Docosahexaenoic acid attenuates Western diet-induced hepatic fibrosis in Ldlr-/- mice by targeting the TGFβ-Smad3 pathway.

DHA (22:6,ω3), but not EPA (20:5,ω3), attenuates Western diet (WD)-induced hepatic fibrosis in a Ldlr(-/-) mouse model of nonalcoholic steatohepatitis. We examined the molecular basis for the differential effect of dietary EPA and DHA on WD-induced hepatic fibrosis. DHA was more effective than EPA at preventing WD-induced effects on hepatic transcripts linked to fibrosis, including collagen 1A1 (Col1A1), transforming growth factor-β (TGFβ) signaling and proteins involved in remodeling the extracellular matrix, including metalloproteases, tissue inhibitors of metalloproteases, and lysyl oxidase subtypes. Examination of the TGFβ pathway showed that mice fed the WD supplemented with either olive oil or EPA had a significant (≥2.5-fold) increase in hepatic nuclear abundance of phospho-mothers against decapentaplegic homolog (Smad)3 when compared with mice fed the reference diet (RD); Smad3 is a key regulator of Col1A1 expression in stellate cells. In contrast, mice fed the WD supplemented with DHA had no increase in phospho-Smad3 when compared with mice fed the RD. Changes in hepatic phospho-Smad3 nuclear content correlated with proCol1A1 mRNA and protein abundance. Pretreatment of human LX2 stellate cells with DHA, but not other unsaturated fatty acids, blocked TGFβ1-mediated induction of Col1A1. In conclusion, DHA attenuates WD-induced fibrosis by targeting the TGFβ-Smad3-Col1A1 pathway in stellate cells.


Animals and diets
All procedures for the use and care of animals for laboratory research were approved by the Institutional Animal Care and Use Committee at Oregon State University. Male Ldlr Ϫ / Ϫ mice (C57BL/6J background, Jackson Laboratories) at 2 months of age were fed one of the following four diets ad libitum for 16 weeks; each group consisted of 8 male mice ( 17 ). The reference diet (RD) (Purina chow 5053) consisted of 13.5% energy as fat, 58.0% energy as carbohydrates, and 28.5% energy as protein.
The WD (D12079B; Research Diets) consisted of 17% energy as protein, 43% energy as carbohydrate, and 41% energy as fat; cholesterol was at 1.5 g%. The WD was supplemented with olive oil (WD + O), EPA (WD + E), or DHA (WD + D). WD supplementation with olive oil, EPA, or DHA increased total fat energy to 44.7% and reduced protein and carbohydrate energy to 15.8% and 39.5%, respectively. Olive oil was added to the WD to ensure a uniform level of energy from fat, protein, and carbohydrate in all WD diets. Preliminary studies established that the addition of olive oil to the WD had no effect on development or progression of diet-induced NAFLD in Ldlr Ϫ / Ϫ mice. The C 20-22 3 PUFA in the WD + E or WD + D diets were at 2% total energy. This dose of C 20-22 3 PUFA is comparable to the dose consumed by patients taking Lovaza TM (GlaxoSmithKline) for treating dyslipidemia ( 38 ), and this dose increases plasma C 20-22 3 PUFA in mice to levels seen in humans consuming 4-6 g/d of C 20-22 3 PUFA (39)(40)(41). A description of the diets and diet effects on mouse plasma and liver lipid composition was previously reported ( 17 ). At the end of the 16 week feeding period, all mice were fasted overnight (18:00 to 08:00 the next day) and euthanized at 08:00 for the collection of blood and liver ( 17,28 ).

Measurement of plasma adipokines and proinfl ammatory cytokines
Plasma adipokines (leptin and adiponectin) were quantifi ed by ELISA (R and D Systems). Plasma proinfl ammatory cytokines were quantifi ed using a cytometric bead array mouse infl ammation kit (BD Biosciences, San Jose, CA). Quantitative measurements of IL-12, TNF ␣ , IFN ␥ , MCP-1, IL10, and IL6 were determined by fl ow cytometry per the manufacturer's protocol. Data were acquired using FACSCalibur (BD Biosciences), and data analyses were done using FCAP Array Software version 3.0 (BD Biosciences). Plasma for these assays was obtained from mice described previously ( 17,18 ).

Analysis of plasma Toll-like receptor-2 and -4 activators
Plasma levels of Toll-like receptor (TLR)-2 and TLR-4 activators used Hek-Blue TM mTLR2 and Hek-blue TM mTLR4 cells (Invivogen), respectively. Cells were grown in 96 well cell culture plates overnight in 200 l of Hek-Blue Detection medium TM (Invivogen) and 10 l of plasma from the previous study ( 17 ). The formation of a blue color, refl ecting activation of the TLR-NF B pathway, was quantifi ed by absorbance at 620 nm. Standard curves with authentic TLR2 (PAM3SK4) and TLR4 (LPS-B5) agonists are shown in supplementary Fig. 1.

RNA extraction and quantitative RT-PCR
Total RNA was extracted from liver, and specifi c transcripts were quantifi ed by quantitative RT-PCR (qRT-PCR) ( 17,42 ). Primers for each transcript are listed in supplementary Table 1. Cyclophilin was used as the internal control for all liver transcripts, and hydroxymethyl bilane synthase was used as the internal control of RNAs isolated from LX2 cells in culture. Transcripts were also quantifi ed by qRT-PCR using the mouse fi brosis array (PAMM-120ZE; SABioscience-Qiagen). Hsp90ab1 was used as collagen subtypes ( 16 ). Fibrosis resulting from hepatic damage is linked to increased extracellular deposition of type 1 collagen [collagen1A1 (Col1A1)], smooth muscle actin, elastin, fi bronectin, and other proteins. Fibrosis is also associated with increased production of proteins from stellate cells and macrophages that are involved in ECM remodeling; these proteins include lysyl oxidase and lysyl oxidase-like subtypes involved in collagen cross-linking, metalloprotease subtypes (MMPs), and tissue inhibitors of metalloproteases (TIMPs). The relative abundance of these proteins affects fi brosis progression and severity.
We also reported that adding EPA (20:5, 3) or DHA (22:6, 3) to the WD affected diet-induced hepatic fi brosis. DHA was more effective than EPA at attenuating WD-induced hepatic fi brosis ( 17,18,28 ). Although EPA and DHA attenuated WD-induced dyslipidemia, neither EPA nor DHA affected WD-induced body weight, the percent of body fat, blood glucose, or plasma endotoxin. The effect of C 20-22 3 PUFA on hepatic fi brosis was established by histology and quantifying the expression of fi brosis markers (i.e., Col1A1 and TIMP1) ( 17,18 ). DHA and EPA are known to attenuate infl ammation, decrease fatty acid synthesis, and increase fatty acid oxidation ( 29 ). The fi nding that DHA attenuated WD-induced hepatic fi brosis in obese mice was unexpected.
Several clinical trials have reported that dietary 3 PUFA supplementation lowered hepatic fat in obese children and adults with NAFLD (30)(31)(32)(33)(34)(35)(36)(37), whereas other investigators report that dietary supplementation with fi sh oil ( 36 ) or EPA-ethyl esters ( 37 ) does not attenuate the histological features of the disease, like fi brosis. As such, human studies using 3 PUFA to treat NAFLD/NASH have yielded mixed results. Our studies have established a difference in how specifi c 3 PUFAs affect clinical features associated with NASH. Herein, we examined potential mechanisms to explain how DHA and EPA differentially affect WD-induced hepatic fi brosis. Our fi ndings establish that dietary DHA interferes with the TGF ␤ -mothers against decapentaplegic homolog (Smad)3 signaling pathway in vivo and attenuates TGF ␤ -mediated induction of Col1A1 expression in human stellate cells. The outcome of these studies has the potential to infl uence the design of clinical studies in children and adults with NAFLD/NASH.

agonist. Feeding Ldlr
Ϫ / Ϫ mice the WD + O diet increased TLR2 and TLR4 agonist activity by ‫ف‬ 60% ( Fig. 1 ). Although the WD + D diet signifi cantly lowered TLR2 agonists in plasma, WD + E and WD + D failed to lower plasma TLR4 agonists. Failure of the 3 PUFA-containing diets to suppress TLR4 agonist is consistent with our previous fi nding, which showed that dietary 3 PUFAs do not attenuate WD-induced endotoxinemia ( 18 ). Hepatic fi brosis is also linked to products derived from adipose tissue and cells involved in innate immunity, including adipokines and cytokines, respectively ( 52,53 ). We quantifi ed plasma leptin, adiponectin, IFN ␥ , interleukins (IL6, IL10, and IL12), monocyte chemoattractant protein-1 (MCP1), and TNF ␣ . Of these, WD + O feeding signifi cantly increased plasma leptin and TNF ␣ ( Fig. 1 ). Addition of EPA or DHA to the WD had no signifi cant effect on plasma levels of TNF ␣ or leptin. The effect on leptin was anticipated because plasma leptin levels parallel adiposity. WD-fed mice are obese, and including EPA or DHA in the WD had no effect on adiposity ( 17 ).

Effects of WD and 3 PUFA on the expression of genes involved in fi brosis
Feeding Ldlr Ϫ / Ϫ mice the WD elevated hepatic expression of Col1A1, TGF ␤ 1, and Timp1, and dietary C 20-22 3 PUFA attenuated this effect ( 17 ). Herein, we used the SA-Bioscience qRT-PCR fi brosis array to get a broad assessment of the impact of diet on hepatic fi brosis. The heat map ( Fig. 2 ) illustrates the results with samples from livers of mice fed the RD, WD + O, WD + E, or WD + D (8 mice/ group). The pie charts represent quantitation of transcripts that changed signifi cantly. When compared with the RD group, 38, 28, and 12 transcripts were signifi cantly induced by the WD + O, WD + E, and WD + D diets, respectively. Only two transcripts [metalloprotease-3 (Mmp3) and caveolin1] were signifi cantly suppressed by the WD + the internal control for array analysis. Transcripts and their acronyms are provided in supplementary Table 2. Results from these studies were analyzed online using RT 2 Profi ler PCR Array Data Analysis version 3.5 (SABioscience).

Immunoblot analysis
The quantitation of hepatic proteins by immunoblotting was previously described ( 17 ). The antibodies used to detect Smad2/3, Smad4, phospho-Smad2/3, and GAPDH were obtained from Cell Signaling Technologies. The antibody for procollagen 1A was obtained from Santa Cruz Biotechnologies.

Hydroxyproline
Hepatic hydroxyproline was quantifi ed using the Hydroxyproline Kit (Sigma, MAK008). The colorimetric assay was carried out according to the manufacturer's instructions.

LX2 cells
LX2 cells were obtained from SL Friedman (Mount Sani Medical School) ( 43,44 ). LX2 cells are activated human hepatic stellate cells; they are maintained in DMEM with 5% FCS containing penicillin, streptomycin, and normocin. Cells were plated on 100 mm plastic petri dishes at ‫ف‬ 100,000 cells/plate and grown to confl uence. Cells were treated with fatty acids (at 25 M) in endotoxin-free BSA (at 10 M) during the growth phase. Fatty acids [oleic acid (18:1, 9), arachidonic acid (20:4, 6), and DHA (22:6, 3)] were obtained from Nu-Chek Prep. Cells were treated with 100 pM recombinant-human TGF-␤ 1 (R and D Systems) in media containing no fatty acids or BSA for 24 h. RNA was extracted for qRT-PCR as previously described ( 18 ). Human PCR primers are listed in supplementary Table 1.
For fatty acid analysis, confl uent cells were washed twice with cold PBS. Cells were extracted for fatty acids and quantifi ed by gas chromatography analysis ( 17,45 ) or applied to a reverse-phase HPLC system for the quantitation of 14 C-labeled fatty acids as described ( 46 ).

Statistical analysis
Statistical analysis of data used one-way ANOVA and Tukey-HSD to detect signifi cant differences between groups. Data were analyzed for homogenous variances by the Levine test. If unequal variances were detected, data were log-transformed. ANOVA analysis was performed on both transformed and untransformed data. The hydroxyproline data were analyzed by Student's t -test. Values are reported as mean ± SD. A P value р 0.05 was considered signifi cantly different.

Effect of the WD and 3 PUFA on plasma factors linked to hepatic infl ammation and fi brosis
Toll-like receptors (TLR2 and TLR4) have been linked to hepatic infl ammation and fi brosis (47)(48)(49)(50). We previously reported that feeding Ldlr Ϫ / Ϫ mice the WD increased plasma endotoxin ( 18 ), a gut-derived microbial factor that activates the TLR4 pathway ( 51 ). WD-induced endotoxinemia was associated with increased hepatic nuclear content of NF B as well as multiple NF B-regulated transcripts linked to hepatic infl ammation ( 17,18 ). EPA or DHA supplementation of the WD did not attenuate WD-induced plasma endotoxin ( 18 ). Herein, we use a cell-based approach to quantify plasma levels of bioactive TLR2 and TLR4 effect on the hepatic abundance of these transcripts, the WD + D diet was more effective than WD + E at attenuating expression of Mmp1a, Mmp2, Timp2, Serpinh1, and Lox.

Collagen crosslinking
Lox mRNA was induced by WD + O feeding and attenuated by WD + D > WD + E ( Figs. 4 and 5 ). Lox is one of several enzymes involved in collagen crosslinking. Lox and the lysyl oxidase-like (LoxL) family of enzymes are expressed in stellate cells and portal fi broblast ( 54 ). Accordingly, we quantifi ed the effects of diet on the expression of Lox and LoxL subtypes (LoxL1, LoxL2, and LoxL3). In RD-fed mice, hepatic LoxL1 and LoxL3 are more abundant than Lox and LoxL2. Lox and all LoxL subtypes were induced ( у 4-fold) by WD + O feeding. Feeding mice the WD + D diet attenuated the expression of Lox, LoxL1, and Loxl2 but not LoxL3. Expression levels of Lox and all LoxL subtypes in WD + O-and WD + E-fed mice were not different.
We also quantifi ed hepatic levels of the product of Lox and LoxL1-3 activity (i.e., hydroxyproline) ( Fig. 5C ). Hydroxyproline levels were highly variable in livers from WD + O-fed mice. Hydroxyproline increased in three of the eight mice to levels >6-fold above the mean value quantifi ed in RD-fed mice and mice fed WD + E or WD + D. WD + E and WD + D attenuated hepatic hydroxyproline content when compared with mice fed WD + O.

TGF-␤ Superfamily
TGF-␤ plays a major role in promoting fi brosis ( 48 ). Accordingly, we examined the TGF ␤ superfamily for diet O diet when compared with RD-fed mice. As illustrated in the heat map and in detailed quantitation of transcripts described below, the WD + D diet was more effective than the WD + E diet at attenuating WD effects on the expression of genes linked to fi brosis.

Collagen subtypes
Although the liver expresses several collagen subtypes ( Fig. 3 ), Col1A1 is the major collagen subtype associated with hepatic fi brosis in WD-fed mice ( 17 ) and humans with NASH ( 16,52 ). In normal mouse liver, Col1A1 and Col7A1 are expressed at low levels when compared with Col1A2 or Col4A. Using the qRT-PCR fi brosis array data ( Fig. 2 ) as well as independent qRT-PCR analysis, WD feeding induced the expression of Col1A1, Col1A2, and Col4A1 by 17-, 5-, and 3-fold, respectively. Of these transcripts, only Col1A1 expression was signifi cantly attenuated by dietary DHA. Dietary EPA did not attenuate WD-mediated induction of any collagen subtype.

ECM remodeling proteins
Fibrosis is associated with the induction of multiple proteins involved in remodeling the ECM, including metalloproteases, protease inhibitors, and enzymes involved in collagen cross-linking ( Fig. 4 ). The mRNA abundance of three metalloproteases (Mmp1a, Mmp2, and Mmp13), four protease inhibitors (Timp1, Timp2, Serpine1, and Serpinh1), and the cross-linking enzyme lysyl oxidase (Lox) was signifi cantly induced by WD + O. Although both WD + E and WD + E signifi cantly attenuated the WD + O WD + O diet also increased expression of the TGF ␤ receptor subunit-2 (Tfgbr2, 2-fold), Smad 6 (Smads are transcription factors and downstream targets of TGF ␤ signaling) ( ‫ف‬ 2-fold), and two proteins that regulate TGF ␤ function [thrombospondins (Thbs1, Thbs2), 3-to 5-fold]. No signifi cant difference was detected in the expression of these transcripts in mice fed the WD + O and the WD + E diets. Feeding mice the WD + D diet, however, signifi cantly attenuated expression of all TGF ␤ subtypes, both TGF ␤ receptor subunits, Smad6, Thbs1, and Thbs2 when compared with WD + O-fed mice. As such, DHA, but not EPA, attenuates multiple transcripts associated with TGF ␤ signaling.

TGF ␤ Signaling
To further explore the impact of the WD and 3 PUFA on TGF ␤ signaling, we examined the pathway for TGF ␤ induction of Col1A1 expression. TGF ␤ binds to type 1 and type II TGF ␤ receptors; TGF ␤ binding promotes recruitment of Smad2 and Smad3 to the TGF ␤ receptor complex where Smad2 and Smad3 are phosphorylated ( 55-60 ) ( Fig. 7A ). Smad2 is phosphorylated at S 465 and S 467 , whereas Smad3 is phosphorylated at S 423 and S 425 . Phospho-Smads move from the cytosol to the nucleus in association with Smad4. Binding of phospho-Smad3 to the Col1A1 promoter is associated with increased Col1A1 gene transcription and the accumulation of Col1A1 mRNA and protein ( 59 ).
Total and phosphorylated Smad2 and Smad3 were quantifi ed in cytosol and nuclear extracts from livers of mice fed the RD or the WD + O, WD + E, or WD + D diets. Total hepatic Smad2 and Smad3 levels remained unchanged by diet. Although the antibody for phospho-Smads recognized both phospho-Smad 2 and phospho-Smad3, the position of the phosphorylated protein in the immunoblot corresponded to phospho-Smad3. Feeding mice the WD + O and WD + E diets signifi cantly increased cytosolic and nuclear phospho-Smad3, refl ecting activation of the TGF-␤ pathway. Feeding mice the WD + D diet, however, showed no increase in cytosolic or nuclear phospho-Smad3. Changes in nuclear phospho-Smad3 paralleled changes in Col1A mRNA ( Fig. 3C ) and proCol1A protein abundance ( Fig. 7B and D ). Although the WD + O and WD + E diets activate the TGF ␤ -Smad3 pathway and promote fi brosis, feeding mice the WD + D diet attenuates this pathway.

DHA attenuates TGF ␤ induction of Col1A1 in LX2 cells
The induction of Col1A1 expression in stellate cells is in response to increased production of TGF ␤ and cytokines (e.g., IL-1 ␤ ) from hepatocytes, Kupffer cells, macrophages, and T-cells ( 53,(56)(57)(58)(59). Herein, we determine whether DHA can act directly on stellate cells to regulate TGF ␤ 1 control of Col1A1. LX2 cells are a human activated stellate cell line ( 43,44 ). We fi rst determined if LX2 cells assimilate and metabolize exogenous fatty acids. Accordingly, LX2 cells were treated with 25 M fatty acids for 96 h to enrich cellular lipids ( Fig. 8A ). Treatment of cells with fatty acids (at 25 M) had no adverse effects on cell growth or morphology. Fatty acid analysis shows that treatment of cells with fatty acids increased the cellular content of fatty effects on gene expression ( Fig. 6 ). Mice fed the RD express three TGF ␤ subtypes in liver; the relative abundance of the three transcripts is: TGF ␤ 1, 1.0 ± 0.1; TGF ␤ 2, 0.031 ± 0.01; TGF ␤ 3, 0.06 ± 0.013. Of these, TGF ␤ 1 is the predominant hepatic TGF ␤ subtype. Feeding mice the WD + O diet induced all three TGF ␤ subtypes (2-to 3-fold). The   The relative expression of hepatic LOX and LOXL subtypes was quantifi ed by qRT-PCR using liver RNA from mice fed the RD; cyclophilin was the reference transcript. The qRT-PCR primers are listed in supplementary Table 1. B: Effect of diet on LOX and LOXL subtype mRNA abundance. Results are expressed as mRNA abundance-fold change (mean ± SD; n = 8). * P р 0.05 versus RD; # P р 0.05 versus WD + O. C: Hepatic hydroxyproline abundance was quantifi ed as described in Materials and Methods and represented as g hydroxyproline/mg protein (mean ± SD; n = 8). P values were calculated using Student's t -test. Fig. 6. Diet effects on expression of the TGF ␤ superfamily. Transcript abundance of proteins involved in TGF ␤ signaling was quantifi ed using data from the qRT-PCR fi brosis array ( Fig. 2 ). Results are represented as mRNA abundance-fold change (mean ± SD; n = 8). * P р 0.05 versus RD; # P р 0.05 versus WD + O.
In this report, we have expanded our analysis of diet effects on hepatic fi brosis by quantifying 84 gene expression markers of fi brosis, four plasma markers linked to fi brosis, hepatic TGF ␤ -Smad signaling, and fatty acid effects on TGF ␤ signaling in human stellate (LX2) cells. Feeding Ldlr Ϫ / Ϫ mice the WD supplemented with DHA (WD + D), but not the WD + E diet, abrogated WD + O-induced accumulation of phospho-Smad3 in hepatic nuclei ( Fig. 7 ). TGF ␤ is a major regulator of hepatic fi brosis ( 48 ). TGF ␤ binding to the TGF ␤ receptor increases Smad3 phosphorylation, and phospho-Smad3 migrates from the cytosol to the nucleus where it binds the Col1A promoter and induces transcription of the Col1A1 gene ( 55,59 ). Although TGF ␤ signaling is operative in multiple hepatic cell types, including hepatocytes ( 64 ), the effect of WD and DHA on phospho-Smad3 correlates with changes in hepatic stellate cell expression of Col1A1 mRNA and proCol1A1 ( Fig. 7 ). Moreover, we show that DHA acts directly on human LX2 stellate cells to block TGF ␤ 1 induction of Col1A1 gene expression ( Fig. 8 ). Taken together, these fi ndings establish DHA as a regulator of TGF ␤ signaling and Col1A1 expression.
Our analysis revealed broad effects of the WD and C 20-22 3 PUFA on the expression of genes involved in fi brosis and TGF ␤ signaling ( Fig. 2-6 ). Feeding mice the WD + O diet induced multiple components associated with TGF ␤ signaling, including several TGF ␤ subtypes (TGF ␤ 1-3), a TGF ␤ receptor subunit (TGF ␤ -R2), and two thrombospondins (Thbs1 and Thbs2). Thbs1 and Thbs2 are expressed in macrophage, whereas Thbs2 is also expressed in endothelial cells ( 65,66 ). Thbs1 induces TGF ␤ Ϫ dependent and TGF ␤ Ϫ independent fi brosis. DHA has been reported to fatty acids de novo and to transport exogenous fatty acids from the media into cells and metabolize these fatty acids.
To assess the impact of fatty acids on stellate cell function, LX2 cells were treated with TGF ␤ 1 (100 pM, overnight) after prior fatty acid treatment, as described above. TGF ␤ treatment of LX2 cells induced Col1A1 expression 4-fold in vehicle-treated cells and cells pretreated with oleic acid (18:1, 9) or arachidonic acid (20:4, 6). DHA pretreatment, however, blocked TGF ␤ 1 induction of Col1A1 mRNA. Thus, DHA has the capacity to act directly on human hepatic stellate cells to attenuate TGF ␤ 1mediated induction of Col1A1.

DISCUSSION
The goal of this report was to identify mechanisms that explain the differential effect of EPA and DHA on WD-induced hepatic fi brosis. This is highly relevant because dietary 3 PUFAs are being evaluated as treatment strategies for NAFLD/NASH in children and adults (30)(31)(32)(33)(34)(35)63 ). Although these clinical studies show that dietary 3 PUFAs reduce liver fat (30)(31)(32)(33)(34)(35)(36)(37), some clinical trials have reported that 3 PUFAs, like fi sh oil ( 36 ) or EPA-ethyl esters ( 37 ), fail to improve fi brosis scores associated with NASH. Our previous report established that DHA was more effective than EPA at attenuating WD-induced hepatic fat, infl ammation, and fi brosis ( 17 ). Hepatic fi brosis in that study examined hepatic expression of three fi brosis gene expression markers (Col1A1, Timp1, and TGF-␤ 1) and histology (i.e., trichrome staining of liver). Mouse liver cytosolic and nuclear extracts were prepared for immunoblotting as described ( 17 ). Antibodies used to detect total and phosphorylated Smad2 and -3 are listed in the Materials and Methods section. A: The pathway for TGF ␤ regulation of Col1A1 gene transcription (56)(57)(58)(59). B: Representative immunoblot of total Smad 2, Smad3, phospho-Smad3, and proCol1A1. The loading control for phospho-Smad3 was total Smad3; while the loading control for proCol1A1 was GAPDH. The number of independent samples for each group was: RD, 2; WD + O, 3; WD + E, 3; WD + D, 3. C and D: Results are quantifi ed for phospho-Smad3 (C) and proCol1A1 (D) and expressed as fold change (mean ± SD). * P р 0.05 versus RD; # P р 0.05 versus WD + O.
Because TLRs and cytokines are expressed in multiple hepatic cell types, we predict that the in vivo mechanism for DHA control of Col1A1 expression likely involves both direct effects of DHA on stellate cells ( Fig. 8 ) and indirect effects mediated by changes in the function of Kupffer cells and macrophages and T-cells infi ltrating the liver. Two factors support this hypothesis. First, the level of TGF ␤ 1-mediated induction of Col1A1 in LX2 stellate cells is modest (4-fold) when compared with the nearly 20-fold induction of Col1A1 seen in vivo. Second, DHA is a pleiotropic mediator of cell function affecting multiple pathways, such as membrane composition as well as multiple cell signaling and transcription regulatory networks ( 29 ). In addition to stellate-macrophage interaction, hepatocyte-stellate cell interaction has been reported ( 72 ). As such, the LX2 cells can serve as a model to investigate how 3 PUFAs regulate factors derived from hepatocytes and Kupffer cell/macrophage to control stellate cell function.

ECM remodeling enzymes
The progression and remission of fi brosis requires a balance in expression of ECM genes and enzymes that remodel the ECM. The WD induces expression of several hepatic collagen subtypes ( Fig. 3 ) as well as enzymes involved in ECM remodeling, including metalloproteases (Mmp1a, Mmp2, and Mmp13), protease inhibitors [Timp1 and Timp2, serine protease inhibitors (serpine1 suppress Thbs1 expression in adipose tissue ( 65 ). Thbs2, in contrast, attenuates fi brosis ( 67,68 ). We are unaware of reports documenting effects of dietary PUFA on Thbs2 expression. The WD + D diet was more effective than the WD + E diet in attenuating WD + O induction of all TGF ␤ subtypes, both TGF ␤ subunits (R1 and R2), Smad6, and Thbs1 and Thbs2.
Our fi ndings go beyond an earlier report by Chen et al. ( 69 ), who showed that DHA attenuated fi brosis in a rat bile duct ligation model for cholestatic liver injury. Although these investigators showed that DHA targeted TGF ␤ , NF B, and Erk signaling, the authors did not assess effects of other fatty acids, Smad phosphorylation, and cytosolic-nuclear traffi cking or TGF ␤ control in human hepatic stellate cells. Although cholestatic liver injury has a different etiology than diet-induced NASH in obese patients, these and other studies ( 70 ) suggest that DHA may have antifi brotic effects in liver and other tissues by controlling TGF ␤ signaling.

Regulators of hepatic fi brosis
Factors promoting diet-induced hepatic fi brosis are derived from multiple sources, including the gut ( 51 ) and multiple cell types, such as adipocytes (leptin), macro- Fig. 8. TGF ␤ 1 and DHA regulation of proCol1A1 expression in human LX2 stellate cells. LX2 cells were plated at ‫ف‬ 10% confl uence on plastic petri dishes in DMEM + 5% FCS and antibiotics. The next day, cells were treated with vehicle (Veh: fatty acid-and endotoxin-free BSA at 10 M) or 25 M fatty acids in BSA-containing media. Media was change every third day. This treatment enriches membrane lipids with exogenous fatty acids. A: Cells were harvested for fatty acid extraction and gas chromatographic quantitation. The results are presented as fatty acid, Mol% (mean ± range of two separate studies). B: Cells were treated with oleic acid (OA), arachidonic acid (ARA), or DHA as described above. After fatty acid pretreatment, cells were treated without or with 100 pM TGF ␤ 1 overnight in the absence of fatty acids. Cells were harvested for RNA for qRT-PCR quantitation of proCol1A1 mRNA; hydroxymethylbilane synthase was the reference RNA. Results are presented as Col1A1 mRNA-fold change (mean ± SD; n = 3). * P р 0.05 versus Veh; # P р 0.05 versus 18:1, 9 + TGF ␤ . cardiovascular disease, clearly showed that feeding humans EPA did not increase blood DHA ( 74 ). In rodents, EPA and DHA suppress the expression of enzymes involved in PUFA synthesis ( 17 ). This mechanism involves the suppression of sterol regulatory element binding protein nuclear abundance ( 28 ), a key transcription factor regulating de novo lipogenesis and PUFA synthesis ( 29 ). As such, this mechanism lowers the capacity of cells to convert dietary C 18 -PUFA precursors to C 20-22 3 and 6 PUFAs. Because dietary DHA is retroconverted to EPA through a peroxisomal mechanism ( 61 ) and EPA is not well converted to DHA in vivo, DHA is a logical choice for usage in clinical trials.
In a recent double-blind, placebo-controlled trial, NAFLD patients received placebo or Lovaza TM at 4 g/d ( ‫ف‬ 50:50 mix of EPA-and DHA-ethyl esters) for 15-18 months ( 35 ). When compared with the placebo-treated group, the Lovaza TM -treated group showed a signifi cant reduction in liver fat without a signifi cant reduction in fi brosis scores. In our studies, C 20-22 3 PUFA in the WD + E or WD + D diets was used at 2% total energy. This dose of C 20-22 3 PUFA is comparable to the dose consumed by patients taking Lovaza TM to treat dyslipidemia ( 38 ). The dose used in our studies increased plasma C 20-22 3 PUFA in mice to levels seen in humans consuming 4-6 g/d of C 20-22 3 PUFA (39)(40)(41). We also used a combination of EPA and DHA to mimic Lovaza TM treatment ( 17 ). Whereas the overall C 20-22 3 PUFA dose was at 2% total energy, EPA and DHA were at 1% total energy each. The combination of EPA + DHA was less effective than using DHA alone at suppressing Col1A1 expression. As such, the preclinical prevention studies with Ldlr Ϫ / Ϫ mice suggest that DHA (at 4-6 g/d) may be more effective than EPA or fi sh oil at preventing diet-induced NASH and fi brosis in humans. A search of clinical trials (www.clinicaltrials.gov; query "NASH and DHA") yields three trials that are recruiting or active. The outcome of these studies may provide evidence in support of the use of dietary DHA in the prevention or treatment of diet-induced fi brosis.

Summary
This study identifi ed mechanisms to explain the differential effects of EPA and DHA on WD-induced hepatic fibrosis. Feeding Ldlr Ϫ / Ϫ mice the WD induced changes in expression of multiple genes associated with hepatic infl ammation and fi brosis. DHA, but not EPA, attenuated WD-induced hepatic fi brosis by targeting the TGF ␤ -Smad3-Col1A1 pathway. Moreover, DHA acted directly on human LX2 stellate cells to block TGF ␤ 1-mediated induction of Col1A1. These outcomes establish DHA as a key regulator of TGF ␤ signaling and hepatic fi brosis. and serpinh1)], and enzymes involved in collagen crosslinking (Lox and LoxL1-3) ( Figs. 4 and 5 ). Mmps and Timps are products of hepatocytes, leukocytes, Kupffer cells, and stellate/myofi brillar cells ( 16 ). A response favoring ECM removal will involve increased expression of Mmps, decreased Timp expression, decreased interaction of Timps with Mmps, and increased IL-10 and IL-13Ra2 (decoy receptor for IL13) expression (supplementary Fig.  3 ) ( 53 ). The WD, however, induced both metalloproteases and protease inhibitors involved in ECM remodeling, whereas C 20-22 3 PUFA attenuated this response. As such, the impact of 3 PUFA on Mmp1a, Mmp2, and Mmp13 expression raises a concern for the capacity of these dietary lipids to promote remission of fi brosis in livers with preexisting fi brosis. Moreover, the WD + O diet induced IL-10, whereas the WD + D diet attenuated this response (supplementary Fig. 3 ). Although our study design was on prevention of diet-induced fi brosis, these concerted effects of DHA on factors involved in ECM removal raise the possibility that DHA may not be an effective NASH therapy that promotes remission of fi brosis. Further studies are required to quantify protein levels of Timps and Mmps, assess their activity, establish their subcellular localization, and examine protein-protein interaction. Such studies may reveal that C 20-22 3 PUFAs change the balance of ECM remodeling enzymes to favor removal of ECM.

Can 3 PUFA be used to treat human NASH?
Our mouse studies provide strong evidence in support of the use of DHA in the prevention of NAFLD/NASH. Omega-3 PUFAs have well-defi ned effects on hepatic lipid metabolism and infl ammation ( 29,73 ), and more recently their effects on hepatic fi brosis have been noted ( 17,28,69 ). Although several human studies have provided evidence in support of using supplemental 3 PUFAs to treat NAFLD (30)(31)(32)(33)(34)(35)(36)(37), other studies suggest there may be limitations for their use in NASH treatment ( 36,37 ). In trials using either fi sh oil or EPA-ethyl ester supplements, investigators report that 3 PUFA supplements failed to improve fi brosis scores. Because DHA attenuates fi brosis in two separate rodent models of liver injury and fi brosis (i.e., WD-fed mice and rats with bile duct ligation) ( 17,28,69 ), failure of 3 PUFAs to attenuate fi brosis in human clinical studies may be due to the type and dose of 3 PU-FAs used in the study. The fi nding that DHA suppressed TGF-␤ signaling and Col1A1 expression in mouse liver ( Figs. 3 and 7 ) and human stellate cells ( Fig. 8 ) suggests that DHA likely functions in humans.
DHA is the main 3-PUFA accumulating in human and rodent tissues. Although ␣ -linolenic acid and EPA are converted to DHA in humans, this process is insuffi cient for the accumulation of tissue levels of DHA needed to affect disease processes ( 74,75 ). Humans and rodents retro-convert ingested DHA to EPA. In Ldlr Ϫ / Ϫ mice, for example, WD supplementation with DHA leads to a 30% and 500% increase in hepatic phospholipid DHA and EPA content, respectively ( 18 ). In contrast, phospholipid DHA levels do not increase in mice fed the WD supplemented with EPA. Moreover, the JELIS trial, a primary prevention trial for