Polyunsaturated fatty acid metabolites as novel lipidomic biomarkers for noninvasive diagnosis of nonalcoholic steatohepatitis.

Lipotoxicity is a key mechanism thought to be responsible for the progression of nonalcoholic fatty liver (NAFL) to nonalcoholic steatohepatitis (NASH). Noninvasive diagnosis of NASH is a major unmet clinical need, and we hypothesized that PUFA metabolites, in particular arachidonic acid (AA)-derived eicosanoids, in plasma would differentiate patients with NAFL from those with NASH. Therefore, we aimed to assess the differences in the plasma eicosanoid lipidomic profile between patients with biopsy-proven NAFL versus NASH versus normal controls without nonalcoholic fatty liver disease (NAFLD; based on MRI fat fraction <5%). We carried out a cross-sectional analysis of a prospective nested case-control study including 10 patients with biopsy-proven NAFL, 9 patients with biopsy-proven NASH, and 10 non-NAFLD MRI-phenotyped normal controls. We quantitatively compared plasma eicosanoid and other PUFA metabolite levels between NAFL versus NASH versus normal controls. Utilizing a uniquely well-characterized cohort, we demonstrated that plasma eicosanoid and other PUFA metabolite profiling can differentiate between NAFL and NASH. The top candidate as a single biomarker for differentiating NAFL from NASH was 11,12-dihydroxy-eicosatrienoic acid (11,12-diHETrE) with an area under the receiver operating characteristic curve (AUROC) of 1. In addition, we also found a panel including 13,14-dihydro-15-keto prostaglandin D2 (dhk PGD2) and 20-carboxy arachidonic acid (20-COOH AA) that demonstrated an AUROC of 1. This proof-of-concept study provides early evidence that 11,12-diHETrE, dhk PGD2, and 20-COOH AA are the leading eicosanoid candidate biomarkers for the noninvasive diagnosis of NASH.

Defi nition of NASH. Patients with biopsy-confi rmed NAFLD who had predominantly zone-3 macrovesicular steatosis and lobular infl ammation and the presence of classic ballooning degeneration were classifi ed as having NASH.
Inclusion criteria for the NAFLD. Inclusion criteria included age at least 18 years during the consent process, ability and willingness to give written informed consent, minimal or no alcohol use history consistent with NAFLD (see Exclusion criteria), and collection of plasma within 90 days of the liver biopsy.
Exclusion criteria. Clinical or histologic evidence of alcoholic liver disease included the following: regular and excessive use of alcohol within the 2 years prior to interview defi ned as alcohol intake >14 drinks per week in a man or >7 drinks per week in a woman. Approximately 10 g of alcohol equals one "drink" unit. One unit equals 1 ounce of distilled spirits, one 12 ounce beer, or one 4 ounce glass of wine. Secondary causes of hepatic steatosis included previous surgeries, bariatric surgery, total parenteral nutrition, short bowel syndrome, steatogenic medications, evidence of chronic hepatitis B as marked by the presence of Hepatitis B surface antigen in serum, evidence of chronic hepatitis C as marked by the presence of anti-Hepatitis C virus antibody (HCV) or HCV RNA in serum, evidence of other causes of liver disease (such as ␣ -1-antitrypsin defi ciency, Wilson disease, glycogen storage disease, dysbetalipoproteinemia, known phenotypic hemochromatosis, autoimmune liver disease, or drug-induced liver injury), or concomitant severe underlying systemic illness that in the opinion of the investigator would interfere with the study.
Defi nition of normal controls. A novel aspect of this study was the inclusion of a uniquely well-characterized non-NAFLD normal control group. Participants were classifi ed as normal non-NAFLD by accurate hepatic fat quantifi cation by MRI-PDFF-derived fat fraction of <5% ( 18,20 ). Liver biopsy is unethical in normal individuals. Other noninvasive measures such as ultrasound and computed tomography are inaccurate and lack sensitivity especially at liver fat fraction between 1% and 10%. Therefore, MRI-PDFF was utilized in this study for accurate diagnosis of absence of hepatic steatosis. MRI-PDFF is highly accurate, sensitive, reproducible, and precise. The detailed description of MRI-PDFF protocol has been published previously (18)(19)(20)(22)(23)(24)(25).
Exclusion criteria included 1 ) age less than 18 years; 2 ) significant systemic illness; 3 ) inability to undergo MRI; and 4 ) evidence of possible liver disease, including any previous liver biopsy, positive hepatitis B surface antigen, hepatitis C viral RNA, or autoimmune serologies, ␣ -1 antitrypsin defi ciency, hemochromatosis genetic testing, or low ceruloplasmin.

Lipid extraction
Plasma samples for lipidomic profi ling were obtained within 90 days of the liver biopsy and MRI-PDFF for cases and controls, respectively. All plasma samples were stored at Ϫ 80°C, thawed once, and immediately used for free fatty acid and eicosanoid isolation as described previously ( 15,17 ). Briefl y, 50 µl plasma was spiked with a cocktail of 26 deuterated internal standards (individually purchased from Cayman Chemicals, Ann Arbor, MI) and brought to a volume of 1 ml with 10% methanol. The samples were then purifi ed by solid phase extraction on Strata-X columns (Phenomenex, Torrance, CA), using an activation procedure consisting of consecutive washes lipidomics techniques, our laboratory has developed a robust and comprehensive approach to the lipidomics analysis of hundreds of fatty acids, acylethanolamines, and infl ammatory eicosanoids, including their numerous metabolites arising from an array of cyclooxygenases, lipoxygenases, cytochrome P450s, and nonenzymatic oxidation-producing isoprostanes, as well as combinations thereof ( 15 ). Particular attention has been focused on the eicosanoids derived from arachidonic acid (AA), and we can now routinely quantify >150 such metabolites and have used this approach to profi le AA and other PUFAs as well as their metabolites in human plasma (15)(16)(17). We have now applied this approach to analyze the plasma of NAFLD patients.
The aim of this proof-of-concept study was to detect if plasma eicosanoid profiling can differentiate wellcharacterized patients with biopsy-proven NASH versus NAFL versus uniquely phenotyped normal controls by documenting liver fat content of <5% by proton-density-fatfraction (MRI-PDFF), a novel MRI-based method.

Study design and participants
This study was a cross-sectional analysis derived from a prospective nested case-control study including three groups of uniquely phenotyped patients with biopsy-proven NAFLD (including NASH and NAFL) and normal non-NAFLD controls. All participants were derived from the University of California at San Diego (UCSD) NAFLD Research Clinic and were seen between January 2011 and November 2012 (18)(19)(20). All participants provided written informed consent and underwent a detailed standardized clinical research visit including medical history, alcohol use and quantifi cation history (using Audit and Skinner questionnaire), physical examination, anthropometrics, fasting biochemical tests, and detailed exclusion of other causes of liver disease (see the inclusion and exclusion criteria described subsequently). A fasting plasma sample was collected in the morning of the clinical research visit and stored at Ϫ 80°C freezer housed in the UCSD NAFLD Translational Research Unit. The study was approved by the UCSD Human Subjects Institutional Review Board.

Description of cohort
All cases of NAFLD included in this study had a liver biopsyconfi rmed diagnosis of NAFLD.
Histologic description. Biopsy was scored by an experienced liver pathologist who was blinded to clinical data and lipidomic and imaging data. The NASH Clinical Research Network histologic scoring system was used to score biopsies ( 21 ). NAFLD activity score (NAS) and fi brosis score were recorded for all patients. NAS score ranges from 0 to 8 and is the summation of the degree of steatosis (0-3), lobular infl ammation (0-3), and hepatocellular ballooning (0-2). Liver fi brosis ranges from 0 to 4 with 0 being no fi brosis and 4 indicating cirrhosis.
Defi nition of NAFL. Patients with biopsy-confi rmed NAFLD who had predominantly zone-3 macrovesicular steatosis with or without minimal infl ammation, absence of ballooning degeneration, and no fi brosis were classifi ed as having NAFL. the eicosanoid precursor ions was achieved with nitrogen as the collision gas with the declustering potential, entrance potential, and collision energy optimized for each metabolite. Eicosanoids were identifi ed by matching their MRM signal and chromatographic retention time with those of pure identical standards.

Quantitation of lipids
Eicosanoids and free fatty acids were quantitated by the stable isotope dilution method. Briefl y, identical amounts of deuterated internal standards were added to each sample and to all the primary standards used to generate standard curves. To calculate the amount of eicosanoids and free fatty acids in a sample, ratios of peak areas between endogenous metabolite and matching deuterated internal standards were calculated. Ratios were converted to absolute amounts by linear regression analysis of standard curves generated under identical conditions.

Statistical analysis
The Chi-square ( 2 ) test was used for comparisons between categorical variables, and the t -test was used for comparisons between continuous variables. We examined differences in the plasma eicosanoid profi les between normal controls, patients with biopsyproven mild NAFL, and patients with biopsy-proven NASH. Finally, we examined the diagnostic accuracy of nine biomarkers that yielded signifi cant differences as biomarkers to differentiate NAFL from NASH. A two-tailed P value р 0.05 was considered statistically signifi cant. Statistical analyses were performed using the SAS statistical software package version 9.4 (SAS Inc., Cary, NC). with 3 ml of 100% methanol followed by 3 ml of water. The eicosanoids were then eluted with 1 ml of 100% methanol, and the eluent was dried under vacuum, dissolved in 50 µl of buffer A [consisting of water-acetonitril-acetic acid, 60:40:0.02 (v/v/v)], and immediately used for analysis as follows: For free fatty acids analysis, 50 l of plasma was spiked with deuterated fatty acid standards, and the free fatty acids were isolated by selective extraction with methanol and isooctane. The extracted fatty acids were derivatized and analyzed by gas chromatography and MS, as described ( 15 ).

Reverse-phase LC/MS
Eicosanoids in plasma were analyzed and quantifi ed by LC/ MS/MS as previously described ( 17,26 ). Briefl y, eicosanoids were separated by reverse-phase chromatography using a 1.7 M 2.1 × 100 mm BEH Shield Column (Waters, Milford, MA) and an Acquity UPLC system (Waters). The column was equilibrated with buffer A, and 5 µl of sample was injected via the autosampler. Samples were eluted with a step gradient starting with 100% buffer A for 1 min, then to 50% buffer B (consisting of 50% acetonitril, 50% isopropanol, and 0.02% acetic acid) over a period of 3 min, and then to 100% buffer B over a period of 1 min. The LC was interfaced with an IonDrive Turbo V ion source, and mass spectral analysis was performed on a triple quadrupole AB SCIEX 6500 QTrap mass spectrometer (AB SCIEX, Framingham, MA). Eicosanoids were measured using electrospray ionization in negative ion mode and multiple reaction monitoring (MRM) using the most abundant and specifi c precursor ion/product ion transitions to build an acquisition method capable of detecting 158 analytes and 26 internal standards. The ionspray voltage was set at Ϫ 4,500 V at a temperature of 550°C. Collisional activation of The P values in bold are statistically signifi cant ( P р 0.05). Differences between groups evaluated with t -test. Alk P, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; Chol, cholesterol; D Bili, direct bilirubin; GGT, gammaglutamyl transferase; Hba1c, hemoglobin a1c; PT, protime; T Bili, total bilirubin; WBC, white blood count. stress contribute to disease progression from steatosis with relatively benign outcome to NASH with risk of cirrhosis and hepatocellular carcinoma. Here we used LC/MS to profi le and quantitate bioactive lipids and lipid peroxidation products in circulation that are characteristic of hepatic infl ammation in NASH patients.

Cohort demographics
This study included 19 patients with NAFLD (10 NAFL cases and 9 cases of NASH) and 10 non-NAFLD normal controls. The detailed baseline characteristics including demographics, BMI, biochemical tests, lipid profi le, MRI-PDFF for controls, and liver biopsy data on patients with NAFLD are described in Table 1 . Non-NAFLD controls were younger, had lower BMI, and had lower serum ALT, AST, GGT, and glucose and insulin levels as expected. Routine liver-related and metabolic tests did not significantly differ between NAFL versus NASH ( Table 1 ), except that plasma triglycerides were marginally higher in patients with NASH. Compared with patients with NAFL, patients with NASH had more severe liver histology with a higher degree of steatosis, ballooning degeneration, lobular infl ammation, and fi brosis.

PUFA and metabolite lipidomics profi ling
At present, there are no noninvasive biomarkers with suffi cient specifi city to distinguish NASH from other fatty liver states. Liver biopsy remains the benchmark to reliably identify NAFL and NASH, but the procedure is invasive and carries certain risks. Thus, there is great demand from the clinical community for the development of noninvasive procedures capable of accurately characterizing and staging NAFLD, as that furnishes valuable information on treatment options and prognosis. Infl ammation and oxidative soluble epoxide hydrolase (sEH) on epoxyeicosatrienoic acids, which are the primary products of the epoxygenase pathway of CYP on the initial substrate AA. A number of biological effects have been ascribed to epoxyeicosatrienoic acids including cardioprotective vasodilation and leukocyte antimigratory and anti-infl ammatory actions ( 28 ). Conversion of the epoxides to their corresponding diols by the sEH decreases their functional levels and thereby diminishes the associated health benefi ts. Similarly, 20-HETE, an AA metabolite synthesized by the CYP hydroxylase pathway, is reported to have important vasoactive properties ( 29 ). We did not detect 20-HETE in plasma, but we found a consistent increase of 20-carboxy arachidonic acid (20-COOH AA) in NASH samples. However, it did not reach statistical significance when comparing NAFL versus NASH ( Table 2 ). The conversion of 20-HETE to 20-COOH AA is catalyzed by CYP enzymes and is responsible for reduced bioactivity.

Identifi cation of a panel of eicosanoids as a diagnostic tool for detecting NASH
Based on Table 2 , we found nine biomarkers to be signifi cant in the assessment of NAFLD. We assessed their individual diagnostic test performances using AUROC human platelets, were detected at low levels in the control samples but were signifi cantly higher in both the NAFL and NASH samples; however, no signifi cant difference was found between NAFL and NASH. In contrast, prostaglandin E 2 (PGE2) was elevated only in the NASH samples, and no differences were observed between the controls and NAFL ( Fig. 1 ). Prostaglandin D 2 (PGD2) was not detected in any of the samples, but the degradation product 13,14-dihydro-15-keto PGD2 (dhk PGD2) was signifi cantly higher in NASH compared with NAFL ( P value <0.0011) or control ( P value <0.0002) (see Fig. 1 ).
LOX-derived metabolites were also increased in NAFLD. Of note, while the AA-derived products of 5-LOX and 12/15-LOX pathways appear to be highest in NAFL ( Fig.  1 ), the metabolites from related PUFAs including linoleic acid, ␣ -linolenic acid, dihomo-␥ -linolenic acid, EPA, and DHA were generally higher in NASH ( Fig. 2 ). Similarly, the AA-derived metabolites of the CYP pathway were predominantly elevated in NASH but unchanged in NAFL compared with healthy controls ( Figs. 1, 2 ). In particular, 11,12-dihydroxy-eicosatrienoic acid (11,12-diHETrE) and 14,15-dihydroxy-eicosatrienoic acid (14, were signifi cantly elevated in NASH compared with NAFL or controls. These metabolites are produced by the action of  Table 2 . PGD2, and 20-COOH AA were greatly elevated in NASH patients versus NAFL versus normal controls in a dosedependent manner. This proof-of-concept study provides early evidence that 11,12-diHETrE, dhk PGD2, and 20-COOH AA are the leading eicosanoid candidate biomarkers for the noninvasive diagnosis of NASH. Because these fi ndings are derived from a pilot, single-center study, further validation studies should include testing of all three potential biomarkers. Thus, these novel fi ndings, although preliminary in nature, provide the justifi cation to carry out a large, well-designed, multicenter biomarker validation study in noninvasive detection of NASH.

In the context of the published literature
Our results were different than those that were observed by Feldstein and colleagues due to differences in the assessment of the metabolites. Feldstein and colleagues ( 13 ) measured total metabolites, that is, both free and those esterifi ed to phospholipids, triglycerides, and so forth. For this they treated plasma with concentrated KOH at high pH to hydrolyze the complex lipids and to liberate the oxidized fatty acids that were then measured. Our approach was completely different; we measured the metabolites that were present in their free form and without KOH treatment. Even though the levels of the free and reported them in Table 3 . The top candidate as a single biomarker for differentiating NAFL from NASH was 11,12-diHETrE with an AUROC of 1. In addition, we also found a panel including dhk PGD2 and 20-COOH AA that demonstrated an AUROC of 1. These novel, hitherto unrecognized biomarkers need to be confi rmed in larger studies and need to be validated.

Main fi ndings
Utilizing a uniquely well-characterized cohort of patients with biopsy-confi rmed NAFLD including NAFL and NASH, as well as non-NAFLD normal controls, we showed herein that plasma eicosanoid profi ling can differentiate between NAFL and NASH. The only previous PUFA metabolites suggested to be elevated in NAFLD over controls were observed for the total plasma content of 9-HODE and 13-HODE, derived by nonenzymatic autooxidation of linoleic acid and linolenic acid, respectively ( 13 ). Although in the present study, we discovered several eicosanoid moieties that were signifi cantly different between normal versus NAFL versus NASH, 11,12-diHETrE, dhk  Table 2 .
Plasma eicosanoid profi ling may provide a novel biomarker candidate for noninvasive detection of NASH. Further research is needed to assess the accuracy and reliability of these candidate biomarkers obtained by a relatively noninvasive blood lipidomic profi ling for the diagnosis of NASH. Specifi cally, it will be important to establish whether a single eicosanoid biomarker will have suffi cient diagnostic capacity or whether it should be used along with other eicosanoids as a panel of biomarkers. Large, multicenter, longitudinal clinical trials will be needed to assess the utility of these lipidomic biomarkers for diagnosis as well as clinical follow-up, and eventually to document whether they predict long-term clinically meaningful outcomes such as development of cirrhosis, hepatocellular carcinoma, or liver-related death.
eicosanoids are much lower than the eicosanoids esterifi ed to lipids, our approach captures many more metabolites and allows for a much broader profi ling strategy. We know from recent studies that KOH destroys all prostaglandins as well as a number of LOX-derived metabolites, thus, it is plausible that Feldstein et al. missed these metabolites in their experiments.
In NAFLD, normal lipid metabolism is disrupted leading to increased levels of free fatty acids and triglyceride synthesis (30)(31)(32). Free fatty acids have been shown to elicit hepatotoxicity and may stimulate the progression from NAFL to NASH via several mechanisms ( 12 ). They can be directly cytotoxic and stimulate the production of infl ammatory pathways in hepatocytes ( 11,12 ). These fatty acids also serve as precursors for infl ammatory eicosanoids. Consistent with this, the plasma levels of several free PUFAs were consistently higher in NAFL and NASH compared with healthy controls ( Figs. 1, 2 ). However, there were no signifi cant differences between NAFL and NASH suggesting that plasma free fatty acids are poor markers for differentiating between the various stages of NAFLD. By contrast, their conversion to eicosanoids may constitute a critical mechanism in disease progression, and the analysis of eicosanoid levels, rather than fatty acid levels, may be a useful clinical tool to discriminate between NAFL and NASH.
The strengths of the study include utilization of wellcharacterized patients with biopsy-confi rmed NAFLD, including patients with NAFL and NASH, as well as MRI-PDFF-confi rmed non-NAFLD controls; plasma having been stored at Ϫ 80°C under identical conditions for all the participants included in the study; and utilization of uniform criteria for the diagnosis of NAFLD. Limitations include the cross-sectional nature of the study, the small sample size of this proof-of-concept study, and the lack of a large validation cohort. We plan to undertake a large, longitudinal validation cohort study in the future to confi rm these fi ndings. Further studies are needed to assess the association between the gene expression of enzymes and their plasma metabolite concentration differentiating patients with NAFL from NASH. Finally, additional studies are needed to assess the association of these biomarkers in differentiating primary NAFLD from secondary causes of NAFLD such as viral hepatitis and alcohol use.