Comparative profiling and comprehensive quantification of stratum corneum ceramides in humans and mice by LC–MS/MS

Ceramides are the predominant lipids in the stratum corneum (SC) and are crucial components for normal skin barrier function. Although the composition of various ceramide classes in the human SC has been reported, that in mice is still unknown, despite mice being widely used as animal models of skin barrier function. Here, we performed LC–MS/MS analyses using recently available ceramide class standards to measure 25 classes of free ceramides and 5 classes of protein-bound ceramides from the human and mouse SC. Phytosphingosine-type ceramides (P-ceramides) and 6-hydroxy sphingosine-type ceramides (H-ceramides), which both contain an additional hydroxyl group, were abundant in human SC (35% and 45% of total ceramides, respectively). In contrast, in mice, P-ceramides and H-ceramides were present at ~1% and undetectable levels, respectively, and sphingosine-type ceramides accounted for ~90%. In humans, ceramides containing α-hydroxy FA were abundant, whereas ceramides containing β-hydroxy FA (B-ceramides) or ω-hydroxy FA were abundant in mice. The hydroxylated β-carbon in B-ceramides was in the ( R ) - configuration. Genetic knockout of β-hydroxy acyl-CoA dehydratases in HAP1 cells increased B-ceramide levels, suggesting that β-hydroxy acyl-CoA, an FA-elongation cycle intermediate in the endoplasmic reticulum, is a substrate for B-ceramide synthesis. We anticipate that our methods and findings will help to elucidate the role of each ceramide class in skin barrier formation and in the pathogenesis of skin disorders.

. Ceramides are classified into 25 classes according to the different combinations of LCBs and FAs they contain (Fig. 1C). Each ceramide class is designated by a combination of abbreviations that represent the structure of the LCB and FA. For example, the most abundant ceramide class in mammals is composed of sphingosine (S) and non-hydroxy FA (N), and is designated by NS. NS and NDS exist in almost all tissues in mammals, while other ceramide classes exist only in specific tissues. In the human SC, sixteen ceramide classes other than SD-and B-ceramides have been identified (5)(6)(7). Alteration of ceramide class composition and reductions in the total amount of ceramides have been observed in the SC of atopic dermatitis patients (15)(16)(17). EO-ceramides are epidermisspecific ceramide classes, and their characteristic EO structure is important for the formation and stabilization of lipid organization in the lipid lamellae (18,19). A portion of EO-ceramides are converted into protein-bound ceramides (P-O-ceramides) (Fig. 1B), which are connected to cross-linked proteins constituting the cornified envelope; a structure located beneath the surface of corneocytes (20)(21)(22). Impairment of either EO-or P-O-ceramide production causes ichthyosis (1,2,8,(21)(22)(23)(24)(25)(26).
Each ceramide class includes a variety of molecular species which differ in FA chain-length and/or the degree of unsaturation (8,27,28). Therefore, more than several hundred ceramide species exist in the human SC (5)(6)(7). The composition of ceramide classes/species in the human SC has previously been analyzed using LC-MS (5-7, 16, 17, 29, 30). However, it is difficult to separately detect and precisely quantify such a tremendous number of ceramide species by LC-MS. Recently, using LC-MS/MS in the MRM mode, which selects both the molecule-related ion of each lipid species and its specific fragment ion generated by collision-induced dissociation, became a standard technique for specifically detecting and quantifying lipid molecules of interest. However, the following problems must be addressed in order to develop an LC-by guest, on May 7, 2020 www.jlr.org Downloaded from MS/MS method for quantifying a variety of ceramide species. First, appropriate internal standards are required. So far, NS with C17:0 FA (C17:0 NS) has generally been used as an internal standard for the quantification of SC ceramides. However, this ceramide is not ideal, since it exists endogenously.
Additionally, the fragment ion pattern of ceramides differs depending on their LCB moiety (5). Therefore, a ceramide standard for each LCB class is required. However, these standards were not commercially available until recently. Second, ionization of the molecule of interest is often suppressed by coexisting molecules (ion suppression or matrix effects), causing decreased detection sensitivity and inaccurate quantification. Although an SC sample is often collected by tape stripping, the matrix effect caused by coextracted compounds from the tape is unknown. Third, during ionization, dehydration of ceramide is often induced by in-source decay (31). If the MRM setting has been determined without the consideration of whether the ceramide species of interest is dehydrated or not, erroneous peak annotation and inaccurate quantification may result. However, the degree of dehydration of each ceramide class has not yet been determined.
Several KO mouse models, lacking genes responsible for the production of ceramides involved in skin barrier formation, have been produced and analyzed (8,32,33). In particular, KO mice lacking the genes responsible for EO-ceramide production [e.g., the FA elongases, Elovl1 and Elovl4 (34,35); the FA ωhydroxylase, Cyp4f39 (the mouse orthologue of human CYP4F22) (26); the ceramide synthase, Cers3 (36); and the transacylase, Pnpla1 (24)] have been used as ichthyosis mouse models to elucidate the pathogenesis of ichthyosis. In addition, the atopic dermatitis mouse model has been used for investigating the correlation between skin barrier function and ceramide composition (37,38). Thus, although mice have been generally utilized as a model animal for studying skin barrier function, the ceramide classes measured have primarily by guest, on May 7, 2020 www.jlr.org Downloaded from 9 (Alabaster, AL). N-ω-Hydroxytriacontanoyl D-erythro-sphingosine (C30:0 OS) was obtained from Cayman Chemical (Ann Arbor, MI). Methanol, acetonitrile, and isopropanol of LC-MS grade, as well as chloroform and ammonium formate of HPLC grade, were all obtained from FUJIFILM Wako Pure Chemical Corporation (Osaka, Japan)

Collection of SC samples by tape stripping
Human SC samples were collected from healthy volunteers (10 males and 9 females; aged 20-50 y). To collect the samples, the inner forearm was first cleaned with water and dried before tape stripping was performed. A 25 mm × 50 mm piece of film masking tape 465#40 (Teraoka Seisakusho, Tokyo, Japan) was stuck to the inner forearm, then pressed and removed. The procedure was repeated using a new piece of tape. The collected tape samples were applied to an OHP film and stored at -30 °C. Mouse SC samples were collected by tape stripping from the back of C57BL/6J mice on the first day after birth using the same procedure as human SC sampling. The second tape strip was used for lipid extraction, since the first tape strip may have included dust, detergents, or lotions.

Lipid analyses by LC-MS/MS
Ceramides were detected and quantified using UPLC coupled with an electrospray ionization tandem quadrupole mass spectrometer (Xevo TQ-S; Waters, Milford, MA). LC separation was conducted using a reverse-phase column [ACQUITY UPLC CSH C18 column (particle size, 1.7 µm; inner diameter, 2.1 mm; length, 100 mm; Waters)] with a binary gradient solvent system as previously described (39). In positive ion mode, ionization was performed using the following parameters: capillary voltage, 2.5 kV; cone voltage,  Table S1).
[M-H2O+H] + was selected as a precursor ion to detect NS, NH, NSD, AS, BS, AH, and ASD ceramides, and [M+H] + was selected to detect NDS, NP, ADS, and AP ceramides. Both [M+H] + and [M-H2O+H] + were selected to detect O-and EO-ceramides. Ceramides were quantified by calculating the ratio of the peak area of each ceramide species compared to that of the deuterium-labeled ceramide (internal standard) corresponding to each ceramide class. SD-ceramides were quantified using a deuterium-labeled S-ceramide standard, since deuterium-labeled SD-ceramide is commercially unavailable. O-and EO-ceramides were quantified using the deuterium-labeled A-ceramide standards that shared common LCB structures.
MassLynx software (Waters) was used for data analysis.

Evaluation of matrix effect
Fresh tape, not including SC, was cut and transferred into a glass tube containing 2 mL of methanol. After sonication (room temperature, 5 min), the tape was removed. The tube was centrifuged (2,600 × g, room temperature, 5 min), and the supernatant was recovered to a new glass tube and evaporated. Tape extracts were dissolved in 100 µL of chloroform/methanol (1:2, v/v) containing deuterium-labeled ceramide standards (10 pmol each), and each ceramide standard was detected by LC-MS/MS.

Lipid extraction
Tape strips, including SC, were cut to a size of 5 mm × 10 mm and transferred into a glass tube containing 400 µL of methanol. As internal standards, deuterium-labeled ceramide standards were added in the To remove the SC pellet, the tube was centrifuged (2,600 × g, room temperature, 5 min). The supernatant was recovered to a new glass tube, and dried (total free lipid samples). The SC pellet was subjected to extraction of P-O-ceramides as follows. To remove free lipids completely, 400 µL of methanol was added, vigorously mixed, and centrifuged (2,600 × g, room temperature, 5 min). After removal of the supernatant, the same procedure was repeated twice and samples were then incubated with 400 µL of 95% methanol (60 °C, 2 h). After centrifugation (20,400 × g, room temperature, 3 min), the supernatant was removed. The pellet was again incubated with 400 µL of 95% methanol (60 °C, 2 h), followed by centrifugation (2,600 × g, room temperature, 5 min) and removal of the supernatant. Lipid extractions from the mouse epidermis were performed as previously described (28). Lipids were extracted from cultured cells as follows. Cells cultured in a six-well plate were washed twice with 1 mL of PBS, detached from the plate using a scraper, and transferred into plastic tubes. After centrifugation (400 × g, 4 °C, 3 min), the supernatants were removed. The cells were suspended in 100 μL of PBS and mixed with 375 μL of chloroform/methanol/12 M formic acid (100:200:1, v/v) and an internal standard (d9-C16:0 AS, 10 pmol). Samples were mixed with 125 μL of chloroform and 125 μL of water, and centrifuged (20,400 × g, room temperature, 3 min). The organic phase was recovered and dried. Lipids were dissolved in 500 µL of chloroform/methanol (1:2, v/v) and subjected to LC-MS/MS analyses.

Cell culture and transfection
HAP1 cells are near-haploid human cells derived from a chronic myelogenous leukemia sample (40).

Establishment of the method that specifically detects ceramide species by LC-MS/MS
In order to develop a method for distinguishing each ceramide species by LC-MS/MS, we first performed product ion scanning using deuterium-labeled ceramide standards (d9-ceramide; containing nine deuteriums instead of protiums in its FA moiety). By fragmentation of NS and NH, product ions of m/z 264 and 280, which correspond to S and H moieties, respectively, were predominantly generated ( Fig. 2A). Dissociation of NP yielded three major product ions (m/z 264, 282, and 300) at similar intensities, of which two product ions (m/z 264 and 282) overlapped with the major and minor product ions of NS, respectively, indicating that the product ion of m/z 300 was specific to P-ceramide. Product ions of m/z 266 and 284 were mainly produced by fragmentation of NDS, and neither were detected in other ceramide classes. Since the NSD standard was not commercially available, we synthesized C16:0 NSD by incubating C16:0-CoA and SD with membrane fraction prepared from HEK 293T cells. NSD was predominantly cleaved to a fragment ion of m/z 262, which corresponded to the SD moiety (Fig. 2B). Next, to investigate whether the difference in a FA moiety affected the fragmentation of ceramide, we performed fragmentation analyses using AS and EOS standards. The MS/MS spectrum of both standards were the same as that of NS (Fig. 2C), indicating that the difference in the FA moiety did not affect the generation of the fragment ion derived from the LCB moiety. Recently, BS, which is composed of a β-hydroxy FA and sphingosine, has been identified in mouse epidermis (11,12). Since AS and BS are structural isomers, in positive ion mode, BS is predicted to be detected at a different retention time using the same MRM settings as AS detection. In LC-MS/MS analysis using the MRM setting to detect C26:0 AS, C26:0 AS was detected in human SC (13.20 min), while a large peak at an earlier retention time (12.86 min), instead of C26:0 AS, was detected in the mouse epidermis by guest, on May 7, 2020 www.jlr.org Downloaded from (Fig. 2D). The MS/MS spectrum of this peak (12.86 min) in positive ion mode was similar to that of NS ( Fig. 2E), indicating that the peak was derived from S-ceramide. In negative ion mode, four product ions (m/z 340, 310, 263, and 237) were generated with the highest intensity at m/z 340. The fragmentation pattern of this peak is consistent with that of C26:0 BS in previous reports (12,42), demonstrating that the peak Ceramide classes with double bond(s) in their LCB (NS, NH, and NSD) were highly dehydrated, while those without double bonds (NDS and NP) were not (Fig. 2F). Dehydrated AS was also detected predominantly, whereas non-dehydrated OS was detected at the same levels as dehydrated OS. The nondehydrated EOS and BS levels were about half of the dehydrated levels. From these results, we quantified ceramides using MRM settings in which a dehydrated type, a non-dehydrated type, or both types were selected, depending on the specific ceramide class. Tape stripping is often conducted to collect SC samples. However, certain ingredients in the tape are coextracted with lipids and may cause a matrix effect. To investigate the effect of the extracts derived from the tape on the detection of ceramides, d9-ceramides were mixed with the tape extracts and detected by LC-MS/MS. Compared to the standard samples only diluted with methanol (external standards), the detection ratio of each d9-ceramide mixed with tape extracts was decreased, especially in the extracts from 200 mm 2 and 400 mm 2 sized pieces of tape (Fig. 2G). The extracts from 25 mm 2 and 50 mm 2 pieces of tape did not affect the detection ratio. From this result, although a matrix effect (suppression of ionization) is caused by the tape-derived co-extracts, this effect can be nullified by reducing the tape size to 50 mm 2 or less.

Different SC ceramide classes and FA composition between humans and mice
Using the extraction protocol and the LC-MS/MS settings determined in Fig. 2, we measured ceramides in the human and mouse SC. The SC specimens were collected via tape stripping from mice on the first day after birth (to avoid hair) and humans. Lipids were extracted, d9-ceramides were added as internal standards,  Table S2). In the human SC, NP, NH, and AH were abundant (24.2%, 23.7%, and 18.0% of total ceramides, respectively), and followed by AP, NDS, NS, AS, EOH, EOS, EOP, and ADS (in descending order, 1-9%). The proportions of other ceramide classes were lower than 1%. In the mouse SC, NS levels were the highest (57.9% of total ceramides), followed by EOS, NDS, OS, and BS in that order by guest, on May 7, 2020 www.jlr.org Downloaded from (4-16%). The proportions of other ceramide classes were lower than 2%. Of the categories of ceramide classes that contain a common LCB, H-and P-ceramides were abundant in the human SC (45.4% and 34.7% of total ceramides, respectively) (Fig. 3C). On the other hand, the proportion of the S-ceramide class was the highest in the mouse SC (87.7%). The proportion of the P-ceramide class was low (1.2%), and the Hceramide class was undetectable in mice. SD-ceramide levels were quite low (~0.4%) both in humans and mice. Of the categories of ceramide classes that have the same type of FA, the proportion of the N-ceramide class was the highest (59.4%) in the human SC, followed by A-ceramides (32.5%), EO-ceramides (6.3%), and O-ceramides (1.6%) (Fig. 3D). Trace amounts of B-ceramides were present in the human SC (0.17%).
Although N-ceramide was the principal ceramide class present in both the mouse (68.9%) and human (59.4%) SC, the proportions of the other ceramide classes were largely different between mice and humans.
In mice, the EO-ceramide, O-ceramide, and B-ceramide classes were 16.9%, 8.7%, and 4.9%, respectively, and A-ceramide, the second highest ceramide class in humans, was quite low (0.7%). Thus, the composition of SC ceramide classes was largely different between humans and mice.
Various ceramide species with different FA chain-lengths exist in each ceramide class. We next compared FA composition among ceramide classes that exceeded 1% of the total ceramides (10 and 6 classes in humans and mice, respectively) (Fig. 3E). Generally, the FA chain-length of ω-hydroxylated ceramides (EO- The total percentage of ceramides with odd-chain FAs were 14-25% in human and 6-10% in mouse ceramide classes (Fig. 3F). Unsaturated FAs in the N-, A-and B-ceramide classes were low (0.3-4% in human, 1-4% in mouse), while those in the EO-and O-ceramide classes were high (5-7% in human, 29-33% in mouse), especially mouse C34:1 EOS (10.5%) (Fig. 3G). Overall, there were no remarkable differences in FA chain-length of ceramides between human and mouse SC.
There were no clear differences in FA composition among P-O-ceramide classes with different LCBs, both

(R)-Configuration of the hydroxylated β-carbon in BS
The structural configuration of the hydroxylated β-carbon in BS, a ceramide class moderately present in the mouse SC (Fig. 3), is unknown. To determine its configuration, we first synthesized C18:0 BS by incubating (R, S) β-hydroxy C18:0 FA and d7-sphingosine with the membrane fraction prepared from HEK 293T cells overexpressing CERS1 (a ceramide synthase that shows high activity against C18:0 acyl-CoA), and performed LC-MS/MS analysis. In the LC chromatogram of the synthesized (R, S) C18:0 BS, two peaks were observed at 8.63 min and 8.78 min, and the latter peak was identical with C18:0 BS in the mouse epidermis (Fig. 5A). Since neither (R) nor (S) β-hydroxy C18:0 FAs were commercially available, we could not determine the structural configuration of the two peaks. Next, we synthesized (R, S) C14:0 BS and (R) C14:0 BS by mixing (R, S) β-hydroxy C14:0 FA and (R) β-hydroxy C14:0 FA, respectively, with d7sphingosine and the membrane fraction prepared from HEK 293T cells overexpressing CERS5 (a ceramide synthase that exhibits high activity against C14:0 acyl-CoA). In the LC chromatogram of (R, S) C14:0 BS, two peaks (6.88 min and 7.04 min) were again detected (Fig. 5B). Although two peaks (6.79 min and 6.96 min) were also observed in (R) C14:0 BS, the latter peak was larger than the former one, indicating that the former and latter peaks corresponded to (S) BS and (R) BS, respectively. The reason why (S) C14:0 BS was detected despite the use of (R) β-hydroxy C14:0 FA may be due to the isomerization of (R) to (S) or the by guest, on May 7, 2020 www.jlr.org Downloaded from 21 contamination of (S) β-hydroxy C14:0 FA in (R) β-hydroxy C14:0 FA. From these results, we concluded that the structural configuration of the hydroxylated β-carbon in BS is in the (R)-configuration.

Involvement of the FA elongation cycle in BS synthesis
The FA elongation cycle in the endoplasmic reticulum (ER) membrane produces FAs with ≥C18 chainlength. In this cycle, FA is elongated in the form of acyl-CoAs, and β-hydroxy acyl-CoA is generated as an intermediate (Fig. 6A) (27,47). Although the synthesis pathway of BS has not yet been identified, we suspected that β-hydroxy acyl-CoAs in the FA elongation cycle are used for BS production, since ceramide synthesis also proceeds in the ER membrane. In the FA elongation cycle, β-hydroxy (3-hydroxy) acyl-CoA dehydratase HACD1 or HACD2 converts β-hydroxy acyl-CoAs to trans-2-enoyl-CoAs (41,48). To investigate the involvement of the FA elongation cycle in BS synthesis, we measured BS levels in HACD1 HACD2 DKO HAP1 cells, in which β-hydroxy acyl-CoAs accumulate. The amount of BS was increased in HACD1 HACD2 DKO cells by 15-fold compared to control cells (Fig. 6B), suggesting that the β-hydroxy FA moiety of BS is derived from the FA elongation cycle. The ratio of each ceramide class to total ceramides in this study was not largely different from earlier studies (Supplemental Table S6). Although the amounts of SD-ceramide and BS were previously unknown, we determined them for the first time ( Fa2h KO mice, neither the skin phenotype nor reduced AS levels in skin were observed (53). In that report, however, it is likely that BS, but not AS, was measured, since AS levels were determined by TLC or MS analysis without LC separation. BS levels were increased by DKO of β-hydroxy acyl-CoA dehydratase genes HACD1 and HACD2 in the FA elongation cycle localized in the ER (Fig. 6B). Since acyl-CoAs with ≥C18 are produced in the FA elongation cycle, the high levels of BS species with ≥C18 FAs (Fig. 3E) may indicate that BS is synthesized in the ER by the condensation of LCB and β-hydroxy acyl-CoA supplied from the FA elongation cycle.
However, further analyses are required, since (R) β-hydroxy acyl-CoAs are also produced as an intermediate in β-oxidation in peroxisomes (54), and since it is still possible that BS is generated by β-hydroxylation of NS by an unidentified hydroxylase. To our knowledge, no experimental evidence has been reported for the structural configuration of β-hydroxy acyl-CoA in FA elongation so far. In this study, we have proven, for by guest, on May 7, 2020 www.jlr.org Downloaded from 24 the first time, that the β-hydroxy acyl-CoA produced in FA elongation has an (R)-configuration.
Earlier studies have reported that ceramides with odd-chain FAs exist at high levels in the SC (Fig. 3F, Supplemental Table S3) (7,55). Odd-chain FAs are produced by α-oxidation (cleavage of one carbon-chain) of α-hydroxy FAs (56). Therefore, the high C25:0 N-and A-ceramide levels are presumed to be derived from the high levels of C26:0 A-ceramide, and odd-chain FAs with ≥C27:0 may be generated by the elongation of C25:0 acyl-CoA through the FA elongation cycle.
The amounts of EO-ceramides were comparable between EOS, EOH, and EOP in the human SC (Fig.   3A), whereas those of P-O-ceramides were different: P-OS was the most abundant, and P-OH and P-OP were half and 1/10 of P-OS, respectively (Fig. 4A). Since P-O-ceramides are derived from EO-ceramides, the difference in LCB composition between EO-and P-O-ceramides indicates that the conversion efficiency of EO-ceramides to P-O-ceramides varies depending on LCB structure (the presence of hydroxylation), presumably due to the low activities against EOP and EOH of the enzymes involved in the production of P-O-ceramides.
In this study, we revealed that the composition of ceramide classes, especially in hydroxylated ceramide classes, are different between the human and mouse SC. Therefore, it is predicted that the use of mice deficient in a hydroxylase gene involved in the production of hydroxylated ceramide may not always be appropriate as a model for human skin disorders. On the other hand, the use of mice is still suitable for elucidating the importance of the common features between human and mice, including EO-ceramide, P-O-ceramide, and ceramides with ultra-long chain FAs (FA with ≥C26), in skin barrier function.
In the present study, we successfully quantified a large number of ceramide species in the SC. Although we herein focused on and quantified ceramide species containing LCBs with a C18 chain-length, LCBs with by guest, on May 7, 2020 www.jlr.org

Data Availability Statement
All data are contained within the manuscript.