A rapid and quantitative LC-MS/MS method to profile sphingolipids.

Sphingolipids comprise a highly diverse and complex class of molecules that serve not only as structural components of membranes but also as signaling molecules. To understand the differential role of sphingolipids in a regulatory network, it is important to use specific and quantitative methods. We developed a novel LC-MS/MS method for the rapid, simultaneous quantification of sphingolipid metabolites, including sphingosine, sphinganine, phyto-sphingosine, di- and trimethyl-sphingosine, sphingosylphosphorylcholine, hexosylceramide, lactosylceramide, ceramide-1-phosphate, and dihydroceramide-1-phosphate. Appropriate internal standards (ISs) were added prior to lipid extraction. In contrast to most published methods based on reversed phase chromatography, we used hydrophilic interaction liquid chromatography and achieved good peak shapes, a short analysis time of 4.5 min, and, most importantly, coelution of analytes and their respective ISs. To avoid an overestimation of species concentrations, peak areas were corrected regarding isotopic overlap where necessary. Quantification was achieved by standard addition of naturally occurring sphingolipid species to the sample matrix. The method showed excellent precision, accuracy, detection limits, and robustness. As an example, sphingolipid species were quantified in fibroblasts treated with myriocin or sphingosine-kinase inhibitor. In summary, this method represents a valuable tool to evaluate the role of sphingolipids in the regulation of cell functions.


Chemicals and solutions
Butanol, methanol (HPLC grade), and formic acid (98-100%, for analysis) were purchased from Merck (Darmstadt, Germany).   mented with L-glutamine and 10% fetal calf serum in a humidifi ed 5% CO 2 atmosphere at 37°C. For lipid analysis, cells were seeded into 6-well plates and grown to confl uence. Cells were rinsed two times with ice-cold PBS and either lysed in 0.2% SDS in water or scraped in PBS. Subsequently, samples were subjected to centrifugation at 240 g for 7 min, and the resulting pellet was homogenized in distilled water by sonication. Fibroblasts treated with myriocin or SKI (Calbiochem) were lysed in 0.2% SDS ( Fig.  4) . Aliquots of the cell homogenates were taken for protein determination. Protein concentrations were measured using bicinchoninic acid as described previously ( 23 ).

Cell culture
Primary human skin fi broblasts were cultured as described previously ( 22 ) in Dulbecco's modifi ed Eagle's medium supple-

Sphingolipid analysis by LC-MS/MS
Sphingolipid analysis was performed by LC-MS/MS. The HPLC equipment consisted of a 1200 series binary pump (G1312B), a 1200 series isocratic pump (G1310A), and a degasser (G1379B) (Agilent, Waldbronn, Germany) connected to an HTC Pal autosampler (CTC Analytics, Zwingen, Switzerland). A hybrid triple quadrupole linear ion trap mass spectrometer API 4000 Q-Trap equipped with a Turbo V source ion spray operating in positive ESI mode was used for detection (Applied Biosystems, Darmstadt, Germany). High purity nitrogen was produced by a nitrogen generator NGM 22-LC/MS (cmc Instruments, Eschborn, Germany).
Gradient chromatographic separation was performed on an Interchim (Montlucan, France) HILIC silica column (50 × 2.1 mm) with a 1.8 µm particle size equipped with a 0.5 µm prefi lter (Upchurch Scientifi c, Oak Harbor, WA). The injection volume was 2 µL and the column was maintained at 50°C. The mobile phase consisted of water containing 0.2% formic acid and 200 mM ammonium formate (eluent A) and acetonitrile containing 0.2% formic acid (eluent B). Gradient elution was performed with 100% B for 0.1 min, a step to 90% B until 0.11 min, a linear increase to 50% B until 2.5 min, 50% B until 3.5 min, and reequilibration from 3.51 to 4.5 min with 100% B. The fl ow rate was set to 800 µl/min. To minimize contamination of the mass spectrometer, the column fl ow was directed only from 1.0 to 3.0 min into the mass spectrometer using a diverter valve. Otherwise, methanol with a fl ow rate of 250 µl/min was delivered into the mass spectrometer.
The turbo ion spray source was operated in the positive ionization mode using the following settings: ion spray voltage = 5,500 V, ion source heater temperature = 400°C, source gas 1 = 40 psi, source gas 2 = 35 psi, and curtain gas setting = 20 psi. Analytes were monitored in the multiple reaction monitoring (MRM) mode, mass transitions and MS parameters are shown in Table 1 . Quadrupoles Q1 and Q3 were working at unit resolution.
Data analysis was performed with Analyst software 1.4.2. (Applied Biosystems). The data were exported to Excel spreadsheets and further processed by self-programmed Excel macros that sorted

Sample preparation
Unless otherwise indicated, aliquots of 100 µg protein from the fi broblast homogenates were used for sphingolipid analysis. A total of 20 µL of an IS mixture containing 20 ng SPH d17:1, 2 ng SPC d17:1, 20 ng GluCer 12:0, 20 ng LacCer 12:0, and 20 ng Cer1P 12:0 was added prior to lipid extraction. We applied a butanolic extraction procedure described by Baker et al. ( 24 ). In brief, 500 µl cell homogenate corresponding to 100 µg of cellular protein were mixed with 60 µL of a buffer containing 200 mM citric acid and 270 mM disodium hydrogenphosphate (pH 4). Extraction was performed with 1 ml of 1-butanol and 500 µL of water-saturated 1-butanol. The recovered butanol phase was Values represent the percent recovery of standards spiked before and after extraction to examine extraction effi ciencies. Matrix effects are calculated from fi broblast lipid extracts corresponding to 50 µg of cellular protein spiked after extraction (corrected by endogenous sphingolipid concentrations) in percent of the same sphingolipid standard mixture used as spike. Each value represents the average of four determinations ± SD. Calibration lines were generated by plotting the ratios of the areas analyte to IS against the spiked concentrations (pmol). Each value represents the average of four determinations ± SD. golipids with a short analysis time and coelution of analyte and IS. The latter is of major importance to compensate for matrix effects and varying ionization effi ciencies, especially during gradient elution. Because reversed phase chromatography shows chain length-dependent separation, coelution of analytes and ISs may not be accomplished ( 13,16,19,27 ). Classical normal phase chromatography offers polar head group-specifi c separation but may be impaired by limited reproducibility and insuffi cient peak shapes. Moreover, the use of apolar solvents may not provide optimal ionization conditions for ESI. Hence, we established an LC separation based on HILIC, which shows lipid head group selectivity along with the use of polar solvents. Using a sub-2 m particle size, we achieved baseline separation for all sphingolipid classes within 2 min and 4.5 min total run time including reequilibration ( Figs. 2 and 3 ). Gradient elution was performed with a mixture of acetonitrile and water including 0.2% formic acid and 200 mM ammonium formate. Addition of formic acid improved the ionization effi ciency; an optimum was found at 0.2%. For optimum performance and reproducibility, it is recommended to use at least a concentration of 10 mmol/L ammonium formate in the mobile phase. Therefore, 200 mmol/L buffer and 0.2% formic acid were added to mobile phase A and 0.2% formic acid to mobile phase B.

Extraction effi ciency and matrix effects
To analyze polar sphingolipids from one lipid extract, we tested a butanolic extraction previously described for S1P analysis ( 18 ). The extraction effi ciency was determined in fi broblast homogenate by adding a sphingolipid standard mixture before and after extraction ( Table 2 ). Mean recoveries were between 60 and 70% and did not vary with concentration of standard added.
We assessed matrix effects by analyzing a standard mixture in methanol and also spiked into fi broblast lipid extracts (Table 2). Addition of fi broblast cell extract either did not infl uence or slightly increased the signals up to 20%.

Quantifi cation of sphingolipid species
To compensate for variations in sample preparation and ionization effi ciency, a set of non-naturally occurring sphingolipids, GluCer 12:0, LacCer 12:0, SPH d17:1, Cer1P 12:0, and SPC d17:1, was added as ISs prior to extraction. The ratio between analyte and IS was used for quantifi cation as indicated in Table 1 . We generated calibration lines by addition of different concentrations of naturally occurring sphingolipids to human skin fi broblasts ( Table 3 ). For glycosylated ceramide species, a possible chain length dependency was addressed by generating two independent calibration lines with a short-chain (16:0) and a long-chain fatty acid (24:0). The obtained standard curves were linear in the tested calibration range. Additional the results, calculated the analyte/IS peak area ratios, generated calibration lines, and calculated sample concentrations. Where necessary, isotopic overlap of the species was corrected based on theoretical isotope distribution according to principles described previously ( 25 ). Analytes and their corresponding ISs are shown in Table 1 .

Sphingolipid fragmentation
To analyze various sphingolipid classes, we applied ESI in the positive ion mode and acquired product ion spectra. The fragmentation patterns obtained were in accordance to previous studies for SPH, SPA, Cer1P, and glycosylated ceramide species (Table 1)  + for further analysis of glycosylated ceramides. As expected, SPC showed only one intense fragment ion at m/z 184 due to the loss of the phosphocholine head group ( 29 ). DimetSPH showed beside fragments resulting from a loss of one water molecule ( m/z 310) or one water molecule and a formaldehyde molecule ( m/z 280), and an ion at m/z 110, possibly a conjugated iminium ion ( Fig. 1A ). TrimetSPH showed only one intense fragment representing a trimethylammonium-ion at m/z 60 ( Fig. 1B ). In contrast to Cer1P species showing a sphingoid base fragment, dihydro-Cer-1P displayed a neutral loss of phosphoric acid in positive ion mode ( Fig. 1C ). Collision-induced dissociation of PhytoSPH showed two prominent fragment ions resulting from the loss of one and two water molecules ( Fig. 1D ).

HILIC of sphingolipids
Due to the relatively low level of the selected sphingolipids in crude lipid extracts, a direct analysis using "shotgun approaches" may be hampered by signal suppression caused by other matrix components ( 12,19,27 ). Therefore, we decided to establish an HPLC separation of sphin- Fibroblast homogenates (100 µg cellular protein) were spiked with increasing amounts of GluCer 24:1. Values represent peak area ratios of GluCer 24:1 and 24:0 to GluCer 12:0. The GluCer 24:0 peak area ratios are shown before and after isotope correction. The displayed values are mean of three independent samples. GluCer 24:0 ( m/z 812.7) with and without isotope correction. Whereas the GluCer 24:0 to IS ratio increased almost 2-fold upon the addition of 200 pmol GluCer 24:1 without correction, no signifi cant increase was detected after correction of isotope overlap ( Table 4 ).

Assay characteristics
Assay accuracy was calculated using three spiked fi broblast lipid extracts at different concentrations, covering the entire calibration range. Accuracy was found between 90 and 110% ( Table 5 ). evidence for the specifi city of the method is derived from the fact that both mass transitions used for SPH and SPA analysis (Table 1) revealed similar results (data not shown).
Due to coelution, monounsaturated species exhibit an overlap of the M+2 isotope peak with the corresponding saturated species. To correct this overlap, we applied a previously described algorithm based on calculated isotope distributions ( 25 ). To test this procedure, we added increasing amounts of GluCer 24:1 ( m/z 810.7) to fi broblast homogenate and calculated analyte to IS ratios of The displayed values are mean concentrations in pmol and the CV of human skin fi broblast lipid extracts corresponding to 25, 50, and 100 µg of cellular protein. A pool of fi broblast homogenates was aliquoted, and lipid extracts were analyzed in series for intraday precision (n = 6) and on 6 different days for interday precision (n = 6). Accuracy is displayed as the mean of the assayed concentration (corrected by endogenous sphingolipid concentrations in human skin fi broblasts) in percent of the spiked concentration. Each value represents the average of three determinations ± SD. Precision was determined in three fi broblast samples containing 25, 50, and 100 µg of cellular protein (Table 5). Coeffi cients of variation (CVs) were below 10% for most species for both intraday and interday precision (Table 5).
Because no analyte free matrix was available, we calculated the limit of detection (LOD), defi ned as a signal to noise ratio of 3. Whereas for most of the analyzed sphingolipid classes, <10 fmol is suffi cient for quantifi cation, PhytoSPH and dhCer1P displayed a LOD up to 50 fmol on column ( Table 1).

Preparation of cell culture samples and sample stability
Because a main application of this method is the analysis of cultured cells, we tested different methods to harvest the cells. First, a precursor ion scan of m/z 264 was applied to check which HexCer, LacCer, and Cer1P species are found in primary human skin fi broblasts. For both HexCer and LacCer, we found 16:0, 22:0, 23:0, 24:0, and 24:1 species; for Cer1P, only 16:0 was detected. To compare sample preparations, fi broblasts were either scraped in PBS and homogenized in water by sonication or lysed in 0.2% SDS. Both sample preparations did not differ in their ionization response, because IS signals were similar (data not shown). Cells lysed in water showed about 10% higher HexCer and PhytoSPH levels as well as slightly decreased LacCer 16:0, 22:0, 24:0, and 15% decreased SPH level ( Table 6 ). For reproducibility, SDS showed advantages compared with water, which gave higher SDs.
Next, we tested the stability of the homogenates. Fibroblast homogenates prepared either in water or SDS were frozen immediately or after 6 h at room temperature. Storage at room temperature showed no effect on most sphingolipid levels, except a slight increase of SPH and Cer1P in SDS and PhytoSPH in water ( Table 7 ).
Myriocin decreased cellular S1P and SPC levels at subnanomolar concentrations to 60% and 40% of the untreated control ( Fig. 4A ). The other analyzed sphingolipid classes showed only minor changes upon treatment with myriocin up to 1 nM ( Fig. 4A , B). The most pronounced effects were observed at 5 nM myriocin, with decreased Cer, HexCer, LacCer, and free sphingoid bases concentrations and a further decline of S1P and SPC level.
SKI treatment of fi broblasts at nanomolar concentrations decreased S1P and SPC by more than 50% ( Fig. 4C ). Micromolar concentrations of SKI resulted in S1P below and the SPC concentration close to the LOD and led to a pronounced increase in the level of the free sphingoid base. Interestingly, increased levels of SPA were paralleled by dihydro-SM ( Fig. 4C , D). SKI treatment in the pharmacological range (0.5-5 µM) ( 34 ) did not change Cer and The displayed values are mean (pmol/mg cellular protein) ± SD of three independent samples.
Although analysis of the "sphingolipidome" by shotgun approaches has been recently demonstrated for yeast ( 35 ), an analysis of a more complex sphingolipid pattern in mammalian systems may be hampered, especially for minor metabolites, by signal-suppressing matrix effects or lack of sensitivity ( 12,19,27 ).
In this study, we present a novel LC-MS/MS method to quantify various sphingolipid species from cultured cells. In contrast to most previous methods using reversed-phase chromatography ( 10,11,15,16,19,21,27 ), we applied HILIC, which allows coelution of analytes and non-natural occurring ISs. This is a key feature of LC-based MS methods, because matrix effects and ionization response may vary during LC separation, especially when gradient meth- Fibroblasts, either homogenized in water by sonication or lysed in 0.2% SDS, were stored immediately at Ϫ 80°C or for 6 h at room temperature. The displayed values are percent of the immediately stored fi broblast cell homogenates. The displayed values are the mean ± SD of three independent samples. SM levels signifi cantly ( Fig. 4 ). Surprisingly, SKI treatment decreased LacCer at low concentrations.
Taken together, these data show that drug treatments that affect enzymes involved in sphingolipid metabolism may not affect only the targeted metabolites but also the whole pathway.

DISCUSSION
Sphingolipid metabolism consists of a dynamic network of molecules, including important bioactive signaling molecules (1)(2)(3)(4)(5)(6)(7)(8)(9). Therefore, to understand the function of sphingolipids, it is necessary to assess a sphingolipid profi le instead of one single metabolite. ods are used. Consequently, only coelution of analytes with adequate ISs may compensate for these effects and prevent misquantifi cation. Due to the coelution of multiple species, an isotopic overlap of species is possible. Therefore, we corrected peak areas according to principles described previously ( 25 ) to avoid an overestimation of species.
Further advantages of our method are a short analysis time of 4.5 min per sample and a simple liquid-liquid extraction as sample preparation. Because the presented method uses the same butanolic extraction and LC components as a previously described method for S1P and lysophosphatidic acid analysis ( 18 ), it is possible to analyze both sets of analytes from one extract. Consequently, one can achieve with two straightforward liquid-liquid extractions (Bligh and Dyer and butanol) a full coverage of the main sphingolipid metabolites ( 25,26 ) as well as glycerophospholipids ( 22,25,36 ) and cholesterol/cholesteryl ester ( 37 ). Calibration was performed in the sample matrix by addition of naturally occurring species prior to lipid extraction. This allows compensation for potential matrix effects on ionization and extraction effi ciency as well as for small retention time differences observed between shortchain and very long-chain species. Moreover, a full validation was performed according to U.S. Food and Drug Administration guidelines ( 28 ). This extensive validation showed excellent precision, accuracy, and sensitivity for all analyzed sphingolipid classes.
First, applications of this method showed that sample preparation methods may infl uence sphingolipid levels, particularly HexCer and free sphingoid bases. Due to reproducibility and handling reasons, we prefer a direct lysis of cultured cells with 0.2% SDS instead of scraping cells. However, immediate freezing of the samples until analysis is advisable. Finally, treatment of fi broblasts with myriocin and SKI demonstrated the importance of methods covering multiple instead of single sphingolipid metabolites because treatment affected not only direct metabolites but almost the whole pathway including unexpected concentration changes of some species.
In summary, we could show that LC-MS/MS-based sphingolipid profi ling using HILIC may provide a powerful tool to understand regulatory and metabolic mechanisms involved in cellular sphingolipid homeostasis. Similar as previously shown for glycerophospholipid metabolism ( 22 ), this method can be also used for metabolic profi ling using stable isotope labeled precursors.
We thank Jolante Aiwanger, Doreen Müller, and Simone Peschel for excellent technical assistance.