Hyperosmolarity-induced lipid droplet formation depends on ceramide production by neutral sphingomyelinase 2.

Hyperosmolarity (HO) imposes a remarkable stress on membranes, especially in tissues in direct contact with the external environment. Our efforts were focused on revealing stress-induced lipid changes that precede the inflammatory cytokine response in human corneal epithelial cells exposed to increasing osmolarity. We used a lipidomic analysis that detected significant and systematic changes in the lipid profile, highly correlated with sodium concentrations in the medium. Ceramides and triglycerides (TGs) were the most-responsive lipid classes, with gradual increases of up to 2- and 3-fold, respectively, when compared with control. The source of ceramide proved to be sphingomyelin hydrolysis, and neutral sphingomyelinase 2 (NSM2) activity showed a 2-fold increase 1 h after HO stress, whereas transcription increased 3-fold. Both TG accumulation and IL-8 secretion were shown to be dependent on ceramide production by specific knock-down of NSM2. In HCE cells, diglyceride acyltransferase 1 was responsible for the TG synthesis, but the enzyme activity had no effect on cytokine secretion. Hence, NSM2 plays a key role in the cellular response to hyperosmolar stress, and its activity regulates both cytokine secretion and lipid droplet formation.

Acquity Ultra Performance liquid chromatograph combined with a Waters Q-Tof Premier mass spectrometer ( 19 ).The column (at 50°C) was an Acquity UPLCTM BEH C18, 2.1 × 100 mm with 1.7 m particles. The solvent system included ultrapure water (1% 1 M NH 4 Ac, 0.1% HCOOH) (A) and (B) LC/MS-grade acetonitrile/isopropanol (1:1, 1% 1M NH4Ac, 0.1% HCOOH). The gradient started from 65% A/35% B, reached 80% B in 2 min, and 100% B in 7 min, and remained there for 7 min. The fl ow rate was 0.4 ml/min, and the injected amount was 2.0 l. The lipid profi ling was carried out using ESI in positive mode, and the data were collected at a mass range of m/z 300-1,200. The data were processed using MZmine2 software ( 20 ). Lipids were identifi ed using an internal spectral library. The data were normalized using internal standards representatives of each class of lipid present in the samples [PC (17: Finally, the data were further normalized by dividing the normalized lipid concentration with the protein content of the sample. The differences between samples were compared using Student's t -test. To account for the multiple testing issues, q values indicating the false discovery rate were calculated. Median values of the lipids were used for calculating fold changes, and if the fold change was <1, the inverse negative of the value was taken. The heat map showing the fold changes was constructed by taking the median value of each group and comparing it to the median value of control samples. Correlation between lipid concentrations and added concentration of NaCl was calculated using Pearson's correlation. Sphingomyelinase activity was measured using the AmplexRed sphingomyelinase assay kit (Invitrogen) under both neutral as well as acidic conditions as recommended by the manufacturer. Cells grown on 6 cm plates were washed and scraped in PBS on ice. The cells were then pelleted by centrifugation at 400 g for 10 min at 40°C and resuspended in distilled water. The wholecell lysate was prepared by sonication of the cell suspension in water for 10 s using a Soniprep 150 (SanyoElectric; Tokyo, Japan). A 1:10 (v/v) dilution of the lysate in sodium acetate buffer (pH 5.0) was used for acid sphingomyelinase activity measurement, whereas for neutral sphingomyelinases, the sample was diluted 1:5 (v/v) in 0.1 M Tris-HCl, 10 mM MgCl 2 (pH 7.4). Reaction mixtures were incubated in the dark at 37°C for 30 to 60 min. Fluorescence was measured with Victor 2 Multilabel Counter (Wallac; Turku, Finland) with excitation at 535 nm and emission at 585 nm. Standard curves were prepared for each reaction, from the positive-control enzyme provided with the kit. Results were normalized to protein concentration and experimental control and presented as fold change.

Gene expression analysis
Total RNA was extracted using the RNeasy Mini Kit (QIAGEN; Hilden, Germany), and 1 µg was converted into cDNA utilizing a fi rst-strand synthesis kit (Fermentas; St. Leon-Rot, Germany) and following the protocol recommended by the manufacturer. Subsequently, 50 ng cDNA were used per reaction in Q-PCR with SYBR Green/ROX FastStart Universal master mix (Roche; Basel, Switzerland). The reaction was performed in 25 µl fi nal volume and 0.2 µM of each primer in an ABI Prism7000 thermo cycler (Life Technologies Corporation; Carlsbad, CA). The sequence for used primers is given in supplementary Table I. As control gene, we have used human actin, and expression was quantifi ed by calculating ddCt and 2-ddCt, graphically represented as fold change.
Sphingosine-1-phosphate (S1P) production was quantifi ed abiding by a protocol previously published ( 21 ). Cells readily take up the [3H]sphingosine (Perkin Elmer; Waltham, MA) and then convert it intracellularly to [3H]S1P and further metabolites. for LD formation, and the recent focus on their structure has led to a broader view of their roles in cell physiology. LDs do not emerge only from a cell's need to store FAs or sterols. Under stress conditions, their formation is induced even in the absence of an extracellular source of lipids. Many environmental challenges, such as cell crowding, apoptosis, or the infl ammatory response, have been described to induce neutral lipid accumulation (11)(12)(13). The cellular function of LDs is considered both detrimental as well as protective. The fi rst is illustrated by eicosanoids production from the FFAs released during triglyceride (TG) hydrolysis ( 14 ), whereas the protective effect was hinted at by longer survival under stress of cells that produce TGs ( 15 ). Membrane phospholipids provide the building blocks for stress-induced LD synthesis ( 16 ).
In the present study, we aimed to assess adjustments in the lipid profi le of HO-stressed cells and to identify the most-dynamic lipid species as well as the key modulators of this stress response. For these purposes, we used a corneal epithelial cell line (HCE2) to illustrate that HO induces remarkable changes in cellular lipid homeostasis and that the sphingolipid signaling pathway plays an important role in the response to this external stimulus. SM hydrolysis is an early event in HO stimulation, and the stress-induced enzyme in the sphingomyelinase pathway is neutral sphingomyelinase 2 (NSM2). Furthermore, HO-induced stress droplet formation, as well as IL-8 secretion, is an event controlled by the same enzymes, and this outcome may be relevant in the clinical context.

Cells
Human corneal epithelial cells (HCEs) were grown and maintained in DMEM/F12 with 15% FBS and 10 ng/ml epithelial growth factor, 5 µg/ml insulin, 1µg/ml L -glutamine, 40 µg/ml gentamicin (all supplied by Invitrogen; Carlsbad, CA), and 0.1 µg/ml cholera toxin (Sigma; St. Louis, MO) at 37°C under 5% CO2 in a humidifi ed incubator ( 17 ). Cells were subcultured every 3 days for 10-15 passages after defrosting and serum-starved for 18 h before experiments. Standard growth medium contained 140 mM NaCl and had an osmolality of 317 mOsm/kg. The osmolarity of the media was increased by adding 5 M NaCl solution. Therefore, the molar concentrations specifi ed in RESULTS, in the range of 50 to 140 mM, represent the added NaCl concentrations above the fi xed salt content of the growth medium. For transfection, cells were subcultured in antibiotic-free medium at preconfl uence density and transfected on the following day.

Lipid extraction and nontargeted lipidomic analysis
Lipids were extracted from cell suspension using the Bligh-Dyer protocol ( 18 ). Briefl y, the cells were grown on 6-well plates, washed with cold PBS, and scraped in 0.9 ml of 2% NaCl solution. An aliquot of 0.1 ml was used for protein concentration measurements, whereas the remaining 0.8 ml was mixed with chloroform and methanol (0.8:1:2, v/v), and the sample was centrifuged to eliminate debris. Next, for the phase separation, 1 ml water and 1 ml chloroform were added and the samples centrifuged. The lower phase was aspirated to a clean glass tube, and solvents were evaporated under nitrogen . Samples were stored at Ϫ 20°C until analysis. The cell extracts were analyzed on an MN). The array measured concomitantly the levels of 36 different cytokines and chemokines (spotted in duplicates on a nitrocellulose paper). Samples were prepared following the manufacturer's instructions. Briefl y, conditioned media from confl uent cells was cleared of fl oating cells and debris and an aliquot of 700 µl was mixed with the kit reagents and assayed immediately.

siRNA transfection
We used DharmaFECT SMART pool, a mixture of four short, interfering RNAs (siRNAs) targeting the human NSM2 gene (accession number NM_018667.3) (Dharmacon; Lafayette, CO). Cells were transfected at 70% confl uence in 12-well plates by adding to each well 25 pmol of the siRNA pool and 3 µl of Dharmafect Transfection Reagent 1. The transfection protocol followed the manufacturer's instructions. Before experiments, cells were serum-starved overnight at 36 h after the transfection. As control for transfection experiments, we used nontargeting siRNA.

Statistical analysis
Statistical analysis was performed using GraphPad Prism 4.03 (GraphPad Software, Inc.; La Jolla, CA). Statistical signifi cance was determined by t -test or ANOVA on values from an average of three independent experiments. Data are reported as mean ± SEM unless otherwise specifi ed. Values of P < 0.05 were considered signifi cant.

RESULTS
Osmotic stress has been associated with infl ammation for almost a decade, and most of its organ-level effects are a consequence of increased secretion of proinfl ammatory molecules ( 23 ). Therefore, we measured HO-induced cytokine secretion by corneal epithelial cells using a protein array ( Fig. 1A ). As control, we used cells incubated in normosmolar medium (317 mOsm/kg with 140 mM NaCl). The osmolarity of the media was increased by addition of 5 M NaCl to reach the specifi ed molar concentrations. The pro-infl ammatory cytokines found to be modulated by HO in HCE cells (IL-8, IL-6, MIF, and GROalpha) have been previously observed in tears of patients with infl ammatory eye diseases ( 24,25 ).
Because IL-8 is regarded as an acute-phase chemokine and has proved to be highly modulated by osmolarity, we used ELISA to determine the time course of the cytokine response ( Fig. 1B ). A signifi cant increase in IL-8 secretion was evident 3 to 4 h after the stress induction.
We next intended to distinguish early lipid alterations that precede the cytokine secretion and accordingly used cells incubated for 2 h in increasing concentrations of NaCl for total lipid extraction and lipidome analysis by LC/MS. Quantifi ed lipids were normalized to internal standards, added to the sample before lipid extraction and to total protein concentration.
The nontargeted lipidomic analysis identifi ed 176 individual lipids, whereas another 274 potential lipid compounds remained unknown. Of the identifi ed lipids, 108 lipid species (61.4%) showed signifi cant positive or negative correlation (58 and 50 compounds, respectively) with the increase in NaCl concentration in the medium (Pearson correlation index above 0.5 or below 0.5). The lipid profi le is presented as a heat map with fold changes in individual lipids in Fig. 2A .
For these measurements, HCEs were cultured on 6-well plates; and after experiments, media was collected, cleared of cells and debris by centrifugation, and mixed with equal volumes of methanol, chloroform, and 0.1 N NaOH. After the phase separation, under alkaline solvent extraction conditions, S1P is negatively charged and partitions into the aqueous phase, whereas other neutral sphingolipids, including sphingosine, are retained in the organic phase. In this manner we could measure both secreted as well as intracellular S1P. For intracellular S1P levels, cells were collected in 1 M NaCl. A small part was saved for protein analysis, whereas the remainder was used for lipid extraction, similarly to the conditioned medium. Counts per minute were then normalized to protein concentration.

Oil Red O staining fl uorescence microscopy and quantifi cation
Cells grown on glass coverslips were used for Oil Red O fl uorescence microscopy. After the media was removed, the cells were carefully washed with PBS and fi xed with 10% formalin in PBS. Oil Red O staining solution was freshly prepared by mixing 3% Oil Red O solution in isopropanol with water (3:2, v/v), and cells were stained for 5 min and then rinsed with water. Samples were visualized on an Axioplan 2 fl uorescence microscope (Carl Zeiss, AG; Germany) with a Plan Neofl uar 40×/0.75 (DIC II) air objective. Images were acquired at room temperature with an AxioCamHR camera. For quantitative data, cells grown in 12-well plates were stained with Oil Red O as described above. After rinsing, the retained Oil Red O dye was extracted with 300 µl of 100% isopropanol, and 250 µl were pipetted into 96-well plates for absorbance measurement at 490 nm (Wallac Victor 2 Multilabel Counter; Turku, Finland). As blank, we used similarly processed cell-free wells. Data are presented as absorbance at 490 nm of samples after blank subtraction.

Lipolysis assay
To measure TG hydrolysis, we used a previously described radiometric method ( 22 ). Cells were cultured on 10 cm-diameter plates, washed, and scraped in PBS. Cells were pelleted and lysed by brief sonication in the assay buffer (0.2 M Tris-HCl buffer, pH 8.4, with 0.1 M NaCl). The lipolytic activity of 0.2 ml of cellular lysate was measured using labeled triolein (30 nmol/assay). The reaction mixture was incubated for 60 min at 37°C and terminated by adding 3.25 ml of methanol-chloroform-heptane (10:9:7, v/v) and 1 ml of 0.14 M potassium carbonate-0.14 M boric acid, pH 10.5. The samples were then centrifuged (1,600 g for 15 min), and the radioactivity in 1 ml of the upper phase was determined by liquid scintillation counting. Results were normalized to protein concentration.

IL-8 ELISA
For IL-8 secretion analysis, a commercial ELISA DuoSet kit (R and D Systems; Minneapolis, MN) was used. The assay plates were prepared as recommended by the manufacturer. Media was collected from culture plates and centrifuged for 10 min at 400 g to eliminate fl oating cells and debris and stored frozen until measured. For assays, 60 µl of media were mixed with an equal volume of reagent diluent, and 100 µl were pipetted onto the plates coated with the capture antibody. Plates were analyzed using the microplate reader Victor 2 Multilabel Counter (Wallac) at a wavelength of 450 nm. To calculate the concentration, the standards were fi tted with a second-order polynomial equation. Concentrations were reported as pg/ml.

Cytokine secretion
For parallel determination of relative levels of secreted cytokines and chemokines, we have used a proteome profi ler array (Human Cytokine Array Panel A, R and D Systems; Minneapolis, increased after 4 h. At this latter time point, NSM and ASM show signifi cant activation in samples with high NaCl concentration ( P < 0.05, as shown in Fig. 3A, B ). Gene expression analysis of these samples highlighted NSM2 as the sphingomyelinase sensitive to HO ( Fig. 3C ). NSM2 transcription was progressively induced by NaCl up to 90 mM NaCl and then gradually decreased back to basal levels ( Fig. 3D ). Intriguingly, at these concentrations, where NSM2 expression seems to decline, the ASM enzyme shows increased enzymatic activity ( Fig. 3B ).
A deeper investigation into the sphingomyelin cycle revealed that 4 h after stress, stimulated cells displayed upregulated expression of neutral ceramidase, whereas its acidic counterpart remained unaffected ( Fig. 3C, D ). An elevated ceramidase activity suggests sphingosine (Sph) and, consequently, S1P production, inasmuch as Sph is highly toxic and its cellular accumulation must be avoided ( 26 ). Hence, we measured S1P production and secretion as well as expression of the S1P1-3 receptors (that we identifi ed as being expressed by corneal epithelial cells in vivo; data not shown). Intracellular concentration of S1P remained relatively constant at 2 and 4 h after stimulation, but secretion decreased in cells stressed for 4 h ( Fig. 3E ). The absence of S1P extracellular signaling was strengthened by the observation that S1P1 receptor expression was upregulated under stress ( Fig. 3C ), because signaling through this receptor is followed by downregulation of its expression ( 27 ). The SM-cycle activation and the accumulation of Cer prompted us to assess the degree of cellular death as a result of the HO stimulation. After 2 h under stress conditions, more than 90% of the cells remained on the plate, whereas after 4 h, approximately 20% of the cells detached from the substrate. Signifi cant cell death (more than 20%) was observed after 18 h in 140 mM of added NaCl solution (see supplementary Fig. I ).
TG accumulation measured by LC-MS/MS could be easily visualized in cells using Oil Red O staining and fl uorescence microscopy ( Fig. 4A-C ) and estimated by Oil Red O isopropanol extraction as well ( Fig. 4D ). Enzymatic measurement of both total cholesterol and free cholesterol showed that the cells do not accumulate cholesterol and that the visualized lipid droplets are mostly composed of TGs (see supplementary Fig. II ). Our fi rst hypothesis was that these stress-induced LDs are active sites of lipolysis, and we therefore measured general TG lipolysis as well as mRNA levels for diglyceride acyltransferase 1and 2 (DGAT1, 2) (as enzymes involved in TG synthesis), adipose tissue triglyceride lipase (ATGL), and hormonesensitive lipase as enzymes involved in TG hydrolysis ( Fig. 4E ). The enzyme responsible for the TG synthesis in our cellular model was DGAT1, as demonstrated by specifi c knockdown with siRNA ( Fig. 4G ). DGAT1 gene silencing increased IL-8 secretion induced by low HO stress (70 mM of added NaCl; Fig. 4H ), suggesting a protective role for LDs under low-stress conditions. DGAT2 silencing by siRNA transfection had no signiffi cant effect on lipid droplet formation or IL-8 secretion (see supplementary Fig. IIIA, B , respectively). The lipolysis assay suggested lipoprotein lipase (LPL) as an effi cient lipolytic enzyme, Most affected lipid classes were the ceramides (Cers) and TGs with signifi cant fold increases from nonstressed cells. Conversely, a few SM species and phosphatidylethanolamine (PE) levels were inversely correlated with the increase in extracellular osmolarity. Phosphatidylcholine species were not affected at this early phase of treatment ( Fig. 2B ).
Considering the suggested correlation between ceramide accumulation and SM decrease, we measured sphingomyelinase activity in stimulated cells at neutral pH as well as under acidic conditions. After 1 h under HO conditions, the acid sphingomyelinase (ASM) remained unaffected, but activity at neutral pH increased with the ionic strength of the media. This effect was conserved at 2 h and because no releasing agent was used to remove LPL from the cell surface. Stressed cells did not differ from controls at transcriptional level, when genes involved in TG metabolism were analyzed ( Fig. 4E ; supplementary Table I).
Considering that the neutral lipids storage occurs in serum-deprived conditions, we searched for the FFA source because TG hydrolysis measured in the absence of LPL stabilizers (heparin and apoC-II as LPL cofactors) was signifi cantly lower. In both cases, lipolytic activity was attenuated by the addition of 90 mM NaCl ( Fig. 4F ). LPL, although an unlikely candidate for intracellular lipolysis, could contribute to the general lipolysis assay results, has suggested that the FFAs for stress-induced TG synthesis arise from phospholipid breakdown ( 16 ). S-bromoenol lactone [(S)-BEL] effi ciently prevented TG accumulation as well as IL-8 secretion ( Fig. 5A, B ), suggesting an involvement of the calcium-independent phospholipase A2 ␤ (iPLA2 ␤ ) in the events triggered by HO. iPLA2 ␤ was proven beforehand to be an enzyme involved in LD formation and, interestingly, as an upstream activator of sphingomyelinase ( 28,29 ). Additionally, by blocking that would induce TG synthesis. First, we investigated whether de novo synthesis was the source of FFAs to be used for TG formation. HCE cells were preincubated for 1 h with C75, a FAS inhibitor, after which HO was induced by adding NaCl to the media and TG droplets were quantifi ed by the Oil Red O dye assay. FAS inhibition did not prevent the neutral lipid accumulation ( Fig. 5A ); however, this result was expected, because the measured effects of HO stress occur relatively quickly (1-2 h). Previous work We employed the SB239063 inhibitor to test the involvement of the p38 MAPK in our experimental setup and found that it strongly inhibited IL-8 secretion upon stress, but did not affect TG droplet formation ( Fig. 5A, B ).

DISCUSSION
The surface of the eye is in constant contact with the external environment and hence subjected to numerous aggressions. HO is a clinically relevant stress factor for the corneal epithelium, and this condition is a consequence of either increased evaporation, decreased production of tears, or changes in the composition of the tear fl uid ( 33 ). HO is accompanied both in vivo and in vitro by a proinfl ammatory response, and we have used IL-8 secretion from HCE2 cells as an infl ammatory marker in our experimental design.
HO causes an undeniable stress on membranes, and our aim was to reveal how lipid homeostasis is affected by this extracellular stimulus. We have employed a lipidomic approach that brought forward ceramides and TGs as altered lipid classes. Here, we show that high osmolarity induces cellular TG build-up as LDs and that furthermore, clathrin-dependent endocytosis with chlorpromazine, TG synthesis was obstructed to the same extent as with (S)-BEL (see supplementary Consequently, we investigated a possible link between sphingomyelinase-mediated ceramide production and LD development. To reliably impede ceramide production, we silenced the NSM2 gene using siRNA transfection. The interfering RNA blocked NSM2 upregulation upon HO stimulation and also signifi cantly decreased IL-8 secretion from the HCE cells ( Fig. 5E, D, respectively). Figure 5E also shows that BEL treatment impeded NSM2 gene upregulation under HO conditions. Cellular TG accumulation was signifi cantly reduced, as verifi ed by Oil Red O isopropanol extraction ( Fig. 5C ). Knock-down effi ciency is presented at mRNA level as well as enzymatic activity at neutral pH in Fig. 5G, H , respectively.
In mammalian cells, p38 MAPK is the HO stress response kinase ( 30 ), and previous studies have linked the activity of this kinase to sphingolipid signaling ( 31,32 ). starting 5 min after the addition of the stimulus (data not shown), but we could measure a signifi cant response in sphingomyelinase activity only after 1 h under HO stress conditions, suggesting that SM pathway activation might be an effector more than a sensor of HO-induced membrane stress. Moreover, we observed an almost-linear induction of enzymatic activity with stress that was refl ected at gene expression level as well. NSM2 gene transcription was linearly induced by the addition of less than 90 mM of NaCl, after which it gradually returned to basal levels. Conversely, ASAH2 transcription was induced by the higher sodium concentrations (>80 mM NaCl).
Considering the regulation and crucial balance in sphingolipid metabolism, information on one compound does not provide a comprehensive representation of cellular response status. Gene expression analysis after 4 h suggested this process is closely coupled to ceramide production through NSM2 activity.
Although ASM and NSMs are known as stress response enzymes, this is, to our knowledge, the fi rst time that the NSM2 is reported to respond to osmolarity variations. The stress-dependent increase in ceramides proved to be a reliable indicator of the SM pathway activation by HO. We investigated SM hydrolysis as the source of ceramides because of the apparent, although not statistically signifi cant, corresponding decrease in SM species. This trend was most evident for the Cer(d18:1/16:0)-SM(d18:1/16:0) pair. Similar trends were more diffi cult to observe for other Cer species if we consider the disproportionality between the lipid species, inasmuch as the highest Cer concentration measured was 10-fold lower than the level of the corresponding SM . We tested for enzymatic activation In the experiments involving inhibitors, cells were preincubated for 1 h in the presence of the inhibitor alone followed by osmolar stress induction with NaCl. * P < 0.05 as compared with control. stress condition by reducing immune cell recruitment as well as the "outside in" effects of IL-8 signaling. Induction of the HO-stress pathway causes complex responses in cells; therefore, identifi cation of participants and branching points is needed in order to start any therapeutic approaches. Hyperosmolar tear fl uid has been associated with decreased cell viability and delayed wound healing ( 36 ); hence, an effi cient DES treatment should aim at restoring tissue homeostasis and at not inducing additional metabolic turbulence.
Nontargeted lipidomic analysis proved to be an important tool, inasmuch as it provided a hypothesis-free approach to this study, eliminating biased search for possible activated pathways. Data analysis revealed that HO stress in epithelial cells is accompanied by signifi cant rearrangements in cell lipidomes and small-scale changes in abundant lipids can translate to large-scale changes in rare lipid classes, such as Cers and TGs. In our experimental model, we showed an important participation of NSM2 in the cellular stress response to increased osmolarity. LD formation is not an event relevant only for our cellular model, but could have implications in other systems as well. Schwartz et al. ( 37 ) revealed that pathogen-related infl ammatory reaction increases the osmolarity of infl ammatory fl uids. Given that infl ammation is an essential cellular response to all stress inducers, its intimate connection with HO becomes an important issue in contexts different from that of the corneal surface or DES. Thus, revealing the details of this signaling pathway is an important near-future target of our research.
The authors thank Jari Metso for valuable technical help. upregulated ceramidase activity, which could defuse the harmful accumulation of proapoptotic ceramides. However, at the osmotic pressure at which both sphingomyelinase and ceramidase expression peaked, intracellular S1P production decreased. This implies that SM hydrolysis to ceramide is not counterbalanced by production of the prosurvival metabolite S1P. An increased S1P lyase activity in stimulated cells that would irreversibly eliminate the sphingolipid could be an alternative, but we found no evidence of this in the gene expression analysis.
Lipidomic analysis revealed that in addition to ceramide accumulation, HO induces LD formation, seen as a linear intracellular build-up of TGs with the added NaCl. This process was independent of de novo synthesis of FFAs, but was very sensitive to BEL treatment and seemingly independent of cyclooxygenase activity. Moreover, BEL prevented NSM2 gene upregulation in response to HO. The suggested iPLA2 ␤ role is defended to some extent in our model by the decrease in PE species measured in the lipidomic analysis and by the inhibitory effect of endocytosis blockage on the TG build-up. PLA2 enzymes are not novel candidates for stress-associated LD biogenesis, and iPLA2 was recently linked to high-salt response in Chlamydomonas ( 28,34 ). LD biogenesis in various types of cellular stress has become a very active fi eld of research during recent years, and it is presently considered a generalized response ( 16 ). This response could be an effi cient way to suppress lipotoxicity induced by FFA production resulting from PLA2 activation, thus increasing cell viability under stress ( 15 ). NSM2 knock-down had effects comparable to BEL treatment in lowering both cytokine secretion and formation of TG-enriched droplets and tightly linked ceramide production to both major outcomes. The delayed NSM2 response to HO stress suggests that SM pathway activation is a downstream event in the HO-induced pathway; however, the molecular sensors for high osmolarity remain to be identifi ed. Phospholipase activity together with the ceramide production might be events targeted at eliminating the pressure imposed on the plasma membrane by the cell shrinkage in the HO environment. This membrane stress is then eliminated by stimulated endocytosis of the "extra" membrane, and phospholipids are then recycled as stress droplets.
The HO-responsive p38 MAPK is a known activator of NSM2 ( 30 ), but surprisingly, its inhibition had no effect on LD formation, affecting only the proinfl ammatory outcome of the stress. This fi nding proposes a downstream position for p38 from NSM2 in our experimental set-up, as other studies have previously suggested ( 35 ).
HO stress is an environmental hazard with important effects on the ocular surface, and a well-concerted cellular response becomes a defi ning factor between healthy and dysfunctional cornea. This is clinically translated into DES, the most-common infl ammatory disease of the cornea. We have shown here that NSM2 blockage inhibits almost completely IL-8 chemokine secretion from HCE2 cells. The same effect was observed after BEL and p38 MAPK inhibitor treatment. This anti-infl ammatory effect would at least minimize the pathological outcome associated with the