Lipidomic and transcriptomic analysis of western diet-induced nonalcoholic steatohepatitis (NASH) in female Ldlr -/- mice

Background Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease worldwide, particularly in obese and type 2 diabetic individuals. NAFLD ranges in severity from benign steatosis to nonalcoholic steatohepatitis (NASH); and NASH can progress to cirrhosis, primary hepatocellular carcinoma (HCC) and liver failure. As such, NAFLD has emerged as a major public health concern. Herein, we used a lipidomic and transcriptomic approach to identify lipid markers associated with western diet (WD) induced NASH in female mice. Methods Female mice (low-density lipoprotein receptor null (Ldlr -/-) were fed a reference or WD diet for 38 and 46 weeks. Transcriptomic and lipidomic approaches, coupled with statistical analyses, were used to identify associations between major NASH markers and transcriptomic & lipidomic markers. Results The WD induced all major hallmarks of NASH in female Ldlr -/- mice, including steatosis (SFA, MUFA, MUFA-containing di- and triacylglycerols), inflammation (TNFα), oxidative stress (Ncf2), and fibrosis (Col1A). The WD also increased transcripts associated with membrane remodeling (LpCat), apoptosis & autophagy (Casp1, CtsS), hedgehog (Taz) & notch signaling (Hey1), epithelial-mesenchymal transition (S1004A) and cancer (Gpc3). WD feeding, however, suppressed the expression of the hedgehog inhibitory protein (Hhip), and enzymes involved in triglyceride catabolism (Tgh/Ces3, Ces1g), as well as the hepatic abundance of C18-22 PUFA-containing phosphoglycerolipids (GpCho, GpEtn, GpSer, GpIns). WD feeding also increased hepatic cyclooxygenase (Cox1 & 2) expression and pro-inflammatory ω6 PUFA-derived oxylipins (PGE2), as well as lipid markers of oxidative stress (8-iso-PGF2α). The WD suppressed the hepatic abundance of reparative oxylipins (19, 20-DiHDPA) as well as the expression of enzymes involved in fatty epoxide metabolism (Cyp2C, Ephx). Conclusion WD-induced NASH in female Ldlr -/- mice was characterized by a massive increase in hepatic neutral and membrane lipids containing SFA and MUFA and a loss of C18-22 PUFA-containing membrane lipids. Moreover, the WD increased hepatic pro-inflammatory oxylipins and suppressed the hepatic abundance of reparative oxylipins. Such global changes in the type and abundance of hepatic lipids likely contributes to tissue remodeling and NASH severity.


Introduction
Nonalcoholic fatty liver disease (NAFLD) is defined as excessive neutral lipid (triglycerides and cholesterol esters) deposition in the liver, i.e., hepatosteatosis [1,2]. The top 4 risk factors for NAFLD are obesity, dyslipidemia, type 2 diabetes mellitus (T2DM) and metabolic syndrome (MetS) [3,4]. NAFLD is a continuum of diseases ranging from benign steatosis to primary hepatocellular cancer (HCC). It is strongly associated with obesity [5,6]; and it is the most common chronic fatty liver disease worldwide [7]. Approximately 30% of the US population is estimated to have some form of chronic fatty liver disease [8].
Ten to 30% of NAFLD patients develop nonalcoholic steatohepatitis (NASH), the progressive form of the disease. NASH patients have hepatic steatosis plus hepatic and systemic inflammation, oxidative stress and liver injury [9]. Excessive liver damage resulting from NASH activates tissue remodeling mechanisms involving deposition of extracellular matrix components (ECM), leading to fibrosis. There is a high prevalence of NASH (�60%) in the T2DM population [10]; and NASH is a risk factor for cardiovascular disease [11][12][13]. NASH patients also have higher mortality rates than NAFLD patients; and both have higher mortality rates than the general population. Twenty to 30% of NASH patients progress to cirrhosis. Over a 10 year period, cirrhosis and liver related deaths occur in 20% and 12% of NASH patients, respectively [9]. By the year 2020, cirrhosis resulting from NASH is projected to be the leading cause of liver transplantation in the United States [14]. Given the increasing prevalence of NASH and its adverse clinical outcomes, NASH is considered a major public health problem [15].
While NAFLD appears to be more prevalent in men than in women [16,17], others report that lean patients with NAFLD are more frequently female [18]. In fact, data from nearly 700 patients from the NASH Clinical Research Network showed that biopsy proven NASH patients were more likely female by a ratio of 2:1, female to male [19]. Despite the occurrence of NAFLD in human males and females, preclinical NAFLD models typically use male rodents [20]. By way of illustrating this point, we examined 100 recent preclinical NAFLD publications appearing in PubMed using the query terms (NAFLD, mice). Eighty nine percent of publications used male rodents; typically mice. One explanation for this gender bias is that female mice on the C57BL/6J background, a common mouse model for diet-induced obesity, display some resistance to high fat diet-induced obesity [21,22]. However, multiple factors contribute AM) prior to euthanasia by CO 2 administration and exsanguination; blood and liver were collected as previously described [31].

Plasma and hepatic measures
Plasma glucose, triglycerides, total and free cholesterol were measured using kits obtained from Wako. Plasma non-esterified fatty acids (NEFA) and β-hydroxybutyrate (β-HB) were measured using kits from Sigma-Aldrich and Zen-Bio, respectively. Plasma aspartate amino transferase (AST) and alanine amino transferase (ALT) were measured using kits from Thermo-Fischer (Waltham, MA) [29,31].

Liver histology
Liver (~100 mg) was fixed in buffered-formalin, paraffin embedded, sliced, and stained with hematoxylin-eosin and sirius red (Nationwide Histology, Veradale, WA) [31]. Each slide contained 2-4 slices/liver. Histological analysis and scoring was performed by a board certified veterinary pathologist using a scoring system established for mouse models of NAFLD/NASH [35].
Steatosis was analyzed as macrovesicular (displacing nucleus) and microvesicular vacuolation and severity was scored as the percentage of affected surface area at 100x using the following scale: 0 (0%), 1 (>5% but <33%), 2 (>33% but <66%), 3 (>66%). Steatosis was objectively quantified as percent surface area occupied by lipid vacuoles by image analysis using Image J (NIH; imagej.nih.gov/ij/index.html). Images were taken at 100x from the subjectively least and most severely affected areas. Measurements are presented as the average over two analyzed images per liver normalized to the average percent surface area occupied by lipid vacuoles in liver sections from control mice.
Fibrosis was scored at 40x as absent, mild, moderate or severe. Distribution of fibrosis was recorded as perisinusoidal (s), periportal (p), pericentral (c), and bridging (b). Fibrosis was objectively quantified as percent surface area occupied by sirius red-stained collagen by image analysis using Image J. Images were taken at 100x from the subjectively least and most severely affected areas. Measurements are presented as the average over two analyzed images per liver normalized to the average percent surface area occupied by collagen in liver sections from control mice. Representative images of hepatic micro-and macro-steatosis, leukocyte infiltration and fibrosis are shown in Supplemental Information (S1 and S2 Figs).

RNA extraction and qRT-PCR
Liver RNA was extracted using Trizol (Life Technologies) and quantified using a Beckman spectrometer at 260 and 280 nm [31,36]. qRT-PCR was performed using the 7900HT fast machine from Applied Bio-systems as previously described [31,36]. S1 Table list all transcripts examined in this study and the corresponding primer pairs used for qRT-PCR. In addition, qRT-PCR arrays were used to profile the expression of genes involved in cell death (Mouse Cell Death RT 2 Profiler PCR Array) according to the manufacturer's protocol (Qiagen, Valencia, CA, USA). Diet affected genes identified by the Qiagen array analysis were further characterized using in-house generated primer pairs and qRTPCR. S1 Table identifies these transcripts with an ( � ). Primers were designed using OligoPerfect Primer Designer (Thermo-Fisher, Waltham, MA). Gene expression analysis in the Qiagen arrays was normalized to Hsp90ab1, while cyclophilin was used as the reference gene in the qRTPCR using "in-house" designed primers (S1 Table). Relative quantification was determined using the delta (Δ) Ct method with cyclophilin as the reference transcript [31]. The ΔCt values were used for all statistical analyses. Diet effects on transcript abundance was graphically presented as Fold Change, Mean + SEM, N = 7-8 per group. The statistical package in MetaboAnalyst 4.0 [http://www.metaboanalyst.ca/MetaboAnalyst/] [37,38] was used for statistical analysis of the transcriptomic data. Accordingly, ANOVA with Tukey's HSD Post-hoc test was used to identify features that were significantly different amongst the 3 groups: RD46, WD35, WD46. A false discovery rate (FDR) value of < 0.05 was the criteria for significance.

Hepatic lipid composition
Total hepatic lipids were extracted, saponified and the fatty acids converted to fatty acid methyl esters (FAMES) as described [31,39]. FAMES were separated and quantified by gas chromatography (GC) [31,39]. GC FAMES standards were purchased from Nu-Chek Prep Inc (Elysian, MN). Hepatic protein content was measured using Pierce BCA Protein Assay Kit with bovine serum albumin (BSA) (Thermo-Fisher, Waltham, MA) as a standard. Levels of hepatic fatty acids were normalized to hepatic protein content.
Sample preparation for un-targeted and targeted lipidomics. Un-targeted lipidomics. All LC-MS/MS analyses described in this report were carried out at the OSU Mass Spectrometry Center (https://mass-spec.science.oregonstate.edu/). The extraction of lipid for lipidomic analysis was adapted from Choi et al [40]. Accordingly, frozen liver (10 mg) was transferred to a 2 ml pre-filled bead rupture tube containing 300 mg of RNAse-and DNAse-free ceramic beads (1.4 mm) (Thomas Scientific) [40,41]. One ml of degassed methylene chloride:isopropanol:methanol (25:10:65, v/v/v; -20˚C) mixture was added to each tube along with a mixture of deuterated lipids (5 μl) (SPLASH Lipidomics Mix, Avanti Polar Lipids) as internal standards. Samples were homogenized using a Precellys 24-bead-based homogenizer for 2 min at 1350 rpm. Samples were incubated at -20˚C for 1 hour and then centrifuged at 13,000 rpm and 4˚C for 10 min [40,41]. Extracts were transferred to glass vials and stored less than 24 h at -80˚C before analysis.
Targeted oxylipidomics and relative quantitation. Oxylipins were extracted from frozen liver (6 mg). Liver was homogenized in 0.5 ml of 10% methanol as described by others [42,43]. A 10 μl sample of the homogenate was transferred to a microfuge tube for protein quantification (Pierce BCA Protein Assay Kit; BSA as standard) [31]. Homogenized livers were incubated at room temperature for 1h followed by centrifugation 13,000 rpm at 4˚C for 10 min.
The supernatant (490 μl) was transferred to a methanol-rinsed 1.5 ml microfuge tube and deuterated internal standards (10 μl) were added to each liver extract. The internal standard solution contained 20 deuterated oxylipins at 5 ng/μl each. Oxylipins were further extracted using Strata-X polymerized solid reverse phase extraction columns (Phenomenex, Torrance, CA) as described [43]. Columns were conditioned with methanol and equilibrated with water. Samples were loaded into the cartridge and washed twice with 500 μl of 10% methanol [43]. Additional methanol washes eluted non-specific hydrophobic chemical species and salts from the sample. Oxylipins were eluted from the column with 100% methanol containing 2% formic acid (1 ml), while lipophilic species that could potentially lead to ion suppression were retained on the column [44]. Lipids were stored in 100% methanol at −80˚C to minimize non-enzymatic oxidation and degradation until analysis. Just before analysis, samples were evaporated under a stream of nitrogen, re-dissolved in 50 μl of acetonitrile, and incubated in wet ice for 1 h. After a short centrifugation, samples (40 μl) were transferred to glass vials with inserts for analysis.
As noted above, all samples were spiked with deuterated internal standards [45] allowing for the monitoring of extraction efficiencies, changes in ionization of different classes of oxylipins and data normalization [46]. Accordingly, we considered a detected peak was valid when its peak height was 2.5-fold greater than the background noise. Peaks that do not meet this requirement were disregarded. In order to pair a metabolite with a particular internal standard, we used previously reported criteria [42]. S2 Table list all deuterated standards used in this study and their analyte assignments.
The TripleTOF instrument was operated in the information-dependent MS/MS acquisition mode under parameters adapted from Kirkwood, et al. [47]. In summary, TOF MS and TOF MS/MS accumulation times were 0.25 and 0.10 s, respectively. The period cycle time was 0.7 s for both TOF MS and TOF MS/MS. The scan range was 70-1700 m/z for TOF MS and 50-1700 m/z for TOF MS/MS. Ion source gas 1 and 2 and curtain gas were set at 50, 45, and 35, respectively. Nitrogen gas was used in all cases. The temperature of the source was set at 550˚C. The ion spray voltage was set at 5.5/-5.5 kV. In order to determine the stability of the instrument during the complete run, as well as biological variance, an equal mixture of all liver samples (n = 24) were injected periodically from the beginning to the end of the batch.
Additionally, in order to determine potential carry over from one to another injection, blank samples were also periodically analyzed. Auto calibrations (AB SCIEX calibration solution) were performed every five samples in order to correct for small mass drifts during the acquisition [47].
Targeted oxy-lipidomics. The same instrumentation and column explained above were used to implement a pseudo multi-reaction monitoring (MRM) method in negative ion ESI mode for the analysis of oxylipins as adapted from Wang et al [43]. Accordingly, the mobile phase consisted of (A) acetonitrile/water/acetic acid (60/40/0.02, v/v) and (B) acetonitrile/isopropyl alcohol (50/50, v/v). Gradient conditions were as follows: 0-4.0 min, 0.1-55% B; 4.0-4.5 min, 55-99% B; 4.5-5.0 min, 99% B [43]. The sample injection volume was 10 μl aliquot and the flow rate was 0.5 ml/min. of each sample was injected onto the column. The column temperature was kept at 40˚C. All samples were kept at 4˚C throughout the analysis. Sixty multi-reaction monitoring (MRM) transitions were achieved by flow injection of pure standards and manual optimization by comparison to literature values. This method contains seven different periods corresponding to elution windows for the different compounds. Inside each window, the quantification of the eluted oxylipins was enhanced, assuring optimal dwell time and sufficient data points per peak [51]. For co-eluting isobars a unique fragment ion was chosen. For example, 8,9-DiHETE (m/z 335!185) and 14,15-DiHETE (m/z 335 ! 111), both eluting at 1.47 minutes in our chromatography conditions, were quantified by selecting unequal fragment ions.
Data processing and statistical analyses for targeted and un-targeted lipidomics. Raw data from targeted oxylipins analyses was imported into MultiQuant software (AB SCIEX) in order to perform the alignment and integration of the peaks (obtaining peak areas). This software allows for the correction of metabolite intensity with the intensity of the internal standards. Data obtained with MultiQuant was imported into MarkerView software (AB SCIEX) for initial data visualization [52]. Student's t-tests were performed with GraphPad Prism Software (GraphPad, La Jolla, CA). Raw data from un-targeted lipidomic analysis was imported into Progenesis QI software (Version 2.3, Nonlinear Dynamics, Waters) in order to perform data normalization, feature detection, peak alignment, and peak integration [52]. A repeated measure analysis of variance (ANOVA) was also performed using Progenesis QI software in order to identify significantly alter features [52]. Similarly, we obtained the q-values based on the p-values in order to adjust for multiple comparisons. The optimized false discovery rate (FDR) approach uses characteristics of the p-value distribution to produce a list of q-values. We considered annotations with a q-value <0.05, which implies a FDR of 5%. The principal component analysis (PCA) for the annotated and significantly affected compounds by the WD was performed in MetaboAnalyst 4.0 [53]. Log transformation and auto-scaling were applied on the data for normality.
Lipid species were confirmed by validated retention times, MS, MS/MS fragmentation, and isotopic distribution using the LipidBlast [54] and Metlin (The Scripps Research Institute) and the Human Metabolome (HMBD) databases as the reference data bases for comparisons [55]. Peak intensities were normalized using the QC pool sample, and the SPLASH Lipidomics Mix (Avanti Polar Lipids, Alabaster, AL).
The statistical package in MetaboAnalyst 4.0 [http://www.metaboanalyst.ca/MetaboAnalyst/] [37,38] was used to log transform the numerical data and prepare heat maps, principal component analyses and pattern analyses of all anthropometric, plasma, histologic, lipidomic and transcriptomic data. ANOVA with Tukey's HSD Post-hoc test was used to identify features that were significantly different amongst the 3 groups: RD46, WD35, WD46. We use a FDRvalue < 0.05 as the criteria for significance. Paired t-tests were also performed in MetaboAnalyst, with annotated compounds in the 3 groups (RD46 vs WD35 and RD46 vs WD46).

Study design and overall statistical analysis
Female Ldlr -/mice at 2 months of age were randomized for weight, divided into 3 groups and fed a reference (RD) or western (WD) diet (Fig 1). RD-fed mice were euthanized 46 wks later (RD46), while WD-fed mice were euthanized after 38 (WD38) and 46 wks (WD46) on the WD. Liver and blood were collected for analysis.
We examined anthropometric, plasma and liver histology, gene expression & lipid features [52]. All quantified data was assembled in an excel spreadsheet and analyzed using the statistical package in MetaboAnalyst [31,37,38]. Principal component analysis (PCA) was used to establish similarities/dissimilarities amongst the 3 groups (Fig 2). Features of mice fed the WD clustered together and were separate from the cluster of the RD-fed mice. As expected, there was overlap in the features of mice fed the WD for 38 and 46 weeks. Separate PCA on only lipid and gene expression markers showed that the two WD groups clustered together and were clearly separate from the RD group (Fig 2 B and 2C). Thus, both lipid and gene expression features contribute to the separation of the RD and WD-fed groups seen in Fig 2A.

Western diet (WD) induced a NASH phenotype in female Ldlr -/mice
Mice fed the RD for 46 wks had a~24% increase in body weight over the course of the study (Table 1), while mice fed the WD for 38 or 46 wks increased body weight by 62 and 68%, respectively. Changes in body weight in the WD fed mice represent a 31 and 36% increase in body weight over that seen in the RD-fed mice. WD fed mice also displayed a significant increase in liver weight (LW), i.e., 5.5-5.8 LW%BW, when represented as a percent (%) of body weight (BW) versus that seen in RD mice (3.8 LW%BW). These features were comparable to changes seen in male Ldlr -/mice fed the WD for 16-32 wks [29][30][31].
Plasma glucose and triglycerides were not increased by WD feeding. In contrast, free and total plasma cholesterol, NEFA, β-hydroxybutyrate, ALT and AST were elevated in mice fed the WD (Table 1). Increased fasting ALT, AST and cholesterol in WD-fed mice reflects hepatic injury and hypercholesterolemia, respectively. Increased fasting NEFA and β-hydroxybutyrate reflects increased fat mobilization from adipose depots and hepatic ketogenesis, respectively. In contrast to our previous studies with male Ldlr -/mice, the female mice did not develop abnormal plasma triglycerides in response to the WD [29][30][31].
Histology of livers from RD and WD-fed mice revealed the major hallmarks of NASH, including micro-and macro-steatosis (hematoxylin/eosin stain) and branching "chicken wire" fibrosis (Sirius red) (Fig 3, S1 and S2 Figs). Quantitation of the histological features described in Materials and Methods revealed heterogeneity in steatosis, inflammation and fibrosis (Fig 3, right panels). Heterogeneity in response to the WD was seen in our previous studies with WD-fed male mice [29][30][31]. This heterogeneity may be linked to variations in the gut microbiome amongst these age-matched mice [56]. Despite this heterogeneity, the majority of WDfed mice displayed a phenotype characteristic of NASH, i.e., hepatic increased steatosis, leukocyte infiltration of the liver (inflammation) and fibrosis.
We next used a statistical approach to identify highly significant features associated with WD-induced NASH. The top 10 highly significant changes in lipids (FDR, �4.4 x 10 −11 ) and gene expression (FDR, � 4.4 x 10 −8 ) are listed in Table 2. WD consumption suppressed hepatic abundance of lipids containing PUFA (ω3 or ω6) and increased lipids containing SFA and MUFA. This change in lipid type has been reported in patients diagnosed with NASH [29][30][31]34]. WD feeding also significantly increased hepatic transcript abundance for enzymes involved in membrane remodeling (LpCat), oxidative stress (Ncf2), inflammation (TNFα), EMT (S1004A), Notch (Hey1) signaling and apoptosis (CtsS), but suppressed hepatic abundance of a transcript encoding an enzyme involved in cholesterol synthesis (Fdps). Suppression  The WD induced the hepatic abundance of mRNAs (�3-fold) encoding two enzymes involved in oxidative stress (Fig 4C), i.e., NADPH-oxidase subunits [Ncf2 (p67phox) and Cybb (cytochrome b245, Gp91 phox, Nox2)]. While Gstα1 mRNA abundance was suppressed (>50%) by WD, Nrf2 mRNA, a major transcriptional regulator of oxidative stress, was not affected by the WD. Oxidative stress is associated with human NAFLD and vitamin E Diet induced NASH in female Ldlr -/mice supplementation has been recommended as one approach to combat NAFLD-associated oxidative stress [57,58] WD-induced fibrosis (Figs 3 and 4D) was associated with increased expression of several fibrosis markers, including (Col1A1, Timp1, Lys Ox, Mmp13 and Tgfβ1). However, not all fibrosis markers were induced by the WD, including smActin and HmgB1 (a transcription factor linked to chronic liver disease and fibrosis [59]. NAFLD is associated with increased apoptosis and autophagy in humans and mice [2,60,61]. Female Ldlr -/mice fed the WD have elevated hepatic mRNA for key proteins linked to apoptosis (CtsB, CtsS, Gadd45, Nol3, Bcl2) and autophagy (Casp1) (Fig 5A). All markers were induced � 2-fold in mice fed the WD. Since apoptosis and autophagy are associated with cellular and tissue remodeling, we examined WD effects on notch & hedgehog signaling and EMT (Fig 5B and 5C) [62,63]. Activation of notch signaling is associated with the �2-fold induction of key transcription factors i.e., Hes1, Hey1 and HeyL. While the mRNAs encoding indian hedgehog (Ihh), a hedgehog ligand, and the transcription factor Taz, a downstream target of hedgehog signaling [64], were induced �2-fold by the WD. The mRNA encoding the hedgehog inhibitory protein (Hhip), however, was suppressed by 50%. Three of the seven EMT markers (Snail1, S1004A, Mpk) were induced in livers of WD-fed mice. Together, these findings indicate that WD feeding induced hepatic mRNAs encoding proteins involved in apoptosis, autophagy, hedgehog & notch signaling and EMT in female Ldlr -/mice. Since NASH has the potential to progress to primary hepatocellular cancer (HCC) in humans [65,66], we asked if female Ldlr -/mice displayed evidence of hepatic cancer. Expression of eight of the eleven cancer markers examined (Cd44, Gpc3, Cd133, Bcl2, Vim, Ki-67, Grip1, Sox4) were affected by the WD [67,68] (Fig 5D). Some cancer markers (Afp, Sox9) were not affected by the WD, while one marker, ID2 (inhibitor of DNA binding) was suppressed by 50% in livers of WD-fed mice. Changes in mouse hepatic ID2, Vim, and S100A4 are consistent with findings reported in human HCC [68,69] and pancreatic cancer [70]. As such, this analysis indicates that female Ldlr -/mice fed the WD for 38 to 46 wks become obese and increase hepatic expression of multiple transcriptomic markers of NASH and HCC.

The WD significantly alters the hepatic lipidome
We [29] and others [71] have examined the hepatic lipidome in an effort to identify potential lipid mediators of NALFD. Herein, we use gas chromatographic (GC) and LC-MS/MS (untargeted and targeted) approaches to identify and quantify hepatic lipids. The GC analysis of free and saponified fatty acids showed that consumption of the WD was associated with a massive increase of hepatic saturated (SFA) and monounsaturated fatty acids (MUFA) (Fig 6),  [29]. As such, in vivo metabolic pathways account for the increase in C 20-22 SFA and MUFA.
To identify possible mechanisms accounting for WD effects on hepatic fatty acyl content, we quantified the abundance of mRNAs encoding transcription factors regulating hepatic fatty acid metabolism (Fig 7A), as well as enzymes involved fatty acid metabolism (Fig 7B and 7C). All mice in this study were fasted overnight in order to obtain fasting blood samples for the quantitation of blood lipids. Some hepatic transcription factors, like SREBP1c, and enzymes involved in de novo lipogenesis (DNL) are induced after refeeding fasted mice [28,73]. However, we examined transcripts in the fasted state only.
Of the genes examined, only SREBP1c (at 46 wks) and PPARγ2 (at 38 & 46 wks) were induced (� 2-fold) by the WD. Others have reported that increased hepatic PPARγ2 expression is associated with diet-induced hepatosteatosis [74]. We examined downstream targets of SREBP1c and found no induction of expression of transcripts encoding enzymes involved in malonyl CoA metabolism (Acc1, Acc2, Mcd) or DNL. In fact, expression of Fasn, a major DNL enzyme and target of SREBP1c, was suppressed (�60%) in mice fed the WD. We previously reported that WD-fed male Ldlr -/mice express low levels of Fasn [29]. Since Fasn mRNA parallels changes in hepatic DNL, increased DNL cannot account for the massive increase in SFA or MUFA (Fig 6) [75,76]. Although the WD has a high sucrose content (~23% total calories, Research Diets), the suppression of Fasn expression by the WD suggests that the massive We also quantified diet effects on mRNAs encoding enzymes involved in fatty acid remodeling, i.e., desaturases (Scd1, Fads1, Fads2) & elongases (Elovl 1-3; Elovl 5-7). Several of these enzymes (Scd1, Fads1, Fads2) are also downstream targets of SREBP1c, PPARα & PPARγ [73]. While both Scd1 and Elovl7 were induced (>2-fold) in livers of WD-fed mice, Fads1 and Elovl3 mRNA were suppressed by 50 and �65%, respectively. Scd1 is the major hepatic desaturase involved in the MUFA synthesis and both Elovl3 and Elovl7 elongate SFA & MUFA [77]. Increased Scd1 and Elovl7 expression likely contribute to the increase in C 20-22 MUFA (Fig 7B).
Fatty acid binding proteins and fatty acyl CoA formation are requisite events associated with fatty acid trafficking and metabolism. While expression of the major hepatic fatty acid binding protein (Fabp1) was unaffected by diet, expression of enzymes involved in acyl CoA formation (Acsl1 & 3) were suppressed by �50% in livers of WD-fed mice (Fig 7C). This change in gene expression may affect the delivery of substrate to downstream pathways involved in neutral and polar lipid synthesis.
A key pathway for fatty acid disposal is mitochondrial fatty acid β-oxidation (FAO). Of the enzymes involved in FAO, i.e., Pdk4, Cpt1A, Slc25a20 and HMG CoA Syn2, only Pdk4 mRNA Diet induced NASH in female Ldlr -/mice was strongly affected by the WD. Hepatic Pdk4 mRNA abundance was suppressed by >75% in mice fed the WD (Fig 7D). Insulin is a strong repressor of Pdk4 [78], while PPAR (α, β, γ) activators induce hepatic Pdk4 [79]. Pdk4 plays a key role in substrate selection by controlling pyruvate dehydrogenase activity, a key step for entry of metabolites into the mitochondrial tricarboxylic acid cycle [78,80]. Attenuated Pdk4 in livers of WD-fed mice may enhance carbohydrate versus fatty acid oxidation. Neutral lipids. Neutral lipids examined in our analysis include diacylglycerols (DG) and triacylglycerols (TG). We annotated 10 DG and the relative abundance of 7 DG was significantly affected by the diet (Fig 8A). Those containing SFA and MUFA, i.e., DG 32:1, DG 34:1, DG 36:1, were significantly increased. Two of the 3 DG containing PUFA (DG 34:2, DG 38:5) were significantly increased, while DG 36:4 was significantly decreased in livers of WD-fed mice. Most of the quantified DG had acyl chains in the sn1 & sn2 positions, while two had acyl chains in the sn1 & sn3 positions; sn1,3 DG are generated as a result of TG catabolism, and specifically involve adipocyte triglyceride lipase (ATGL) [81]. To gain insight into mechanisms mediating changes in neutral lipid composition, we quantified transcripts of enzymes involved in TG synthesis and catabolism (Fig 9). Enzymes involved in DG & TG synthesis include Gly Kin, Gpatm, Mogat1, Lipin1 and Dgat 1 & 2 ( Fig  9A). The expression of these enzymes changed little in response to the WD. Expression levels of 2 of 3 enzymes involved in neutral lipid catabolism (Ces1g and Tgh), however, were reduced by � 50% in livers of WD-fed mice ( Fig 9B). Interestingly, ablation of Tgh protected mice from diet-induced hepatosteatosis [82], while Ces1g deficiency promoted diet-induced fatty liver [83]. TG catabolism is required for assembly and lipidation of ApoB for VLDL assembly and secretion [84,85]. The lack of fasting hypertriglyceridemia in WD-fed Ldlr -/mice (Table 1) may be an indirect indicator of impaired TG catabolism and/or VLDL assembly & secretion. More studies are required to verify this mechanism. Polar lipids. Phosphatidyl cholines (GpCho) are prominent lipids in the outer leaflet of the plasma membrane, while phosphatidyl ethanolamines (GpEtn), phosphatidyl serines (GpSer) and phosphatidyl inositols (GpIns) are prominent glycerophospholipids in inner plasma membranes and mitochondrial matrix membranes [86]. Four of the 13 annotated GpCho were affected by the WD (Fig 10A). WD feeding was associated with a � 40% decline in 3 prominent GpCho containing MUFA (GpCho 36:2) and PUFA (GpCho 36:4, GpCho 40:6). Only one GpCho (32:1) increased in response to the WD.

Diet effects on neutral and polar lipids
Of the 19 annotated GpEtn annotated (Fig 10B) (Fig 11A and 11B). While GpIns 38:3 increased in response to the WD, other GpSer and GpIns were lower (� 50%) in livers of WD-fed mice.
Lysophospholipids (LysoP) are intermediates of phospholipid synthesis (Kennedy pathway) [90] and also arise from phospholipase action during membrane remodeling (Lands Pathway) [91]. Our analysis of LysoP did not distinguish between sn1 vs sn2 acyl chain positions ( Fig  12A). Of the seven annotated LysoP, six increased in livers of WD-fed mice. All LysoP increasing in response to the WD contained SFA, MUFA or ω6 PUFA; none contain ω3 PUFA. Taken together, these findings suggest that increases in hepatic SFA and MUFA promotes remodeling of membrane glycerophospholipids acyl chain content resulting in a loss of ω3 PUFA acyl chains from GpCho, GpEtn, GpSer, GpIns, but increases of MUFA and ω3 & ω6 PUFA in GpGro. Such changes in glycerophospholipid PUFA content affects membrane fluidity and signaling from membranes. In addition, the increase of PUFA in cardiolipin precursors may sensitize mitochondria to oxidative damage. Diet induced NASH in female Ldlr -/mice Lastly, we annotated 3 sphingomyelin (SM) species (Fig 12B). SM are key plasma membrane lipids, often found in lipid rafts in association with cholesterol [92]. Sm 34:2 was the most abundant Sm detected in liver; and SM 34:2 was~50% lower in livers of mice in the WD46 group.
To assess potential mechanisms accounting for changes in membrane lipids, we examined hepatic expression of enzymes associated with glycerophospholipid and sphingolipid metabolism ( Fig 9A). Two enzymes associated with the Kennedy pathway for de novo phospholipid synthesis are also involved in neutral lipid synthesis, e.g., Gly Kin & Gpatm. Only Gpatm expression was suppressed in livers of WD-fed mice. In contrast, enzymes involved in membrane remodeling (LpCat 1, 2 & 4) and sphingolipid synthesis (SpTlc 2) were induced �2-fold in livers of WD-fed mice. LpCat2 was the most responsive (5-fold induction) to the WD. Expression of enzymes involved in cardiolipin synthesis and turnover (Crls1, Taf) were attenuated <50%. Expression of SpTlc2 was the only enzyme involved in sphingolipid metabolism affected by the WD-fed mice. Changes in the mRNA abundance encoding these enzymes cannot fully explain the changes in glycerophospholipid and sphingolipid abundance and composition. As such, other mechanisms, such as substrate availability and post-translational mechanisms regulating enzyme activity will need to be examined for their involvement in WD-induced changes in membrane lipid composition.

Diet effects on oxylipins
Oxylipins represent the third class of lipids examined in our analysis (Figs 13 and 14). Oxylipins play key regulatory roles in multiple pathways [93]. Our targeted LC-MS/MS analysis of oxylipins included only the non-esterified fraction of hepatic lipids. Both enzymatic and nonenzymatic pathways generate these oxylipins. Oxylipins annotated in this analysis were derived from ω6 PUFA, i.e., linoleic acid (LA) & arachidonic acid (ARA) and ω3 PUFA, e.g., eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA). A total of hepatic 28 oxylipins were annotated and 20 oxylipins were significantly (q < 0.05) affected by the WD. The heat map (S4 Fig) provides an overview of the WD effects on oxylipin content in female Ldlr -/mouse liver.
In contrast to LA-derived oxylipins, most ARA-derived oxylipins are in low abundance in mouse liver (Fig 13A). Of these, the prostaglandins (PGE2, 6-keto PGF1α and 8-iso PGF2α) are significantly increased (� 6-fold) in livers of WD-fed mice (Fig 13B). Cyclooxygenases (Cox1 & Cox2) are key enzymes required to initiate prostaglandin synthesis; and PGE2 is a key prostaglandin involved in inflammation. In contrast, 6-keto PGF1α is a non-enzymatic degradation product of prostacyclin (PGI2), while 8-iso PGF2α is an isoprostane (IsoP2), a nonenzymatic oxidation product of ARA. Both are generated in states of oxidative stress by ROS, like O 2 -(superoxide radical), OH (hydroxyl radical or H 2 O 2 (hydrogen peroxide). The increase in these ARA-derived pro-inflammatory and oxidative stress markers is associated with Diet induced NASH in female Ldlr -/mice increases in transcriptomic markers of hepatic inflammation and oxidative stress in WD-fed mice (Fig 4B and 4C). Leukotrienes are also low abundance oxylipins in mouse liver. Leukotrienes (5-, 12-and 15-HETE) are products of specific lipoxygenases, while 20-HETE is a cytochrome P450 (Cyp450) product. Of these, only 12-HETE increased in livers of WD-fed mice. However, the variance was too high to be significant. The DiHETEs are epoxide hydrolase products of epoxy-fatty acids. Epoxy fatty acids are enzymatic products of ARA catalyzed by Cyp2C or Cyp5J. These oxidized PUFA are generally regarded as beneficial in promoting organ and tissue regeneration and attenuating inflammation [96,97]. Epoxy-fatty acids are converted to dihydroxy fatty acids (DiHETE) by soluble epoxide hydrolase (Ephx2) activity. While Ephx1 is associated with microsomes, Ephx2 is soluble. We detected no ω6-PUFA derived epoxide Diet induced NASH in female Ldlr -/mice products of ARA. However, we quantified several DiHETrE, derived from epoxy-derivatives of ARA. Most DiHETrE were lower in livers of WD-fed mice.
All of the EPA-and DHA-derived epoxy and dihydroxy oxylipins were significantly lower (� 50%) in livers from the WD-fed mice. There was no significant diet effect on EPA or DHAderived selective pro-resolving mediators [resolvins (Rev E1, Rev D1), protectin DX] or the EPA-derived isoprostane, 8-Iso-PGF3α.

Associations between major NASH markers and specific hepatic features
Finally, we took advantage of the heterogeneity in response to the WD reported in Fig 3 to carry out an unbiased statistical approach to identify associations between key NASH markers and transcriptomic and lipidomic features. Accordingly, "Pattern Hunter" in Metabolanalyst was used to identify associations (Tables 3-6 and Fig 16) [38]. NASH markers used in this analysis included steatosis (total hepatic fatty acyls), inflammation (TNFα), oxidative stress (Ncf2), fibrosis (Col1A1), notch signaling (Hey1), EMT (S1004A), cancer (Gpc3), ω6 PUFAderived oxylipins (PGE2) and ω3 PUFA-derived oxylipins (19,. Results in Tables  3-6 include the top 10 features that were positively and negatively associated with each NASH marker. The features are rank ordered based on the correlation coefficient (CC) and statistical significance (FDR value).
Steatosis. Features positively associated with steatosis included SFA, MUFA and MUFAcontaining DG & TG (Table 3 and Fig 16). Features negatively associated with hepatic steatosis include the relative abundance of PUFA (ω3 and ω6 index) and PUFA-containing TG, GpCho, GpEtn, GpSer & Fdps. The decline in membrane PUFA gives the appearance of essential fatty acid deficiency. However, the WD is an essential fatty acid sufficient diet [29]. Markers of essential fatty acid deficiency include increased DNL and mead acid (20:3,ω9) synthesis [98].  (Figs 10 and 11). Formation of 20:3,ω9 is likely due to the high abundance of 18:1,ω9 in the livers of WD-fed mice ( Fig 5) and off-target actions of Fads1 & Fads2 [72].
Other mechanisms contributing to hepatosteatosis include suppression of FAO, increased fatty acid uptake, neutral lipid synthesis and attenuated VLDL secretion. The WD did not affect hepatic expression of PPARα or its target genes involved in FAO (Cpt1A & HMG CoA Syn2), fatty acid binding protein (Fabp1) (Fig 7C) or increased neutral lipid synthesis ( Fig 9A). Instead, the expression of two enzymes involved in TG catabolism, i.e., Ces1g and Tgh, was suppressed �50% by the WD (Fig 9B). Ablation of Tgh [82] or Ces1g [83] affects hepatic neutral lipid content. We also examined the expression of a third enzyme involved in TG catabolism, i.e., ATGL (Fig 9B). Others have reported that mice fed high fat diets display low levels of ATGL expression in adipose tissue [99]. This effect was linked to high fat diet-induced Snail1; and Snail suppressed ATGL expression [99]. While the WD induced Snail1 expression in Ldlr -/mouse liver (Fig 5C), there was no significant effect of the WD on hepatic ATGL expression ( Fig 9B). As such, WD-mediated suppression of Tgh and Ces1g may be one explanation for increased neutral lipid content in livers of WD-fed Ldlr -/mice. As noted above, TG catabolism is required for assembly and lipidation of ApoB for VLDL assembly and secretion [84,85]. Decreased TG catabolism, coupled with decreased VLDL secretion may account for the increased in hepatic neutral lipid in WD-fed mice. Diacylglycerols and triacylglycerols were enriched in MUFA in livers of mice fed the WD (Fig 8). The increase in hepatic MUFA, can be explained, at least in part, by increased Scd1 expression (Fig 7), as well as the high dietary content of SFA & MUFA [29]. Scd1 expression is regulated by multiple transcription factors, including, SREBP1, ChREBP, PPARα, PPARγ2 and LXR [73,74]. The lack of diet effects on Cpt1a or HMG CoA Syn2, two PPARα target genes, argues against PPARα activation in the control of Scd1. Moreover, previous studies indicated that hepatic ChREBP nuclear abundance was unaffected in WD-fed Ldlr -/mice [28]. Herein, we report that SREBP1c and PPARγ2 were induced in livers of WD-fed mice ( Fig 7A). These results are consistent with reports of others [74,100]; PPARγ2 expression is associated with increased hepatic neutral lipid storage. While we did not measure LXR (α or β) directly, we present evidence (5S Fig) suggesting that LXR was activated by the WD. Specifically, the LXR target gene, Cyp7α, was induced � 1.5-fold in livers of mice fed the WD for 38 & 46 wks. LXR is activated by oxysterols [101] and the WD contains moderately high (0.15% w/w) cholesterol, which likely increases hepatic oxysterol formation. As such, the massive increase in hepatic SFA and MUFA and diacylglycerols & triacylglycerols is likely due to: a) high SFA and MUFA content of the WD; b) induction/activation of transcription factors (SREBP1c, PPARγ2, LXR) controlling lipid synthesis; c) modulation in TG catabolism through (Tgh and Ces3) and 4) possibly impaired VLDL secretion.
Inflammation, oxidative stress and fibrosis. Expression levels of TNFα, Ncf2 and Col1A1 were used as markers for inflammation, oxidative stress and fibrosis, respectively (Tables 3  and 4 and Fig 16). The correlation analyses revealed that these pathways share many common features. The common features positively associated with these markers include enzymes involved in membrane remodeling (LpCat1 & LpCat2), a chemokine (Mcp1), neutral lipids containing SFA and MUFA, and a phosphatidylglycerol containing 18:1,ω9 and 20:4,ω6 (GpGro 38:5). Phosphatidylglycerols are cardiolipin precursors; and cardiolipins are complex phospholipids found in the mitochondrial matrix. The correlation coefficients for these associations are high (r > 0.84). Common features that were negatively associated with inflammation, oxidative stress and fibrosis include the ω3 and ω6 index and several PUFA-containing glycerophospholipids (GpCho, GpEtn, GpSer). While a negative association between PUFA, particularly ω3 PUFA, and inflammation has been reported previously [102,103], these new findings identify specific PUFA-containing membrane lipids affected by the WD.
Particularly relevant is the major change in fatty acid type in membrane lipids. Inflammation, oxidative stress and fibrosis are positively associated with membrane lipids enriched in SFA and MUFA and negatively associated with phospholipids containing PUFA. This is particularly evident in GpEtn and GpSer, which are typically associated with the inner plasma membrane and intracellular membranes, e.g., ER, golgi, etc. This change in membrane lipid composition can have broad effects on hepatic physiology resulting from changes in trafficking and membrane-associated cell signaling.
Notch signaling, EMT and hepatic cancer. Notch signaling is a highly conserved morphogenic signaling mechanisms that control cell fate decisions, morphogenesis, proliferation and apoptosis during development, EMT and tissue repair [62,104]. We combined the analysis of Notch signaling with EMT & hepatic cancer markers since these pathways are upregulated in hepatic malignancies, such as primary hepatocellular cancer (HCC) and intrahepatic cholangiocarcinoma [62,[68][69][70]105]. Accordingly, Hey1, S100A4 and Gpc3 were used as markers for notch signaling, EMT and hepatic cancer, respectively (Table 5 and Fig 16).
Hey1 and HeyL, notch-associated transcription factors, were positively associated with both EMT and cancer. Three lipids (GpCho 36:5, TG 58:9 & 11,12-EpETE and 8,9-EpETE) were negatively associated with the notch, EMT and cancer (Table 5 and Fig 16). These lipids are either complex lipids containing PUFA or oxidized metabolites derived from PUFA. An emerging theme from this analysis is that hepatic PUFA-containing phosphoglyceroholipids, e.g., GpCho, GpEtn, GpSer, are negatively associated with an increased expression of hepatic markers associated with steatosis, inflammation, oxidative stress, fibrosis (Tables 3 and 4), Notch signaling, EMT and cancer (Table 5). Whether these changes in membrane lipids are causally linked to WD-induced hepatic pathology remains to be established.
ω6 PUFA-derived oxylipins. Features positively and negatively associated with hepatic PGE2, a cyclooxygenase product, are presented in Table 6. Elevated hepatic PGE2 is positively associated with oxidative stress markers (Ncf2, 8-isoPGF2α, 6-ketoPGF1α), membrane remodeling (LpCat2 & 4), inflammation (TNFα), liver weight & liver weight as a % of body weight. Elevated hepatic PGE2 is negatively associated with glycerophospholipids (GpCho, GpEtn, GpSer) and TG containing PUFA. Neither Cox1 nor Cox2 are expressed in hepatocytes, but these enzymes are expressed in other liver cells, like macrophage (Kupffer cells) and leukocytes infiltrating liver in response to the WD [106] (Fig 3). Products of these enzymes serve as ligands that bind G-protein receptors and typically play a role in inflammation. However, PGE2 has been reported to inhibit expression of lipogenic enzymes in isolated rat primary hepatocytes through EP3 receptors [106]. Thus, the suppression of Fasn (Fig 7B) may be due to elevated hepatic PGE2.
Epoxide products of PUFA are generally regarded as beneficial and play a role in reparative events [96]. Fatty epoxides, however, are rapidly degraded to dihydroxy products by epoxide hydrolases, e.g., Ephx2 (soluble) [96,97,107,108]. Expression of hepatic enzymes involved in the formation of epoxides and dihydroxy fatty acids was suppressed by~50% in livers of mice consuming the WD (Fig 15). Hepatic levels of 19, 20-DiHDPA, which are dependent on Cyp2C production of the epoxide precursors, are associated with PUFA-enriched membrane lipids and negatively associated with hepatic inflammation, oxidative stress and repair.

Limitations and conclusions
This report provides new information on the capacity of the WD to promote NASH in a female preclinical mouse model. The WD promotes obesity & dyslipidemia and all of the major hallmarks of NASH, including hepatosteatosis, inflammation, oxidative stress and fibrosis (Table 1 and Figs 3, 4 and 6). Others have reported that NASH is associated with increases in hepatic abundance of transcripts involved in tissue remodeling, namely apoptosis & autophagy [2,60,61], hedgehog & notch signaling [62,63] and cancer [65,66]. Female Ldlr -/fed the WD have increased hepatic abundance of transcripts encoding proteins involved in these same pathways (Figs 5 and 16).
A key feature accompanying tissue remodeling is the change in the type and abundance of tissue lipid. Our studies provide compelling evidence for WD-induced changes in the acyl chain composition of neutral and polar lipids, particularly hepatic membrane lipids. While this analysis is largely a descriptive, it provides insight into possible mechanisms contributing to the onset and progression of NASH. Chief amongst these is the major shift from PUFA-enriched membrane lipids to SFA & MUFA enriched membrane lipids. Likely mechanisms for this change in membrane composition include the induction of membrane remodeling, i.e., the Land's pathway, through induction of enzymes like LpCat 1, 2 & 4 (Figs 9-12) and the high substrate availability of SFA & MUFA arising from the diet and disruption of hepatic lipid metabolism. These changes in hepatic lipids have the capacity to affect several regulatory pathways. For example, the increase in hepatic DG containing SFA and MUFA may serve as regulatory ligands controlling protein kinase C subtypes, e.g., PKCε, and tissue insulin sensitivity [109]. The decrease in membrane glycerophospholipids containing PUFA will have effects on membrane fluidity and will affect membrane-associated signaling mechanisms [110] (Fig 16). Compounding these dramatic changes in membrane lipids is the change in the type and abundance of PUFA-derived regulatory oxylipins appearing in the NASH liver. Clearly, the increase in hepatic PGE2 and the ARA-derived oxidative stress marker (8-iso-PGF2α) exacerbates disease progression through induction of inflammatory mechanisms. These changes in hepatic lipid composition are likely major contributors to the onset & progression of WD-induced NASH since supplementing the WD with DHA (at 2% total calories) both restores hepatic C 20-22 ω3 PUFA content and attenuates disease severity in prevention and remission studies [29,31,103].
In conclusion, the female Ldlr -/mouse is a suitable preclinical model to examine NAFLD, from benign steatosis to early stage HCC. Our studies go on to establish that the major markers of NASH are strongly associated with major changes in hepatic neutral and polar lipid acyl chain composition, as well as ω3 and ω6 PUFA-derived oxylipins. The outcome of these studies sets the stage for future mechanistic studies.