Selective knockdown of ceramide synthases reveals complex interregulation of sphingolipid metabolism.

Mammalian ceramide synthases 1 to 6 (CerS1-6) generate Cer in an acyl-CoA-dependent manner, and expression of individual CerS has been shown to enhance the synthesis of ceramides with particular acyl chain lengths. However, the contribution of each CerS to steady-state levels of specific Cer species has not been evaluated. We investigated the knockdown of individual CerS in the MCF-7 human breast adenocarcinoma cell line by using small-interfering RNA (siRNA). We found that siRNA-induced downregulation of each CerS resulted in counter-regulation of nontargeted CerS. Additionally, each CerS knockdown produced unique effects on the levels of multiple sphingolipid species. For example, downregulation of CerS2 decreased very long-chain Cer but increased levels of CerS4, CerS5, and CerS6 expression and upregulated long-chain and medium-long-chain sphingolipids. Conversely, CerS6 knockdown decreased C16:0-Cer but increased CerS5 expression and caused non-C16:0 sphingolipids to be upregulated. Knockdown of individual CerS failed to decrease total sphingolipids or upregulate sphingoid bases. Treatment with siRNAs targeting combined CerS, CerS2, CerS5, and CerS6, did not change overall Cer or sphingomyelin mass but caused upregulation of dihydroceramide and hexosyl-ceramide and promoted endoplasmic reticulum stress. These data suggest that sphingolipid metabolism is robustly regulated by both redundancy in CerS-mediated Cer synthesis and counter-regulation of CerS expression.

ceramide synthesis as a gain-of-function from these proteins ( 13-15, 20, 25 ). How individual CerS contribute to the steady-state sphingolipid pool in mammalian cells and whether specifi c CerS are necessary for the maintenance of sphingolipids with specifi c acyl chains are still poorly understood. Furthermore, little is known about the regulation and interregulation of these proteins with regard to maintaining overall sphingolipid homeostasis. Therefore, we hypothesized that knockdown of individual CerS would downregulate specifi c populations of sphingolipids in a manner consistent with the known acyl-CoA specifi cities of these enzymes.
In this study, we have used small interfering RNAs (siRNAs) against the six CerS family members in order to specifi cally reduce the expression of each isoform in MCF-7 breast adenocarcinoma cells. Knockdown of individual CerS resulted in a wide range of effects on nontargeted CerS expression. These effects were associated with both increases and decreases in multiple sphingolipid species, often with little change in total Cer levels. To our surprise, knockdown of some CerS (e.g., CerS5) or knockdown of combined CerS2, CerS5, and CerS6 produced little or no decrease in sphingolipid levels but did cause several lipids, especially dHCer and HexCer, to increase. These data also indicate that changes in CerS activity are met with a shift in sphingolipid metabolism suffi cient to allow ceramide and/or sphingoid base levels to be maintained at the expense of accumulating glycosphingolipids.

MATERIALS AND METHODS
Cell culture MCF-7 human breast adenocarcinoma cells were maintained in RPMI medium (Gibco-Invitrogen, Carlsbad, CA) supplemented with L -glutamine, 10% (v/v) fetal bovine serum. Cells were kept in a humidifi ed incubator at 37°C with 5% C O 2 .

siRNA transfection
Cells were plated at 4.0 × 10 5 cells/dish (10 cm dish) and incubated for 24 h. At 24 h, cells were transfected with doublestranded RNA oligomers, using Oligofectamine (Invitrogen, Carlsbad, CA) according to the manufacturer's protocol. Fortyeight h posttransfection, cells were harvested by scraping into ice-cold phosphate-buffered saline (PBS) and pelleted by centrifugation at 250 g . Cell pellets were then frozen at Ϫ 80°C until used for further analysis. siRNA sequences used in this study are provided in Table 1 . We have determined empirically that CerS siRNA oligomers (siCerS) are effective for reducing target expression (>50%) in the 5-20 nM range, and therefore, for most experiments, 5 nM siCerS or negative control siRNA (siControl) was used.

Real-time quantitative PCR analysis
RNA was extracted using a Qiagen RNeasy® kit according to the manufacturer's protocol. RNA concentration was determined by using a Quant-iT™ RiboGreen® RNA assay kit (Invitrogen, Valencia, CA). One microgram of RNA was used to produce cDNA, using a SuperScript fi rst-strand synthesis system (Invitrogen). The resultant cDNA was used for real-time quantitative PCR (q-PCR) using a QuantiTect SYBR Green PCR kit (Invitrogen) on an ABI 7300 quantitative (q)-PCR system (Applied (SM) and glycosphingolipids. Cer may also be produced via hydrolysis of complex sphingolipids or reacylation of free Sph by CerS through what has been termed the salvage or recycling pathway ( 4 ).
The acylation of free sphingoid bases by CerS occurs through an acyl-CoA-dependent process ( 5 ). In addition to regulating sphingolipid synthesis, CerS activity has been shown to regulate numerous aspects of cell biology, including cell growth, apoptosis, and autophagy (6)(7)(8). The abilities of CerS to control de novo pathways and to salvage pathways of Cer synthesis make the proteins that catalyze these reactions of unique biochemical and biological interest.
Although acyl-CoA-dependent Cer synthesis was fi rst described in 1966, more than 30 years passed before the fi rst genes responsible for dHCer/Cer synthesis were identifi ed in Saccharomyces cerevisiae ( 9,10 ). The fi rst gene was originally termed the longevity assurance gene ( LAG1 ) because of its ability to increase yeast replicative life span ( 11 ). Several years later, LAG1 and its homolog, the longevity assurance gene ( LAC1 ) cognate, were shown to be necessary for the synthesis of yeast dHCer ( 9 ). A search for genes containing the characteristic Lag1p motif of LAG1 and LAC1 led to the identifi cation of homologs in several organisms, including a human gene product, CerS1 (formerly known as Lag1pHs, Lass1, and UOG1) ( 12 ). Overexpression of this gene in mammalian cells led to an increase in CerS activity and Cer ( 13 ). Additional mammalian homologs (CerS2-6, formerly known as Lass2-6) were subsequently discovered and shown to regulate the synthesis of Cer as well ( 14,15 ).
Mammalian CerS proteins share many features. Each appears to localize in the endoplasmic reticulum (ER); each has at least fi ve transmembrane domains; and CerS2, CerS5, and CerS6 are N -glycosylated near their N termini (13)(14)(15). All CerS, as well as nonmammalian Lag1p homologs, possess a Lag1p homology domain that has been demonstrated to be essential for enzymatic activity in CerS1 and CerS5 ( 12,16 ). All mammalian CerS, except for CerS1, also contain a Hox-like domain the function of which is not known ( 17 ). Furthermore, purifi ed CerS5 has been shown in vitro to possess CerS activity, which strongly supports the notion that mammalian CerS proteins possess bona fi de acyl-CoA-dependent CerS activity ( 18 ).
Most of the studies investigating the roles of CerS in the regulation of sphingolipids have relied on plasmid-mediated overexpression of CerS proteins to show increased

In vitro CerS activity
In vitro CerS activity was performed essentially as described previously ( 29 ). Cells were treated as indicated and then harvested by scraping into ice-cold PBS. Cells were pelleted and resuspended in a solution of 20 mM HEPES (pH 7.4), 250 mM sucrose, 2 mM KCl, and 2 mM MgCl 2 and then sonicated, and protein content was determined by using the Bradford assay. A reaction mixture (100 l, fi nal volume) containing 15 M 17C-dHSph and 50 M palmitoyl-CoA or lignoceroyl-CoA in 25 mM potassium phosphate buffer (pH 7.4) was prewarmed at 37°C for 5 min, followed by addition of 100 g of whole-cell lysate to start the reaction. Reaction time was 15 min, after which the reaction mixture was transferred to a glass tube containing 2 ml of extraction solvent (ethyl acetate/2-propanol/water, 60:30:10 {v/v/v}), which stopped the reaction. Lipids were extracted as previously described, and 17C16:0-dHCer or 17C24:0-dHCer content was determined by HPLC/MS ( 16,27 ).

Data presentation and statistical analysis
Heat maps displaying log 2 -fold change of siCerS over siControl were generated using Matrix2png software (http://chibi.ubc.ca/ matrix2png/index.html) ( 30 ). Statistical analyses were performed using Microsoft Excel. Statistical signifi cance was determined using a two-tailed, unpaired Student's t -test using a P value of < 0.05 as our criterion for signifi cance.

CerS expression in MCF-7 breast adenocarcinoma cells
In order to study the roles of CerS in regulating sphingolipid levels, we chose a cell system, MCF-7 breast adenocarcinoma cells, in which multiple CerS transcripts were present and able to be knocked down using siRNA. We were further interested in this cell line because of recent publications demonstrating altered CerS expression in human breast cancer (31)(32)(33). We fi rst used real-time q-PCR to determine the expression of CerS1-6 in MCF-7 cells ( Fig. 1 ). The transcript of CerS2 was the most abundant, followed by those of CerS6, CerS5, CerS1, and CerS4. CerS3 mRNA levels were near the limit of detection, which is consistent with reports of CerS3 having a limited tissue distribution and being found mainly in the testis and skin ( 14,20 ). Because CerS2 and CerS6 were the most abundant transcripts, the majority of this report will focus on the effects observed upon siRNA-mediated knockdown of these enzymes. Data concerning the knockdown of the less abundant CerS can be found in the supplementary data section.
Biosystems, Foster City, CA) as described by the manufacturer. Primers used in this study are described in Table 2 .
Ceramide analysis by high-performance liquid chromatography/mass spectrometry MCF-7 cells were harvested by scraping into 10 ml of ice-cold PBS and pelleted. Lipids were extracted and analyzed by the Lipidomics Core facility at the Medical University of South Carolina, using quantitative high-performance liquid chromatography/ mass spectrometry (HPLC/MS), as described previously ( 27 ). Our analysis was limited to sphingoid bases, dHCer, Cer, SM, HexCer (consisting of glucosylceramide {GlcCer} and galactosylceramide {GalCer}), and lactosylceramides (LacCer) possessing an 18 carbon sphingoid base backbone and ceramide-containing lipids whose acyl chains were nonhydroxylated and 14 to 26 carbons in length. Sphingolipid levels were normalized to total cellular lipid phosphate, which was determined as described previously ( 28 ).   ticular CerS could lead to upregulation of one or more other CerS.
We next tested the ability of siCerS to reduce the protein levels of CerS. Of several commercially available antibodies tested, only those against CerS2 and CerS6 recognized proteins within the range of predicted molecular masses of CerS1-6 ( ‫ف‬ 40-46 kDa). Fig. 2C shows knockdowns of CerS2 and CerS6 at the protein level. Knockdowns of CerS3, CerS4, and CerS5 showed a slight elevation of CerS2, whereas knockdown of Cer2 or CerS4 showed an elevation in CerS6. These effects were in large part similar to those seen at the mRNA level ( Figs

Effect of CerS1-6 on acyl chain lengths of sphingolipids
We sought to determine the effects of CerS knockdown on steady-state sphingolipid levels in MCF-7 cells. Following siRNA treatment, we analyzed sphingolipid levels by HPLC/MS and compiled data into a heat map showing log 2 -fold change of each lipid species in siCerS-treated cells over siControl-treated cells ( Fig. 3A ). To screen for signifi cant changes in lipids, we used unpaired, two-tailed Student's t -tests to compare each siCerS to siControl ( supplementary Fig. I ).
The sphingolipid distribution in control cells was such that the most abundant sphingolipid class analyzed was

CerS knockdown by targeted siRNA causes multiple changes in nontargeted CerS mRNA levels
We next established the effi cacy of siRNA-induced knockdown on the transcript levels of each CerS in MCF-7 cells. As indicated in Fig. 2A , B , knockdown of CerS2 and CerS6, the two most abundant CerS, reduced expression of these targets at the mRNA level. Knockdown of the other CerS (CerS1, CerS3, CerS4, and CerS5) is reported in supplementary Fig. IA-D) . Overall, we found that the siRNAs (5 nM) were more effective at reducing the expression of their targets than the siControl.
Interestingly, the targeting of certain CerS resulted in the elevation of nontargeted CerS. Knockdown of CerS2 resulted in increased mRNA levels of CerS4, CerS5, and CerS6 ( Fig. 2A ). siCerS6, on the other hand, upregulated levels of the CerS5 message but did not change levels of the other CerS ( Fig. 2B ). Both siCerS2 and siCerS4 increased CerS6 mRNA levels ( Fig. 2A and supplementary  Fig. IC ). siCerS5 increased mRNA levels of CerS2 and CerS4 (supplementary Fig. ID ). siCerS3, which further decreased the already low levels of its target in MCF-7 cells, also increased CerS4 mRNA ( Fig. IB ). From these data, we concluded that siRNA-mediated CerS knockdown was effective at the mRNA level and that the targeting of a par- CerS (black bars) or siControl (white bars) for 48 h. Cells were harvested, and RNA was extracted for q-PCR analysis of expression of CerS1-6. q-PCR data are normalized to ␤ -actin mRNA expression, and data are means ± standard errors of the mean (SEM) for three independent experiments. A: Effects of siCerS2 on CerS1-6 expression. B: Effects of siCerS6 on CerS1-6 expression. C: Western blot analysis of CerS2, CerS6, and ␤ -actin protein expression following transfection with siCerS1-6. CerS2 and CerS6 were detected using monoclonal antibodies specifi c for these proteins. ␤ -actin protein levels were used as a control for equal gel loading. *, P < 0.05; **, P < 0.01; ***, P < 0.01 versus siControl. C16:0 ( Figs. 3A, 4A ), which was similar to the effects seen when this enzyme was downregulated in SMS-KCNR neuroblastoma cells ( 8 ). In MCF-7 cells, C14:0-Cer increased by ‫ف‬ 4.5-fold, and C16:0-Cer nearly doubled. The increase in Cer was refl ected in a dramatic increase in C14:0-and C16:0-HexCer and LacCer ( Fig. 4D, E ). C16:0-SM, the most abundant sphingolipid species in our analysis, increased by 35.2% ± 6.4% or ‫ف‬ 4 pmol/nmol lipid phosphate, thus representing the largest sphingolipid change induced by the knockdown of a CerS. However, despite the effects of CerS2 knockdown on C16:0 lipids and CerS5 and CerS6 expression, siCerS2 did not increase the in vitro C16:0-CerS activity ( supplementary Fig. IIIB ). Therefore, CerS2 is not only a very-long-chain CerS but its absence also causes the accumulation of long-chain sphingolipids.

CerS2 downregulation results in shift of sphingolipid distribution to predominantly long-chain species
Because CerS2 was the most abundant CerS in MCF-7 cells, we expected that its downregulation would cause the most dramatic changes in sphingolipids. We found that siCerS2 was able to downregulate in vitro C24:0-CerS activity (supplementary Fig. IIIA) and Cer containing C22-C24 saturated and monounsaturated fatty acids ( Figs. 3A and 4A ). The greatest decrease was in C22:0-Cer (34.1% ± 3.0% of siControl), and the most abundant ceramide, C24:0-Cer, was decreased to 52.2% ± 4.4% of that of siControl. Interestingly, very-long-chain dhCer species were not as affected as very-long-chain Cer ( Fig. 4B ). There was a trend toward a decrease in several very-long-chain SM species, but only the decrease in C22:0-SM was statistically significant ( Fig. 4C ). Surprisingly, very-long-chain HexCer and LacCer levels were not signifi cantly 6affected ( Fig. 4D, E ). Therefore, although very-long-chain Cer were sensitive to CerS2 depletion, very long-chain SM and HexCer were largely unaffected.
On the other hand, downregulation of CerS2 resulted in large increases in the long-chain Cer species C14:0 and Despite increases in C16:0-SM ( Fig. 4C ), the most abundant sphingolipid we measured, siCerS2 caused no overall increase in SM ( Fig. 6A ). This may be explained by the presence of small decreases in multiple very-long-chain SM species being balanced out by the larger change in C16:0-SM. These data also indicate that cells can maintain overall SM levels despite a large shift in acyl chain composition.
Pharmacological inhibition of CerS or genetic deficiency of CerS2 in mice can induce the accumulation of the sphingoid bases dHSph and Sph ( 34,35 ). We therefore asked whether individual knockdowns of CerS would cause an accumulation of these lipids in MCF-7 cells. We analyzed dHSph and Sph levels following CerS1-6 knockdowns and found that none of the knockdowns was capable of signifi cantly increasing sphingoid base levels ( Fig.  6B and supplementary Fig. VIII ). Of note, several siCerS appeared to have slight stimulatory effects on sphingosine levels; however, none of these values reached our criterion for signifi cance, and further studies using higher doses of siRNA failed to demonstrate Sph increases (data not shown). dHS1P and S1P were near or below detectable levels in the analyses performed and could not be reliably quantifi ed. Based on these data, it is likely that even when CerS are knocked down, there is enough residual activity and redundancy to prevent sphingoid bases from accumulating.

Combined knockdown of CerS2, CerS5, and CerS6 increases HexCer, Sph, and S1P
We observed that knockdown of CerS2 or CerS6 individually caused decreases in particular Cer levels as well as alterations in the expression of other CerS such as CerS5, but only CerS2 knockdown showed an appreciable altered by CerS6 knockdown (supplementary Fig. IIID ). Based on the increases in very-long-chain sphingolipids and the decreases in C16:0-dHCer/-Cer, CerS6 knockdown had a clear effect on reducing the amount of C16:0 sphingolipids as a proportion of all acyl chain lengths. Therefore, it is likely that CerS6 is the predominant C16:0-CerS in MCF-7 cells and additionally that it controls the production of very-long-chain sphingolipids.

Downregulation of less abundant CerS
Knockdowns of CerS1, CerS3, CerS4, and CerS5 did not cause robust decreases in any sphingolipid species, but there were several unanticipated increases in several sphingolipids and sphingolipid classes ( Fig. 3A ). For example, downregulation of CerS1 upregulated several HexCer and LacCer species, especially those possessing a C14:0 acyl chain. A more detailed description of these changes is provided in supplementary Figs. II-VII ).

Knockdown of CerS increases total levels of different sphingolipid classes but fails to signifi cantly increase sphingoid base levels
Because many of the decreases in sphingolipids we saw were accompanied by increases in other sphingolipids, we asked whether CerS knockdown affected the total levels of each sphingolipid class ( Fig. 6A ). The only signifi cant decrease we observed was a slight decrease in ceramide by siCerS2, which can be attributed to the effect of siCerS2 on very-long-chain ceramide ( Fig. 6A ). CerS6, on the other hand, did not signifi cantly decrease Cer or any other sphingolipid class.  Analysis of total lipid changes showed no overall changes in Cer or SM levels ( Fig. 7B ) but did show a shift of chain length distribution, predominantly among medium-longchain species (data not shown). Total dHCer, on the other hand, was increased, and this was attributable to increases in long-chain and very-long-chain species. Interestingly, total HexCer was enhanced by more than 2-fold, and every species we analyzed showed a signifi cant increase ( Fig.  7C ). We also found Sph levels to be elevated, and levels of S1P, although not routinely detectable in MCF-7, were increased following siCerS2/5/6 treatment ( Fig. 7D ). These data suggest that Cer synthesis was inhibited suffi ciently to cause accumulation of sphingoid bases; however, dHSph and dHS1P remained near or below the limit of detection in these samples. Indeed, because multiple CerS are likely metabolizing the same pool of sphingoid base substrates, suffi cient knockdown of more than one enzyme is required to see any increase in dHSph or Sph (or their phosphates). These data also suggest that inhibition of multiple CerS is insuffi cient to drive the total levels of any sphingolipid down. Instead, the increases suggest the presence of a counter-regulatory mechanism that allows the maintenance of Cer and SM levels at the expense of dHCer and HexCer.

Combined knockdown of siCerS2/5/6 upregulates CHOP
Previous work has shown that downregulation of CerS2 or CerS6 can induce ER stress via upregulation of the C/EBP homologous protein (CHOP) ( 8,36 ). We asked whether the siCerS2/5/6 changes in sphingolipids would lead to upregulation of CHOP. We discovered that siCerS2/ 5/6 indeed caused an upregulation of CHOP expression, indicative of an ER stress response. These data suggest that dysregulation of sphingolipid metabolism, even with preserved total Cer and SM levels, can promote an ER stress response.
On the other hand, knockdown of less abundant CerS such as CerS1 or CerS5 largely failed to produce the decrease in total Cer levels. The paucity of effects seen with individual CerS knockdowns on total levels might be explained by compensation from nontargeted CerS, especially when a particular CerS is upregulated (e.g., CerS5 upregulation by siCerS2 and siCerS6). We therefore hypothesized that combined knockdown of the major CerS CerS2 and CerS6, as well as prevention of CerS5 counterelevation, would cause a detectable decrease in both total and individual ceramide levels due to an overall decrease in CerS activity. We used a treatment that combined siCerS2 (10 nM), siCerS6 (10 nM), and siCerS5 (20 nM) (siCerS2/5/6) and found that siCerS2/5/6 resulted in a decrease in CerS2 and CerS6 expression levels ( Fig. 7A ). Interestingly, CerS5 mRNA levels, which were increased by either CerS2 or CerS6 alone ( Fig. 2A, B ), were maintained at control levels by treatment with siCerS2/5/6 ( Fig. 7A ).   Fig. 7. Treatment of of siCerS2, siCerS5, and siCerS6 combined increases dHCer and HexCer levels and causes elevation of Sph and S1P. The effects of combined treatment of siCerS2 (10 nM), siCerS5 (20 nM), and siCerS6 (10 nM) (siCerS2/5/6) on CerS2, CerS5, and CerS6 expression compared to siControl (40 nM) were analyzed by real-time q-PCR (A). Data are means ± SEM for six independent experiments. B: The effects of siCerS2/5/6 on total levels of dHCer, Cer, SM, HexCer, and LacCer were assessed by HPLC/MS. Data are means ± SEM for six independent experiments. C: Analysis of changes in individual HexCer species in response to siCerS2/5/6. Data are means ± SEM for six independent experiments. D: Changes in Sph and S1P content of cells treated with siCerS2/5/6 or siControl. Data are means ± SEM for six independent experiments. E: Effect of siCerS2/5/6 treatment on CHOP expression. Data are means ± SEM for fi ve independent experiments. *, P < 0.05; **, P < 0.01; ***, P < 0.01 versus siControl.
found that knockdown caused the increase of nearly all HexCer species, except for C18:0-HexCer ( Fig. 3A and  supplementary Fig. IVB ). One interpretation of these data is that there was an overall increase in HexCer synthesis that allowed cells to maintain a normal level of C18:0-HexCer despite decreased availability of CerS1-derived C18:0-Cer. However, we did not detect a signifi cant decrease in C18:0-Cer, so it is diffi cult to say whether CerS1 is required for C18:0-Cer synthesis in MCF-7 cells. siCerS3 also caused slight increases in multiple HexCer species that culminated in an increase in total HexCer levels (supplementary Fig. IV ). The mechanism of this change is unknown, but it suggests that even though CerS3 is expressed at low levels, it may have a subtle role in regulating HexCer metabolism. siCerS4 caused an increase in C26:0-dHCer/-Cer (supplementary Fig. VIB ), and siCerS5 increased C26:1sphingolipids (supplementary Fig. VIIB ). Whether these two latter effects are related is unknown. According to Laviad et al. ( 25 ), CerS2 overexpression can increase activity toward C26:0-CoA and C26:1-CoA, so it is possible that CerS2 is somehow contributing to these increases. However, other than the slight increases in CerS2 expression induced by siCerS5, no evidence supports increases in CerS2 activity as a mechanism for elevated C26 sphingolipids. Overall, the changes induced by knockdowns of CerS3-5 were largely unexpected and are not readily explained by the corresponding changes in CerS expression.
Both siCerS6 and siCerS5 caused an unanticipated increase in total SM ( Fig. 6A and supplementary Fig. VIIA , respectively), but knockdown of CerS6 had more widespread effects including increased SM and HexCer, as well as elevations in CerS5 message levels ( Fig. 2B ). In a manner analogous to that of siCerS2, siCerS6 induced a shift in acyl chain compositions of SM ( Fig. 5B ) and HexCer ( Fig.  5C ), favoring the increase of nearly all species except C16:0-SM and C16:0-Cer, while total Cer and dHCer remained unchanged ( Fig. 6A ). Also, the only evidence suggestive of increased CerS5 expression at the sphingolipid level was an increase in C14:0-HexCer ( Fig. 5C ). The upregulation of CerS5 may indeed be a compensatory mechanism for protecting C16:0-sphingolipid levels from being depleted. Taken together, the effects of CerS6 knockdown likely refl ect a shunting of excess sphingoid base substrates into the next most available CerS (e.g., CerS2, CerS5, and others), which leads to the accumulation of non-C16:0sphingolipids.
The ability of CerS knockdown to cause increases in sphingolipids was exemplifi ed by our attempted simultaneous knockdown with CerS2/5/6 ( Fig. 7 ). The resulting increases in dHCer and HexCer imply that there is a signifi cant response of cells to the reduced CerS expression or activity that involves the accumulation of these lipids ( Fig. 7B ). Unlike individual knockdowns of CerS2 and CerS6, these data cannot be explained by sphingoid base shunting alone. This notion is supported by the fact that nearly all HexCer species increased ( Fig. 7C ). It also should be noted that medium-long-chain species (presumably generated collectively by CerS1, CerS4, and CerS5) showed the largest increases and are likely the result of expected decreases in their expected products, C18:0-and C16:0-sphingolipids, respectively ( Figs. 4A, 5A ). The one exception was a slight reduction in C18:0-SM levels by siCerS1 treatment, and this siRNA decreased C18:1-Cer levels as well ( Fig. 3A and supplementary Fig. IVA ). CerS4 knockdown was predicted to affect C20:0-Cer, but we failed to observe any changes in this lipid and instead saw a decrease in C24:1-Cer ( Fig. 3A and supplementary Fig. VIA ). As for CerS3, no major changes were expected because of its low expression in MCF-7 cells ( Fig. 1 ), but we did observe an overall increase in HexCer species ( Fig. 3A and  supplementary Figs. V, IX ).
Knockdown of CerS2 produced several unexpected changes. siCerS2 treatment caused an upregulation of CerS4, CerS5, and CerS6 mRNA levels ( Fig. 2B ). This was associated with a shift of dHCer and Cer synthesis from very-long-chain species to long-chain species. Interestingly, there was hardly any change in total dHCer or Cer level, suggesting that shifting of synthesis to long-chain species was suffi cient to maintain the overall levels of these lipid classes. Of note, CerS2 knockdown did not increase in vitro C16:0-CerS activity ( supplementary Fig. IIIB ), which supports our previous data ( 8 ) indicating that although expression of CerS5 and CerS6 is increased by siCerS2, this does not necessarily translate into increased C16:0-CerS activity. Furthermore, it supports the hypothesis that long-chain Cer/dHCer accumulation is likely due to shunting of sphingoid bases away from CerS2-mediated Cer/ dHCer synthesis. Alternatively, the regulation of other enzymes (e.g., forward or reverse Cer activities) may also play a role ( 8 ). The mechanism whereby CerS5 and CerS6 mRNA upregulation fails to contribute to increased C16:0-CerS activity is unclear, but there may be a role for ER stress in altering the production or activity of these enzymes ( 8 ).
siCerS2 treatment also induced the accumulation of C16:0-SM, long-chain HexCer, and long-chain LacCer, which could be attributed to the accumulation of their Cer precursors ( Figs. 3A and 4A-D ). Despite an increase in C16:0-SM ( ‫ف‬ 4 pmol/nmol lipid phosphate), which was the largest mass change we observed in this study, there was no overall increase in SM ( Fig. 6A ). This may be explained by the presence of small decreases in multiple very-long-chain SM species being balanced out by the larger change in C16:0-SM. These data also indicate that cells can maintain overall SM levels despite a large shift in acyl chain composition. HexCer and LacCer, on the other hand, did exhibit increases in total levels following siCerS2 treatment, suggesting that these lipids are not subject to the same control as Cer and SM. It is surprising that very-long-chain HexCer species do not decrease in response to siCerS2 despite the decrease in Cer levels. This could indicate that steady-state very-long-chain HexCer, consisting of GlcCer and GalCer, are not synthesized and degrade readily enough to show a decrease in response to decreased Cer. Alternatively, the activities of GlcCerS and GalCerS may be altered to enhance or maintain production of HexCer despite decreasing substrate levels.
Other unexpected increases were observed with CerS1, CerS3, CerS4, and CerS5 knockdowns. For CerS1, we have previously shown that downregulation of CerS2 induced ER stress and autophagy in SMS-KCNR neuroblastoma cells and MCF-7 cells ( 8 ). Senkal et al. ( 36 ) have also shown that CerS6 downregulation results in decreased C16:0-Cer and ER stress via CHOP upregulation in head and neck squamous carcinoma cell lines. In this study, we found that siCerS2/5/6 treatment of MCF-7 cells also produced a characteristic sphingolipid profi le (i.e., high HexCer and dHCer levels with an increased proportion of medium-long-chain sphingolipids) that was associated with CHOP upregulation. The evidence that dysregulation of CerS activities elicits changes in both other enzymes of sphingolipid metabolism (including other CerS) as well as stress responses suggests that Cer synthesis is highly regulated and is a determinant of ER homeostasis. The challenge remains to fully triangulate and defi ne the lipids that specifi cally regulate the changes in CerS expression and the ER stress response.
In this study, we have demonstrated that downregulation of CerS causes multiple changes in sphingolipid metabolism and residual CerS expression. Many changes in sphingolipids refl ect the unique acyl-CoA preferences of CerS, but the robust maintenance of Cer and SM levels at the expense of HexCer, as well as the multiple changes seen in nontargeted CerS expression, suggests a high degree of redundancy and interregulation. Taken together, it is clear that CerS are not only important mediators of sphingolipid metabolism, but they are emerging as key regulators of ER homeostasis.
shunting. However, it is also likely that other non-CerS enzymes are being regulated to effect these changes in total dHCer and HexCer while maintaining Cer and SM levels. These results are in contrast to the effects of fumonisin B 1 (FB 1 ), an inhibitor of CerS ( 34 ), which depletes ceramide levels in MCF-7 cells (our unpublished observations).
According to our data, overall sphingolipid metabolism appears to be regulated such that total Cer levels are maintained at a constant level at the expense of accumulating other lipids, especially HexCer. In order for Cer levels to be maintained, it is possible that substrate supply may be increased by enhanced de novo synthesis of dHSph or augmented Sph production via the salvage pathway. Both of these hypotheses are plausible given the fact that siCerS2/5/6 treatment increased dHCer and Sph levels. However, Sph would have to be salvaged from a sphingolipid that is not included in our analysis in order for there to be a net addition to the sphingolipid pool. Supporting the de novo synthesis hypothesis are data showing that treatment of mice with FB 1 upregulates liver expression of the obligatory subunits of SPT, SPTLC1, and SPTLC2 ( 37 ). Alternatively, reduced sphingoid base degradation via decreased S1P lyase activity could result in increased S1P and Sph and allow maintenance of Cer/SM levels despite reduced CerS activity ( 38 ).
Only a few studies have examined how the loss of these proteins infl uences sphingolipid metabolism. Laviad et al. ( 25 ) showed that loss of CerS2 decreases activity of verylong-chain ceramide synthase activity and levels of verylong-chain ceramides ( 25 ). Mizutani et al. ( 26 ) knocked down CerS2 in HeLa cells overexpressing fatty acid 2-hydroxylase (FA2H) and found that very-long-chain HexCer were decreased along with a corresponding increase in long-chain HexCer species ( 26 ). In that latter study, CerS5 knockdown was shown to downregulate steady-state levels of long-chain HexCer species. More recently, our laboratory demonstrated that CerS2 knockdown in SMS-KCNR neuroblastoma cells increased long-chain Cer and SM ( 8 ).
Our understanding of the roles of CerS in sphingolipid metabolism has been advanced very recently by the generation of CerS2 knockout mice ( 35,39 ). These mice exhibit profound defects in very-long-chain sphingolipid synthesis in multiple tissues, and adult mice develop neuronal and hepatic dysfunctions. In the liver, CerS2 deficiency leads to a drop in total Cer, HexCer, and SM levels, and the chain length distribution shifts to one favoring C16:0 and other long-chain fatty acids ( 39 ). In the brain and kidney, however, total Cer, HexCer, and SM levels are much less affected by a loss of CerS2. This is likely due to the reduced dependence of these tissues on CerS2 (which is refl ected in their sphingolipid acyl chain composition) and the ability of residual CerS (e.g., CerS1 in the brain and CerS5 and CerS6 in the kidney) to compensate for the absence of CerS2 activity. Notably, CerS2 defi ciency was also associated with an increase in CerS5 and glucosylceramide synthase GCS expression in the liver ( 35 ).