels appeared unaffected in cells treated with the pro-apoptotic drugs suggested that CERT localization was affected by cleavage of CERT itself. To address caspase cleavage of CERT, we analysed the protein by immunoblotting with an antibody that recognizes an epitope near the middle of the protein (between amino acids 300 and 350). We found that the level of full-length CERT decreased after treatment with TNFα and anisomycin (Figure 3A). The full-length protein level decreased by 45 and 26% in cells treated for 4 h with anisomycin or TNFα, respectively (Figure 3B). A cleavage product of approximately 47 kDa was observed when blots were exposed for long durations (Supplementary Figure S2 available at http://www.BiochemJ.org/bj/442/bj4420391add.htm). Given that long exposures were required to detect a CERT cleavage product, we concluded that the fragment containing the antibody epitope was short-lived.
Endogenous full-length CERT decreases after treatment with pro-apoptotic stimuli
We pre-treated HeLa cells with the caspase inhibitor Q-VD-OPh for 1 h, and then subsequently for an additional 4 h in the presence of TNFα or anisomycin. Treatment with the caspase inhibitor prevented reduction of full-length CERT in TNFα and anisomycin-treated cells (Figure 3B), suggesting that the decrease in full-length CERT in the presence of pro-apoptotic stimuli was mediated by caspase cleavage. Therefore it is likely that reduced CERT staining at the Golgi region (Figure 2) can be partly attributed to cleavage of the full-length CERT.
CERT is a caspase substrate
To directly test whether CERT is cleaved by caspases, we incubated in vitro transcribed and translated 35S-labelled CERT with recombinant caspases 2, 3, 6, 7, 8, 9 and 10. Incubation with caspases 2, 3 and 9 produced fragments of similar size, approximately 47 and 22 kDa (asterisks, Figure 4A). Caspase 9 also appeared to cleave CERT near one of the termini, resulting in a band slightly smaller than the full-length protein. There was no detectable cleavage with caspases 6, 7, 8 and 10 (results not shown). As both CERT and caspase 2 are partly localized at the Golgi region [4,23], we further characterized cleavage of CERT by caspase 2. To identify the caspase 2 cleavage site in CERT, we mutated aspartate residues at several potential sites to alanine. Incubation of in vitro transcribed and translated CERT mutant proteins with recombinant caspase 2 identified Asp213 in the sequence TTRSD213 as the preferred caspase 2 cleavage site in CERT (Figure 4B). kcat/Km values were calculated for cleavage of both wild-type and D213A mutant CERT by caspase 2. The kcat/Km value for wild-type CERT was 4.02×103 M−1·s−1, whereas that of D213A mutant was 0.007×103 M−1·s−1 (Figure 4C; see the Experimental section). However, CERT with either the D197A or D213A mutation was still cleaved by caspases 3 and 9 (Supplementary Figure S3 available at http://www.BiochemJ.org/bj/442/bj4420391add.htm), suggesting that cleavage by these caspases occurred at other sites. Caspase 3 and/or 9 could cleave CERT near Asp213, or at a site towards the C-terminus that would produce fragments of size similar to those produced by caspase 2 cleavage at Asp213. We conclude that it is possible that CERT is cleaved at Asp213 by caspase 2 at the Golgi during pro-apoptotic stress induced by TNFα and anisomycin. However, cleavage of CERT by caspases 3 and 9 is also likely to contribute to fragmentation of the protein in apoptotic cells.
CERT is cleaved by caspases in vitro
Caspase 2 cleavage is predicted to inactivate CERT
Cleavage of CERT by caspase 2 at Asp213 would generate two fragments of CERT: an N-terminal fragment (amino acids 1–213) and a C-terminal fragment (amino acids 214–598) (Figure 5A). The N-terminal fragment of CERT contains the Golgi-interacting PH domain of CERT and part of the middle region, including the serine repeat motif that can be phosphorylated [22,35,36]. The C-terminal fragment encompasses the ceramide-extracting START domain of CERT as well as a part of the middle region that interacts with the ER . We predicted that the two fragments of CERT would be targeted to different organelles and thus would not function efficiently in ceramide transport. We expressed epitope-tagged N-terminal and C-terminal fragments of CERT in HeLa cells and determined their subcellular localization using immunofluorescence microscopy (Figure 5B). Wild-type full-length CERT localized to the Golgi, with additional punctae present around the Golgi region  (Figure 5B). We hypothesized previously  that the punctae may represent ER–Golgi contact sites [25,26], where efficient ceramide transport occurs. In comparison, the N-terminal fragment of CERT localized exclusively to the Golgi region without additional punctae (Figure 5B). The C-terminal fragment of CERT showed a dispersed ER-like staining pattern (Figure 5B). The C-terminal fragment was poorly expressed, but the localization pattern was similar in each of the rare cells expressing this truncation. Not surprisingly, the staining pattern of the N-terminal and C-terminal fragments of CERT resemble point mutants where the ER-binding or the Golgi-binding domains of CERT were inactivated, as described previously . Thus it is likely that the fragments of CERT produced by caspase 2 cleavage would be mislocalized and rendered non-functional for efficient ER to Golgi trafficking of ceramide.
Altered localization of CERT fragments corresponding to those generated by caspase 2 cleavage
Cleavage of CERT at the Asp213 site in transfected cells
To analyse cleavage of CERT at Asp213, we expressed epitope-tagged wild-type CERT or the D213A mutant in HeLa cells. We chose to use an N-terminal epitope tag (V5), since the C-terminal CERT fragment appeared to be extremely unstable. After 4 h of anisomycin or TNFα treatment, very low levels of CERT cleavage products were produced in transfected cells (results not shown). However, after 8 or 16 h of treatment, N-terminal fragments could be detected after blotting with the anti-V5 antibody. In cells expressing wild-type CERT, bands at 25, 39 and 52 kDa were present along with the full-length V5–CERT (shown at 16 h of treatment in Figure 6). The 25 kDa form is the size predicted for the V5-tagged N-terminus after cleavage by caspase 2 at Asp213. In cells expressing the D213A mutant, the 25 kDa band was absent (as was the 52 kDa band), but the 39 kDa form was still produced (Figure 6). These results indicate that CERT is cleaved at Asp213 during pro-apoptotic stress, but also at other sites as well. Since we did not observe a fragment corresponding to the 39 kDa form in the in vitro cleavage assays with any of the caspases we tested, it is possible that post-translational modifications (not present in the in vitro synthesized protein) are required for caspase cleavage at this site, or that other proteases are responsible. The 52 kDa form containing the N-terminal tag could be produced by cleavage by caspase 3 or 9 at a site towards the C-terminus of the protein. Interestingly, this form was only produced when the Asp213 site was intact. Perhaps this downstream site is poorly accessible in the D213A mutant. Regardless of the order of cleavage and the caspase responsible, CERT cleavage at any of these potential sites would be predicted to reduce the efficiency of CERT-mediated ceramide transport.
Cleavage of N-terminally tagged CERT in apoptotic cells
Treatment with pro-apoptotic stimuli impairs CERT function
Morphological alteration of the Golgi complex, reduced localization of CERT at the Golgi region and cleavage of CERT after pro-apoptotic stimuli could each lead to reduction of SM synthesis at the trans-Golgi. We thus analysed de novo SM synthesis in treated cells. HeLa cells were treated with anisomycin, TNFα or vehicle control for 4 h in the absence or presence of the caspase inhibitor Q-VD-OPh, and labelled with [3H]serine for the last 1 h of treatment. Lipids were extracted and analysed by TLC and phosphorimaging. SM production was significantly reduced in cells treated with anisomycin or TNFα (Figure 7A). This decrease was completely (anisomycin) or partially (TNFα) abrogated by the caspase inhibitor. The decrease in SM synthesis induced by anisomycin and TNFα was not due to a block in ceramide synthesis (Figure 7B). Anisomycin treatment resulted in a small increase in 3H-labelled ceramide, whereas newly synthesized ceramide was decreased by approximately 19% in cells treated with TNFα. However, the decrease in de novo ceramide in TNFα-treated cells cannot completely account for the much larger decrease (~82%) in SM synthesis. We performed several control experiments to rule out effects of the drugs on SM synthase. The pro-apoptotic drugs did not directly inhibit SM synthase, as shown by using an in vitro assay in cell lysates (Figure 7C). We also found that SM synthase activity in lysates from anisomycin- and TNFα-treated cells was not significantly altered from the control, suggesting that SM synthase is not inhibited through signalling events caused by pro-apoptotic treatments (Figure 7D). Thus disassembly of the Golgi complex, reduction of CERT at the Golgi region and cleavage of CERT induced by pro-apoptotic drugs are accompanied by a decrease in SM synthesis, all of which are largely caspase-dependent.
Reduction in de novo SM synthesis in cells treated with pro-apoptotic stimuli
Since the fragments corresponding to caspase 2 cleavage of CERT lacked either the N-terminal Golgi-targeting PH domain or the C-terminal ceramide-extraction domain (Figure 5A), it is likely that these fragments also lacked efficient ER–Golgi transport activity. Although overexpression of the START domain can support ceramide transport [22,35], this activity may not be efficient for fragments expressed at endogenous levels . To examine the function of cleaved CERT, we depleted HeLa cells of endogenous CERT by RNAi, expressed either full-length CERT or the N-terminal fragment of CERT, and monitored SM synthesis. We were unable to successfully express sufficient levels of the C-terminal fragment of CERT to include it in this assay. As shown in Figure 8(A), depletion of CERT by RNAi resulted in a 74% decrease in endogenous protein by Western blot at 84 h post-transfection. Expression of epitope-tagged full-length CERT or the N-terminal fragment of CERT corresponding to caspase 2 cleavage resulted in an approximately 4-fold increase in protein compared with the endogenous level (Figure 8A). Analysis of SM synthesis in cells treated with CERT RNAi revealed a 45% reduction in newly synthesized SM (Figure 8B). This decrease was rescued upon expression of Myc-tagged wild-type CERT to a similar level as the control (Figure 8B). Given that only approximately 25% of the depleted cells expressed the Myc-tagged CERT, it is not unexpected that SM synthesis was not higher in the rescued cells. However, expression of the Myc-tagged N-terminal fragment of CERT did not rescue de novo SM synthesis (Figure 8B). Surprisingly, expression of the N-terminal fragment of CERT in mock-depleted control cells showed a statistically significant decrease in the level of newly synthesized SM of approximately 30% (P=0.01) when compared with mock-depleted control cells alone. This suggests that the N-terminal fragment of CERT inhibits endogenous CERT activity, possibly by competing for binding sites on Golgi membranes and may thus act in a dominant-negative manner. Although we could not test it, we predict that the C-terminal fragment of CERT would also be inefficient at transporting ceramide from the ER to the Golgi, given its mislocalization and instability.
The N-terminal fragment of CERT cannot rescue de novo SM synthesis in CERT-depleted cells
The mammalian Golgi apparatus plays a central role in the secretory pathway and in sphingolipid synthesis. The complex architecture of the mammalian Golgi ribbon with ER–trans-Golgi contact sites is thought to facilitate delivery of ceramide from the ER for synthesis of SM at the trans-Golgi [25,26]. We showed previously that certain Golgi morphologies (induced by drugs that perturb Golgi structure) promote efficient ceramide trafficking and SM synthesis, whereas others do not . Efficient SM synthesis correlated with CERT localization at Golgi membranes. A recent study also found a dramatic decrease in SM synthesis in cells deficient in COG2 (conserved oligomeric Golgi complex-2), a component of a retrograde trafficking complex that is required for proper localization of certain glycosyltransferases . The morphology of the Golgi complex in COG2-null cells is disrupted, which may interfere with proper CERT localization. Thus Golgi architecture is important for efficient SM synthesis. In the present study, we showed that induction of pro-apoptotic pathways by TNFα and anisomycin results in disassembly of the Golgi complex, probably due to cleavage of Golgi structural proteins including Golgin-160, giantin, p115 and GM130 by activated caspases [4–7]. In addition to Golgi disassembly, treatment with pro-apoptotic drugs resulted in a decrease of CERT at the Golgi region, caspase cleavage of CERT and a large reduction in SM synthesis.
CERT was cleaved by caspases 2, 3 and 9 in vitro, and we mapped the caspase 2 cleavage site to Asp213 in the sequence TTRSD. Although this sequence is not predicted to be a good caspase 2 target , the best-characterized caspase 2 cleavage sequence (GESPD in Golgin-160) does not fit the prediction either. The kcat/Km value for cleavage of CERT by caspase 2 was 4.02×103 M−1·s−1, and the kcat/Km value for cleavage of Golgin-160 was 4.38×103 M−1·s−1. Thus CERT is almost as good a substrate for caspase 2 as Golgin-160. The kcat/Km values measured by us for cleavage of Golgin-160 by caspase 2 was less than the kcat/Km value (33×103 M−1·s−1) reported previously . This discrepancy could result from our use of the N-terminal portion instead of the full-length Golgin-160 substrate, or a different preparation of recombinant caspase 2. Although we did not map the sites of cleavage for caspases 3 and 9, we found that CERT-D213A was still cleaved in cells treated with pro-apoptotic drugs at sites downstream of Asp213. Thus, even though caspase 2 may contribute to inactivation of CERT, other caspases can also perform this role if the caspase 2 site is absent or if activation of procaspase 2 does not occur.
Interestingly, although anisomycin and TNFα treatments resulted in approximately 45 and 26% decreases in full-length endogenous CERT, newly synthesized SM levels decreased by approximately 71 and 82% respectively. Thus it is likely that Golgi disassembly and CERT mislocalization also contribute to the impairment of SM synthesis, thereby causing a greater reduction in newly synthesized SM than would be expected with just CERT cleavage alone.
Changes in Golgi morphology during early times (1 h) of pro-apoptotic treatment were subtle, and we observed CERT mislocalization, reduction in full-length protein and impairment of SM synthesis only after 4 h of pro-apoptotic treatment when Golgi alterations were more significant. Activation of initiator caspases and cleavage of a subset of Golgi substrates at early times may lead to small changes in Golgi organization. Subsequent activation of effector caspases may then result in gross morphological changes of the Golgi apparatus, including generation of dispersed Golgi membranes seen at later times. It is also possible that gross morphological changes at later time points correlate with dysregulation of cellular ceramide levels, thus amplifying the pro-apoptotic signalling pathway. Ceramide is also thought to play a role in the maintenance of structural integrity of the ER and the Golgi , as well as in apoptotic signalling pathways . Cells experiencing mild stress may activate survival responses (e.g. the unfolded protein response). However, cells exposed to prolonged or severe stress may activate caspases and commit to the apoptotic pathway . One route for this could be inhibition of SM synthesis and an increase in the ceramide pool. However, at 4 h of TNFα and anisomycin treatments we did not observe a significant increase in newly synthesized ceramide. Since we were only analysing de novo synthesis of ceramide during the pro-apoptotic treatments, we cannot rule out generation of unlabelled ceramide from SM, sphingosine and other sphingolipids in different regions of the cell. Previous studies have indicated that TNFα treatment in particular is associated with a rapid increase in ceramide levels at the plasma membrane by the activation of sphingomyelinase [40,41]. Thus understanding how cells sense ceramide levels at the ER and the Golgi  and integrate pro-apoptotic signalling is imperative. This would help us to examine the possibility that the Golgi apparatus acts as a platform to integrate pro-apoptotic and sphingolipid signalling pathways via caspases and CERT.
Inactivation of CERT and inhibition of SM synthesis results in decreased utilization of substrates ceramide and PC (phosphatidylcholine). Although the consequence of increased ceramide levels in cells has been studied (discussed above), it is not clear how cells may handle PC when SM synthesis is inhibited during pro-apoptotic treatments . When SM is synthesized in the trans-Golgi from the substrates ceramide and PC, DAG (diacylglycerol) is produced [43,44]. DAG at Golgi membranes serves to recruit PKD (protein kinase D), which regulates vesicle production and thus plays a role in modulating cargo trafficking through the organelle [45–47]. If SM synthesis decreases due to inactivation of CERT, DAG and consequently PKD recruitment to Golgi membranes would also presumably decrease, suggesting that CERT may function indirectly in cargo trafficking through the Golgi apparatus . Additionally, altering SM synthesis in the Golgi and accordingly SM levels along the secretory pathway may also alter the SM/cholesterol ratio , thereby affecting partitioning and trafficking of lipids and proteins to their target destination (e.g. proteins targeted to lipid rafts in the plasma membrane). Thus CERT may play an important role in regulating both lipids and proteins when cells are under severe stress.
In the present study, we have shown that both Golgi disassembly and inactivation of CERT contribute to decreased SM synthesis during pro-apoptotic stress. A more comprehensive approach for determining the order in which apoptotic signalling, ceramide accumulation and Golgi disassembly occur will be required to understand how the Golgi apparatus contributes to cellular stress sensing and integration of apoptosis signalling.
ceramide transport protein
conserved oligomeric Golgi complex
galactosyltransferase fused to Discosoma striata red fluorescent protein
cis-Golgi matrix protein of 130 kDa
protein kinase D
steroidogenic acute response protein-related lipid transfer
tumour necrosis factor α
Suchismita Chandran performed the experiments, analysed the data and wrote the paper. Carolyn Machamer analysed the data and edited the paper before submission.
We thank Winny Yun for contributing to the construction of the plasmid encoding V5-tagged CERT, the Johns Hopkins University School of Medicine Microscope Facility for assistance with confocal microscopy and Antony Rosen for purified caspase 2. We also thank Travis Ruch and David Zuckerman for useful comments on the paper.
This work was supported by the National Institutes of Health [grant number GM42522 (to C.E.M.)].