Serum metabolomic profiling reveals an increase in homocitrulline in Chinese patients with nonalcoholic fatty liver disease: a retrospective study

Study population and sample Collection

One hundred subjects, including 50 healthy controls and 50 NAFLD patients admitted to the outpatient clinic at Jinling Hospital in Nanjing, China, were enrolled in this study from January 2015 to December 2018. The inclusion criteria for the NAFLD group were as follows: (1) hepatic steatosis diagnosed by imaging; (2) age between 14 and 75 years; and (3) history of alcohol consumption of <210 g/week in men and <140 g/week in women over 2 years prior to the diagnosis of hepatic steatosis. The exclusion criteria for the NAFLD group were as follows: (1) positive results of a serum test for hepatitis B virus surface antigen and hepatitis C virus antibodies; (2) patients with alcoholic liver disease, drug-induced liver injury, total parenteral nutrition, hepatolenticular degeneration, autoimmune liver disease, and other specific diseases that can cause fatty liver (Chalasani et al., 2012); (3) patients who have taken nonsteroidal anti-inflammatory drugs, anticoagulants, antibiotics, and proton pump inhibitors in the past month; (4) patients who have lost weight through diet or vigorous exercise within the past month; (5) patients with serious diseases, such as heart, lung, brain, or kidney diseases; and (6) patients with malignant tumors or autoimmune diseases.

Blood samples were collected from the enrolled subjects after at least 12 h of fasting. Serum samples were obtained by centrifugation (3,500 rpm, 6 min) and divided into two parts; one part was stored at −80 °C until analysis, and another part was used for the detection of albumin, aspartate aminotransferase (AST), alanine aminotransferase (ALT), uric acid, triglycerides, fasting blood glucose, and total cholesterol at Nanjing Jinling Hospital Laboratory by a Hitachi 7600-110 automatic biochemical analyzer (Hitachi, Tokyo, Japan).

Serum sample pretreatment for LC-MS analysis

Before LC/MS analysis, 200 µL of serum samples, which were thawed at room temperature for 15 min, were mixed with 600 µL of methanol in the presence of 20 µg/mL DL-o-chlorophenyl alanine and vigorously vortexed for 30 s. The mixtures were centrifuged at 12,000 rpm for 15 min at 4 °C. A 200 µL aliquot of the supernatant was used for LC-MS analysis.

LC-MS analysis

The samples were analyzed on an Agilent LC-Q/TOF-MS system (Agilent Technologies, Santa Clara, CA, USA), which consisted of an Agilent 1290 liquid chromatography system and an Agilent 6530 time-of-flight mass spectrometer. The samples were injected onto an Agilent C18 particle column (100 ×2.1 mm, 1.8 µm). The injected sample volume was 4 μL, and the flow rate was 0.35 mL/min. The column temperature was maintained at 45 °C. Solvent A consisted of 0.1% formic acid in water, and solvent B consisted of 0.1% formic acid in acetonitrile. The gradient of the mobile phase is shown in Table S1. An Agilent 6530 Accurate-Mass Q-TOF/MS (Agilent Technologies, CA, USA) equipped with an electrospray ionization (ESI) source in both negative mode and positive mode was used to perform the mass spectrometry assays. Nitrogen was used as a nebulizer gas. The measurement conditions were as follows: capillary voltage, −3.5 kV in ESI− and 4 kV in ESI+; sampling cone voltage, 50 kV in ESI− and 35 kV in ESI +; dissolving gas flow rate, 700 L/h in ESI− and 600 L/h in ESI+; source temperature, 100 °C in ESI− and ESI+; dissolving gas temperature, 350 °C in ESI− and ESI+; cone gas flow rate, 50 L/h in ESI− and ESI+; and extraction cone voltage, 4 kV in ESI− and ESI+. Centroid data were collected from 50 to 1,000 m/z, and the scan time was 0.03 s with an interscan delay of 0.02 s. The pooled quality control (QC) samples were used to ensure the stability and repeatability of the HPLC-Q-TOF system. QC was a mixture of 10 µl of each sample and was staggered after every ten samples; thus, the stability of the instrument could be investigated based on the overlap of QC chromatograms. The total ion current (TIC) chromatograms of the QC samples overlapped, as shown in Fig. S1.

Statistical analysis

Two data sets from LC-QTOF/MS ESI+ and ESI− were used for peak selection, filtering, and filling by XCMS software. The differences in the metabolic features between the two groups were identified by the unsupervised method (principal component analysis, PCA) and supervised method (orthogonal partial least squares-discriminant analysis, OPLS-DA) using SIMCA-P software (Umetrics AB, Umea, Sweden). The Mann–Whitney U test was performed to identify differential metabolites between the two groups based on the false discovery rate (FDR). We matched the experimental tandem MS spectrum, retention time, and accurate mass of the metabolic features with spectral databases to identify the metabolites. Differential metabolites were characterized by search of an online database (HMDB) and comparison of mass spectra based on the mass-to-charge ratio or exact molecular mass. the SPSS software version 22.0 for Windows (SPSS, Inc., Chicago, IL, USA) was used to analyze the clinical data. Nonparametric data were analyzed using the Wilcoxon and Kruskal-Wallis tests and were expressed as the median with ranges, including ALT, AST, TG, TC, TBS and UA. Categorical data were analyzed using Fisher’s exact test, such as gender. Continuous variables, such as age, was analyzed using the two sample Student’s t-test and were expressed as mean ±standard deviation. MetaboAnalyst 4.0 (http://www.metaboanalyst.ca/) and Mbrole (http://csbg.cnb.csic.es/mbrole2/analysis.php) were used to perform metabolic pathway analysis. The significance level was set to a bilateral asymptotic p-value of <0.05.

Ethics and consent

Informed consent was obtained from all individuals included in this study. Research involving human subjects was approved by the Institutional Review Board of Jinling Hospital (2014NZKY-007-01).

Clinical characteristics of patients

In the present study, 100 serum samples were analyzed using LC-MS to determine the metabolic profiles of 50 healthy controls and 50 NAFLD patients. The biochemical parameters are summarized in Table 1. There were no significant differences in age between patients and the control subjects (p = 0.304). The levels of serum triglycerides, total cholesterol, uric acid, AST, and ALT were significantly higher (p < 0.001) in the NAFLD group than those in the control group. The fasting blood glucose (FBG) level was higher in the NAFLD group (p < 0.001). The number of men was higher in the NAFLD group than in the control group (80% vs 42%, p < 0.001). We used unsupervised PCA to analyze associations of the serum metabolic profiles with sex. The results showed that metabolic profiles were similar in males and females which were shown in Fig. 1.

Multivariate analysis of differences between the nafld and control groups

The matrix of detected peaks obtained using XCMS was used to perform a multivariate statistical analysis to detect the differences between the NAFLD and control groups. A total of 1,645 variables were included in the matrix in the positive ion mode and 1,463 variables were included in the negative ion mode. Then, the matrix was imported into SIMCA, and all variables were analyzed by unsupervised PCA to determine the general relationships between the two groups, as shown in Figs. 2A and 2B (R2X = 0.507, Q2 = 0.475 in the positive ion mode; R2X = 0.521, Q2 = 0.477 in the negative ion mode), revealing a clear separation trend. Then, orthogonal partial least squares-discriminant analysis (OPLS-DA) was performed, and the results indicated significant separations with valid model fitting (R2X = 0.47, R2Y = 0.899, Q2 = 0.867 in the positive ion mode; R2X = 0.485, R2Y = 0.882, Q2Y = 0.862 in the negative ion mode), as shown in Figs. 3A and 3B. The parameters of the OPLS model indicated that these results were able to reliably and predictably discriminate between the two groups because the R2Y values in the ESI+ and ESI− modes were >0.4. The model did not have an overfitting problem because the R2Y and Q2 values were high and the differences between the R2Y and Q2 values were lower than 0.2.

PCA showed a trend of separation of the groups on the score plot and was able to detect and exclude some outliers, which were defined as observations located outside the 95% confidence region of the model. In our study, NAFLD patients were clearly separated from the healthy controls. Moreover, OPLS-DA models indicated clear separations between the NAFLD and healthy control groups.

Identification of metabolites in the altered profiles

Thus, a panel of 89 variables significantly discriminated the NAFLD and healthy control groups (FDR <0.05), and the volcano plot (Figs. 4A and 4B) showed alterations in 53 metabolites in the ESI+ mode and 41 metabolites in the ESI− mode in serum from NAFLD patients. Furthermore, the serum concentrations of 55 metabolites were increased and the serum concentrations of 39 metabolites were decreased in NAFLD patients compared to those in healthy controls. The families of the changed metabolites contained glycerophospholipids, including phosphatidic acid (PA), phosphatidylcholines (PC), phosphatidylethanolamines (PE), phosphatidylserine (PS), lysophosphatidylethanolamine (LPE), phosphatidylglycerol (PG), cyclic phosphatidic acid (CPA), lysophosphatidylcholines (LPC), lysobisphosphatidic acids (LPA), phosphatidylinositol (PI), fatty acyls, amino acids, bile acids, organic acids, sphingomyelins, dipeptides, purines, and other metabolites. Analysis of all 89 variables indicated that the most abundant altered metabolite families were phosphatidylcholine ≈ phosphatidic acid >phosphatidylethanolamines >lysophosphatidylethanolamine >phosphatidylglycerol, as shown in Fig. 4C.

The error between the qualitative estimate of the compound and the actual molecular weight of the compound was described by Δppm. Metabolites with m/z within 5 ppm and retention time (RT) within 50 min were selected for further study. In total, 35 metabolites were accurately recognized based on this standard, as summarized in Table 2. Analysis of these metabolites indicated that the contents of amino acids, including L-homocitrulline and N-succinyl-L-diaminopimelic acid, were increased in the NAFLD group. The level of glutaconic acid, which was classified as dicarboxylic acid, was increased in the NAFLD group. The levels of fatty acid esters, including L-acetylcarnitine and propionylcarnitine, were increased in the NAFLD group. The changes in glycerophospholipids were variable because of diverse types of fatty acyl chains. Additionally, the changes in fatty acids and their conjugates were variable. The level of 2-isopropylmalic acid was decreased in the NAFLD group, and the level of 20-COOH-leukotriene B4 was increased in the NAFLD group. Table 3 shows the metabolites with higher area under the curve (AUC). Analysis of these metabolites indicated that L-acetylcarnitine, L-homocitrulline, and glutamic acid were the top 3 metabolites ranked by VIP score (VIP 1.9, 1.94, 1.9, respectively) and had favorable effective functions (AUC 0.9952, 95% CI [0.985–1.000]; 0.9908, 95% CI [0.966–1.000]; 0.9884 95% CI [0.973–1.000], respectively, p < 0.001) for diagnosis.

Pathway analysis of altered profiles

A total of 89 metabolites altered in the NAFLD group versus healthy control group were selected for metabolomic pathway analysis (MetPA). The relevant pathways for the NAFLD patients and healthy controls were visualized by an interactive visualization framework in Fig. 5. Metabolic pathways with the impact values >0.1 or −log(p) >10 was considered the most relevant pathways involved in the studied conditions [10]. In the present study, 14 metabolic pathways were selected as potential metabolic pathways for NAFLD patients and healthy controls based on their impact value, as shown in Table 4. In these pathways, some biological modules were involved in the lipid metabolism, including glycerophospholipid metabolism, linoleic acid metabolism, alpha-linolenic acid metabolism, and ether lipid metabolism. Some biological modules were involved in the carbohydrate metabolism, including pyruvate metabolism, glycolysis/gluconeogenesis, and glyoxylate and dicarboxylate metabolism. Glycerophospholipid metabolism was the most relevant pathway.