Hepatocytes release ceramide-enriched pro-inflammatory extracellular vesicles in an IRE1α-dependent manner[S]

Nonalcoholic steatohepatitis (NASH) is a lipotoxic disease wherein activation of endoplasmic reticulum (ER) stress response and macrophage-mediated hepatic inflammation are key pathogenic features. However, the lipid mediators linking these two observations remain elusive. We postulated that ER stress-regulated release of pro-inflammatory extracellular vesicles (EVs) from lipotoxic hepatocytes may be this link. EVs were isolated from cell culture supernatants of hepatocytes treated with palmitate (PA) to induce lipotoxic ER stress, characterized by immunofluorescence, Western blotting, electron microscopy, and nanoparticle tracking analysis. Sphingolipids were measured by tandem mass spectrometry. EVs were employed in macrophage chemotaxis assays. PA induced significant EV release. Because PA activates ER stress, we used KO hepatocytes to demonstrate that PA-induced EV release was mediated by inositol requiring enzyme 1α (IRE1α)/X-box binding protein-1. PA-induced EVs were enriched in C16:0 ceramide in an IRE1α-dependent manner, and activated macrophage chemotaxis via formation of sphingosine-1-phosphate (S1P) from C16:0 ceramide. This chemotaxis was blocked by sphingosine kinase inhibitors and S1P receptor inhibitors. Lastly, elevated circulating EVs in experimental and human NASH demonstrated increased C16:0 ceramide. PA induces C16:0 ceramide-enriched EV release in an IRE1α-dependent manner. The ceramide metabolite, S1P, activates macrophage chemotaxis, a potential mechanism for the recruitment of macrophages to the liver under lipotoxic conditions.

this pathway and/or S1P signaling may be salubrious in NASH.

Cells
Previously described immortalized mouse hepatocyte (IMH) cell lines derived from IRE1 ␣ KO and WT mice ( 18 ), ATF6 ␣ -WT and KO IMH cell lines, and eIF2 ␣ S51A phosphorylation resistant (AA) and WT (SS) IMH cell lines were a kind gift of Dr. Randal Kaufman. The IMH cell lines were cultured in DMEM (Life Technologies) supplemented with 10% FBS, penicillin, and streptomycin. Huh7 cells were cultured in DMEM supplemented with 10% FBS, penicillin, and streptomycin. Primary mouse hepatocytes were isolated by collagenase perfusion followed by Percoll purifi cation, cultured as previously described by us ( 19 ), and used in experiments if viability exceeded 95%. For EV isolation, cells were cultured in growth medium supplemented with 5% EV-depleted FBS [prepared by overnight centrifugation at 100,000 g at 4°C according to standard protocols ( 20 )] and 1% BSA, with the addition of either PA or oleate (OA) as previously described by us ( 4 ), thapsigargin (Tg), or vehicle, respectively. Bone marrow-derived macrophages (BMDM s) were isolated from the long leg bones of C57Bl/6J mice, as previously described by us ( 16 ). Briefl y, dissected long bones were fl ushed with serum-free RPMI and cells were differentiated in 20% L929 cell conditioned medium containing RPMI and used on day 7 after differentiation.

EV isolation
Cells were grown to 90% confl uency on 150 mm tissue culture plastic dishes. Before treating, cells were washed twice with PBS and then the medium was changed to assay medium supplemented with PA, OA, or Tg ( 4 ). After 14-16 h, supernatant was recovered and sequential low-speed centrifugation was performed to deplete cells and cellular debris at 2,000 g for 20 min followed by 20,000 g for 30 min ( 20 ). The supernatant was further ultracentrifuged at 100,000 g for 90 min to pellet EVs, which were washed once by resuspending in PBS followed by ultracentrifugation at 100,000 g for 90 min. The fi nal EV pellet was resuspended in PBS and either used for downstream experiments or stored at Ϫ 80°C. For each experimental condition, isolated EVs were normalized to cell number and expressed relative to the vehicle-treated condition, unless indicated otherwise. Circulating vesicles were isolated from platelet-poor plasma by differential ultracentrifugation, as previously described ( 20 ). For human plasma samples, 900 l each were used. For mouse plasma samples, 100 l each were used. Briefl y, plasma was diluted with an equal volume of PBS and centrifuged at 2,000 g for 30 min at 4°C. The clear supernatant was transferred to new tubes and centrifuged at 12,000 g for 30 min at 4°C. EVs were isolated from the second spin supernatant by ultracentrifugation at 110,000 g for 120 min. Isolated vesicles were characterized by nanoparticle tracking analysis and stored frozen until further analyses.

Nanoparticle tracking analysis
The Nanosight NTA NS300 (Malvern Instruments, UK) equipped with a fast video capture and nanoparticle tracking analysis (NTA) software was used to characterize EV particle size and concentration ( 21 ). The instrument was calibrated according to the manufacturer's protocol. EV samples were diluted in PBS to perform measurements in the linear dynamic range of the instrument (2E+08 to 8E+08 particles/ml). Each sample was sphingolipid synthesis and important structural components of eukaryotic cell membranes ( 6 ). They also play important roles in signaling. In the context of obesityassociated disorders, ceramides are known to accumulate in adipose tissues, and the inhibition of ceramide accumulation ameliorates diabetes and atherosclerosis in lipotoxic disorders ( 6,7 ). Recent studies have implicated hepatic C16:0 ceramide accumulation in the pathogenesis of insulin resistance and NASH ( 8,9 ). However, the exact pathways by which C16:0 ceramides mediate liver injury and infl ammation remain incompletely defi ned.
Extracellular vesicles (EVs) are cell-derived membranedefi ned circulating nano-particles that are shed both basally and under stress conditions ( 10 ). EVs are heterogeneous and composed of particles shed by different mechanisms and of variable sizes. Depending on these characteristics, they can be further classifi ed into exosomes, microparticles, oncosomes, etc. Until advancements permit distinction between these particle types, the allencompassing term, EV, is most appropriate. EVs are a route for cell-derived cargo to be shed from cells, and a mechanism for delivery of specifi c cargoes from donor cells to recipient cells, thus acting as carriers of a stress-stimulated message. Recently, studies have demonstrated an increase in circulating EV release under in vitro lipotoxic conditions, circulating EVs in mouse models of NASH and ischemia/reperfusion injury, and in patients with chronic hepatitis C and NASH (11)(12)(13). Furthermore, it has been reported that ceramides are needed for the formation of EVs via the multivesicular body (MVB) endosomal traffi cking pathway ( 14 ). However, whether PA-induced ceramide biosynthesis drives an EV release response is not known.
Hepatocytes are enriched in the endoplasmic reticulum (ER), a membrane-bound organelle that serves as the subcellular site for lipid and sterol synthesis, as well as the folding factory for secreted proteins ( 15 ). Perturbations in ER function result in ER stress. ER stress is observed in NAFLD and PA-induced lipotoxicity ( 15,16 ). Indeed, PA can directly activate all three ER stress sensors: i ) inositol requiring enzyme 1 ␣ (IRE1 ␣ ); ii ) activating transcription factor 6 ␣ (ATF6 ␣ ); and iii ) protein kinase-like ER kinase (PERK). Because PA-driven de novo ceramide synthesis occurs at the ER, we asked whether PA-induced lipotoxic ER stress would lead to an EV response, and whether any of the three ER stress sensors would mediate PA-induced EV release.
Although a full spectrum of infl ammatory cells exists in the injured liver, macrophages have received the most attention in NASH, as they are recruited to the liver during lipotoxicity ( 17 ). Furthermore, inhibition of macrophage activation, recruitment, or accumulation in the liver ameliorates steatohepatitis ( 16 ). Herein, we report that PA-induced ER stress leads to EV release. Furthermore, ceramides are enriched in PA-induced EVs, and this phenomenon occurs in an IRE1 ␣ -dependent manner. The ceramide-enriched vesicles also contain sphingosine-1-phosphate (S1P), a known infl ammatory mediator and activator of macrophages. We speculate that inhibition of Quantitative real-time PCR After assay medium was removed, cells were washed two times with PBS, collected in 1 ml Trizol, and the aqueous phase, after addition of 200 l chloroform and centrifugation, was applied to the G-eliminator column (RNeasy Plus; Qiagen) and RNA was extracted from the fl ow-through according to the manufacturer's instructions. Quantity and quality of RNA was measured spectrophotometrically using a NanoDrop ND1000 (Thermo Scientifi c), and RNA was reverse transcribed into cDNA by the iScript cDNA synthesis kit (Bio-Rad). Quantitative real-time PCR reactions were run on the LightCycler 480 (Roche), using the LightCycler 480 SYBR Green 1 Master Mix (Roche) and previously published primers ( 16 ).

Ceramide treatment
WT or IRE1 ␣ KO IMH cells were cultured to 90% confl uency on 150 mm dishes and treated with C16 ceramide, as described ( 24 ). After washing with PBS, the medium was changed to assay medium supplemented with 10 M C2 ceramide (Enzo, BML-SL100) or 10 M C16 ceramide (Cayman, item number 10681), and then medium was recovered after 16 h for measuring EVs. Cells were cultured for 2-4 h under high-dose C16 ceramide (50 M or 100 M) conditions and the medium recovered for isolation of EVs.

CRISPR/Cas9 gene editing
The lentiCRISPRv1 and lentiCRISPRv2 plasmids were a gift from Dr. Feng Zhang and obtained from Addgene ( 25 ). The online tool (http://crispr.mit.edu) was used to design the single guide RNA spanning the fi rst exon of human XBP-1 (sgRNA 5 ′ -CACC-GGCGCTGTCGCTTGCGCGCC-3 ′ ), which is conserved between total and spliced XBP-1. Annealed oligonucleotides of sgRNA were used to generate a lentivirus following subcloning into the lentiCRISPRv1 plasmid (Addgene), and transfected into HEK 293T cells along with packaging plasmids, pCMV-VSV-G and pCMV-dR8.2 dvpr (Addgene), using Lipofectamine LTX reagent (Invitrogen). Virus was harvested 48 h after transfection and passed through at 0.45 M pore cellulose acetate fi lter (Millipore). Huh7 cells were infected with XBP-1-targeting lentivirus in the presence of 8 g/ml Polybrene. Successfully infected cells were selected under 2 g/ml puromycin (Invitrogen) selection perfused through the sample chamber at a constant rate of 25 l/min using a syringe pump. The light scatter and Brownian motion of each sample of nanoparticles was recorded at least three times, 30 s each at constant room temperature (22.5°C); particle tracks were analyzed by NanoSight software to measure the concentration of the particles (particles per milliliter) and size (in nanometers).

Lipidomics
Ceramides and nonesterifi ed fatty acids were measured using mass spectrometry at the Mayo Clinic Metabolomics Core Laboratory. Briefl y, ceramides were extracted from EVs or cell pellets suspended in 1× PBS after the addition of internal standards and sonication. The extracts were measured against a standard curve on the Thermo TSQ Quantum Ultra mass spectrometer (Thermo Scientifi c, West Palm Beach, FL) coupled with a Waters Acquity UPLC system (Waters, Milford, MA), as previously described ( 22 ). EVs isolated from equal numbers of cells treated with vehicle or PA were used to quantify changes in EV ceramides. Cell pellet ceramides were normalized to protein content. Ceramides in plasma EVs were measured in EVs isolated from equal volumes of plasma across experimental groups.

Electron microscopy
Whole mount preparations of isolated EVs were prepared for electron microscopy according to established methods ( 20 ). Briefl y, EVs suspended in PBS were fi xed with 2% paraformaldehyde, applied to Formvar carbon-coated EM grid, fi xed with 1% glutaraldehyde, contrasted with uranyl-oxalate solution, embedded in a mixture of uranyl acetate and methylcellulose, air dried, and observed under the JEOL 1400 electron microscope (JEOL USA, Peabody, MA).

Immunofl uorescence
Cells were cultured on cover slips in 6-well plates. After removing medium, cells were washed three times with PBS and fi xed (0.1 M PIPES, 1.0 mM EGTA, 3.0 mM MgSO 4 , and 2.5% formaldehyde). After washing with PBS, cells were permeabilized by Triton X-100. Blocking buffer (5% goat serum, 5% glycerol, and 0.04% sodium azide) was applied to fi xed and permeabilized cells for 1 h followed by primary antibody overnight at 4°C. After washing with PBS, cells were incubated with secondary antibody for 1 h, washed with PBS, and mounted with 20 l Prolong Anti-Fade (Life Technologies) on a clean glass slide. In the case of CD63 (Santa Cruz Biotechnologies), we used poly-L-lysine-coated cover slips (Sigma) for growing cells. Fluorescence was observed with confocal microscopy (LSM 780 Zeiss). graded in a blinded manner by an experienced hepatopathologist ( 28 ). For lipidomic analyses, 20-25 mg frozen liver samples were powdered in liquid nitrogen, weighed, and homogenized in PBS to make a homogenate (10 l/mg tissue). Fifty microliters of homogenates (5 mg of liver tissue) were used for sphingolipid and ceramide analysis.

Statistical and data analyses
Data represent mean ± SEM from three or more experiments unless indicated otherwise. Student's t -test and ANOVA were used for statistical analyses in GraphPad Prism version 6.00 for Windows (GraphPad Software, LaJolla, CA; www.graphpad.com). P < 0.05 was considered signifi cant. JMP software (SAS Institute) was used for regression analyses.

PA induces the release of EVs
PA, a toxic saturated free fatty acid, induces hepatocellular apoptosis in a concentration-and time-dependent manner ( 4,29 ). In order to measure cellular EV release prior to the onset of apoptosis, we established the concentration and duration for treating IMH and Huh7 cell lines with PA and Tg, a known inducer of ER stress signaling ( 15 ), such that there was no increase in cell death (supplementary Fig. 1A, B). Next, we measured EV release under these conditions. For each experimental condition, isolated EVs were normalized to cell number and expressed relative to the vehicle-treated condition. Following treatment of IMH cells with Tg or PA, a 2.9fold increase in EV release in Tg-treated cells and a 7.7fold increase in PA-treated cells were observed ( Fig. 1A ). We did not observe a signifi cant EV response in cells treated with the nontoxic monounsaturated fatty acid, OA ( Fig. 1A ); therefore, we further analyzed Tg-induced and PA-induced EVs. EVs released under Tg or PA were similar in size to vehicle-treated cells, 117-124 nm (particle size, mode, Fig. 1B ). Thus, the size distribution was not altered, although the magnitude of EVs released was signifi cantly greater in cells treated with Tg or PA ( Fig. 1C ). We confi rmed these fi ndings in the human hepatoma cell line, Huh7 ( Fig. 1D-F ). Electron microscopy of IMH cells confi rmed the characteristic morphology of EVs released under all three conditions ( Fig. 1G ). These particles expressed characteristic extracellular microparticle markers, Tsg101 and CD63 ( Fig. 1H ) ( 10 ). Furthermore, when cultured on poly-L-lysine-coated coverslips, thereby increasing adhesion between EVs and the coverslip, we could detect extracellular CD63-expressing particles (supplementary Fig. 1C). A signifi cant increase in PA-induced EV release was also observed in primary mouse hepatocytes ( Fig. 2A ). All together, these data demonstrate a signifi cant increase in EV release from mouse and human hepatocyte cell lines and primary mouse hepatocytes treated with either Tg or PA. The consistency of the response between Tg and PA suggested that an ER stress sensor was promoting the stress-induced EV release. pressure. XBP-1 KO was confi rmed by Western blotting of nuclear extracts from cells treated with tunicamycin to induce ER stress.

Chemotaxis assays
Modifi ed Boyden chambers (Neuro Probe, BW200S) were used to perform chemotaxis assays using 5 m pore size polycarbonate track-etch membranes (Neuro Probe, PFA5), which were coated with 0.1% gelatin (Sigma, G1393) for 30 min and then air-dried. BMDM s were serum starved in 2% exosome-free FBS containing RPMI for 1 h and then pretreated with the appropriate treatment conditions for 1 h. Cells were detached in treatment conditions, counted, and diluted to a concentration of 250,000 cells/ml. Chambers were assembled with chemo-attractants on the bottom in 0.2 ml volume and pretreated cells were placed on top of the membrane in 0.2 ml volume. EVs isolated from equal numbers of vehicle-or PA-treated cells were used as chemoattractants. PF543 (5 M; Selleckchem) and ABC294640 (1 M; Activebiochem) were employed to inhibit sphingosine kinase (SphK)1 and SphK2, respectively. W146 (1 M; Cayman Chemical) or FTY720 (0.25 M; Cayman Chemical) were used to antagonize S1P receptors. Cells were allowed to migrate for 4 h, fi xed in 10% neutral buffered formalin for 20 min at room temperature, rinsed once in PBS and once in water, and mounted in Prolong Gold antifade reagent with DAPI (Life Technologies). Migrated cells were counted on a confocal microscope ((LSM 780 Zeiss) using a UV laser.

Mouse studies
All animal studies and use were approved by the institutional care and animal use committee of the Mayo Clinic and were conducted in accordance with the public health policy on the humane use and care of laboratory animals. Mice were housed in standard pathogen-free facilities with 12 h day-night circadian cycles and unrestricted access to food and water. C56Bl/6J mice were purchased from Jackson Laboratory (Bar Harbor, ME). Twelve-week-old mice were fed either chow or a diet high in saturated fat, fructose, and cholesterol (FFC) for 24 weeks, as previously described ( 26 ). This diet recapitulates the pathologic features of human NASH. Upon completion of the feeding study, blood was collected by cardiac puncture, and platelet-poor plasma was isolated by centrifugation at 1,200 g for 20 min at room temperature, followed by 13,000 g at 4°C for 2 min. Plasma was stored at Ϫ 20°C until further analyses.

Human studies
This study was approved by the Institutional Review Board of the Mayo Clinic and all subjects provided written informed consent. Available archived liver samples from a previously published study were utilized ( 27 ). Subjects who had undergone liver biopsy during bariatric surgery for medically complicated obesity with body mass index of >30 kg/m 2 [at Mayo Clinic, Rochester, MN, with histologic categorization of the biopsies according to established NASH criteria by an experienced hepatopathologist into three groups] were included ( 28 ). The three groups were obese normal (normal liver biopsies, n = 4; plasma, n = 11), simple steatosis (liver biopsies, n = 4; plasma, n = 16), and NASH with early fi brosis (F0-1) (liver biopsies, n = 6; plasma, n = 16). Subjects with secondary causes of steatohepatitis and other chronic liver diseases were excluded [drugs, prior gastric surgery for obesity, excessive alcohol consumption, viral hepatitis (B, C), cholestatic liver disease, hemochromatosis, Wilson disease, druginduced liver disease, and ␣ -1-antitrypsin defi ciency]. NAFLD activity score (NAS) was calculated as described, using the sum of steatosis, lobular infl ammation, and hepatocyte ballooning phosphorylation of eIF2 ␣ and transcriptional upregulation of CHOP in hepatocytes treated with PA (supplementary Fig. 2A-C). To ascertain which of the three ER stress

IRE1 ␣ mediates the release of EVs
We confi rmed activation of IRE1 ␣ signaling via the splicing of XBP-1 and activation of PERK signaling via * P < 0.05 for PA-treated cells compared with vehicle. E: XBP-1 was knocked down in IMH cells using siRNA (siXBP1) and a nontargeting siRNA was used as a control (siCont). EVs were isolated using a commercially available polymer-based reagent. * P < 0.05 for PA-treated si-Cont cells compared with vehicle. ** P < 0.05 for PA-treated siCont cells compared with siXBP1. F: EVs from XBP-1 deleted Huh7 (XBP1 Ϫ / Ϫ ) cells using the CRISPR/Cas9 genome editing technology, treated with 400 M PA or vehicle for 16 h, were isolated using a commercially available polymer-based reagent. * P < 0.05 for PA-treated WT cells compared with vehicle. ** P < 0.05 for PA-treated WT cells compared with XBP1 Ϫ / Ϫ . sensors might be contributing to the enhanced EV release, we employed a series of genetically defi cient cell lines. Furthermore, as Tg is not a pathophysiologically relevant ER stress-inducing agent and due to the pathophysiologic importance of PA-induced lipotoxicity in the pathogenesis of NASH, we focused our subsequent experiments on PA, due to its obvious disease relevance. Initially, we isolated primary mouse hepatocytes from WT and IRE1 ␣ hepatocyte-specifi c KO mice, treated them with PA, and measured the EV response. The PA-induced EV response was signifi cantly reduced in IRE1 ␣ -KO hepatocytes ( Fig. 2A ). The IRE1 ␣ -dependent release of PA-induced EV release in both IRE1 ␣ WT and KO cells ( Fig. 3E ). Lastly, treating cells with a short chain C2 ceramide did not lead to an EV response, demonstrating specifi city for physiologically relevant ceramides in this process ( Fig. 3F ). Thus, IRE1 ␣dependent accumulation of C16:0 ceramide leads to the release of C16:0 ceramide-enriched EVs in PA-treated hepatocytes.

IRE1 ␣ regulates SPT1
To elucidate the mechanism by which PA induces an IRE1 ␣ -dependent EV response, we examined the regulation of SPT1. SPT1 is one of the three subunits that are required for the formation of the functional heterodimeric enzyme. SPT1 mRNA expression was induced by PA treatment in IRE1 ␣ WT cells. This induction of SPT1 mRNA did not occur in IRE1 ␣ KO cells ( Fig. 3G ). To determine whether SPT1 mediated PA-induced EV release, we used a siRNA approach; validation of the SPT1 knockdown was confi rmed by immunoblot analysis ( Fig. 3H ). There was a signifi cant reduction in PA-induced EV release in cells defi cient in SPT1 ( Fig. 3H ). As anticipated, exogenous application of C16:0 ceramide to cells silenced in SPT1 led to a restoration of EV secretion ( Fig. 3H ), similar to the data in IRE1 ␣ KO cells. Thus, PA-induced ceramide biosynthesis occurs via IRE1 ␣ induction of SPT1 expression.

PA-induced EVs are chemotactic to macrophages
To determine a biologic role for lipotoxic EVs, we focused on macrophage chemotaxis, as macrophages recruited to the liver are known to be pro-infl ammatory in NASH ( 17 ). We treated primary BMDM s with EVs derived from PA-treated or vehicle-treated cells. PA-derived EVs induced signifi cant chemotaxis in BMDM s ( Fig. 4A ). To elucidate a mechanism for lipotoxic C16:0 ceramideenriched EV-induced macrophage chemotaxis, we asked whether S1P signaling might mediate macrophage migration. S1P is formed by the phosphorylation of C16:0 ceramide-derived sphingosine by SphK1 and SphK2 ( 33 ). Newly formed S1P is released by cells into the extracellular milieu where it binds cell membrane receptors and stimulates downstream signaling, the classical inside-out autocrine signaling ( 33 ). Therefore, we reasoned that if ceramide transfer to macrophages was providing a substrate for S1P generation, inhibitors of both SphKs and S1P receptors (S1PRs) would inhibit macrophage migration. We fi rst used pharmacologic inhibitors of SphK1 and SphK2. Both inhibitors signifi cantly inhibited macrophage migration toward PA-stimulated lipotoxic EVs ( Fig. 4B ), suggesting that the synthesis of S1P by macrophages was necessary for lipotoxic EV-induced macrophage migration. Next, we used S1PR inhibitors to assess their effect on macrophage migration. We employed FTY720, which binds S1P receptors 1, 3, 4, and 5 (though functionally antagonizing S1P 1 ), and W146, a specifi c antagonist of S1P 1 receptor ( 34 ). Both inhibitors signifi cantly reduced PAstimulated EV-induced migration of both Raw264.7 cells ( Fig. 4C ) and BMDM s ( Fig. 4D ). Next, to confi rm that S1P 1 mediated macrophage migration, we employed RNA silencing. BMDM s defi cient in S1P 1 demonstrated a was further tested in the IMH cell lines. Consistent with the data from the primary mouse hepatocytes, the deletion of IRE1 ␣ led to a signifi cant reduction in the PAinduced EV response ( Fig. 2B ). To determine whether this phenomenon was a consequence of the general unfolded protein response or specifi c to IRE1 ␣ , we treated cell lines genetically deleted in ATF6 ␣ or cell lines mutated at serine 51 (to alanine), rendering a phosphorylation-resistant eIF2 ␣ (thus inactive) with PA ( 30,31 ). Both, ATF6 ␣ WT and KO cells had an equivalent EV response to PA ( Fig.  2C ). Similarly, interrogation of the PERK signaling pathway was accomplished by employing eIF2 ␣ WT and phosphorylation-resistant cell lines, both of which had comparable induction of an EV response to PA ( Fig. 2D ). Thus, these experiments confi rmed that PA-induced EV release specifi cally involves the IRE1 ␣ branch of the ER stress response.
EV release requires the canonical IRE1 ␣ target, XBP-1 IRE1 ␣ activates its canonical target, XBP-1, a potent transcription factor that mediates its signaling function ( 15 ). Thus, we next examined the need for XBP-1 signaling in the IRE1 ␣ dependence of PA-induced EV release. We approached this by generating XBP-1 deleted cell lines using RNA interference or the CRISPR/Cas9 genome editing technology ( 25 ). In IMH cells with XBP-1 knockdown using siRNA, we observed an attenuation of PA-induced EV release ( Fig. 2E ). Next, the human Huh7 cells were deleted in XBP-1 (XBP1 Ϫ / Ϫ ) and treated with PA. The PAinduced EV release was signifi cantly reduced in the XBP1 Ϫ / Ϫ Huh7 cell line ( Fig. 2F ). All together, these studies suggest that XBP-1 mediates PA-induced EV release following IRE1 ␣ activation.

EVs are enriched in C16:0 ceramide
Given the role for ceramides in EV generation, we next measured EV ceramide and sphingolipid content by tandem mass spectrometry ( Fig. 3A ). There was a signifi cant enrichment in C16:0 ceramide in IRE1 ␣ WT IMHs treated with PA ( Fig. 3A ). Consistent with a lack of EV response in cells lacking IRE1 ␣ , an increase in EV C16:0 ceramide was not observed in IRE1 ␣ KO IMHs. An increase in whole cell C16:0 ceramide in PA-treated IRE1 ␣ WT IMH cells was also observed, with lack of an increase in IRE1 ␣ KO cells ( Fig. 3B ). The formation of 3-ketosphinganine from PA and serine, the rate-limiting step in de novo ceramide synthesis, is catalyzed by serine palmitoyltransferase (SPT) ( 32 ). Therefore, we hypothesized that inhibition of this enzyme would mitigate an EV response. Cells treated with myriocin, a pharmacologic inhibitor of SPT, displayed reduced PA-mediated EV release ( Fig. 3C ). Conversely, we treated cells with C16:0 ceramide to determine whether exogenous loading of C16:0 ceramide would lead to EV release. We predicted that C16:0 ceramide loading would restore an EV release response in IRE1 ␣ KO cells by bypassing the need for ceramide biosynthesis. Indeed, in both WT and IRE1 ␣ KO cells, C16:0 ceramide treatment led to a signifi cant increase in EV release ( Fig. 3D ). The response to C16:0 ceramide was concentration dependent The effi ciency of silencing was analyzed by SPT1 Western blotting (inset). EVs were isolated from PA-or vehicle-treated cells using a commercially available polymer, and from C16 ceramide-treated cells (100 M, 4 h) with ultracentrifugation. * P < 0.05 for PA-treated cells; ns, not signifi cant. content of circulating EVs from both chow-fed and FFCfed mice. C16:0 ceramide was signifi cantly enriched in mouse plasma EVs in FFC-fed mice ( Fig. 5B ). Corresponding liver samples demonstrated a signifi cant increase in C16:0 ceramide in FFC-fed livers ( Fig. 5D ). Though the S1P content was increased in FFC-fed circulating EVs and mouse liver samples, this did not reach statistical significance ( Fig. 5C, E ). To expand our observations on the ceramide and S1P content of circulating mouse EVs in NASH to human subjects, we analyzed plasma EVs from obese normal, simple steatosis, and NASH with early fi brosis. Macrophage recruitment to the liver is an early event in NASH pathogenesis, and as we are interested in understanding these early "liver homing" signals in disease pathogenesis, we included only NASH with early fi brosis signifi cant attenuation in lipotoxic EV-induced migration ( Fig. 4E ). These data suggest a potential pro-infl ammatory role for PA-induced EVs by promoting macrophage recruitment to the liver via S1P 1 receptor-mediated macrophage chemotaxis.

Ceramides are enriched in circulating EVs derived from murine and human NASH
Having demonstrated an EV response under PA-induced lipotoxic conditions with a concomitant ER stress response, we next asked whether circulating EVs were increased in a diet-induced model of murine NASH. Mice were fed either chow or a diet high in FFC for 24 weeks. Circulating EVs were signifi cantly increased in FFC-fed mice ( Fig. 5A ). We measured the C16:0 ceramide and S1P The inverted gel image shows S1P 1 expression in BMDM s treated with nontargeting siRNA (siCtrl) or S1P 1 (siS1P 1 ). * P < 0.05 and ** P < 0.001 for the comparisons shown.
conditions. The principal fi ndings of this study are: i ) PAtreated hepatocytes release EVs in an IRE1 ␣ -dependent manner; ii ) lipotoxic EVs are enriched in C16:0 ceramide, and IRE1 ␣ -regulated de novo ceramide biosynthesis is necessary for PA-stimulated EV release; iii ) lipotoxic EVs stimulate macrophage chemotaxis via S1P generation; and iv ) ceramide-enriched circulating EVs are increased in mouse and human NASH. These data provide a mechanistic link between lipotoxic ER stress and disease pathogenesis, suggesting that the observed IRE1 ␣ activation in NASH livers may promote secretion of C16:0 ceramideenriched EVs from steatotic hepatocytes, thus likely promoting macrophage recruitment to the liver.
PA induces ER stress and activates all three of the unfolded protein response sensors ( 16,35 ). We have extended these observations by demonstrating that PAmediated ER stress stimulates EV release from hepatocytes. The PA-stimulated EV release was specifi c for the ER stress sensor, IRE1 ␣ . In cells lacking ATF6 ␣ or cells resistant to PERK-induced eIF2 ␣ phosphorylation due to a phosphorylation resistant mutation at serine 51 of eIF2 ␣ , the PA-induced EV response was intact. Next, we defi ned subjects in this analysis. The NAS was signifi cantly higher in subjects with NASH compared with simple steatosis ( Fig. 6A ); obese normal subjects has normal liver histology. Consistent with the mouse fi ndings, circulating EVs were signifi cantly elevated in human NASH, and were signifi cantly enriched in C16:0 ceramide and S1P ( Fig.  6B-D ). Interestingly, plasma C16:0 ceramide and S1P exhibited a reciprocal relationship ( Fig. 6E, F ; supplementary Fig. 3) with EV C16:0 ceramide and S1P, such that in NASH subjects there was depletion in circulating plasma levels of these bio-active lipids, though an increase in their enrichment in EVs. Thus, there may be a selective partitioning of hepatocyte-derived sphingolipids into EVs in NASH. These fi ndings would need substantiation in larger cohorts. Furthermore, we found an increase in C16:0 ceramide and not S1P in corresponding liver biopsy samples ( Fig. 6G, H ).

DISCUSSION
In this study, we have elucidated a mechanistic basis for the release of pro-infl ammatory EVs under hepatic lipotoxic and S1P (C) were measured in EVs isolated from 100 l of plasma from the mice described in (A). C16:0 ceramide (D) and S1P (E) were measured in liver samples from the mice described in (A). * P < 0.05 for the comparisons shown.
ing pathways could contribute to the formation and release of PA-induced vesicles. Furthermore, ceramide generation via acid and neutral sphingomyelinases is known to mediate vesicle release, especially vesicles derived from the MVB ( 14,36 ), and play a role in the sorting of proteo-lipid cargo to vesicles. To our knowledge, this is the fi rst demonstration that de novo ceramide biosynthesis via SPT is necessary for the biogenesis of EVs. Whether these EVs are derived from membrane shedding or from the mechanism for this observation by demonstrating that XBP-1 mediates this EV response by transcriptionally upregulating its target, SPT, the rate-limiting enzyme of de novo ceramide biosynthesis ( 33 ).
As endosomes mature from early to late, thus forming MVBs, they interact intimately with the ER and ceramides synthesized in the ER are rapidly transported and transferred to the Golgi complex and eventually to the plasma membrane; thus, either or both of these cellular traffi ck- Fig. 6. C16:0 ceramide and S1P in human NASH. A: NAS for subjects with simple steatosis (n = 4) and early NASH F0-1 (n = 6). B: Circulating EVs were measured in 900 l of plasma obtained from obese normal (n = 11), simple steatosis (n = 16), and early NASH F0-1 (n = 16). Particle concentration is expressed relative to the averaged obese normal. C16:0 ceramide (C) and S1P (D) were measured in EVs isolated from 900 l of plasma from the subjects described in (B). C16:0 ceramide (E) and S1P (F) were measured in 25 l of plasma samples from the subjects described in (A), in whom matching liver biopsy samples were available: obese normal (n = 4), simple steatosis (n = 4), and early NASH F0-1 (n = 6). C16:0 ceramide (G) and S1P (H) were measured in available liver biopsy samples from the subjects described in (E). * P < 0.05 by ANOVA for comparisons involving all three groups; ** P < 0.05 for the comparisons shown across two experimental conditions. MVB fusion to the plasma membrane is a question we will address in future studies.
EVs shed from cells can serve many roles, including the activation of target cells via the cargo they carry. They can activate target cells such as macrophages ( 37 ), enhance angiogenesis ( 11 ), promote tumor metastases ( 37 ), and transfer transcriptionally competent and regulatory RNA molecules to recipient cells ( 10 ). Herein, we demonstrate a chemotactic role for PA-induced EVs in our in vitro system. Our data implicate the activation of the S1P 1 receptor on macrophages by ceramide-derived S1P on lipotoxic EVs as the chemotactic stimulus. Though pharmacologic inhibitors can be nonspecifi c, FTY720 is known to functionally antagonize S1P 1 , and W146 is highly selective for S1P 1 at the concentrations used ( 34 ). Additionally, using RNA interference, we show that macrophage migration is mediated, in part, by S1P 1 , though other receptors may contribute. The S1P and SIP 1 signaling axis has been implicated in lymphocyte traffi cking in sterile infl ammatory conditions ( 33 ); here we advance this observation to potentially mediating macrophage recruitment to the steatotic liver .
Our fi ndings of plasma EV and hepatic S1P and C16:0 ceramide enrichment in both mouse and human NASH are an essential biologic correlate to our in vitro fi ndings. Recent mouse studies have demonstrated a critical pathogenic role for hepatic C16:0 ceramide in the pathogenesis of diet-induced obesity-related fatty liver. Our data might provide a mechanism by which C16:0 ceramide and its derivative, S1P, are pro-infl ammatory by leading to the recruitment of myeloid cells to the liver, and potentially to other target organs. We acknowledge that a limitation of our human study is its small size. However, others also have independently demonstrated increased hepatic ceramides in models of NAFLD ( 38 ). Interestingly, we noted an increase in C16:0 ceramide and S1P in EVs from subjects with early NASH compared with simple steatosis and obese normal livers, suggesting that these may be an early liver-homing signal, attracting macrophages to the steatotic liver. It is also possible that EVs shed under these conditions perpetuate, rather than initiate, liver infl ammation.
EVs are an emerging biomarker ( 10 ). They have been investigated in numerous human diseases and in mouse models of disease ( 10 ), and an increase in circulating EVs is described in a mouse model of NASH ( 39 ). Our fi ndings further these observations, and suggest that the ceramide and S1P content of EVs might serve as an additional bioactive biomarker in NASH patients.
In conclusion, this is the fi rst report linking PA-induced IRE1 ␣ activation to the release of pro-infl ammatory EVs by lipotoxic hepatocytes. Our data suggest a mechanism for myeloid cell recruitment to the liver and the use of EVs as a bio-active biomarker. Additionally, several potential pathways for therapeutic intervention are identifi ed, e.g., interference with the S1P signaling axis in macrophages.
The authors are grateful to Dr. Gregory Gores for critical review of the manuscript; Dr. Randal Kaufman for the gift of cell lines, mice, and reagents; Dr. Emanuel Strehler for the plasma membrane Ca 2+ ATPase antibody; Mr. Steven F. Bronk for mouse