Lipid polarity gradient formed by ω-hydroxy lipids in tear film prevents dry eye disease

Meibum lipids form a lipid layer on the outermost side of the tear film and function to prevent water evaporation and reduce surface tension. (O-Acyl)-ω-hydroxy fatty acids (OAHFAs), a subclass of these lipids, are thought to be involved in connecting the lipid and aqueous layers in tears, although their actual function and synthesis pathway have to date remained unclear. Here, we reveal that the fatty acid ω-hydroxylase Cyp4f39 is involved in OAHFA production. Cyp4f39-deficient mice exhibited damaged corneal epithelium and shortening of tear film break-up time, both indicative of dry eye disease. In addition, tears accumulated on the lower eyelid side, indicating increased tear surface tension. In Cyp4f39-deficient mice, the production of wax diesters (type 1ω and 2ω) and cholesteryl OAHFAs was also impaired. These OAHFA derivatives show intermediate polarity among meibum lipids, suggesting that OAHFAs and their derivatives contribute to lipid polarity gradient formation for tear film stabilization.


Introduction
The tear film maintains visual function by eliminating foreign materials, supplying oxygen and nutrients to the ocular surface (cornea and conjunctiva), and reducing friction between the eyelid and the ocular globe (Ohashi et al., 2006). Tear film consists of three layers: in order from the outside, these are the tear film lipid layer (TFLL), the aqueous layer, and the glycocalyx layer ( Figure 1A; Gipson, 2004;Cwiklik, 2016). The TFLL contributes to the suppression of water evaporation from the aqueous layer and reduces the surface tension of tears (Butovich, 2013). Since the TFLL lipids are secreted from the meibomian glands, which are distributed behind the eyelids, they are collectively called meibum lipids. The aqueous layer contains nutrients, electrolytes, and bioactive molecules. The mucin-rich glycocalyx layer has a role in maintaining the aqueous layer on the corneal surface.
Dry eye disease is caused by destabilization of the tear film and is accompanied by symptoms of eye discomfort and visual dysfunction, and potentially by ocular surface damage (Gayton, 2009). The prevalence of dry eye disease varies among countries and regions (7-33% of the population) and is increasing year by year (Gayton, 2009;Dana et al., 2019). Dry eye disease is roughly classified into two types: aqueous-deficient dry eye (ADDE) and evaporative dry eye (EDE) (Bron and Tiffany, 2004;Craig et al., 2017). Most cases of EDE are caused by meibomian gland dysfunction (MGD). MGD is the most common cause of dry eye disease: one study reported that 87% of dry eye patients suffer from MGD (either MGD alone or MGD with ADDE) (Horwath-Winter, 2003).
ceramides, w-O-acylceramides, which is essential for skin permeability barrier function (Kihara, 2016), and CYP4F22/Cyp4f39 mutations cause failure to produce w-O-acylceramides (Ohno et al., 2015;Miyamoto et al., 2020). w-O-Acylceramides contain !C30 w-OH FAs, as seen in OAHFAs. Therefore, we hypothesize that CYP4F22/Cyp4f39 are also involved in FA w-hydroxylation in the OAHFA biosynthesis pathway. Whole-body Cyf4f39 gene knockout is neonatal lethal (Miyamoto et al., 2020), so we created Cyp4f39 -/-Tg (IVL-Cyp4f39) mice (hereafter, Tg-Cyp4f39 -/-), in which the epidermal barrier defect was rescued by the transgenic expression of Cyp4f39 in the epidermis. Using these mice, we examined the involvement of Cyp4f39 in the production of OAHFAs and OAHFA derivatives, and their roles in tear film stabilization.

Results
Palpebral ptosis and abnormal tear covering on the eyeball surface in Cyp4f39-deficient mice To examine the involvement of Cyp4f39 in OAHFA production in the meibomian glands, the neonatal lethality caused by whole-body Cyp4f39 disruption must be circumvented. For this purpose, we created Tg-Cyp4f39 -/mice, in which all tissues except the epidermis lacked Cyp4f39 expression, by expressing a 3ÂFLAG tagged Cyp4f39 transgene under the control of the epidermis-specific involucrin (IVL) promoter ( Figure 1B). Tg-Cyp4f39 -/mice grew normally to adulthood, and their body weights were comparable to those of Tg-Cyp4f39 +/+ and Tg-Cyp4f39 +/mice ( Figure 1C). To confirm the epidermis-specific expression of 3ÂFLAG-Cyp4f39 protein, total cell lysates were prepared from several tissues of Tg-Cyp4f39 -/mice and subjected to immunoblotting with anti-FLAG antibody. 3ÂFLAG-Cyp4f39 was highly expressed in the epidermis, and weak expression was observed in the cornea and meibomian glands ( Figure 1D). It is known that the IVL promoter is active in the cornea (Adhikary et al., 2005). The detection of 3ÂFLAG-Cyp4f39 protein in the meibomian gland is probably due to expression in the keratinized epithelial cells that constitute the meibomian gland ducts rather than to the production of meibum lipids by meibocytes. Palpebral ptosis was observed in Tg-Cyp4f39 -/mice ( Figure 1E,F). In addition, tears did not spread normally over the entire eyeball surface in these mice, but rather accumulated on the lower eyelid side, suggesting that tear film surface tension was higher than normal.
Cyp4f39 deficiency causes dry eye with plugging of the meibomian gland orifices We have previously reported that mice lacking the FA elongase gene Elovl1 show dry eye accompanied by palpebral ptosis and increased eye blinking, due to shortening of the chain length of meibum lipids (Sassa et al., 2018). An increase in blinking frequency has also been reported among human dry eye patients (Su et al., 2018). Blinking frequency was measured in 1-12-month-old Tg-Cyp4f39 -/mice, and we found that it was greatly increased relative to blinking frequency in Tg-Cyp4f39 +/+ and Tg-Cyp4f39 +/mice: the average frequencies in Tg-Cyp4f39 +/+ and Tg-Cyp4f39 +/mice were <0.6 times/min at all ages, whereas in Tg-Cyp4f39 -/mice it was 3.3, 7.4, 7.3, and 11.3 times/min at 1-3, 4-6, 7-9, and 10-12 months, respectively ( Figure 2A). There were no significant differences in these frequencies among the different ages, suggesting that dry eye does not progress with age.
In dry eye disease patients, tear film break-up time (BUT) is shortened due to tear film destabilization (Tsubota, 2018). To measure BUT in Tg-Cyp4f39 -/mice, fluorescein solution was loaded onto the eye surface and observed under a slit lamp microscope. The average BUT in the Tg-Cyp4f39 +/+ mice was 7.4 s, whereas that in Tg-Cyp4f39 -/mice was 2.5 s,~1/3 of that in Tg-Cyp4f39 +/+ mice ( Figure 2B). We then scored corneal epithelial damage and found that Tg-Cyp4f39 -/mice scores were~2.6 times higher than those of Tg-Cyp4f39 +/+ mice ( Figure 2C). These results indicate that Tg-Cyp4f39 -/mice exhibit a dry eye phenotype with tear film destabilization and corneal epithelial damage.
Dry eye disease is classified as ADDE or EDE, depending on the cause of the pathology (Bron and Tiffany, 2004). The majority of EDE is caused by MGD and is often accompanied by obstruction at the orifices of the meibomian glands. Obstruction of meibomian gland orifices with white, semi-liquid plugging was observed in all Tg-Cyp4f39 -/mice (8 out of 8 mice) examined between the ages of 6-17 months (6 months, two mice; 12 months, three mice; 14 months, one mouse; and 17 months, two mice) ( Figure 2D). The degree of plugging did not appear to change with age. On the other hand, no such plugging was observed in any Tg-Cyp4f39 +/+ mice examined between the ages of 6-12 months (0 out of 9 mice). The meibomian glands of Tg-Cyp4f39 -/mice were not swollen compared to those of the control mice. There were no differences in tear volume among Tg-Cyp4f39 +/+ , Tg-Cyp4f39 +/-, and Tg-Cyp4f39 -/mice ( Figure 2E). On the basis of these results, we conclude that Tg-Cyp4f39 -/mice exhibit EDE with meibomian gland plugging.
We next examined the expression levels of meibum lipid-related genes (Awat1, Awat2, Far1, Far2, Soat1, and Cyp4f39) in Tg-Cyp4f39 -/mice by real-time quantitative RT-PCR. The acyl-CoA wax alcohol acyltransferases Awat1 and Awat2 synthesize WEs from an acyl-CoA and a fatty alcohol (FAl) (Turkish et al., 2005). The fatty acyl-CoA reductases Far1 and Far2 catalyze the production of FAls (Cheng and Russell, 2004). The sterol O-acyltransferase Soat1 is involved in CE production in the meibomian glands (Meiner et al., 1996;Yagyu et al., 2000). There were no differences in the expression levels of any of these genes between Tg-Cyp4f39 +/+ and Tg-Cyp4f39 -/mice ( Figure 3C). These results indicate that Cyp4f39 deficiency does not affect the expression of meibum lipid-related genes other than Cyp4f39.
In Tg-Cyp4f39 -/mice, the amounts of C16:1 OAHFAs were lower than in control mice, regardless of chain length and degree of unsaturation, and the total amount was~20% of that of control mice ( Figure 4C,F). Regarding C18:1 OAHFAs, only the fraction containing w-OH C32:1 FA was significantly lower in Tg-Cyp4f39 -/mice than in the control ( Figure 4D). Although the total amount of C18:1 OAHFAs was somewhat lower in Tg-Cyp4f39 -/mice, this difference was not significant ( Figure 4F). Neither the amount of any C18:2 OAHFA species nor the total amount of OAHFAs differed between control and Tg-Cyp4f39 -/mice ( Figure 4E,F). Thus, Cyp4f39 deficiency had no or almost no effect on C18:2 and C18:1 OAHFA levels. However, as these OAHFAs exist in the epidermis (Hirabayashi et al., 2019), they may have been derived from the keratinized epithelial cells that constitute the meibomian gland ducts or from epidermis cells that contaminated the samples, rather than from meibocytes producing meibum lipids. Therefore, we speculate that C16:1 OAHFAs, which were reduced in Tg-Cyp4f39 -/mice, are the major meibum lipids in mice.
In the OAHFAs in the meibomian glands of control mice, C16:1 FA was more abundant than C18:1 FA ( Figure 4F). To determine whether the abundance of FAs present in the meibomian glands was reflective of the OAHFA composition, we next measured the amounts of FAs in the meibomian glands of Tg-Cyp4f39 +/+ mice. The levels of C18:1 FA were highest, followed by those of C18:2, C16:0, C16:1, and C18:0 FA in descending order ( Figure 4G). This result suggests that the higher levels of C16:1 OAHFA compared to those of C18:1 OAHFA were due to the substrate specificity of the unknown acyltransferase that produces OAHFAs in mice. The FA levels in Tg-Cyp4f39 -/mice were not significantly different from those in Tg-Cyp4f39 +/+ mice.

Cyp4f39 deficiency causes loss of type 2w wax diesters in meibum lipids
WEs are molecules in which FA(s) and FAl(s) are ester-bonded ( Figure 5A). Among WEs, those containing one and two ester bonds are called wax monoesters (referred to as WEs here) and wax diesters (WdiEs), respectively. WdiEs are further classified into type 1 and type 2 (Nikkari, 1974): in type 1 WdiEs, the backbone hydroxy FA is esterified with a FA and a FAl, whereas in type 2 WdiEs, the backbone fatty diol forms ester bonds with two FAs. In addition, both type 1 and 2 WdiEs have positional isomers: those with ester bonds at the C1 and w positions of the backbone are type 1/2w WdiEs, and those with these bonds at the C1 and a positions are type 1/2a WdiEs. Although the presence of type 1 and 2 WdiEs in meibum lipids has been suggested (Chen et al., 2013;Butovich, 2017), their ester bond positions (a or w) have not been determined.
To determine the ester bond positions of type 2 WdiEs in meibum lipids and to quantify their levels, standards are required. As neither type 2w nor type 2a WdiE standards are commercially available, we chemically synthesized type 2w WdiE (     supplements 2 and 3). The retention time of (1,w-O-C18:1)-C16:0 WdiE in the LC column was 13.44 min, whereas that of (1,a-O-C18:1)-C16:0 WdiE was 14.31 min ( Figure 5D). This difference in retention time (0.87 min) indicates that type 2w and type 2a WdiEs with the same molecular weight can be separated by LC.
These results indicate that Cyp4f39 is essential for type 2w WdiE production. In addition to type 2w WdiEs, type 2a WdiEs were also detected in control mice ( Figure 5G). Type 2a WdiEs have a C20-C30 fatty diol (87% saturated and 13% monounsaturated) and two FAs (almost exclusively C16:1) ( Figure 5G and Table 1). In Tg-Cyp4f39 -/mice, the total amount of type 2a WdiEs was approximately 1.5-fold that in control mice, with larger increases in the type 2a WdiE species with a chain length of !C56 (total carbon number) ( Figure 5G). Therefore, Cyp4f39 is not involved in type 2a WdiE synthesis. This result is reasonable, considering that the substrates of Cyp4f39 are !C30 FAs and that the chain length of the fatty diol moiety of type 2a WdiEs is C30.
Loss of type 1w WdiEs and Chl-OAHFAs in meibum lipids due to Cyp4f39 deficiency In the product ion scanning of C66:3 type 2w WdiEs, we detected another peak (retention time, 16.72 min) in addition to the peak for type 2w WdiEs (17.70 min) in the LC ( Figure 6A). When this peak was subjected to fragment ion analysis, in addition to the precursor ion (m/z = 981.68), two product ions, with m/z values of 237.14 and 741.46, were detected ( Figure 6B). The former m/z value corresponds to the ion derived from C16:1 FA ([C16:1 FA-OH] + ). The latter value matches the molecular weight of the precursor ion minus that of C16:1 FAl, suggesting that this ion is [M + H-C16:1 FAl] + . This suggests that the unknown peak corresponds to type 1 WdiE, consisting of C16:1 FA, w-or a-OH C34:1 FA, and C16:1 FAl ( Figure 6C). We then analyzed the type 1 WdiEs in meibum lipids by LC-MS/MS in MRM mode, setting (M + H-C16:1 FAl) + as the fragment ion to detect. In control mice, type 1 WdiEs containing C16:1 FAl, C16:1 or C18:1 FA, and C30-36 hydroxylated FAs (the sum of the carbon chain lengths of the FA and hydroxy FA moieties is C42-C52) were detected ( Figure 6D). In Tg-Cyp4f39 -/mice, all of the type 1 WdiE species detected in the control mice were greatly reduced, and the total amount of type 1 WdiEs was~4% of that in control mice. On the basis of this result, combined with the fact that Cyp4f39 is a FA w-hydroxylase, we concluded that the type 1 WdiEs detected in the control mice were type 1w WdiEs.
The existence of Chl-OAHFAs in meibum lipids has previously been reported (Chen et al., 2013;Butovich, 2017). To examine whether Cyp4f39 is involved in the production of these species, we analyzed Chl-OAHFA levels in meibum lipids from control and Tg-Cyp4f39 -/mice by LC-MS/MS. Chl-OAHFAs with C46-C54 in the OAHFA moiety were detected in control mice, but they were almost absent in Tg-Cyp4f39 -/mice ( Figure 6E). As Cyp4f39 is involved in the production of OAH-FAs composed of a C30-C36 w-OH FA and a C16:1 FA ( Figure 4C), the majority of the Chl-OAHFAs detected in the control mice may also contain a C30-C36 w-OH FA and a C16:1 FA (although they also contain a small fraction of C18:1 FA). In summary, Cyp4f39 is involved not only in the production of OAHFAs and type 2w WdiEs, but also in that of type 1w WdiEs and Chl-OAHFAs.

Increases in CE and WE levels in meibomian glands of Cyp4f39deficient mice
To examine the effects of Cyp4f39 deficiency on meibum lipids other than OAHFAs and their derivatives (type 1/2w WdiEs and Chl-OAHFAs), meibum lipids prepared from Tg-Cyp4f39 +/+ and Tg-Cyp4f39 -/mice were separated by normal-phase thin-layer chromatography (TLC), in which the mobility of lipids with lower polarities is greater. We used two resolving solutions for the TLC: one suitable for separating total meibum lipids, and one for the separation of CEs and WEs. CEs and WEs are the least polar of the meibum lipids, and they were detected in the uppermost two bands on the TLC plates ( Figure 7A). We confirmed the reduction of OAHFAs and their derivatives in Tg-Cyp4f39 -/mice ( Figure 7A). Judging from their mobility in the TLC, WdiEs and Chl-OAHFAs showed similar degrees of polarity, and these were intermediate between the polarities of CEs/WEs and OAHFAs. Both CEs and WEs were more abundant in Tg-Cyp4f39 -/mice than in Tg-Cyp4f39 +/+ mice ( Figure 7A). Next, we quantified CE levels by LC-MS/MS. CEs containing C16-C30 saturated or C16-C36 monounsaturated FAs were detected in the meibum lipids of control mice ( Figure 7B). In Tg-Cyp4f39 -/mice, CEs of many chain lengths were more abundant. In particular, the levels of !C32 species were higher in Tg-Cyp4f39 -/mice than in the control mice: the total amount was about 1.7 times higher. These results suggest that !C32 FAs that were not w-hydroxylated by Cyp4f39 were used for the synthesis of CEs.

Discussion
CEs and WEs, which are the most abundant components of the TFLL, form a nonpolar lipid sublayer that is in contact with the external environment. OAHFAs constitute an amphiphilic lipid sublayer and are thought to have a role in stabilizing the tear film by creating an interface between the nonpolar lipid sublayer and the aqueous layer (Butovich et al., 2009;King-Smith et al., 2013). However, it has not yet been possible to create a model organism that lacks OAHFAs, due to a lack of knowledge about their biosynthesis pathway. Therefore, the actual function of OAHFAs has not yet been elucidated. In the present study, we found that FA w-hydroxylase Cyp4f39 is involved in OAHFA production (Figure 4). Cyp4f39-deficient mice exhibited dry eye accompanied by meibomian gland obstruction and tear film destabilization ( Figure 2). In these mice, not only OAHFAs but also their derivatives (type 1w WdiEs, type 2w WdiEs, and Chl-OAHFAs) were reduced ( Figures 5  and 6). These results suggest that the observed dry eye was caused by the combined effects of decreased production of OAHFAs and their derivatives.
Cyp4f39 and its human ortholog, CYP4F22, show high w-hydroxylase activity toward !C30 FAs (Ohno et al., 2015;Miyamoto et al., 2020). Alkyl chains are classified according to their chain length: those with C11-C20 are long-chain, whereas those with !C21 are very-long-chain (VLC). Long-chain acyl-CoAs are converted to VLC acyl-CoAs through the FA elongation cycle, which comprises four reactions: condensation, reduction, dehydration, and reduction (Kihara, 2012). The condensation reaction, which is the rate-limiting step of the FA elongation cycle, is catalyzed by FA elongases. Mammals have seven FA elongase isozymes (ELOVL1-7) that have different substrate specificities (Kihara, 2012). It is likely that ELOVL1, ELOVL3, and ELOVL4 are involved in the production of !C30 monounsaturated very long chain fatty acids (VLCFAs), the components of OAHFAs. ELOVL1 and ELOVL3 are active toward saturated and monounsaturated C22-C24 acyl-CoAs (Ohno et al., 2010;Sassa et al., 2018), whereas ELOVL4 shows activity toward !C24 acyl-CoAs (Vasireddy et al., 2007;Agbaga et al., 2008). The contribution of ELOVL1 to FA elongation is high for saturated VLC acyl-CoAs but low for monounsaturated VLC acyl-CoAs (Sassa et al., 2013;Sassa et al., 2018). Since most of OAHFAs are composed of monounsaturated VLCFAs (Figure 4), OAHFAs are not decreased in Elovl1-deficient mice, in which the expression of Elovl3 is increased as a compensatory mechanism (Sassa et al., 2018). However, CEs and WEs, which are rich in saturated VLCFAs and VLCFAls, respectively, are shortened in Elovl1-deficient mice, leading to a dry eye-like phenotype (Sassa et al., 2018). In addition, plugging at the meibomian gland orifice was observed in Elovl4 mutant mice (McMahon et al., 2014).
We predict that the OAHFA biosynthesis pathway in the meibomian gland is as follows. First, long-chain acyl-CoAs are elongated by ELOVL1 and ELOVL3 to generate VLC acyl-CoAs, and then they are further elongated to !C30 acyl-CoAs by ELOVL4 ( Figure 8A). After the removal of CoA, the resulting !C30 VLCFAs are w-hydroxylated by CYP4F22 (Cyp4f39 in mice). Finally, an C16:1/ C18:1 acyl group is transferred from C16:1/C18:1-CoA to the w-OH VLCFAs, generating OAHFAs. In these reactions, the thioesterase that catalyzes the CoA removal and the acyltransferase that catalyzes the final ester bond formation have not yet been identified.
Although the existence of type 2a/w WdiEs in meibum lipids has been suggested (Chen et al., 2013;Butovich, 2017), it has not been determined whether they are a or w positional isomers. In the present study, we found that both type 2w and type 2a WdiEs are present, and that type 2w WdiEs are synthesized in a Cyp4f39-dependent manner ( Figure 5). Furthermore, we revealed that Cyp4f39 is also involved in the production of type 1w WdiEs and Chl-OAHFAs ( Figure 6). We propose that the synthesis pathway for these OAHFA derivatives is as follows ( Figure 8A). First, OAH-FAs are converted to (O-acyl)-w-OH (OAH) acyl-CoAs. Next, OAH acyl-CoAs are transferred to FAls or cholesterol, producing type 1w-WdiEs or Chl-OAHFAs, respectively. Alternatively, OAH acyl-CoAs are converted to OAHFAls by acyl-CoA reductase, followed by ester bond formation with FAs, generating type 2w WdiEs. However, another pathway is also possible for type 2w WdiE synthesis, in which w-OH VLCFAs are reduced to a,w-fatty diols, followed by conversion to OAHFAls and then to type 2w WdiEs.
At present, the acyltransferase involved in C16:1 OAHFA production (i.e., forming the ester bond between a C16:1 FA and an w-OH VLCFA) is unknown. However, it is possible that C18:2 OAHFA, which was detected at higher levels than C16:1 OAHFA in our meibum lipid preparation (Figure 4), is derived from w-O-acylceramides via degradation. w-O-Acylceramides, which are important for the skin barrier and exist specifically in the epidermis (Kihara, 2016), consist of a long-chain base, an w-OH FA (C30-C36), and linoleic acid (C18:2); the conjugate of the latter two components is C18:2 OAHFA. The transacylase PNPLA1 catalyzes the formation of an ester bond between w-OH FA and linoleic acid using triglyceride as the linoleic acid donor (Ohno et al., 2017). We speculate that the C18:2 OAHFA that was detected in meibum lipids is derived from the keratinized epithelial cells constituting the meibomian gland ducts or from epidermis contamination in the samples. On the other hand, since PNPLA1 can produce few w-O-acylceramides containing C16:1 FA (Hirabayashi et al., 2019), it is unlikely that it is involved in the production of C16:1 OAHFA in meibocytes.
Tears are secreted when the eyes are closed, and the formation of the tear film begins when the eyes open and involves two processes (Brown and Dervichian, 1969;Willcox et al., 2017). First, the liquid components of the aqueous layer that has accumulated on the lower eyelid side extend upward due to the negative pressure in the upper eyelid rim that is generated when the eyes are opened. At that time, the TFLL lipids (meibum lipids) remain on the lower eyelid side but are hardly present on the upper eyelid side, creating a surface pressure gradient. Subsequently, the TFLL lipids spread on the aqueous layer from the lower eyelid side to the upper eyelid side, cancelling out the surface pressure gradient (this is known as the Gibbs-Marangoni effect) (Brown and Dervichian, 1969;King-Smith et al., 2009). For the TFLL to extend, the TFLL and aqueous layers must interact with each other. However, CEs and WEs, which constitute most of the TFLL, are only weakly polar and can barely interact with the aqueous layer, whereas the polar phospholipids are not able to interact sufficiently with the nonpolar CEs and WEs. Therefore, OAHFAs, which have a polar group as well as a long hydrophobic chain, may play an important role in connecting the aqueous layer to the nonpolar lipid sublayer composed of CEs and WEs. Furthermore, in the present study, we have revealed that type 1w WdiEs, type 2w WdiEs, and Chl-OAHFAs, all derived from OAHFAs, have polarities intermediate between those of OAHFAs and CEs/WEs. This suggests that these OAHFA derivatives are located between OAHFAs and CEs/WEs and function to connect them. Therefore, we speculate that a lipid polarity gradient is formed from the aqueous layer/TFLL boundary to the external environment (from phospholipids, to OAHFAs, then OAHFA derivatives, and finally CEs/ WEs), and that this gradient keeps the TFLL stably on top of the aqueous layer ( Figure 8B). In fact, Tg-Cyp4f39 -/mice showed tear accumulation on the lower eyelid side ( Figure 1F), which indicates increased surface tension and decreased interaction between the aqueous layer and the TFLL. In addition, these mice exhibited plugging at the orifices of the meibomian glands ( Figure 2D). It is likely that the decreased interaction between the aqueous layer and the TFLL causes accumulation of meibum lipids at the eyelid rim, leading to abnormal secretion of meibum lipids from the meibomian glands. The plugging observed in Tg-Cyp4f39 -/mice was mild (semi-liquid state), and their meibomian glands were not enlarged. If secretion of meibum lipids had been strongly inhibited, the meibomian glands would have been expected to swell. Therefore, we speculate that that most meibum lipids are secreted in Tg-Cyp4f39 -/mice. However, we cannot exclude the possibility that the slight reduction in the secretion of total meibum lipids (not only OAHFA and its derivatives, but also other meibum lipids) due to the plugging causes further progression of dry eye in Tg-Cyp4f39 -/mice.
In the present study, we have revealed that Cyp4f39/CYP4F22 is involved in the production of OAHFAs and their derivatives, and have obtained some clues concerning their biosynthesis pathways. Furthermore, on the basis of these findings, we propose a model in which the lipid polarity gradient in the TFLL is important for the stabilization of the TFLL on top of the aqueous layer. Although dry eye drugs targeting the aqueous and glycocalyx layers exist, there are no drugs targeting the TFLL. Considering that the majority of dry eye disease is caused by abnormalities in the TFLL, future studies are needed to develop drugs that target this layer, such as eye drops containing OAHFAs/OAHFA derivatives or drugs that promote their production.

Mice
The 3ÂFLAG-Cyp4f39 transgene (Tg-Cyp4f39), under the control of the IVL promoter, was constructed as follows. Cyp4f39 was amplified by PCR from cDNA from mouse testis and cloned into the 3ÂFLAG vector pCE-puro 3ÂFLAG-1 (Kihara et al., 2006), producing the pCE-puro 3ÂFLAG-Cyp4f39 plasmid. The 3ÂFLAG-Cyp4f39 fragment in the pCE-puro 3ÂFLAG-Cyp4f39 plasmid was then transferred to the pH3700-pL2 vector containing a human IVL promoter (Carroll et al., 1993), generating the pH3700-pL2À3ÂFLAG-Cyp4f39 plasmid. The IVL-Tg-Cyp4f39 fragment in the pH3700-pL2À3ÂFLAG-Cyp4f39 plasmid was excised, purified, and microinjected into fertilized eggs from C57BL/6J mice. Genomic DNA samples were prepared from the mice thus generated, and the presence of the Tg-Cyp4f39 transgene was detected by genomic PCR using primers (Involucrin-Tg-F and Cyp4f39-R; Supplementary file 1) to amplify the transgene fragment (678 bp). The transgene was present in four founder mice, and their epidermis was subjected to immunoblotting with anti-FLAG antibody to validate the expression of 3ÂFLAG-Cyp4f39 protein. The transgenic mice expressing 3ÂFLAG-Cyp4f39 protein (Tg-Cyp4f39 +/+ mice) were crossed with Cyp4f39 +/mice (Miyamoto et al., 2020) to produce Tg-Cyp4f39 +/mice. This strain was maintained by repeated back-crossing with C57BL/6J mice. Tg-Cyp4f39 -/mice were prepared by crossing Tg-Cyp4f39 +/mice. Genotyping for Tg-Cyp4f39 was performed as described above, and that for Cyp4f39 was conducted by genomic PCR using p1 and p2 primers as described previously (Miyamoto et al., 2020). The mice were housed in a specific pathogen-free, controlled environment (room temperature, 23 ± 1˚C; humidity 50 ± 5%; 12 hr light/12 hr dark cycle). They were fed with normal diet (PicoLab Rodent Diet 20; LabDiet, St. Louis, MO) and given water ad libitum. All animal experiments were approved by the institutional animal care and use committee of Hokkaido University (Permit Number: 17-0017).

BUT measurement and corneal damage scoring
Soon after 3 mL of 0.5% liquid sodium fluorescein (Tokyo Chemical Industry, Tokyo, Japan) had been loaded onto the mouse eyes, the mice were manually forced to blink three times, to ensure that the fluorescein solution covered the entire surface of the eyeballs. The mouse eye surfaces were then observed under a slit lamp microscope (Slit lamp RO800; Luneau Technology Operations, Pont-del'Arche, France) with a cobalt blue filter. BUT represents the elapsed time (in seconds) from the moment when the eyeball surface was covered with the fluorescein solution until the uniform staining was destroyed. Scoring of corneal epithelial damage was performed immediately after measurement of BUT. Mice were anesthetized by intraperitoneal injection of 0.05 mg/g (body weight) pentobarbital (Tokyo Chemical Industry), and then their eye surfaces were observed under a slit lamp microscope with cobalt blue and yellow filters. Scoring of corneal damage was performed according to a previous report (Lemp, 1995). The eyeball surface was divided into five sections, and the corneal damage in each section was evaluated as grade 0 to 3. The corneal damage score was calculated as the sum of the section scores. The grades were determined on the basis of the fraction of the area over which punctate staining was observed: 0, <1/3 area; 1, 1/3 to <2/3 area; 2, 2/3 to <3/3 area; 3, entire area or presence of filamentary keratitis.

Tear quantity measurement
After the mice had been anesthetized through inhalation of isoflurane (FUJIFILM Wako Pure Chemical), tear quantities were measured using the phenol red-thread test using Zone-Quick (Showa Yakuhin Kako, Tokyo, Japan), according to the manufacturer's manual.

Hematoxylin and eosin staining
The eyeballs and eyelids were fixed with Super Fix (Kurabo, Osaka, Japan) at 4˚C for !24 hr. Preparation of paraffin sections and staining with hematoxylin and eosin were performed as described previously (Sassa et al., 2013). Brightfield images were observed under a Leica DM5000 B microscope (Leica Microsystems, Wetzlar, Germany).

Lipid analyses by LC-MS/MS
Meibomian glands were removed from the upper and lower eyelids of the mice under a stereomicroscope and subjected to lipid extraction as follows. Meibomian glands (2.0-3.8 mg) were transferred to zirconia-bead-containing tubes (SARSTEDT, Nü mbrecht, Germany) and suspended in 450 mL of chloroform/methanol/12 M formic acid (100:200:1, v/v). Samples were vigorously mixed (4˚C, 5000 rpm, 2 min) using a Micro Smash MS-100 (TOMY Seiko, Tokyo, Japan). After removing the zirconia beads, the samples were mixed with 150 mL of chloroform and 270 mL of water. Phase separation was performed by centrifugation (20,400 Â g, room temperature, 5 min), and the organic phase was recovered. The aqueous phase was again subjected to lipid extraction by mixing with 150 mL of chloroform, followed by centrifugation. The organic phase was recovered and combined with the previous one. The extracted lipids were dried and dissolved in 200 mL of hexane. Next, the lipids were subjected to a second phase separation by mixing with 200 mL of water and centrifugation (20,400 Â g, room temperature, 3 min). The organic phase was recovered, and the lipids were reextracted from the aqueous phase by mixing with 200 mL of hexane, followed by centrifugation. The combined lipid extracts (meibum lipid fraction) were dried. FAs were prepared from the meibum lipid fraction by alkaline treatment as described previously (Jojima et al., 2020). FAs and OAHFAs were derivatized with AMPP using an AMP+ MaxSpec Kit (Cayman Chemical, Ann Arbor, MI) as described previously (Hancock et al., 2018). Meibum lipids were dissolved in chloroform/methanol (1:2, v/v) at a concentration of 0.1 mg tissue/mL, and 0.1 mL (FAs) or 1.0 mL (OAHFAs) of each sample was subjected to derivatization, according to the manufacturer's manual. After the reaction, acetonitrile was added so that the sample volume became 100 mL.
LC-MS/MS analyses were performed using ultra-performance LC (UPLC) coupled with electrospray ionization tandem triple-quadrupole mass spectrometry (Xevo TQ-S; Waters, Milford, MA). Lipids were separated by UPLC using a reverse phase column (Acquity UPLC CSH C18 column; length, 100 mm; particle size, 1.7 mm; inner diameter, 2.1 mm; Waters) at 55˚C. The amounts of meibum lipids injected onto the UPLC column were 0.5 mg (tissue weight) for AMPP-FAs, 5 mg for AMPP-OAH-FAs, and 50 mg for each of type 2a/w WdiEs, type 1a/w WdiEs, Chl-OAHFAs, and CEs. Lipid separation by UPLC was performed at a flow rate of 0.3 mL/min using a gradient system in which mobile phase A (acetonitrile/water [3:2, v/v] containing 5 mM ammonium formate) and mobile phase B (acetonitrile/2-propanol [1:9, v/v] containing 5 mM ammonium formate) were mixed. The gradient conditions for AMPP-OAHFA measurement were as follows: 0 min, 40% B; 0-18 min, gradient to 100% B; 18-23 min, 100% B; 23-30 min, gradient to 40% B. For the type 2a/w WdiEs, type 1a/w WdiEs, Chl-OAHFAs, and CEs, the gradient conditions were as follows: 0 min, 60% B; 0-21 min, gradient to 100% B; 21-25 min, 100% B; 25-30 min, gradient to 60% B. Ionization of lipids was performed by electrospray ionization. Quantitative analyses were performed by MRM in positive ion mode. The cone voltage was set at 35 V, and an appropriate collision energy was set for each lipid (AMPP-OAHFA, 60 eV; type 1a/w and type 2a/w WdiEs, 20 eV; Chl-OAHFAs and CEs, 15 eV). The m/z values of the precursor and product ions of each lipid species were set in the mass filters Q1 and Q3, respectively (Yamamoto et al., 2020; Supplementary files 2-6). Product ion scanning was performed for fragment ion analyses. In the analyses of OAHFAs, [M -H]was detected as the product in negative ion mode, but in the analyses of type 1 and 2 WdiEs, [M + H] + was detected in positive ion mode. Data analyses were performed using MassLynx software (Waters).

Lipid analyses by TLC
The upper and lower eyelids were removed from the mice and cooled on ice. Meibomian gland contents were squeezed out from the orifices of the meibomian glands using two pairs of tweezers under a stereomicroscope, while keeping the tissue on ice as described previously (Butovich et al., 2012). The extracted contents were dissolved in chloroform/methanol (1:2, v/v) and dried. They were then dissolved in 450 mL of chloroform/methanol/12 M formic acid (100:200:1, v/v), mixed with 150 mL of chloroform and 270 mL of water, and centrifuged at 20,400 Â g for 5 min at room temperature. Lipids were recovered from the organic phase, dried, suspended in chloroform/methanol (1:2, v/v), and separated using normal-phase TLC using high-performance TLC plates (Glass HPTLC Silica gel 60 plates; Merck) with solution 1 (hexane/diethyl ether/acetic acid [90:25:1, v/v]) or solution 2 (hexane/toluene [1:1, v/v]) as the resolving solution. Lipids were detected by spraying the plates with a copper phosphate reagent (3% CuSO 4 [w/v] in 8% [v/v] aqueous phosphoric acid solution), followed by heating at 180˚C.

Chemical synthesis of OAHFA and type 2 WdiEs
The OAHFA (O-C18:1)-w-OH C22:0 FA was synthesized as follows. 22-Hydroxy docosanoic acid (w-OH C22:0 FA; 1.2 mg, 3.4 mmol; Larodan AB, Solna, Sweden) was dissolved in 400 mL of tetrahydrofuran, mixed with triethylamine (3.7 mL, 27.0 mmol; FUJIFILM Wako Pure Chemical) and oleoyl chloride (4.5 mL, 13.5 mmol; FUJIFILM Wako Pure Chemical) on ice, and incubated at room temperature for 48 hr while being mixed (Figure 4-figure supplement 1). The octadec-10-enoic-22-(octadic-10enoyloxy)-docosanoic anhydride thus produced was hydrolyzed by incubating it with 200 mL of saturated aqueous sodium bicarbonate at room temperature for half a day, producing sodium 22-(octadec-10-enoyloxy)-docosanoate. Next, (O-C18:1)-w-OH C22:0 FA was generated by adding 1 M hydrochloric acid to the above sodium salt until the pH of the sample was less than 3. Phase separation was performed by adding 300 mL of chloroform to the reaction solution, followed by centrifugation at 20,400 Â g for 3 min at room temperature. The organic phase containing the OAHFA was recovered and dried.

Quantification and statistical analyses
Data are presented as means ± SD. The significance of differences between groups was evaluated using Student's t-test in Microsoft Excel (Microsoft, Redmond, WA) or Tukey-Kramer's test in JMP 13 software (SAS Institute, Cary, NC). A P-value of <0.05 was considered significant.