PNPLA1 is a transacylase essential for the generation of the skin barrier lipid ω-O-acylceramide

Lipids are the primary components of the skin permeability barrier, which is the body's most powerful defensive mechanism against pathogens. Acylceramide (ω-O-acylceramide) is a specialized lipid essential for skin barrier formation. Here, we identify PNPLA1 as the long-sought gene involved in the final step of acylceramide synthesis, esterification of ω-hydroxyceramide with linoleic acid, by cell-based assays. We show that increasing triglyceride levels by overproduction of the diacylglycerol acyltransferase DGAT2 stimulates acylceramide production, suggesting that triglyceride may act as a linoleic acid donor. Indeed, the in vitro analyses confirm that PNPLA1 catalyses acylceramide synthesis using triglyceride as a substrate. Mutant forms of PNPLA1 found in patients with ichthyosis exhibit reduced or no enzyme activity in either cell-based or in vitro assays. Altogether, our results indicate that PNPLA1 is directly involved in acylceramide synthesis as a transacylase, and provide important insights into the molecular mechanisms of skin barrier formation and of ichthyosis pathogenesis.

T he body surface epidermis forms a permeability barrier, which has essential roles in protecting terrestrial animals from invasion of pathogens and harmful substances such as allergens and pollutants as well as from internal water loss. Accordingly, several cutaneous disorders-such as ichthyosis, atopic dermatitis, infectious diseases and dry skin-are characterized by alterations or defects of this skin barrier 1,2 . The principal compound family in the skin permeability barrier is lipids. Lipids form multi-layered structures (lipid lamellae) extracellularly in the stratum corneum, the outermost layer of epidermis, and their high hydrophobicity inhibits the invasion of external materials and water loss from inside the body 1,2 .
To carry out this special, barrier-creation function, lipid lamellae contain unusual lipids. Approximately half of stratum corneum lipids are ceramide, which is the backbone of sphingolipids, and epidermis-specific ceramide species such as acylceramide (o-O-acylceramide) exist [3][4][5][6] . Acylceramide is especially important for skin barrier formation. Loss of acylceramide due to mutations in acylceramide synthesis genes leads to autosomal recessive congenital ichthyosis (ARCI) in humans and neonatal lethality in mouse models, where gene loss causes similar skin barrier defects as in humans [7][8][9][10][11][12][13] . Ichthyosis is characterized by dry, thickened and scaly skin, and the skin barrier defect in ARCI is the most severe among several types of ichthyoses 12 .
The structure of acylceramide is quite unique. Although normal ceramides contain two hydrophobic chains, a long-chain base and a fatty acid (FA) with carbon chain-length of C16-24 (refs 6,14), acylceramide has an additional hydrophobic chain, linoleic acid (Fig. 1a). Furthermore, the chain-length of the FA moiety is extremely long (C28-C36) (refs 15,16). Therefore, acylceramide is one of the most hydrophobic lipids in mammalian bodies. The characteristic structure of acylceramide plays a pivotal role in organizing lipid lamellae 17 . In addition, acylceramide is also important as a precursor of protein-bound ceramide, which connects lipid lamellae and corneocytes 18 .
Despite the physiological and pathological importance of acylceramide, the elucidation of the molecular mechanism by which it is created has not been completely resolved. Although recent studies have identified the genes involved in the acylceramide synthesis-specific reactions-such as the FA elongases ELOVL1 and ELOVL4, which are involved in the elongation of very-long-chain (VLC) FAs (VLCFAs; ZC21) to ultra-longchain (ULC) FAs (ULCFAs; ZC26), the FA o-hydroxylase CYP4F22, which hydroxylases the o-carbon of ULCFAs, and the ceramide synthase CERS3, which catalyses an amide bond formation between a long-chain base and a ULCFA (refs 7,9,11,13,19,20) (Fig. 1a)-the gene involved in the final step of acylceramide production, that is, ester bond formation between o-hydroxyceramide and linoleic acid, has not been identified. This means the molecular mechanism by which the skin barrier is created is still unclear.
Human cannot synthesize the acylceramide component linoleic acid. Therefore, linoleic acid is an essential FA that must be supplied from diet. Essential FA deficiency causes several skin symptoms including ichthyosis due to impairment of normal acylceramide production 3,18 . Large decreases in or loss of acylceramide due to mutations in the genes involved in acylceramide synthesis (such as CERS3, CYP4F22 and ELOVL4) causes non-syndromic ARCI (CERS3 and CYP4F22) as described above, or a syndromic form of ichthyosis (ELOVL4) (refs 5,8,10,12,21). Furthermore, decreases in acylceramide levels are also observed in atopic dermatitis patients 22,23 .
Many ichthyosis-causative genes have been identified, and some of them have been shown to be involved in acylceramide synthesis (as described above) or protein-bound ceramide production. However, there still remain several genes whose functions or pathogenic roles have not been revealed. For example, the functions of the ARCI-causative genes NIPAL4 (NIPA-like domain-containing protein 4) and PNPLA1 (patatinlike phospholipase domain-containing protein 1) are currently unclear 12,24,25 . In the present study, we aim to identify the missing gene responsible for the final step of acylceramide production (ester bond formation between o-hydroxyceramide and linoleic acid).
Our results indicate that PNPLA1 encodes the transacylase that catalyses acylceramide production using a triglyceride (TG) as the donor of the substrate linoleic acid. Thus, our findings provide important insights into the molecular mechanism of acylceramide production and into the function and pathogenic role of the ichthyosis-causative gene PNPLA1.

Results
PNPLA1 is involved in acylceramide production. To identify the acyltransferase or transacylase involved in acylceramide production, a proper assay system that can detect its activity or product is necessary. However, the prior lack of such assay systems meant there was no way of identifying the responsible acyltransferase/transacylase for a long time. Cell-based assays had been unsuccessful, since most mammalian cells cannot produce ULC o-hydroxyceramides, the precursors of acylceramides. However, we recently established a cell system that produces ULC o-hydroxyceramides by overproducing the FA elongase ELOVL4, the ceramide synthase CERS3 and the FA o-hydroxylase CYP4F22 in HEK 293T cells 13 , opening the door to identify the acyltransferase/transacylase of interest. In this cell system, when lipids prepared from HEK 293T cells labelled with [ 3 H]sphingosine, the sphingolipid precursor, were separated by normal phase thin layer chromatography (TLC), only long-chain (LC; C11-C20) and VLC ceramides were detected as ceramide species (Fig. 1b). Overexpression of CERS3 and ELOVL4 caused cells to produce ULC ceramides, and further co-overproduction of CYP4F22 leads to production of ULC o-hydroxyceramides, as we have described previously 13 (Fig. 1b). Using this cell system, we examined the involvement of several candidate genes in acylceramide production. The selected candidate genes were ABHD5 (a/b hydrolase domain containing 5)/CGI-58 (comparative gene identification-58), LIPN (lipase, family member N), PNPLA1, PLA2G15 (phospholipase A 2 , group XV)/LLPL (lecithin:cholesterol acyltransferase-like lysophospholipase), LCAT (lecithin:cholesterol acyltransferase) and DGAT2 (diacylglycerol O-acyltransferase 2). ABHD5 is a causative gene of Chanarin-Dorfman syndrome (also known as neutral lipid storage disease with ichthyosis (NLSD-I)), an autosomal recessive disease accompanied by ichthyosis, steatosis and other symptoms 26 . LIPN and PNPLA1 are ARCI-causative genes 12,24,25,27 , but their roles in skin barrier formation have not yet been revealed. Reasoning that since ABHD5, LIPN and PNPLA1 contain phospholipase/hydrolase domains and some proteins containing such domains act as acyltransferases or transacylases 28 , we chose them as candidates for acylceramide synthetic acyltransferases/transacylases. PLA2G15 has been implicated in the synthesis of 1-O-acylceramide, another type of acylceramide with unknown function 29 . We examined the possibility that PLA2G15 (and its homologue LCAT) is also involved in acylceramide (o-O-acylceramide) synthesis. DGAT2 encodes a diacylglycerol acyltransferase, and Dgat2 knockout mice exhibit a skin-barrier-defect phenotype 30 .
Expression of PNPLA1 caused acylceramide production, while that of ABHD5, LIPN, PLA2G15 or LCAT had no effect (Fig. 1b). Expression of DGAT2 also caused acylceramide synthesis, albeit weakly. DGAT2 is involved in TG synthesis 30,31 , suggesting that TGs are somehow involved in acylceramide production, perhaps as a linoleic acid donor. To discriminate whether the produced acylceramide was o-O-acylceramide or 1-O-acylceramide, we next performed the [ 3 H]sphingosine labelling assay in the presence or absence of the FA o-hydroxylase CYP4F22. Acylceramide production induced by PNPLA1 and by DGAT2 expression was CYP4F22-dependent in both cases (Fig. 1c): in other words, it was FA o-hydroxylation-dependent, indicating that the produced acylceramide was o-O-acylceramide. Liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis revealed that acylceramides with carbon chain-length of C28-36, mainly C30-36, were produced by expression of PNPLA1 and DGAT2 (Fig. 1d). PNPLA1 and DGAT2 expression increased acylceramide levels by 7.6-or 3.8-fold compared to control, respectively (Fig. 1d, inset).
To confirm that PNPLA1 is indeed involved in acylceramide production, we performed knockdown analysis using human keratinocytes and a lentiviral vector-encoding shRNA system. Infection with a lentivirus bearing the shRNA specific for PNPLA1 (shPNPLA1) caused a large decrease in PNPLA1 mRNA levels compared with the control (Fig. 2a). [ 3 H]Sphingosine labelling experiments revealed that differentiated keratinocytes treated with control shRNA generated acylceramide and its derivative acyl-glucosylceramide (Fig. 2b). Treatment of keratinocytes with shPNPLA1 caused large decreases in acylceramide/acyl-glucosylceramide levels. LC-MS/MS analysis confirmed that acylceramide levels were decreased by shPNPLA1 treatment, irrespective of chain-lengths (Fig. 2c). Total acylceramide levels in shPNPLA1-treated cells were 7.1% of those in control shRNA-treated cells (Fig. 2c, inset). These results indicate that PNPLA1 is indeed involved in acylceramide production.
PNPLA1 is not involved in TG metabolism. Stimulation of acylceramide synthesis by DGAT2 suggests that TGs are involved in acylceramide production. We next examined the possibility that PNPLA1 also stimulates acylceramide synthesis through an increase in TG levels by [ 14 C]linoleic acid labelling assay. DGAT2 expression increased TG levels as expected, whereas PNPLA1 had no effect (Fig. 3a). We next measured linoleic acid-containing TG levels by LC-MS/MS. The amounts of all of the TGs examined were increased by DGAT2 overexpression (Fig. 3b). In contrast, PNPLA1 again had almost no effect, whereas slight increases were observed for some TG species with shorter chain-lengths for unknown reasons. However, these slight changes might not be able to cause the acylceramide increase. Thus, these results suggest that PNPLA1 does not affect acylceramide production through increasing TG levels. PNPLA1 belongs to the patatin-like phospholipase domaincontaining protein (PNPLA) family. The PNPLA family members are known to exhibit phospholipase, TG hydrolase or transacylase activity, or a combination 28 . Considering these functions of the PNPLA family members, PNPLA1 was expected to be involved in acylceramide production as a TG hydrolase/phospholipase supplying linoleic acid from TG/phospholipids to an unknown acyltransferase, or as a transacylase catalysing acylceramide production directly using the linoleic acid in TGs as a substrate. If the former possibility were true, an increase in cellular linoleic acid levels caused by adding linoleic acid exogenously might bypass the otherwise required step of PNPLA1 in acylceramide production. However, addition of linoleic acid in the medium did not cause an increase in acylceramide in the absence of PNPLA1 (Fig. 4). On the other hand, linoleic acid stimulated acylceramide synthesis in a dose-dependent manner in cells overproducing PNPLA1 (Fig. 4). These results suggest that the role of PNPLA1 in acylceramide synthesis is not to supply linoleic acid as a TG hydrolase/phospholipase, and instead that PNPLA1 may be directly involved in acylceramide synthesis. The conclusionthat it is unlikely PNPLA1 acts as a TG hydrolase-is consistent with the MS data, which showed that PNPLA1 expression did not reduce TG levels (Fig. 3b).
PNPLA1 is a transacylase using TG as a substrate. To prove that PNPLA1 directly catalyses acylceramide production as a transacylase using TG as a substrate, we performed in vitro assays. For this purpose, PNPLA1 was translated using a wheat germ cell-free translation system. Since PNPLA1 is a membrane protein 32 , we added liposomes to the translation reaction mixture. Recently, several membrane proteins have successfully been inserted directly into the lipid bilayer of liposomes by similar cell-free translation systems [33][34][35] . After translation of PNPLA1 mRNA, the resulting proteoliposomes were recovered by centrifugation and used for further analyses. We confirmed production of PNPLA1 by immunoblotting (Fig. 5a). Then, we subjected the proteoliposomes to an in vitro acylceramide synthesis assay, where linoleoyl-CoA (C18:2-CoA), TG and o-hydroxyceramide were used as substrates in different combinations. The highest activity was observed when proteoliposomes containing PNPLA1, TG and o-hydroxyceramide were used (Fig. 5b). Diglyceride levels were also increased ( Supplementary Fig. 1), indicating that the linoleic acid portion of TG was transferred to o-hydroxyceramide to produce acylceramide. Low levels of acylceramides were also produced by PNPLA1 in the presence of o-hydroxyceramide alone (without exogenous TG). This is probably due to a supply of TG from the wheat germ lysates used in the cell-free translation system, as we confirmed by LC-MS/MS analysis. Inclusion of linoleoyl-CoA (C18:2-CoA) in the proteoliposomes containing PNPLA1 and o-hydroxyceramide did not cause a further increase in acylceramide levels. These results indicate that PNPLA1 is a transacylase using TG as a substrate rather than an acyltransferase using linoleoyl-CoA.
Correlation between PNPLA1 activity and ichthyosis pathology. PNPLA1 is one of the genes known to cause ARCI (refs 24,25). Two missense mutations, which cause amino acid substitution (A34T or A59V), and one nonsense mutation (E131X) have been found in the PNPLA1 of ichthyosis patients 24,25 . The mutated residues (Ala34 and Ala59) are located in the patatin domain. We expressed wild type and mutant forms of PNPLA1 in HEK 293T cells together with ELOVL4, CERS3 and CYP4F22 and examined the expression and acylceramide production activities of these mutant PNPLA1 proteins. The point mutants PNPLA1 A34T and A59V were expressed at equivalent levels to the wild-type protein (Fig. 6a). [ 3 H]Sphingosine labelling assay revealed that their activities were decreased to B20% of wild-type protein activity (Fig. 6b). The nonsense mutant protein PNPLA1 E131X was detected as a truncated protein of 14 kDa (Fig. 6a). The acylceramide levels in PNPLA1 E131X-producing cells were indistinguishable from those in vector-transfected cells (Fig. 6b), indicating that the truncated PNPLA1 protein (E131X) had no acylceramide production activity.
Next, we directly measured the transacylase activities of the ichthyosis mutants of PNPLA1 in vitro. PNPLA1 mutants were properly expressed by the cell-free translation system in a manner equivalent to the wild-type protein (Fig. 6c). An acylceramide synthesis assay in the presence of TG and o-hydroxyceramide revealed that both of the point mutants (PNPLA1 A34T and A59V) exhibited reduced activities compared to wild type. The truncated mutant E131X had no activity (Fig. 6d), consistent with the results obtained from the cell-based assay (Fig. 6b). These results show a clear relationship between PNPLA1 activity and ichthyosis pathology. Although it has been unclear why PNPLA1 mutations cause ichthyosis, our results suggest that decreases in acylceramide levels are the cause of skin barrier defects and lead to ichthyosis.

Discussion
Over 30 years have passed since acylceramide and acylglucosylceramide were discovered and their structures determined 36,37 . However, the synthetic genes of acylceramide remained unclear for a long time. There had long been only limited knowledge about candidate genes involved in acylceramide production. However, recent determination of ichthyosis-causative genes gave a clue to the identification of acylceramide synthetic genes. For example, the ichthyosiscausative genes CERS3, ELOVL4 and CYP4F22 are all involved in acylceramide production [8][9][10]13,[20][21] . In addition, we recently succeeded in establishing a cell system to produce ULC o-hydroxyceramide, the substrate of acylceramide, by overproducing CERS3, ELOVL4 and CYP4F22 in HEK 293T cells 13 . Using this system again here, we revealed that the ARCIcausative gene PNPLA1 is involved in acylceramide production ( Fig. 1b-d). Furthermore, we demonstrated that PNPLA1 catalyses the transacylation of the linoleic acid portion of TG to ULC o-hydroxyceramide for acylceramide production in vitro (Fig. 5). Murakami and his colleagues reached the same conclusion using Pnpla1 knockout mice 38 . Their mice exhibited neonatal lethality due to skin barrier defects. Acylceramide was not produced in Pnpla1 knockout mice, but its substrate ULC o-hydroxyceramide accumulated. During the preparation of our manuscript, similar in vivo results were also reported by another group 39 . In that study, decreases in acylceramide levels in the differentiated keratinocytes from an ARCI patient were also reported. Taking our cell-assay-based and biochemical in vitro results with the in vivo results from the two groups, we conclude that PNPLA1 plays an essential role in the final step of acylceramide production, esterification of ULC ohydroxyceramide with linoleic acid. In addition to PNPLA1, overproduction of the diacylglycerol acyltransferase DGAT2 also caused an increase in acylceramide levels in our cell system, although the effect was weaker than PNPLA1 (Fig. 1b-d). Since DGAT2 is involved in TG synthesis 30 , it is likely that an increase in the levels of TG, as the substrate of PNPLA1, by overproduction of DGAT2 (Fig. 3) indirectly enhances acylceramide production. Dgat2 knockout mice exhibit a skin-barrier-defect phenotype 30 , although the mechanism remains unclear. From our results, we speculate that the decreased TG causes impairment of acylceramide synthesis, leading to the skin barrier defect.
PNPLA1 is known as an ARCI-causative gene 24,25 . However, it was previously unclear by what mechanism PNPLA1 mutations cause ARCI. In the present study, we revealed that the PNPLA1 mutant proteins exhibited weak or no acylceramide production activity (Fig. 6). Thus, our results suggest that the pathology of ARCI associated with PNPLA1 mutation is caused by reduced acylceramide levels. Similar correlations among ichthyosis pathology, enzyme activities and acylceramide levels have been observed for ARCI caused by CERS3 and CYP4F22 mutations as well 10,13 .
Our results indicate that PNPLA1 is the transacylase that acts at the final step of acylceramide production, esterification between ULC o-hydroxyceramide and linoleic acid. Since acylceramide is essential to maintain skin barrier integrity, our findings constitute important information by which we can understand the molecular mechanisms behind skin barrier formation. At present, there are no therapeutic agents for the causal treatment of ichthyosis or atopic dermatitis. Elucidation of the molecular mechanisms behind skin barrier formation may lead to the development of such new therapeutic medicines.
Lipid labelling assay. HEK 293T cells were transfected with plasmids according to test group. Twenty-three hours and thirty minutes after transfection, medium was changed to DMEM (without FBS). After 30 min incubation, cells were labelled with 0. The resulting organic (lower) phase was recovered, dried and dissolved in chloroform/methanol  Table 3. Collision energy was set at 20 V. TGs were quantified using a standard curve plotted from serial dilutions of the TG 1-stearin-2-olein-3-linolein (Larodan Fine Chemicals AB, Malmo, Sweden). Data analysis and quantification were performed using MassLynx software (Waters).
Gene knockdown. The lentiviral vector pNS64 was constructed by modifying the restriction sites of pGFP-C-shLenti (OriGene Technologies, Rockville, MD, USA). The pNS72 plasmid encoding shPNPLA1 was constructed as follows. Oligo DNAs containing the PNPLA1 shRNA target sequence (shPNPLA1-F and -R; Supplementary Table 1) were annealed and cloned into pAK1072, the vector for shRNA production under the U6 promoter, generating the pNS56 plasmid. The U6-shPNPLA1 region in the pNS56 was digested and two copies of the digested fragment were tandemly inserted into the pNS56 plasmid, generating pNS68 plasmid. The resulting total of three copies of U6-shPNPLA1 in the pNS56 plasmid were digested and cloned into the pNS64 vector, generating the pNS72 plasmid. Twenty-four hours after seeding in a six-well plate (2.0 Â 10 6 cells per well), Lenti-X 293T cells were transfected with 1 mg of control shRNA vector (OriGene Technologies) or the pNS72 plasmid, together with 0.75 mg of the lentiviral packaging plasmid psPAX2 (Addgene, Cambridge, MA, USA) and 0.5 mg of the VSV-G envelope-expressing plasmid pMD2.G (Addgene). Medium was changed to fresh DMEM containing 10% FBS at 24 and 48 h after transfection, and the media collected at 48 and 72 h after transfection were pooled and centrifuged (1,500g, 4°C, 10 min). The supernatant was centrifuged (50,000g, 4°C, 2 h), and the resultant pellets were suspended in 300 ml of CnT prime Epidermal Keratinocyte Medium and used as the viral solution.
Human primary keratinocytes were seeded at 2.0 Â 10 4 cells per well in a 12-well plate and cultured in CnT prime Epidermal Keratinocyte Medium for 24 h at 37°C. The culture medium was replaced with 0.5 ml of the same medium but containing 25 ml of virus solution and 8 mg ml À 1 polybrene (Nacalai tesque, Kyoto, Japan). Six hours after incubation, the medium was changed to the one without the virus solution and polybrene, and incubated for 3 days. Differentiation was induced by incubating the infected cells with CnT-Prime 3D Barrier medium (CELLnTEC) for 7 days.