Deep sphingolipidomic and metabolomic analyses of ceramide synthase 2 null mice reveal complex pathway-specific effects

Abstract

The sphingolipidome contains thousands of structurally distinct sphingolipid (SL) species. This enormous diversity is generated by the combination of different long-chain bases (LCBs), N-acyl chains and head groups. In mammals, LCBs are N-acylated with different fatty acids (from C14 to C32, with different degrees of saturation) by six ceramide synthases (CerS1-6) to generate dihydroceramides (DHCer), with each CerS exhibiting specificity toward acyl-Coenzyme As of defined chain length. CerS2 synthesizes very-long chain dihydroceramide, and mice in which CerS2 has been deleted display a number of pathologies. We now expand previous analyses of the mouse sphingolipidome by examining 259 individual SL species in 18 different tissues, building an extensive SL tissue atlas of WT and CerS2 null mice. Although many of the changes in SL levels were similar to those reported earlier, a number of unexpected findings in CerS2 null mouse tissues were observed, such as the decrease in ceramide 1-phosphate levels in the brain, the increase in C26-SL levels in the lung, and no changes in levels of ceramides containing t18:0-LCBs (phytosphinganine). Furthermore, analysis of levels of other metabolites revealed changes in at least six major metabolic pathways, including some that impinge upon the SL metabolism. Together, these data highlight the complex changes that occur in the lipidome and metabolome upon depletion of CerS2, indicating how sphingolipids are connected to many other pathways and that care must be taken when assigning a relationship between tissue pathology and one or other specific SL species.

The sphingolipid (SL) metabolic pathway has received much attention over the last couple of decades due to discoveries that SLs play vital roles in metabolism, in intracellular signaling and in human diseases (1). The latter include pathologies in which mutations affecting SL synthesis or degradation lead directly to disease, such as the monogenic lysosomal storage diseases (2), but also diseases associated with changes in SL levels, for which there is a less well-understood mechanistic pathway. For instance, SLs have been implicated in metabolic diseases (3) with SL levels affected in insulin resistance, obesity, dyslipidemia, inflammation, and steatosis (4, 5, 6, 7, 8, 9, 10, 11). The elevation of ceramides in blood is predictive of cardiovascular mortality, where d18:1/C16:0-, d18:1/C18:0-, d18:1/C24:0- and d18:1/C24:1-ceramide are independently associated with major adverse cardiovascular events and are used as biomarkers in the clinic (12).

Much of the work attempting to understand how changes in SL levels impact physiology has been conducted on genetically modified mice. Among these are mice defective in pathways of SL degradation, such as models of Gaucher or Tay-Sachs disease (13). In addition, mice defective in the biosynthetic pathway have been generated, in particular mice lacking one or other of the six mammalian ceramide synthases (CerSs) (14). The CerS have been associated with a number of human diseases. For instance, haploinsufficiency of CerS2 confers susceptibility to diet-induced steatohepatitis and insulin resistance (15). Other disease or syndromes associated with the CerS include skin disorders, various types of cancer, and multiple sclerosis (14).

The CerS have distinct activities because they generate different (dihydro)ceramide species by preferentially using acyl-CoAs of defined chain lengths (16). With the exception of CerS5 and CerS6, which can both utilize C16-CoA, the other CerS all have a relatively tight specificity toward acyl-CoAs (16). Among these are CerS2, which use C22-C24-CoAs (17). Although the preference of CerS for acyl-CoAs is clear, only limited data are available regarding the use of various long-chain bases (LCBs) (18, 19) which can contain between 16 and 20 carbon atoms and 0 to 2 double bonds in mammals (Fig. 1) (20).

Fig. 1.

The structure of some of the sphingoid LCBs analyzed in this study. The length of the LCB can vary between 16 and 20 carbons and can contain one or more double bonds. Hydroxyl moieties (blue) are indicated; m, 3-monohydroxy; d, 1,3-dihydroxy; t, 1,3,4-trihydroxy. The common names of the lipids are indicated. LCBs, long-chain bases.

CerS2 was the first CerS for which a genetically modified mouse became available (21, 22, 23). Multiple studies using this mouse have helped delineate a number of roles that CerS2 plays in cell physiology (14). Among these are roles in drug-induced liver injury (24), oxidative stress (25), hepatic insulin resistance (26) and pheochromocytoma (27). Although the molecular mechanism for some of these phenotypes has been determined (28), for others it is less understood, but clearly begins with the alterations in SL levels that occur upon CerS2 deletion.

Three main changes in SL levels have been observed in CerS2 null mice (21). First, as expected based on the acyl-CoA specificity of CerS2, levels of very-long chain (VLC) ceramides and the corresponding downstream SLs are reduced. Second, levels of the LCB, sphinganine (d18:0), one of the substrates of the CerS reaction, are elevated by up to ∼50-fold in the liver (21) and in the kidney (22) and ∼2 fold in the colon and the brain (27, 28). Thirdly, and unexpectedly, levels of long chain (LC) ceramides (i.e. C16-18), and the corresponding complex SLs, are elevated such that total ceramide levels (LC and VLC) are often unaltered. Although there is some variation in changes in the levels of ceramide and SL species in different tissues [i.e. brain, see (28)], these three main changes typically characterize most of the alterations described in CerS2 null mice.

In the current study, we extend our analysis of the sphingolipidome in 18 tissues isolated from WT and CerS2 null mice by quantifying 259 SL species using a comprehensive targeted LC-MS/MS approach (29), defining the tissue-specific quantitative profile not only of canonical SLs but also molecular species that were not measured previously in CerS2 null mice, such as phytosphingolipids, deoxysphingolipids (deoxySLs) (Fig. 1), ceramide-1-phosphate (C1P) and globosides. In addition, we perform metabolomics analysis of six CerS2 null mouse tissues and show changes in some pathways directly related to SL metabolism and changes in some pathways that do not appear to be directly related to SL metabolism. This study adds to the growing notion that altering one step of SL metabolism results in multiple changes in SL levels, which is presumably due to the complex modes of regulation of this important pathway, along with changes in levels of other lipids and metabolites. Although the latter might sound axiomatic, metabolic pathways are often considered in isolation, whereas a more holistic approach would be to take into account the relationship of the SL pathway to other metabolic pathways (30, 31) and to consider that some of the diseases caused by or related to defects in one step (i.e. one enzyme) of SL metabolism may actually be caused by changes in metabolites in a distinct pathway.

Materials and Methods

Mice

CerS2 null mice [C57BL/6 (B6) x 129SvEv] were generated as described (21, 22). Mice were maintained under special pathogen-free conditions and treated according to Animal Care Guidelines of the Weizmann Institute of Science Animal Care Committee and the National Institutes of Health's Guidelines for Animal Care. The experimental protocols were approved by the Weizmann Institute of Science’s IACUC, with animals treated according to the IACUC Animal Care Guidelines and the National Institutes of Health’s Guidelines for Animal Care. Mouse tissues were collected from 27-28 week-old male WT (n = 3) and CerS2 null mice (n = 4). Tissues were isolated and homogenates prepared using a Dounce homogenizer in double distilled water containing protease inhibitors. Homogenates were lyophilized and dry weight recorded.

Lipid extraction

Extraction was performed as described (32) with some modifications. Around 1 mg of dry homogenized tissue was transferred to 2 ml Eppendorf safe-lock tubes and 225 μl of methanol and 10 μl of internal standard (ISTD) were added followed by vortexing for 20 s. The ISTD mixture contained ceramide (4.7 μM; d18:1/8:0-Cer), dihydroceramide (2.33 μM; d18:0/8:0-Cer), deoxyceramide (1.06 μM; m18:0/12:0-Cer), hexosylceramide (3.40 μM; d18:1/8:0-HexCer), deuterated dihexosylceramide (2 μM; d18:1/16:0-Hex2Cer-d3), deuterated trihexosylceramide (1 μM; d18:1/18:0-Hex3Cer-d3), deuterated monosialodihexosylganglioside (5 μM; d18:1/18:0-GM3-d3), sphingomyelin (SM) (7.1 μM; d18:1/6:0-SM), deuterated sphinganine (1.62 μM; d18:0-LCB-d7), deuterated sphingosine (1.63 μM; d18:1-LCB-d7) and ceramide 1-phosphate (0.95 μM; d18:1/8:0-C1P). Samples were sonicated and vortexed for 20 s. The freeze-thaw cycle was repeated 3 times. Subsequently, 750 μl of methyl-tert-butyl-ether (MTBE) was added and samples were incubated in a thermomixer for 1 h at 25°C at 750 r.p.m. Tubes were vortexed for 20 s and 188 μl water was added to cause phase separation. The tubes were centrifuged for 5 min at 10,000 g and 600 μl of the upper phase was collected in a 2 ml Eppendorf safe-lock tube. The upper phase was dried and re-suspended in 200 μl Mobile Phase B (see next section) for lipidomic analysis while the lower phase was kept for metabolomics. The lipid extract (40 μl) was transferred to auto-sampler vials for analysis. Pooled extracts from each tissue were used as technical quality control samples and analyzed at regular intervals during the experiments.

LC-MS/MS analysis of lipids

The method used for our analyses was adapted from previously published methods (33, 34). Chromatographic separation was performed on an Agilent ZORBAX Eclipse plus Rapid Resolution HD column, C18 (2.1 × 100 mm, 1.8 μm) at 40°C in an Agilent 1290 UHPLC system. The flow rate was 0.4 ml/min for Mobile Phase A (60% methanol, 40% water, 0.2% formic acid and 10 mM ammonium acetate) and B (60% methanol, 40% isopropanol, 0.2% formic acid and 10 mM ammonium acetate) using the following gradient of concentrations: from 0% B to 10% B in 3 min, to 40% B at 5 min, 55% B at 5.30 min, 60% B at 8 min, 80% B at 8.50 min, 80% B at 10.50 min, 90% B at 16 min, 100% B at 22 min, 0% B at 22.10 min until 25.00 min. The Agilent 1290 UHPLC system was connected to an Agilent 6495 QQQ MS. AJS ESI source parameters were: dry gas temperature and flow at 200°C and 15 L/min respectively, nebulizer pressure 25 p.s.i., sheath gas temperature and flow were set to 200°C and 12 L/min respectively, capillary voltage and nozzle voltage were set to 3500 V and 500 V respectively, and the delta EMV was 200 V. Positive high/low pressure RF of the iFunnel was set to 210/110 and negative high/low pressure RF was 150/60.

The MS was operated in both positive and negative ionization modes and a dynamic multiple reaction monitoring method was used for the analysis. This method measured ceramide 1-phosphates, ceramides (containing LCBs d16:0, d16:1, d17:0, d17:1, d18:0, d18:1, d18:2, d20:1, t18:0), deoxyceramides (m18:0, m18:1), Hex4Cer (d18:1, d18:2), Hex3Cer (d16:1, d18:1, d18:2), Hex2Cer (d16:1, d18:1, d18:2), HexCer (d16:1, d18:0, d18:1, d18:2), GM3 (d18:1, d18:2), SM and LCB (d16:1, d18:0, d18:1, d18:2). The acyl chain length coverage was from C14 to C26, with the number of double bonds ranging from 0-3. The method included 622 transitions, with 303 transitions used as quantifiers and 319 transitions as qualifiers. Retention time windows were set either to 1 or 1.5 min and the cycle time was 750 msec. Collision energies varied based on the type of SL. Two microliters of tissue extract was injected for each analysis.

Lipidomic data analysis

MS data were analyzed using Agilent MassHunter Quantitative Analysis version B.08.00 and Agilent MassHunter Qualitative Analysis version B.07.00. Previously validated retention times of each lipid species were used as a reference to confirm peak identities. The qualifier peaks served as an additional reference. However, for certain low abundant lipids, the data were integrated based solely on the quantifier ion and on retention times, as the qualifier was below the detection limit. A fragment corresponding to the hydroxylated fatty acyl-amide was monitored to confirm identification of species containing hydroxylated fatty acids. For this study, the ceramides that contained unsaturated long chain bases (d16:1, d18:1, d18:2, d17:1 and d20:1) were normalized to d18:1/8:0-Cer ISTD. The ceramides that contained saturated long chain bases (d16:0, d18:0) were normalized to d18:0/8:0-Cer; deoxyceramides (m18:0, m18:1) were normalized to m18:0/12:0-Cer; phytoceramides (t18:0) were normalized to d18:0/8:0-Cer; HexCer species were normalized to d18:1/8:0-HexCer; Hex2Cer species were normalized to deuterated d18:1/16:0-Hex2Cer-d3; Hex3Cer and Hex4Cer species were normalized to deuterated d18:1/18:0-Hex3Cer-d3; GM3 species were normalized to deuterated d18:1/18:0-GM3-d3; SM species were normalized to deuterated d18:1/6:0-SM; d18:1-, d18:2-, d16:1- and d17:1-LCBs were normalized to deuterated d18:1-LCB-d7; d18:0-LCB was normalized to d18:0-LCB d7; C1P species were normalized to d18:1/8:0-C1P. As only a limited number of ISTD species was available for each subclass, our results are semi-quantitative. Technical quality control (TQC) samples were prepared by pooling lipid extracts from study samples. TQC samples were prepared for each tissue lipidomic analysis. TQC samples were injected every three or four samples, along with two at the beginning and another two at the end of the run. Signal-to-noise ratios were calculated using the raw peak areas in TQC samples and processed blanks. Lipids that showed signal-to-noise ratios <10, coefficient of variation (CoV) > 30%, and linearity of R2 <0.8 were discarded (CoV and linearity were calculated from TQCs). Some biologically important lipid species in specific tissues were retained if they did not pass only one of the quality control criteria (supplemental Data S1).

Metabolomic analysis

The lower phase of the methanol/methyl-tert-butyl-ether extracts generated for lipidomics was dried and reconstituted with 5% methanol and 95% water. Untargeted metabolomics data were acquired in both positive and negative ionization modes using a SCIEX TripleTOF® 6600. Mass spectrometry settings were as follows: gas1 50 V, gas2 60 V, curtain gas 25 V, source temperature 500°C, IonSpray voltage 5500 V, and collision energy 30 V. Metabolites were separated using an Agilent 1,290 Infinity II LC system on an Agilent BC-Poroshell HPH-C18 column (2.1 × 100 mm, 1.9 μm) at 45°C. The gradient elution was conducted with Mobile Phase A (water with 10 mM ammonium formate) and Mobile Phase B (50% methanol, 45% acetonitrile, 5% isopropanol, and 10 mM ammonium formate) at a flow rate of 0.2 ml/min. The LC gradient was as follows: 0 min 1% B, 1.5 min 1% B, 4.5 min 15% B, 8 min 50% B, 12 min 95% B, 13 min 95% B, 13.1 min 1% B, and 15 min 1% B. A hybrid workflow was employed in this project, combining information-dependent acquisition for metabolite annotation and SWATH for semiquantitative analysis. MetaboKit (35) was used for data processing. Compounds were identified at two confidence levels: MS/MS-level spectral matches through public libraries (NIST, MSDIAL, HMDB, and LipidBlast) and MS1-level precursor ion matches. A customized spectral library from information-dependent acquisition experiments was used to extract data features from SWATH data, which were then filtered based on a CoV <30% and detection frequencies.

Statistics

Lipidomics results were analyzed using R version 1.4.1106. For statistical analyses, data were transformed to a log₁₀ scale (nonavailable (NA) data, i.e. values that were below the limit of detection, were removed). Statistical significance was evaluated using a t test. Pearson correlation was evaluated and presented as a heatmap. Heatmaps and boxplots were created using the “ggplot2” package. For each lipid species, log₂ fold change (FC) was calculated as the base-2 log of the mean measured abundance in CerS2 null mice divided by the mean measured abundance in WT mice.

Results

Overview of changes in SL species in the CerS2 null mouse

In the current study, we measured levels of 259 individual SL species in 18 tissues from CerS2 null and WT mice (Supplemental Data 1). The SL molecular species included those that were previously analyzed in this mouse model, such as canonical LCBs, dihydroceramides, ceramides, hexosylceramides (HexCers), and SMs (17, 21, 22). We also measured a number of additional SLs, some of which were included in our recent determination of a mouse SL tissue atlas (29), along with others now measured using a more comprehensive semitargeted approach, such as SLs containing noncanonical LCBs (Fig. 1), N-acyl chains with an odd number of carbons and hydroxyl groups, C1Ps and complex glycosphingolipids (GSLs). Thus, these data not only help determine the changes that occur upon deleting a specific CerS but also supplement our previous characterization of SL levels in tissues of WT mice (29). Moreover, we describe how SL distribution changes in a tissue-specific manner, which likely depends on the structure of the SL and on levels of CerS2 expression, although the relationship between levels of mRNA encoding a specific CerS and the corresponding enzyme activity is not always linear (17).

Figure 2 shows the FC for some of these classes in eight representative tissues. Most of them displayed a diagonal spatter for all SL subtypes, indicating a negative correlation between the length of the acyl chain and the FC in CerS2 null mice, such that levels of VLC-SLs were reduced and levels of LC-SLs increased, consistent with previous data on the brain, liver, lung, and adrenal gland (21, 27, 28, 36). However, levels of SLs only detected in specific tissues, such as C1P, were reduced in the CerS2 null mouse, with no increase in concentration of LC-SLs. Levels of LCBs were elevated in 7 of the 8 tissues, with skin as the exception.

Fig. 2.

Changes in SL levels in selected CerS2 null mouse tissues. The x-axis shows the log₂ fold change (FC) for CerS2 null mice versus WT mice. A positive FC indicates lipids that were elevated in CerS2 null mice whereas a negative FC indicates a reduction. The y-axis of each plot is grouped and color-coded by lipid class, ordered by the length of the N-acyl chain, with the shorter N-acyl chains at the top of each lipid class and the longer chains on the bottom of each class. The size of the point is proportional to the P value of the fold change, with larger circles indicating greater significance. All chain lengths and hydroxylation variants are shown. LCB, long-chain base; Cer, ceramide; SM, sphingomyelin; HexCer, hexosylceramide; Hex2Cer, dihexosylceramide; Hex3Cer, trihexosylceramide; Hex4Cer, tetrahexosylceramide; GM3, monosialodihexosylganglioside; C1P, ceramide 1-phosphate. Prepared using BioRender. SL, sphingolipid.

Analysis of individual SL classes

SL synthesis begins with serine palmitoyltransferase (SPT), which generates LCBs that can vary in length and in the number of hydroxyl moieties (Fig. 1), followed by several enzymatic steps that result in the synthesis of complex SLs and GSLs (30, 31). We now describe SL changes observed in the CerS2 null mouse, following the order of their biosynthesis.

Long-chain bases

Previous analyses on CerS2 null mice only measured d18:0- and d18:1-LCBs in a few tissues, namely the brain (28), liver (21), kidney (22) and colon (37) in which levels of d18:1-LCBs were moderately increased (with the exception of the brain), while the extent of elevation of d18:0-LCBs in the liver, kidney and colon was much higher (21, 37). In the current study, changes in LCB levels in CerS2 null mice were similar to those previously reported and likewise in tissues that were not previously measured (Fig. 3 and supplemental Fig. S1). Thus, the increase in levels of d18:0-LCBs is greater than that of d18:1-LCBs, in line with our previous suggestion that this is due to alterations in the de novo synthesis pathway (38) rather than in the recycling pathway (39). Of note, the highest increase was in the lung, where d18:0- and d18:1-LCBs were elevated ∼220- and ∼15-fold in CerS2 null mice (Fig. 3). This is of considerable interest due to the role that LCBs may play in preventing viral infection in this tissue (40). Similarly, d18:0- and d18:1-LCBs were elevated in the spleen, colon, and kidney. Noncanonical LCBs (d16:1, d17:0, d17:1, and d18:2) were detected in 7 of the 18 tissues analyzed (Fig. 3 and supplemental Fig. S1) and showed a trend similar to the canonical ones. The skin differed from the other tissues since (i) d17:1-LCB was found at similar levels to d18:0- and d18:1-LCBs (Fig. 3), and (ii) it was not elevated in the CerS2 null mouse. The reason for this is unknown but it should be noted that the skin has a unique SL composition, rich in odd, branched, and hydroxylated chains, and it is particularly sensitive to changes in SL levels due to the critical roles that these lipids play in maintaining the skin permeability barrier (41). The d18:2-LCBs were detected at low levels in some tissues with the exception of the spleen and they did not increase in the CerS2 null mouse (Fig. 3).

Fig. 3.

LCB levels in selected tissues. The box represents the lower and upper quartile. The whiskers represent the minimum and maximum values, up to 1.5 times the interquartile range from the bottom or the top of the box to the furthest data point within that distance, thus excluding outliers. The median is shown in black and the mean is in red. ‘Null’ refers to the CerS2 null mouse. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; ∗∗∗∗P < 0.0001. Additional data are included in supplemental Fig. S1. LCB, long-chain base.

Dihydroceramides, ceramides, and deoxyceramides

Ceramides were considered according to their two main structural components, the LCB (m-, d-, and t-LCBs; Fig. 1, Table 1 and supplemental Table S1) and the N-acyl chain (Fig. 4). As expected (Table 1 and supplemental Table S1), d18:1-Cer is the predominant species in all tissues, with the exception of the skin, where d17:1-Cer was found at similar levels as d18:1-Cer.

Table 1.

Levels of ceramides containing different LCBs

Tissue	Sample	Total ceramides (pmol lipid/mg dry tissue Weight)										Total Cer
		d16.0	d16.1	d17.1	d18.0	d18.1	d18.2	d20.1	m18.0	m18.1	t18.0
Cerebellum	Null	N/A	N/A	3.11 ± 0.4	34.0 ± 3.0	626 ± 97.6	39.1 ± 11.9	147 ± 21.8	2.18 ± 0.4	1.25 ± 0.4	N/A	852 ± 130
	WT	N/A	N/A	1.19 ± 0.2	27.0 ± 9.6	792 ± 328	39.7 ± 11.9	205 ± 92.3	1.09 ± 0.3	1.94 ± 0.6	N/A	1,067 ± 440
Cortex	Null	N/A	1.30 ± 0.3	2.57 ± 0.6	50.6 ± 8.6	1,845 ± 419	97.76 ± 28.0	89.3 ± 12.2	3.06 ± 0.5	3.02 ± 0.5	4.93 ± 1.3	2,098 ± 462
	WT	N/A	0.83 ± 0.1	0.69 ± 0.1	16.8 ± 2.2	1,344 ± 264	92.4 ± 15.3	129.2 ± 6.8	0.50 ± 0.0	0.60 ± 0.3	5.72 ± 1.2	1,591 ± 275
Heart	Null	1.83 ± 0.1	1.21 ± 0.1	14.3 ± 0.8	18.7 ± 2.4	273 ± 37.9	79.0 ± 11.1	N/A	2.26 ± 0.3	3.49 ± 0.5	1.97 ± 0.1	396 ± 50.2
	WT	0.46 ± 0.2	0.56 ± 0.1	4.84 ± 0.5	3.29 ± 0.9	125 ± 21.1	49.3 ± 10.0	N/A	0.21 ± 0.0	0.57 ± 0.4	1.45 ± 0.1	186 ± 32.5
Kidney	Null	12.9 ± 3.6	1.49 ± 0.6	7.10 ± 1.0	24.1 ± 4.3	350 ± 99.1	86.4 ± 17.9	4.61 ± 1.1	1.28 ± 0.3	5.74 ± 2.7	4.77 ± 0.7	219 ± 49.9
	WT	1.47 ± 0.5	1.61 ± 1.5	5.31 ± 1.4	7.81 ± 1.9	412 ± 91.9	278 ± 97.7	5.86 ± 0.8	0.42 ± 0.0	8.43 ± 2.4	18.7 ± 4.7	291 ±85.7
Colon	Null	32.2 ± 10.2	2.89 ± 1.6	16.3 ± 6.3	57.2 ± 15.6	384 ± 197.0	19.1 ± 4.8	0.78 ± 0.4	1.70 ± 0.1	15.9 ± 11.7	30.6 ± 7.3	499 ± 112
	WT	11.32 ± 1.4	1.90 ± 0.7	9.03 ± 5.2	23.1 ± 2.1	433 ± 206	51.9 ± 27.5	1.00 ± 0.4	0.94 ± 0.3	8.86 ± 7.5	46.2 ± 1.7	740 ± 193
Liver	Null	14.2 ± 2.4	0.16 ± 0.1	5.16 ± 2.0	23.3 ± 5.3	173 ± 50.6	8.12 ± 2.7	N/A	0.81 ± 0.2	0.84 ± 0.5	0.77 ± 0.2	563 ± 223
	WT	0.84 ± 0.0	0.85 ± 0.1	13.2 ± 2.7	6.80 ± 1.0	283 ± 14.2	30.2 ± 5.5	N/A	0.52 ± 0.1	0.45 ± 0.3	1.21 ± 0.2	587 ± 231
Lung	Null	34.3 ± 6.2	0.48 ± 0.1	2.00 ± 0.2	56.8 ± 8.4	546 ± 35.6	39.1 ± 5.3	0.34 ± 0.0	3.22 ± 0.7	1.71 ± 0.5	1.71 ± 0.3	227 ± 62.5
	WT	1.44 ± 0.2	0.54 ± 0.0	2.71 ± 0.3	10.3 ± 2.6	343 ± 28.1	71.1 ± 8.7	1.65 ± 0.1	0.48 ± 0.0	2.63 ± 1.1	8.25 ± 0.5	337 ± 16.7
Plasma	Null	2.65 ± 1.6	N/A	0.32 ± 0.0	3.77 ± 2.0	17.31 ± 1.4	1.42 ± 1.4	N/A	0.23 ± 0.0	0.20 ± 0.2	1.38 ± 0.4	685 ± 49.3
	WT	0.29 ± 0.1	N/A	1.05 ± 0.2	1.26 ± 0.4	34.6 ± 9.2	3.50 ± 0.8	N/A	0.09 ± 0.0	0.19 ± 0.1	1.40 ± 0.1	442 ± 33.5
Skin	Null	92.2 ± 28.1	71.9 ± 19.8	742 ± 210	160 ± 45.7	694 ± 181	13.2 ± 3.9	N/A	2.28 ± 0.5	0.86 ± 0.2	N/A	1,776 ± 484
	WT	114 ± 25.7	92.4 ± 12.7	988 ± 197	233 ± 45.9	1,159 ± 209	69.4 ± 7.9	N/A	1.16 ± 0.1	0.77 ± 0.2	N/A	26,586 ± 452
Small intestine	Null	48.0 ± 35.5	0.21 ± 0.1	6.73 ± 4.5	76.0 ± 57.3	257 ± 172	21.7 ± 14.0	N/A	3.30 ± 1.3	0.51 ± 0.2	N/A	414 ± 282
	WT	5.09 ± 0.3	0.20 ± 0.1	5.13 ± 2.1	16.0 ± 4.9	184 ± 60.9	24.0 ± 12.2	N/A	1.45 ± 0.1	1.52 ± 0.5	N/A	238 ± 71.1
Spleen	Null	17.6 ± 4.7	0.09 ± 0.0	3.56 ± 0.5	30.3 ± 6.4	287 ± 59.4	10.1 ± 2.0	N/A	3.08 ± 0.6	1.74 ± 0.4	1.53 ± 0.5	355 ± 71.7
	WT	2.56 ± 1.5	0.14 ± 0.1	2.71 ± 1.3	7.22 ± 4.4	189 ± 86.3	15.0 ± 8.2	0.07 ± 0.0	1.23 ± 0.6	2.48 ± 2.1	1.01 ± 0.5	222 ± 104

LCB, long-chain base.

Levels of ceramide (containing all N-acyl species) with various LCBs are shown for selected tissues. See supplemental Table S1 for additional tissues. N/A, not available (below limit of detection). Data are averages +SD, n = 3–4.

Fig. 4.

m18:1-ceramides are less affected than d18:1- and t18:1-ceramides in CerS2 null mice. log₂ fold change (FC) of (A), m-, d-, and t-18:0-ceramides and of (B) m-, d-, and t-18:1-ceramides in CerS2 null mice versus WT mice. Each point represents levels in a specific tissue, where the size of the point is proportional to the P value of the FC, with larger circles indicating greater significance as indicated in the inset. Points are colored according to the number of hydroxyl groups on the LCB, namely m, green; d, orange; t, purple. Additional data are included in supplemental Fig. S5. (C) m18:0- and m18:1-ceramide levels in the liver of WT and CerS2 null mice. The box represents the lower and upper quartile. The whiskers represent the minimum and maximum values, up to 1.5 times the interquartile range from the bottom or the top of the box to the furthest data point within that distance, thus excluding outliers. The median is shown in black and the mean is shown in red. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001. LCB, long-chain base.

Results for ceramides in the 18 tissues measured (Table 1, supplemental Table S1 and supplemental Fig. S2A, B) were generally consistent with previous studies (21, 22, 27, 28, 36, 37) since levels of VLC-ceramides decreased, while levels of LC-ceramides increased in CerS2 null mice (supplemental Fig. S2). Ceramides containing saturated LCBs (ie, d16:0 and d18:0) were elevated in most tissues while, in contrast, ceramides containing t18:0-LCBs (phytosphinganine), detected in about half of the tissues, were unaltered or reduced (Table 1). Phytosphinganine-based SLs (t18:0) are not major species in mammals and their levels were much lower than the other ceramides (Table 1).

Sphingadiene (d18:2)-based ceramides were the second most abundant species, ranging from ∼2 to 37% of the total ceramides depending on the tissue (Table 1 and supplemental Table S1), while d20:1-Cer was mainly found in the brain (cerebellum and cortex) and its levels were unaltered in CerS2 null mice (Table 1 and supplemental Table S1). The skin showed the greatest number of ceramide species, in decreasing order of abundance as d17:1>d18:1>d17:0>d18:0>d16:0>d16:1-Cer and their levels were essentially unaltered in the CerS2 null mouse. The small intestine contained a relatively high level of d16:0-Cer compared to other tissues.

The distribution of ceramide N-acyl chains suggested that the LCB structure did not have a large effect on the elevation of LC-Cer and in the decrease in VLC-Cer species in CerS null mice (Fig. 4A, B), with the exception of SLs containing mLCBs (ie deoxyceramides), whose levels did not follow the pattern of ceramides containing d- or t-LCBs (Fig. 4C). Deoxyceramides (m18:0-Cer and m18:1-Cer) are generated upon use of alanine (or glycine) by SPT rather than the canonical serine, and accumulate as toxic compounds under various pathological conditions such as neuropathy and diabetes (42). m18:0-Cer in WT tissues was predominantly found in C22-C24-ceramides, while m18:1-C20-Cer was the most abundant deoxyceramide in both WT and CerS2 null mice (supplemental Data). In general, deoxyceramides increased or showed no significant change in CerS2 null mice in several tissues (Table 1 and supplemental Table S1), with only C24:1-deoxyceramides (especially the m18:0 class) displaying a decrease. This is illustrated in Fig. 4C, which shows levels of m18:0/LC-Cer and m18:0/VLC-Cer elevated or not changed in the liver of the CerS2 null mouse. This suggests that deoxyceramides may be regulated differently from ceramides, implying that other enzymes of SL metabolism impinge upon the level of m18:0-VLC-Cer.

Interestingly, although levels of d18:2-Cer containing VLC-fatty acids were decreased in the CerS2 null mouse in most tissues as expected, the corresponding increase in d18:2/LC-Cer levels was observed only in the skeletal muscle and lung (supplemental Fig. S2C). Considering that the addition of the second double bond (14,15; diene) occurs due to dehydrogenation of ceramide via the action of FADS3 (43), this finding could suggest a novel mode of regulation of LC d18:2-Cer levels.

Ceramide 1-phosphate

C1P, which is generated by the phosphorylation of ceramide by CERK (44), was not previously measured in CerS2 null mice. C1P is a critical regulator of cell proliferation and apoptosis (45, 46), phagocytosis (47, 48), and inflammation (49, 50). We detected two C1P species, d18:1/16:0 and d18:1/C18:0 in the brain (cortex and cerebellum). Unexpectedly, and in contrast to their nonphosphorylated derivatives, levels of LC-C1P decreased in CerS2 null mice, with d18:1/C18:0-C1P decreasing by ∼100 fold in the cerebellum (Fig. 5). These data suggest a novel mode of C1P regulation that is independent of the availability of their precursor (ceramide).

Fig. 5.

C1P levels are reduced in CerS2 null mice brain tissues. The box represents the lower and upper quartile. The whiskers represent the minimum and maximum values, up to 1.5 times the interquartile range from the bottom or the top of the box to the furthest data point within that distance, thus excluding outliers. The median is shown in black and the mean is shown in red. ∗∗P < 0.01; ∗∗∗P < 0.001. C1P, ceramide 1-phosphate.

Sphingomyelins

Our analyses covered SM molecular species containing C32-C44 and 0–3 double bonds (supplemental Fig. S3). In general, the changes in SM levels were consistent with previous studies and paralleled the observed changes in ceramide levels. The highest levels of SM were measured in the kidney and the lowest levels in skeletal muscle. In the majority of tissues, SM 34:1 (most likely d18:1/16:0 SM) was the most abundant species. C32-38 SM species were elevated in CerS2 null mice (although this effect was not significant for SM containing d18:2-LCB) while C39-44-SM was reduced, irrespective of the LCB backbone (d18:0, d18:1, d18:2) (supplemental Fig. S3).

Glycosphingolipids

Due to technical limitations of our methodology based on reversed-phase chromatography, we were unable to distinguish between GlcCer and GalCer and therefore we report these lipid species as Hex(n)Cer. LC- and VLC-HexCer were detected in most tissues. Since the extent of the decrease of VLC-HexCer levels was larger than the compensatory increase of LC-HexCer (supplemental Fig. S4A), the total levels of HexCer decreased in most of the tissues analyzed in CerS2 null mice (Fig. 6A). Interestingly, the opposite was true for Hex2Cer. These findings are also probably due to the different distributions of fatty acyl chains in Hex- and Hex2Cer. Although in most tissues the most abundant HexCer species contain VLC-fatty acids, Hex2Cer mainly contains LC-fatty acids (29). Thus, even a large decrease of VLC-Hex2Cer in the CerS2 null mouse will not be enough to decrease their total levels, except in specific tissues such as the cerebellum and skin, while the corresponding increase of the LC species would cause an increase in total Hex2Cer levels (supplemental Fig. S4E).

Fig. 6.

GSL levels in CerS2 null mice. A: Total HexCer levels are shown for WT (black) and CerS2 null mice (gray). Data are mean ± SD. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001. B: Ratios of levels of d18:1/C26:0-HexCer in CerS2 null mice tissues versus WT. Data are mean ± SD. ∗P < 0.05; ∗∗P < 0.01. C: Levels of GM3 shown as a log₂ fold change (FC) between WT and CerS2 null mice. Data are shown as a heatmap where red corresponds to increased levels in the CerS2 null mouse, blue indicates a decrease and gray indicates lipids that were not detected. •, P < 0.05. GSL, glycosphingolipid; HexCer, hexosylceramide.

Due to their relative abundance and ionization properties, only a small number of Hex3Cer and Hex4Cer species could be quantified and only in a small number of tissues. The kidney and lungs contain high levels of Hex3Cer in WT mice. Ten or more molecular species of Hex3Cer were observed in the kidney, while these lipids were not quantifiable in the cerebellum due to their low concentration. The inguinal lymph nodes, kidney, and spleen contained more Hex4Cer species than other tissues (supplemental Fig. S4). For these GSLs, the behavior of the CerS2 null model was consistent, with a more modest compensatory increase of LC species which was independent of the LCB backbone. Likewise, GM3 containing either d18:1 or d18:2-LCBs were detected in only few tissues, with levels of VLC-GM3 decreasing and levels of LC-GM3 increasing in the CerS2 null mouse (Fig. 6C).

Analysis of low abundant and noncanonical SLs

SLs containing C26:0 and C26:1-fatty acyl chains in the lung

Although the expression level of CerS2 is very high in the lung, this tissue showed an anomalous profile compared to all the others, since levels of Cer, HexCer, and C26:0-SM and C26:1-SM were all elevated, irrespective of the LCB (d18:0 or d18:1) in CerS2 null mice (Fig. 7). We suggest that C26-Cer is mainly generated by CerS3 (51), and even though C26-Cer levels are much lower than other ceramide species, CerS3 may be activated upon loss of the other VLC-SL synthesized by CerS2 as a compensatory mechanism. The relevance of these specific ceramides in the lung has not been investigated but the SL acyl chain composition contributes to the T cell-mediated inflammatory response in asthma (52).

Fig. 7.

SLs with C26 N-acyl chains are elevated in CerS2 null mouse lung. Levels of C26-ceramide, -HexCer and -SM in the lung of WT and CerS2 null mice. The box represents the lower and upper quartile. The whiskers represent the minimum and maximum values, up to 1.5 times the interquartile range from the bottom or the top of the box to the furthest data point within that distance, thus excluding outliers. The median is shown in black and the mean is shown in red. Data are pmol lipid/mg dry tissue weight. ∗P < 0.05; ∗∗P < 0.01. HexCer, hexosylceramide; SL, sphingolipid.

SLs containing odd chain LCBs or odd chain N-acyl groups

In the presence of the SPTLC3 subunit, SPT can generate straight and branched LCBs with distinct properties, at least in cultured cells (53). In our study, when considering the SPTLC3-specific LCBs, d17:1-SLs were the most abundant (Supplemental Data 1) in almost all tissues. As we were not able to quantify d19-containing SLs, we report here only the relative concentration of d17-based ceramides and SMs. In the CerS2 null mouse, the behavior of odd chain LCB-containing ceramides (supplemental Fig. S5) and SM (supplemental Fig. S6) was the same as that displayed by the other species, with an increase of LC and a significant decrease of VLC ceramides and SM.

When considering SLs containing odd chain N-acyl groups, we were able to measure ceramides (supplemental Fig. S2) and SM (supplemental Fig. S6) containing C17, C19, C23, and C25 fatty acids. Their distribution and behavior in the CerS2 null mouse recapitulated data for species containing an even number of carbons (supplemental Figs. S2 and S6).

SLs containing hydroxylated LCB and FAs

SLs can be hydroxylated on the LCB (to generate phytosphingosine, t18:0) or on the N-acyl chain. We measured a number of hydroxylated N-acyl chains in ceramides and in HexCer in seven representative tissues: plasma, brain, heart, kidney, liver, skin and small intestine from WT (n = 3) and CerS2-deficient mice (n = 4). A targeted list of potential hydroxylated species was generated with corresponding the retention times as a function of the bound chain length. When nonhydroxylated Cers were thereby taken as reference, species containing one hydroxyl group showed increased polarity and a systematic shift to earlier retention times. Isomers carrying a hydroxyl group on the sphingoid base or on the FA coeluted in our method but were distinguished by the different mass of the precursor ion (2 Da). The multiple reaction monitoring transitions used to identify and quantify hydroxylated SL in positive mode included the sphingoid base and the fatty acyl-amide fragments, enabling the identification of the FA type.

Overall, 6 phytoCer, 12 Cer(OH), and 11 HexCer(OH) species were identified. The species measured included only d18:1-LCBs or phytosphingosine (t18:0). Other sphingoid bases were not analyzed and, unexpectedly, no hydroxylated SM species could be identified. Small intestine and skin contain particularly high proportions of hydroxylated sphingolipids, including phytoceramides (Supplemental Data 1). For single molecular species, C16(OH) FAs were the most abundant (Fig. 8), except for the kidney and brain where the distribution of hydroxylated FAs was more heterogeneous (Supplemental Data 1). A large variety of hydroxylated FAs was present in HexCer(OH) although different total amounts of hydroxylated SL were recorded in different CerS2 null tissues (Supplemental Data 1). The total concentration of both Cer(OH) and HexCer(OH) in the brain decreased in the CerS2 null mice relative to the WT mice. In previous analyses (28) d18:1/C18:0(OH)-HexCer levels increased in the brain of CerS2 null mice; this SL species was also elevated in CerS2 null brain tissue in the present study. Thus, the decrease in total HexCer(OH) concentration in the brain appears to be due to the large reduction in the levels of HexCer d18:0/24:1(OH), and to a lesser extent HexCer d18:1/22:0(OH) (Supplemental Data 1).

Fig. 8.

Levels of d18:1/C16:0 and d18:1/C16:0(OH) SLs in selected tissues. A: Levels of d18:1/C16:0-Cer (top) and concentration of d18:1/C16:0(OH)-Cer (bottom) in selected tissues of WT and CerS2 null mice. B: Levels of d18:1/C16:0-HexCer (top) and concentration of d18:1/C16:0(OH)-HexCer (bottom) in selected tissues of WT and CerS2 null mice. Bars indicate the mean. Error bars show a single standard deviation. HexCer, hexosylceramide; SL, sphingolipid.

Levels of polar metabolites in CerS2 null mouse

To determine the relationship between altered SL levels and the metabolome, we next measured levels of ∼800 metabolites, using a mass spectrometry-based untargeted approach in six WT and six CerS2 null tissues: brain, lung, liver, kidney, spleen, and heart. These metabolites belonged to 50 different metabolic pathways. Principal component analysis and partial least squares-discriminant analysis plots illustrate that the differences in metabolite levels between organs are more pronounced than that between WT and CerS2 null mice (supplemental Fig. S7A), highlighting the specific metabolic fingerprint of each tissue. Several pathways were selected for further analysis if at least 3 metabolites (in that pathway) showed significant changes between WT and CerS2 null in at least one tissue (supplemental Fig. S7B).

Analysis of lysophospholipids revealed significant increases in levels of polyunsaturated lysophosphatidylglycerol, lysophosphatidylcholine, lysophosphatidylethanolamine, and lysophosphatidylserine whereas saturated and monounsaturated species were detected at significantly lower amounts, with the most pronounced effects in the brain, liver, lung, and kidney (Fig. 9). Levels of short-acyl chain acylcarnitines also displayed significant changes, with very-short chain acylcarnitines accumulating in the brain and long chain acylcarnitines increasing in the lung (supplemental Fig. S7C). Dysregulation of acylcarnitine levels (15, 54) in CerS2 null mice is consistent with a previously observed impaired mitochondrial function (25) that may also be caused by the accumulation of toxic deoxysphingolipids (55).

Fig. 9.

Lysophospholipids in CerS2 null mice. Lysophospholipid levels were measured in seven tissues and are shown as a volcano plot with the x-axis as log₂ fold change (FC) and the y-axis as minus log₁₀ (P-value). Lysophospholipids with a linear FC > 1.5 (indicated by vertical dashed gray lines) and P < 0.05 (indicated by horizontal solid gray line) are indicated.

Of the pathways chosen for further analysis, six were particularly prominent (Fig. 10) and included inositol phosphate, linoleic acid, galactose, methionine, purine, and pyrimidine metabolic pathways. In the brain, levels of myo-inositol phosphates significantly decreased in CerS2 null mice (Fig. 10A), perhaps implying impaired neural function and cell signaling. Activation of the alpha-linolenic and linoleic acid metabolic pathways was observed in the lung, with a notable increase in levels of linolenic acid (18:3 fatty acid) and stearidonic acid (18:4 fatty acid), linoleic acid (18:2 fatty acid), arachidonic acid (20:4 fatty acid), adrenic acid (22:4 fatty acid), docosahexaenoic acid (22:6 fatty acid), and docosapentaenoic acid (22:5 fatty acid) (Fig. 10B). These omega-6 and omega-3 polyunsaturated fatty acids (PUFAs) have both proinflammatory and anti-inflammatory effects (56, 57, 58), and their accumulation in the lung may indicate a response to airway inflammation (40). In contrast, the levels of many PUFAs decreased in the heart (Fig. 10B). A noticeable decrease in levels of many components of the galactose metabolic pathway (galactose, sucrose, raffinose, stachyose, galactitol, and alpha-lactose) was detected in the liver of CerS2 null mice (Fig. 10C), consistent with earlier transcriptomic analysis (22) in which a number of genes associated with galactose metabolism were downregulated. Whether any of the changes in galactose metabolic pathway are associated with possible changes in galactosylceramide levels cannot be determined since we were unable to separate Glc- and Gal-containing sphingolipids. Methionine metabolism was perturbed in the kidney of CerS2 null mice, where levels of methionine, 5′-methylthioadenosine and betaine were increased (Fig. 10D). Purine and pyrimidine metabolism were altered in a number of organs, including the liver, spleen, brain, and lung of CerS2 null mice (Fig. 10E, F).

Fig. 10.

Analysis of metabolomics pathways in CerS2 null mice. Metabolite levels were measured in seven tissues across six pathways involved in the metabolism of: (A) inositol phosphate, (B) alpha linolenic acid and linoleic acid, (C) galactose, (D) methionine, (E) purine, and (F) pyrimidine. The log₂ fold change (FC) of metabolite levels are shown as a heatmap for each pathway, red indicates an increase and blue indicates a decrease according to the inset legend. The pathways shown here are those that were significantly altered in CerS2 null compared to WT mice (see supplemental Fig. S7). The number in the bracket indicates the number of metabolic pathways that the metabolite is involved in. A smaller number suggests that the metabolite is more specific to the corresponding pathway. ∗P < 0.05; ∗∗P < 0.01. #, metabolites identified at the MS1 level.

Recently, we introduced the concept of the anteome, defined as those metabolic pathways which directly impinge upon the metabolic SL pathway under study (30, 31). Thus, the anteome includes pathways that generate, for instance, acyl CoA, pyridoxal phosphate, NADPH, phosphatidylcholine, and the amino acids required for SPT activity (30, 31). In the lung, at least one metabolite from each of these pathways is decreased in CerS2 null mice. Few changes are observed in the heart, whereas all of the significant changes in the kidney pointed toward an increase in the level of metabolites (Fig. 11). Interestingly, while the variation in SL levels induced by the absence of CerS2 is reproducibly observed across tissues, the anteome variations are extremely tissue-specific (supplemental Fig. S7A). This first attempt to associate variations in the SL pathway and the anteome shows that there is no clear association between them across tissues, and that this kind of study might be better served using targeted rather than nontargeted approaches, to increase the molecular coverage of the pathways of interest.

Fig. 11.

Metabolomics pathway analysis for components of the SL biosynthetic pathway anteome. Selected pathways produce metabolites which serve as direct substrates for SL biosynthesis (substrates are indicated by labels and arrows). Pathways shown here were significantly altered in the CerS2 null mouse compared to WT (see supplemental Fig. S7). Metabolite levels were measured in 7 tissues. The log₂ fold change (FC) of metabolite levels are shown as a heatmap for each pathway, red indicates an increase and blue indicates a decrease according to the inset legend. ∗P < 0.05; ∗∗P < 0.01. SL, sphingolipid.

Discussion

The huge advances in lipidomic analysis over the past 10–15 years (59) have greatly increased our appreciation of the vast variety of lipid species in biological tissues (60) and how levels of lipids change in different physiological and pathophysiological settings. In the case of SLs, the number of species that can be isolated and measured has increased substantially, such that in the current study we were able to measure 259 individual SL species. A previous study generated an atlas of SL species in WT mouse tissues (29), and in addition to extending this atlas, we now show changes in levels of many of these 259 SL species in 18 tissues from the CerS2 null mouse (21). Moreover, while there is some logic in assuming that many of the pathologies observed in a CerS null mice might be caused by changes in levels of the relevant SLs, this is not necessarily the case, since metabolic pathways are tightly connected and it is self-evident that changes in one pathway are likely to cause or result in changes in other pathways. With this in mind, we examined a large number of metabolites, including ones that are associated with the anteome (30, 31) of the SL biosynthetic pathway.

In general, most of the changes reported previously (21) were also observed in the current study, namely a highly significant reduction in levels of VLC-SLs and an increase in LC-SLs and sphinganine (d18:0-LCB). More in detail, a higher FC for LC-SLs affected species containing a saturated LCB (d16:0 and d18:0) and the ones containing the 16:0 N-acyl group. However, the generally observed compensatory increase of LC-SLs was absent in specific tissues, such as Colon, Skin and White Blood Cells (supplemental Fig. S8), where we report a decrease of these sphingolipids. However, upon analysis of levels of SLs that were not measured previously, a number of unexpected findings were obtained. Among these is the large decrease in C1P levels in the two brain regions in which C1P was detected. Unexpectedly, d18:1/-, d18:0/C16:0, d18:1/- and d18:0/18:0-C1P levels were decreased, whereas levels of the corresponding non-phosphorylated ceramides were significantly increased. Whether this is also relevant to the observation that C1P/sphingosine-1-phosphate axis is a key target for neurodegenerative diseases (61) is not known, but the data do suggest that the pathway of C1P generation is regulated in such a way that levels of C1P are not directly related to their precursors level (C16 or C18-ceramide), both of which are elevated in the CerS2 null mouse brain.

Some other changes in levels of SLs in the brain might also suggest that our earlier hypothesis about the cause of brain pathology in this model may need to be revised. Thus, our assumption (28) that the encephalopathy associated with very low levels of VLC-ceramides is likely caused by loss of galactosylceramide appears somewhat simplistic in light of the many changes in brain lipids observed herein. Levels of other lipids and metabolites, such as the inositol phosphates, were also altered in the brain but whether these are linked to myelin loss is not known.

The lung is also a tissue of great interest in SL research, partly due to the role of SLs in combatting infection (40). Levels of sphinganine (d18:0-LCB), sphingosine (d18:1-LCB) and d18:2-LCB were highly elevated in the lung of CerS2 null mice, as were levels of ceramide with C26:0 or C26:1 acyl chains. Our metabolomic analysis revealed that levels of lysophospholipids, especially those containing PUFAs, were significantly elevated in our model. Lysophosphatidylcholines are generated in the cell membranes of epithelial cells and leukocytes, they can induce inflammatory responses and cytokine production (62), with monocyte chemotaxis and macrophage activation (63, 64) in the lung, explaining this part of the phenotype. The hydrolysis of phospholipids by secreted phospholipases (sPLA2) is activated in the inflamed lung, causing extracellular surfactant damage (65). Phosphatidylcholine and phosphatidylglycerol predominate in surfactant phospholipid composition (80% and 10% of total lipid, respectively) and multiple studies have demonstrated the ability of lysophospholipids, generated by sPLA2, to disrupt the surface activity of phospholipid mixtures in inflamed lungs (66). Along with changes in levels of acylcarnitines, linoleic acid metabolism, and galactose metabolism, the lung is an excellent example to illustrate the multiple changes affecting lipid levels in a tissue upon knocking out CerS2 and also illustrates that attempts to assign a pathologic phenotype to one or the other lipid or metabolite are likely to fail.

This latter point is also suggested by evaluating changes in the skin upon depletion of CerS2. Skin contains the largest spectrum of SLs, both canonical and noncanonical, such that d17:1-SL are found at levels similar to those of d18:1-SL. The skin behaves differently from most of the tissues in the null mice, showing no decrease in levels of VLC-containing species. The skin is one of the few tissues apart from the brain that contains high levels of GSLs, including HexCer and GM3. The unique roles of glycosylated and ultra-long SLs in maintaining the skin permeability barrier likely adds to the tight regulation of all SLs in this tissue (67). In this case, deficiency in ceramide synthase 3 (CerS3), the enzyme required for ultra-long SL synthesis, results in complete loss of skin barrier and lethal phenotype (41, 68). We hypothesize the existence of a compensatory mechanism that supports the synthesis of VLC-SL in the skin of CerS2 null mice. Similar compensatory shifts in SL acyl chain lengths have been observed in CerS6 null mice (69, 70) and CerS5/6 knockdown MCF-7 cells (71), with suggested compensatory mechanisms including increased transcription of nontargeted CerS isoforms and increased usage of excess long chain bases to protect total SL levels.

The expected variation in LC versus VLC in CerS2 null tissues was also not seen for deoxyceramides, which increased or showed no significant change (Table 1 and supplemental Table S1). These toxic species seem to accumulate in the CerS2 null mouse, independent of their structural characteristics. This hints at a different regulation of deoxyceramide synthesis when compared to canonical ceramides. Their accumulation could also explain the impaired mitochondrial activity previously observed in this mouse (25). DeoxySLs tend to accumulate in the mitochondria and trigger fission of these organelles, disrupting their activity (55).

We also report here a panel of odd chain-containing SLs, whose levels changed in the CerS2 null mouse similarly to their even chain counterparts. Diet and microbiota also contribute to the odd chain LCB profile of the host. Although d19-LCBs are formed as two isomeric species, with SPTLC1 in combination with SPTLC3 generating anteiso-branched-C18 LCBs (meC18LCB) from anteiso-methyl-palmitate (synthesized from the isoleucine catabolic pathway) in addition to a linear C19 LCB generated from heptadecanoate (C17:0), the more abundant linear d17 LCB is formed directly from pentadecanoate (C15:0). In this study, we detected several d17-SL species, while we were not able to quantify any d19-SLs, supporting previous findings that reported presence of methyl branched chain LCB only in human plasma and not in mice. However, the presence of these odd chain species might affect cellular processes, such for genetic variants of SPTLC3 that are associated with metabolic conditions such as dyslipidemia and cardiac disease risk (53).

Our study has some limitations. For instance, the mice analyzed only included males (27-28-week-old); thus, we were unable to determine the effect of sex and age, both of which are known to affect SL levels (29, 72). In addition, since our method was not able to separate GSL isomers, i.e. GlcCer and GalCer, we were limited to describing changes in GSL levels using a generic designation, namely HexCer. Finally, we did not analyze sphingosine-1-phosphate levels as the analyses did not pass a quality control filter.

In summary, we have demonstrated that deleting one enzyme in the SL biosynthetic pathway has multiple effects on levels of both SLs, other lipids and on apparently unrelated metabolites. Although it may appear obvious that metabolic pathways are tightly interconnected, most studies tend to focus on the metabolites or pathways under study, which could lead to misinterpretation of, for instance, the relationship between changes in levels of a limited number of metabolites and disease pathophysiology.

Data availability

All data were deposited in https://data.mendeley.com/preview/f9zyrzcfpd?a=61258894-8fa9-41e6-a1ca-6f8009c0fbbc.

Supplemental data

This article contains supplemental data.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.