Metabolomics of alcoholic liver disease: a clinical discovery study

Abstract

Alcoholic liver disease (ALD) is associated with poor health, diseases and dysfunctions worldwide. Unfortunately, current biomarkers, including PCIII, IV-C, LN, and HA levels, are expensive and lack sensitivity in ALD detection. Because these biomarkers are invasive, time-consuming or expensive, ALD is usually diagnosed at late stages, for which there are no effective therapies. Thus, biomarkers for the early detection of ALD are urgently needed. Thankfully, metabolomics is a powerful technology that allows the assessment of global low-molecular-weight metabolites in a biological system and shows great potential in biomarker discovery. Analysis of the key metabolites in body fluids has become an important part of improving the diagnosis, prognosis, and therapy of diseases. Urine biomarkers may be a more attractive option, but none can currently detect ALD with the required accuracy. Herein, we describe our metabolomics approach for detecting ALD in a group of 206 patients. A total of six urinary differential metabolites that contributed to ALD progress were identified, and more importantly we discovered three of them with an accuracy of more than 95%. The biomarker panel may be sensitive to early diagnosis, prognosis and therapy of ALD.

---

1. Introduction

Alcohol misuse is a major public health problem worldwide and accounts for elevated social and economic costs.¹ Alcoholic liver disease (ALD), which is caused by alcohol consumption, is a common complication of alcohol misuse. By 2050, an estimated 165 million individuals worldwide will have ALD, consuming an estimated $120 trillion in health care costs per year.² Unfortunately, current biomarkers (including PCIII, IV-C, LN, and HA) for early disease detection are expensive and lack sensitivity in ALD detection.^3–6 The high cost associated with these technologies is a significant barrier to widespread use in clinical practice. They have only been shown to be effective for confirming the diagnosing of the diseases after the ALD symptoms have surfaced.⁷ The challenge is that currently there exists no way to identify which people are at risk of developing ALD. Thus, it is important to develop more effective methods for noninvasive early diagnosis of this disease process; moreover, there is an urgent need for biomarkers that diagnose ALD. Fortunately, metabolomics technology has been used to explore the particular metabolites, which are potentially diagnostic biomarkers for deep understanding of the essence of diseases.^8–10

At the end of the 20th century, genomics wrote out the ‘script of life’; proteomics decoded the script; and metabolomics came into bloom.¹¹ Metabolomics, the endpoints of genotype functions and biochemical phenotype in the body, are linked closely to alterations in the functions, and incorporates a ‘top-down’ strategy to reflect the terminal symptoms of a whole system and facilitate biomarker discovery.¹² Urine is an ideal bio-medium for disease study because it is readily available, easily obtained and less complex than other body fluids.^13,14 Ease of collection allows for serial sampling to monitor disease and therapeutic response. Numerous researchers have revealed the potential role of plasma metabolomics for investigating biomarkers predictive of therapeutic responses.¹⁵ Despite this expansion, there is no report of integrative study on urine metabolomics of ALD. In an attempt to address this issue, we used the urine metabolomics approach to analyze and detect ALD in a group of 206 ​samples. This study was performed by high-throughput UPLC/ESI-Q/TOF-MS metabolomics combined with pattern recognition analysis multiplatform, which were used to select the marker metabolites.

2. Materials and methods

2.1 Reagents

Acetonitrile and methanol were purchased from Fisher Scientific Corporation (New Jersey, USA). High purity formic acid (99%) was purchased from Honeywell Company (Morristown, New Jersey, USA). Water was produced by a Milli-Q Ultra-pure water system (Millipore, Billerica, USA); leucine enkephalin was purchased from Sigma-Aldrich. All the other reagents were of HPLC grade.

2.2 Human subjects

The clinical specimens, including 206 human urine samples from ALD patients and corresponding 101 normal urine samples, were obtained from the First Affiliated Hospital, Heilongjiang University of Chinese Medicine. The diagnoses of these samples were verified by pathologists. All the samples were obtained with the informed consent of the patients and approved by the Ethics Committee of Heilongjiang University of Chinese Medicine (HUCM-2013-3152) and the collection was conducted according to the Declaration of Helsinki.

2.3 Preparation of urine samples

Briefly, the urine samples were thawed on ice and vortexed at 10 000 rpm for 10 minutes at 4 °C to remove fine particulates. The supernatant was transferred to a glass vial and then filtered through a 0.22 μm syringe filter, and 3 μL of the supernatant were injected into liquid chromatography-electrospray ionization quadrupole time-of-flight mass spectrometer (LC-ESI-QTOF-MS) for analysis.

2.4 LC-ESI-QTOF-MS analysis

All the samples were analysed using an LC system (Waters Corp., Milford, USA), equipped with an ACQUITY BEH C₁₈ chromatography column (100 mm × 2.1 mm i.d., 1.7 μm, Waters Corporation, Milford, USA). The column temperature was maintained at 45 °C, and the gradient mobile phase was composed of phase A (water with 0.1% formic acid) and phase B (acetonitrile containing 0.1% formic acid). The gradient for the urine sample was as follows: 0–5 min, 1–55% B; 5–9.5 min, 55–50% B; 9.5–9.6 min, 50–1% B; 9.6–11 min, 1% B; 11–11.1 min, 1–99% B; 11.1–13 min, 99% B. The injection volume was 3 μL, and the flow rate of the LC system was 0.4 mL min⁻¹. After every 10 sample injections, a pooled blank (quality control) sample was injected to ensure the stability and repeatability of the LC-MS system. All the samples were maintained at 4 °C during the analysis.

Mass spectrometry was performed on a quadrupole time-of-flight (Q-TOF) instrument (Waters Corp., Milford, USA), operating in either negative (ESI⁻) or positive (ESI⁺) electrospray ionization mode with a capillary voltage of 3200 V in positive mode and 2800 V in negative mode and a sampling cone voltage of 30 V in both modes. The source temperature was set at 120 °C. The desolvation temperature was set at 350 °C, desolvation gas flow rate was set at 500 L h⁻¹, and cone gas flow was set at 50 L h⁻¹. Accurate mass was maintained by the introduction of a lock–spray interface of leucine-enkephalin (556.2771 [M + H]⁺ or 554.2615 [M − H]⁻) at a concentration of 0.2 ng mL⁻¹ at a flow rate of 100 μL min⁻¹. Data were acquired in centroid MS mode from 100 to 1500 m/z mass range for TOF-MS scanning as single injection per sample, and the batch acquisition was repeated to check experimental reproducibility. For the metabolomics profiling experiments, pooled quality control samples (generated by taking an equal aliquot of all the samples included in the experiment) were run at the beginning of the sample queue for column conditioning and every ten injections thereafter to assess inconsistencies that are particularly evident in large batch acquisitions in terms of retention time drifts and variation in ion intensity over time.

2.5 Statistical analyses

All the LC-MS raw files were converted to EZinfo software (which is included in MarkerLynx Application Manager and can be applied directly) for compound statistics (principal component analysis (PCA) and orthogonal partial least square discriminant analysis (OPLS-DA)), correlation analysis and compound validation. The combining VIP-plot from the OPLS-DA was carried out to select distinct variables as potential markers. Metabolite identification was determined as follows: first, the possible fragment mechanism was searched by MassFragment™ manager (Waters Corp., Milford, USA), which was used to facilitate the MS fragment ion analysis process by way of chemically intelligent peak-matching algorithms, and then the tandem mass was conducted and structure information of metabolites was matched from HMDB and METLIN; finally, the chemical structures of the candidate metabolites were confirmed. The classification performance of the selected metabolites was assessed using the area under the ROC curve (AUC). The ROC can be understood as a plot of the probability of classifying correctly the positive samples against the rate of incorrectly classifying true negative samples. The AUC measure of an ROC plot is a measure of predictive accuracy. All statistical analyses were performed using the Student's t-test. Differences with a P-value of 0.05 or less were considered significant.

3. Results

3.1 Urine metabolite profiles

In this study, a total of 206 human urine samples from ALD patients and corresponding 101 normal urine samples were collected and analyzed by LC-ESI-QTOF-MS. Fig. 1 presents the total ion current chromatograms of the control subjects and ALD patients in positive mode. BPI exhibited the ideal separation result under the optimized gradient procedure. For further analysis of the metabolic differences between the ALD and control group, all the raw data from LC/MS ions were imported into the EZinfo 2.0 package.

Fig. 1 Typical total ion current chromatograms of control subjects (up) and ALD patients (down).

3.2 Multivariate statistical analysis of metabolite profiles

Multivariate data analysis was performed using the score plot of PCA, and an obvious separation between the clustering of the ALD and control groups was observed (see Fig. 2A and B), which suggested that biochemical perturbation significantly occurred in the ALD group. For further analysis of feature ions, the S-plot combined VIP plot from the OPLS was used to select variables as potential markers for distinguishing ALD patients from controls (Fig. 2C and D). We generated VIP plots from the OPLS-DA with a threshold of 1.5 to identify the metabolites that significantly contribute to the clustering between the groups. Six differentially expressed metabolites from ALD patients were distinguished from those of the controls (P < 0.05, VIP > 1.5, two ions in the positive mode and four ions in negative mode). The VIP plot displayed six ions as differentiating metabolites according to their VIP values and considered them to be potential markers that represent the metabolic characteristics (Table 1). According to the protocol detailed above, six endogenous metabolites were finally identified as markers and are listed in Table 1. Overall, these metabolites displayed considerable differences between control and ALD samples, including sebacic acid, 3-hydroxytetradecanedioic acid, isocitric acid, suberic acid, isoamyl salicylate, and 6-methylquinoline. These metabolites were unambiguously identified using tandem mass spectrometry. It was found that, among them, four compounds were upregulated and two compounds were downregulated.

Fig. 2 Metabolomic profiling of alcoholic liver fibrosis. PCA score plots of urine samples collected from normal (green) and ALD (red) groups in positive ion mode (A) and negative ion mode (B). Panel (C) and (D) show the combination of VIP-score plots constructed from the supervised OPLS-DA analysis of urine in positive mode and negative ion mode, respectively.

Table 1 Putative metabolite markers resulting from ALD in positive and negative mode^a

No.	Compound ID	Adducts	Formula	Mass error (ppm)	m/z	Retention time (min)	Compound	Trend	VIP
a The markers were chosen on the basis of significant predictive value as determined by VIP-score plots constructed from the supervised OPLS-DA. Arrows indicate upregulation or downregulation in the comparison group as compared to the control participants.
1	HMDB00792	M − H	C10H18O4	−0.42482	201.1126	4.947166667	Sebacic acid	↓	1.7847
2	HMDB00394	M − H	C14H26O5	3.431132	273.1711	7.0795	3-Hydroxytetradecanedioic acid	↑	1.5822
3	HMDB00193	M − H	C6H8O7	−4.35433	191.0183	0.780166667	Isocitric acid	↑	1.5375
4	HMDB00893	M − H	C8H14O4	2.088976	173.0817	5.322166667	Suberic acid	↑	1.5035
5	HMDB40225	M + H	C12H16O3	−1.57926	209.1174	7.189166667	Isoamyl salicylate	↓	2.2786
6	HMDB33115	M + H	C10H9N	−3.11727	144.0809	2.340266667	6-Methylquinoline	↑	2.1561

---

3.3 Confirmation of clinical diagnosis for marker metabolites

We enrolled 206 ALD participants that met criteria (ESI Table 1†), aged 30 and older and otherwise healthy, into this 2 year observational study. The validation samples were obtained from these clinically defined ALD subjects. The clinical estimation of markers with ROC analysis was determined in urine samples from ALD patients and 101 volunteers to evaluate the metabolite profile for diagnosing ALD. We examined the urine samples from the participants for untargeted metabolomic analysis. The samples were processed and analyzed using the same UPLC-MS technique as in the discovery phase. Metabolomic profiling yielded 6200 positive-mode features and 4700 negative-mode features. A notable finding of this targeted metabolomic analysis was the identification of a set of six metabolites, comprising sebacic acid, 3-hydroxytetradecanedioic acid, isocitric acid, suberic acid, isoamyl salicylate, and 6-methylquinoline. Studies showed decreased urine sebacic acid level and increased 3-hydroxy tetradecanedioic acid and isocitric acid metabolites in patients with ALD.¹⁶ To detect the expression of a set of metabolites in human ALD urine, the LC-MS assay was performed on urine samples from ALD patients and that from healthy individual. The targeted quantitative analysis of the validation set revealed the levels for the six-metabolite panel (Fig. 3) as were observed in the discovery samples. We used receiver operating characteristic (ROC) analysis to assess the performance of the classifier models for group classification. The ROC analysis revealed 3-hydroxytetradecanedioic acid, isocitric acid, and sebacic acid to be potent discriminators of the control and ALD groups (Fig. 4). For the ALD group classification, the initial identified metabolites 3-hydroxytetradecanedioic acid, isocitric acid, and sebacic acid yielded a robust area under the curve (AUC) of 0.997 (Fig. 4A), 0.993 (Fig. 4B), and 0.978 (Fig. 4C) for ALD group classification. The predictive value of these biomarkers in preclinical patients is strong, suggesting that these markers may be useful for the confirmation of clinical diagnosis in the near future.

Fig. 3 Box plots show UPLC-MS relative signal intensities for metabolites in control and ALD groups. The targeted analysis of the six metabolites in the discovery phase and the application of the metabolite panel developed from the targeted discovery phase in the independent validation phase.

Fig. 4 ROC results for the metabolomics analyses. ROC plots represent sensitivity versus (1 − specificity), and the blue line represents the AUC.

4. Discussion

ALD is a major cause of alcohol-related morbidity and mortality.¹⁷ Analysis of the key metabolites in body fluids has become an important part of improving the diagnosis, prognosis, and therapy of diseases.¹⁸ First, in liver histology combined with biochemical results, we successfully established liver fibrosis in animal models. A panel of biomarkers to characterise disease could be useful for ALD diagnostics. In this study, LC-MS combined with pattern recognition analysis approach were used to simplify and quicken the identification of the metabolites of ALD. LC-MS based metabolomics could be an advanced tool to help us find metabolites with regards to its capacity of processing large datasets, and classifying of sample groups, as well as its indiscriminative nature of metabolites.^19,20 Using our metabolomics platform, PCA revealed a significant separation between the ALD and control samples. The OPLS model was built to find biomarkers of ALD, and six statistically important variables with VIP > 1.5 were defined, many of which were in various stages of progress in ALD. We used the metabolomic data from the untargeted analysis to build separate linear classifier models that would distinguish the control and ALD group.

We used ROC analysis to reveal 3-hydroxytetradecanedioic acid, isocitric acid, and sebacic acid to be potent discriminators between control and ALD groups. We found that these biomarkers were linked with the breakdown of the citrate cycle (TCA cycle), glyoxylate and dicarboxylate metabolism, and may give rise to subtle and early changes. Taking these biomarkers as screening indexes, the biomarker panel was validated in a cohort of normal control subjects and subjects with ALD with an AUC of >95%. To the best of our knowledge, this is the first published report of a urine-based biomarker panel with very high accuracy for detecting preclinical ALD. Herein, we present the discovery and validation of urine metabolite changes that distinguish normal participants from ALD in the near future. The accuracy for detection is high, as well as urine is easier to obtain and costs less to acquire, making this method more useful for screening large-scale clinical trials and for future clinical use. This biomarker panel requires external validation using similar rigorous clinical classification before further development for clinical use.

In this study, LC/MS urine metabolomics has been successfully established for biomarker studies in ALD. Herein, we describe our metabolomics approach for detecting ALD disease in a group of 206 ​samples. In conclusion, a total of six urinary differential metabolites that contributed to ALD progress were identified, and more importantly we discovered and validated three of them from urine with an accuracy of more than 95%. The biomarker panel may be sensitive to the early diagnosis of ALD disease. This is the first published report of a urine-based biomarker panel with very high accuracy for detecting ALD.

Conflict of interest

The authors declare no competing financial interests.

Abbreviations

ALD	Alcoholic liver disease
OPLS	Orthogonal projection to latent structures
PCA	Unsupervised principal component analysis
PLS-DA	Supervised partial least squares-discriminant analysis

Acknowledgements

This study was supported by grants from the Key Program of Natural Science Foundation of State (Grant No. 81470196, 81302905) and the Natural Science Foundation of Heilongjiang Province of China (H2015038).

References

1. O. George, C. Sanders, J. Freiling, E. Grigoryan, S. Vu, C. D. Allen, E. Crawford, C. D. Mandyam and G. F. Koob, Recruitment of medial prefrontal cortex neurons during alcohol withdrawal predicts cognitive impairment and excessive alcohol drinking, Proc. Natl. Acad. Sci. U. S. A., 2012, 109(44), 18156–18161.
2. N. A. Nasrallah, J. J. Clark, A. L. Collins, C. A. Akers, P. E. Phillips and I. L. Bernstein, Risk preference following adolescent alcohol use is associated with corrupted encoding of costs but not rewards by mesolimbic dopamine, Proc. Natl. Acad. Sci. U. S. A., 2011, 108(13), 5466–5471.
3. S. F. Kendrick, G. O'Boyle, J. Mann, M. Zeybel, J. Palmer, D. E. Jones and C. P. Day, Acetate, the key modulator of inflammatory responses in acute alcoholic hepatitis, Hepatology, 2010, 51(6), 1988–1997.
4. R. P. Mookerjee, M. Malaki, N. A. Davies, S. J. Hodges, R. N. Dalton, C. Turner, S. Sen, R. Williams, J. Leiper, P. Vallance and R. Jalan, Increasing dimethylarginine levels are associated with adverse clinical outcome in severe alcoholic hepatitis, Hepatology, 2007, 45(1), 62–71.
5. W. R. Kim, J. B. Gross Jr, J. J. Poterucha, G. R. Locke 3rd and E. R. Dickson, Outcome of hospital care of liver disease associated with hepatitis C in the United States, Hepatology, 2001, 33(1), 201–206.
6. A. Sreekumar, L. M. Poisson, T. M. Rajendiran, A. P. Khan, Q. Cao, J. Yu, B. Laxman, R. Mehra, R. J. Lonigro, Y. Li, M. K. Nyati, A. Ahsan, S. Kalyana-Sundaram, B. Han, X. Cao, J. Byun, G. S. Omenn, D. Ghosh, S. Pennathur, D. C. Alexander, A. Berger, J. R. Shuster, J. T. Wei, S. Varambally, C. Beecher and A. M. Chinnaiyan, Metabolomic profiles delineate potential role for sarcosine in prostate cancer progression, Nature, 2009, 457, 910–914.
7. K. Suhre, S. Y. Shin, A. K. Petersen, R. P. Mohney, D. Meredith, B. Wägele, E. Altmaier, D. P. CARDIoGRAM, J. Erdmann, E. Grundberg, C. J. Hammond, M. H. de Angelis, G. Kastenmüller, A. Köttgen, F. Kronenberg, M. Mangino, C. Meisinger, T. Meitinger, H. W. Mewes, M. V. Milburn, C. Prehn, J. Raffler, J. S. Ried, W. Römisch-Margl, N. J. Samani, K. S. Small, H. E. Wichmann, G. Zhai, T. Illig, T. D. Spector, J. Adamski, N. Soranzo, C. Gieger, T. L. Assimes, P. Deloukas, J. Erdmann, H. Holm, S. Kathiresan, I. R. König, R. McPherson, M. P. Reilly, R. Roberts, N. J. Samani, H. Schunkert and A. F. Stewart, Human metabolic individuality in biomedical and pharmaceutical research, Nature, 2011, 477, 54–60.
8. Q. Liang, C. Wang, B. Li and A. H. Zhang, Metabolic fingerprinting to understand therapeutic effects and mechanisms of silybin on acute liver damage in rat, Pharmacogn. Mag., 2015, 11(43), 586–593.
9. A. K. Arakaki, J. Skolnick and J. F. McDonald, Marker metabolites can be therapeutic targets as well, Nature, 2008, 456(7221), 443.
10. T. J. Wang, M. G. Larson, R. S. Vasan, S. Cheng, E. P. Rhee, E. McCabe, G. D. Lewis, C. S. Fox, P. F. Jacques, C. Fernandez, C. J. O'Donnell, S. A. Carr, V. K. Mootha, J. C. Florez, A. Souza, O. Melander, C. B. Clish and R. E. Gerszten, Metabolite profiles and the risk of developing diabetes, Nat. Med., 2011, 17(4), 448–453.
11. G. J. Patti, O. Yanes, L. P. Shriver, J. P. Courade, R. Tautenhahn, M. Manchester and G. Siuzdak, Metabolomics implicates altered sphingolipids in chronic pain of neuropathic origin, Nat. Chem. Biol., 2012, 8(3), 232–234.
12. M. Tomita and K. Kami, Cancer. Systems biology, metabolomics, and cancer metabolism, Science, 2012, 336(6084), 990–991.
13. J. P. Ioannidis and M. J. Khoury, Improving validation practices in "omics" research, Science, 2011, 334(6060), 1230–1232.
14. Q. Liang, C. Wang and B. Li, Metabolomic Analysis Using Liquid Chromatography/Mass Spectrometry for Gastric Cancer, Appl. Biochem. Biotechnol., 2015, 176(8), 2170–2184.
15. M. Mapstone, A. K. Cheema, M. S. Fiandaca, X. Zhong, T. R. Mhyre, L. H. Macarthur, W. J. Hall, S. G. Fisher, D. R. Peterson, J. M. Haley, M. D. Nazar, S. A. Rich, D. J. Berlau, C. B. Peltz, M. T. Tan, C. H. Kawas and H. J. Federoff, Plasma phospholipids identify antecedent memory impairment in older adults, Nat. Med., 2014, 20(4), 415–418.
16. S. K. Manna, A. D. Patterson, Q. Yang, K. W. Krausz, J. R. Idle, A. J. Fornace and F. J. Gonzalez, UPLC-MS-based urine metabolomics reveals indole-3-lactic acid and phenyllactic acid as conserved biomarkers for alcohol-induced liver disease in the Ppara-null mouse model, J. Proteome Res., 2011, 10(9), 4120–4133.
17. A. H. Zhang, H. Sun, Y. Han, G. L. Yan, Y. Yuan, G. C. Song, X. X. Yuan, N. Xie and X. J. Wang, Ultraperformance liquid chromatography-mass spectrometry based comprehensive metabolomics combined with pattern recognition and network analysis methods for characterization of metabolites and metabolic pathways from biological data sets, Anal. Chem., 2013, 85(15), 7606–7612.
18. K. T. Kotz, W. Xiao, C. Miller-Graziano, W. J. Qian, A. Russom, E. A. Warner, L. L. Moldawer, A. de, P. E. Bankey, B. O. Petritis, D. G. Camp 2nd, A. E. Rosenbach, J. Goverman, S. P. Fagan, B. H. Brownstein, D. Irimia, W. Xu, J. Wilhelmy, M. N. Mindrinos, R. D. Smith, R. W. Davis, R. G. Tompkins and M. Toner, Inflammation and the Host Response to Injury Collaborative Research Program. Clinical microfluidics for neutrophil genomics and proteomics, Nat. Med., 2010, 16(9), 1042–1047.
19. A. Zhang, H. Sun, Y. Han, G. Yan and X. Wang, Urinary metabolic biomarker and pathway study of hepatitis B virus infected patients based on UPLC-MS system, PLoS One, 2013, 8(5), e64381.
20. T. J. Wang, M. G. Larson, R. S. Vasan, S. Cheng, E. P. Rhee, E. McCabe, G. D. Lewis, C. S. Fox, P. F. Jacques, C. Fernandez, C. J. O'Donnell, S. A. Carr, V. K. Mootha, J. C. Florez, A. Souza, O. Melander, C. B. Clish and R. E. Gerszten, Metabolite profiles and the risk of developing diabetes, Nat. Med., 2011, 17(4), 448–453.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5ra13417j

---