Abstract
Severe alcoholic hepatitis (SAH) has a high mortality rate, and corticosteroid therapy is effective in 60% patients. This study aimed to investigate a baseline metabolic phenotype that could help stratify patients not likely to respond to steroid therapy and to have an unfavorable outcome. Baseline urine metabolome was studied in patients with SAH using ultra‐high performance liquid chromatography and high‐resolution mass spectrometry. Patients were categorized as responders (Rs, n = 52) and nonresponders (NRs, n = 8) at day 7 according to the Lille score. Multivariate projection analysis identified metabolites in the discovery cohort (n = 60) and assessed these in a validation cohort of 80 patients (60 Rs, 20 NRs). A total of 212 features were annotated by using metabolomic/biochemical/spectral databases for metabolite identification. After a stringent selection procedure, a total of nine urinary metabolites linked to mitochondrial functions significantly discriminated nonresponders, most importantly by increased acetyl‐L‐carnitine (12‐fold), octanoylcarnitine (4‐fold), decanoylcarnitine (4‐fold), and alpha‐ketoglutaric acid (2‐fold) levels. Additionally, urinary acetyl‐L‐carnitine and 3‐hydroxysebasic acid discriminated nonsurvivors (P < 0.01). These urinary metabolites significantly correlated to severity indices and mortality (r > 0.3; P < 0.01) and were associated with nonresponse (odds ratio >3.0; P < 0.001). In the validation cohort, baseline urinary acetyl‐L‐carnitine documented an area under the receiver operating curve of 0.96 (0.85‐0.99) for nonresponse prediction and a hazard ratio of 3.5 (1.5‐8.3) for the prediction of mortality in patients with SAH. Acetyl‐L‐carnitine at a level of >2,500 ng/mL reliably segregated survivors from nonsurvivors (P < 0.01, log‐rank test) in our study cohort. Conclusion: Urinary metabolome signatures related to mitochondrial functions can predict pretherapy steroid response and disease outcome in patients with SAH. (Hepatology Communications 2018;2:628‐643)
Abbreviations
- AUROC
area under the receiver operating characteristic
- ELISA
enzyme‐linked immunosorbent assay
- GPCR
G‐protein‐coupled receptor
- MDF
Maddrey's discriminant function
- MELD
Model for End‐Stage Liver disease
- MS
mass spectrometry
- MSTUS
mass spectrum total useful signal
- NR
nonresponder to corticosteroid
- PBMC
peripheral blood mononuclear cell
- R
responder to corticosteroids
- rs
regression coefficients
- ROC
receiver operating characteristic
- SAH
severe alcoholic hepatitis
Alcoholic hepatitis is a common ailment and is associated with systemic inflammatory response syndrome, organ failure, and short‐term mortality of up to 50%.1 The pathophysiology of severe alcoholic hepatitis (SAH), however, is poorly understood because of the lack of appropriate animal models and limited translational studies.2, 3 Severity of SAH is assessed based on histologic features, although many noninvasive scoring systems, such as Maddrey's discriminant function (MDF) ≥32 and the Model for End‐Stage Liver Disease (MELD) score,2, 3 have been developed for prognostication of SAH. It is important to identify patients with SAH at a high risk of mortality before considering specific therapies. Corticosteroid therapy, although controversial, remains the only option to improve the morbidity and short‐term mortality in SAH.2, 4, 5, 6, 7, 8, 9 While the precise mechanisms of action of steroids in SAH are unknown, inhibition of inflammatory reactions and immune‐mediated hepatic destruction play a dominant role.8 However, the anti‐anabolic effects of steroids may suppress hepatic regeneration and healing.10 Corticosteroid therapy can prove deleterious in patients with clinical manifestations similar to patients with SAH (10%‐30%).6 Further, continuing corticosteroids in the nonresponsive patients could result in predisposition to secondary bacterial infections, spontaneous bacterial peritonitis, and increased mortality.11 Early identification of nonresponders to corticosteroid therapy, which may be around 40%, is therefore essential. A Lille score of ≥0.45 is used to define steroid nonresponse at day 7.7 However, waiting for 7 days leads to unnecessary exposure to steroids in the eventual steroid‐nonresponsive patients. Thus, there is an urgent need of identifying novel indicators for differentiating nonresponders from responders prior to the start of therapy. Severity and progression of alcoholic hepatitis also needs to have better markers, preferably noninvasive ones.5, 8, 10, 12, 13, 14
Urine as a biofluid has gained importance for the identification of putative biomarkers because it is mostly sterile in nature, less complex, easy to obtain in large volume, and largely free from interfering proteins or lipids.15 In addition, ease of urine sample processing makes it a favored biofluid for identifying altered metabolic pathways associated with disease/therapy. Metabolomics is a powerful technology that allows assessment of global metabolic profiles in biofluids.16, 17 In order to explore new indicators of steroid nonresponse, we studied the urine metabolome profile at baseline before corticosteroid therapy. We also investigated whether urinary metabolites correlate with disease severity and mortality. Finally, we developed an approach integrating urine metabolomics and liver transcriptomics in order to explore the possible links between urine metabolites and liver genes and enhance our understanding of SAH pathophysiology.
Patients and Methods
Patients with SAH seen between 2013 and 2015 at the Department of Hepatology, Institute of Liver and Biliary Science, New Delhi, India, and confirmed to have MDF ≥32, recent onset of jaundice, chronic alcohol abuse, and liver biochemistry and histologic features of SAH (n = 180) were screened for corticosteroid therapy.6 All 180 patients underwent transjugular liver biopsy, and a minimum of 10 portal spaces were analyzed before characterization of a patient as SAH. Patients with hepatocellular carcinoma (n = 10), portal vein thrombosis (n = 15), or recent variceal bleed (n = 12) were excluded from analysis. In addition, patients with hepatitis B virus, hepatitis C virus, and human immunodeficiency virus infection were excluded. The remaining 140 patients with SAH were enrolled in the study, and written informed consent was obtained from every patient. The study was approved by the institutional ethics committee.
Baseline demographic profiles were recorded and early morning fasting urine samples were collected before start of prednisolone at 40 mg/day. Patients were characterized as responders (Rs) or nonresponders (NRs) at day 7 using the Lille score.6 At baseline, none of the enrolled patients with SAH documented high serum creatinine level, suggesting normal functioning of the kidneys. Further, serum creatinine‐based estimation of glomerular filtration rate18 was found to be >90 mL/minute/1.73 m2 in all patients, confirming the absence of kidney injury in these patients. The laboratory staff performing the experiments was unaware of the clinical details. Patients were managed according to the standard of care, including intensive care monitoring, high calorie diet (35‐40 cal/kg/day), intravenous albumin, and broad‐spectrum antibiotics. Severity of liver disease was assessed by MDF, Child‐Pugh, and MELD scores at the initial presentation, and steroid responsiveness was assessed by the Lille score during follow‐up. Among the 140 patients, the first 60 patients (enrolled during 2013) formed the discovery cohort and the subsequent 80 patients (enrolled in 2014 and 2015) constituted the validation cohort.
URINE METABOLOMICS
Urine metabolomics was performed in the discovery cohort. About 20 mL of early morning urine sample was aliquoted and stored at –80°C. Urine samples were centrifuged at 1,430g for 5 minutes, diluted at 1:5 in 5% acetonitrile:95% water, spiked with internal standards at known concentrations, and subjected to reverse‐phase chromatography on an ultra‐high performance liquid chromatographic system followed by high‐resolution mass spectrometry (MS) as detailed in the http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full.
MEASUREMENTS OF URINARY ACETYL‐L‐CARNITINE
The determination of acetyl‐L‐carnitine concentrations in the urine samples was performed using the acetyl‐L‐carnitine detection kit (cat. no. CEO400Ge) in both the discovery and the validation cohort (details in the http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full).
QUANTIFICATION AND STATISTICAL ANALYSIS
All statistical tests were two‐tailed with P < 0.05. Statistical analyses were performed using SPSS version 20. Baseline clinical parameters were represented as median (range) or proportions.
Metabolomics and Pathway Analysis
To analyze the metabolomics data, filtered features of the XCMS peak tables were normalized using mass spectrum total useful signal (MSTUS) normalization, which works on the variation in urine volume and diuresis and is much more effective than creatinine normalization.19, 20 This normalization method has been introduced into the Metaboanalyst 3.0 (http://www.metaboanalyst.ca) server21, 22 and into SIMCA P12 software (Umetrix, Sweden) for multivariate projection analyses, such as principal component analysis and partial least square discriminant analysis. A three‐step statistical filtering of the metabolites was carried out, as detailed in http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full and the http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). Pathway enrichment patterns were analyzed using Metaboanalyst,22 a web‐based tool designed for untargeted metabolomics data and pathway analysis.
Prediction of Nonresponse to Corticosteroids and Mortality
Receiver operating characteristic (ROC) curves for predicting nonresponse to corticosteroid with metabolites or other variables were generated by computing sensitivity and specificity at each observed cutoff for the variable of interest. Area under the ROC (AUROC) curve was calculated for each variable of interest. Data on time to death were estimated with the Kaplan–Meier method23 and were compared between groups by the log‐rank test, with hazard ratios and 95% confidence limits estimated by the Cox model.24
Gene–Metabolite Integration Analysis
We recently published results of high‐throughput transcriptomics in liver and corresponding peripheral blood mononuclear cells (PBMCs) of 32 patients with SAH before they received corticosteroid therapy (i.e., under baseline conditions).25 These patients were subsequently classified as NR (n = 16) or R (n = 16) after 7 days of corticosteroid therapy, according to the Lille score.25 In brief, genes overexpressed in the liver compared to the PBMCs were found to be liver specific (henceforth, liver‐specific genes), while genes underexpressed in the liver compared to the PBMCs were considered mainly related to immune cell functions (henceforth, immune‐cell‐related genes). Because patients who had transcriptomic results were also enrolled in the present metabolomics study, we combined “omics” data sets using the following strategy: first, among the sets of liver‐specific genes and immune‐cell‐related genes, we identified genes that were differentially expressed between NRs and Rs; second, we used hierarchical clustering to identify gene clusters that accounted for differences between NRs and Rs, according to the method by Li et al.26 We used Gene Set Enrichment Analysis (http://software.broadinstitute.org/gsea/index.jsp)27 to query the open source databases of Kyoto Encyclopedia of Genes and Genomes (http://www.genome.ad.jp/kegg/),28 REACTOME (https://reactome.org/), and Gene Ontology (http://www.geneontology.org), with the aim to functionally characterize gene clusters. Gene sets or pathways were considered as relevant when they included at least five genes and P < 0.05 and the false discovery rate was <0.05. Next, for patients with results of both transcriptomics and metabolomics, as described,26 a mean value was calculated for each cluster intensity and each cluster intensity was regressed against each value of the validated metabolites, using stepwise linear regression and Spearman correlation.
Results
PATIENTS
Patient characteristics at enrollment for the discovery and validation cohorts according to their response to corticosteroid therapy are shown in Table 1. The number of NRs was 8 (13%) and 20 (25%) in the discovery and validation cohort, respectively (Table 1). In each cohort, the clinical profile, including age, proportion of males, and indices of severity of liver disease, was similar in NRs and Rs (Table 1). However, in each cohort, the baseline leukocyte and platelet counts were significantly higher in NRs than Rs. The 90‐day mortality was also higher in NRs in both cohorts (Table 1).
Table 1.
Baseline Clinical Parameters of Responders and Nonresponders
| Parameters | Discovery Cohort | P value | Validation Cohort | P value | ||
|---|---|---|---|---|---|---|
| Responders | Nonresponders | Responders | Nonresponders | |||
| n = 52 | n = 8 | n = 60 | n = 20 | |||
| Age (years) | 39 (29‐59) | 37 (26‐64) | 0.62 | 36 (25‐60) | 34 (26‐68) | 0.52 |
| Sex (No. males/total number) (%) | 51/52 (98) | 8/8 (100) | 0.21 | 59/60 (98) | 19/20 (95) | 0.14 |
| BMI (kg/m2) | 24.8 (15.2‐34.1) | 26.7 (19‐40) | 0.72 | 24 (14.2‐36.1) | 29 (20.1‐45) | 0.24 |
| Age of onset of alcohol (years) | 26 (14‐43) | 25 (21‐35) | 0.76 | 22 (15‐48) | 27 (20‐38) | 0.55 |
| Jaundice duration (days) | 33 (7‐90) | 39.5 (21‐60) | 0.46 | 32 (6‐100) | 39 (28‐50) | 0.63 |
| Ascites duration (days) | 12 (0‐75) | 18 (1‐45) | 0.35 | 10 (0‐80) | 20 (1‐50) | 0.54 |
| Jaundice to ascites interval (days) | 11 (0‐90) | 3.5 (0‐59) | 0.36 | 14 (0‐80) | 5 (0‐50) | 0.64 |
| Alcohol to steroid interval (days) | 31 (1‐90) | 30 (10‐50) | 0.52 | 30 (1‐70) | 31 (12‐49) | 0.15 |
| Total bilirubin (mg/dL) | 17.3 (5‐45.4) | 22.2 (9.1‐33.6) | 0.72 | 19 (5‐43.4) | 25.2 (9.1‐45.6) | 0.24 |
| Direct bilirubin (mg/dL) | 10.6 (1.6‐31) | 13.0 (3.3‐23.7) | 0.63 | 12 (1.8‐31) | 12.0 (2.3‐24.4) | 0.28 |
| AST (IU) | 122 (51‐374) | 196 (55‐332) | 0.06 | 119 (55‐380) | 189 (58‐342) | 0.05 |
| ALT(IU) | 43.5 (8‐151) | 63.5 (34‐146) | 0.29 | 41.5 (10‐155) | 65.5 (32‐139) | 0.89 |
| AST/ALT ratio | 2.5 (1.30‐10.2) | 2.4 (1.5‐5) | 0.72 | 2.9 (1.4‐11.2) | 3 (1.5‐6.0) | 0.24 |
| Total protein (g/dL) | 7.0 (3.4‐8.9) | 6.7 (5.3‐7.5) | 0.18 | 6.0 (3.0‐7.5) | 6.3 (5.0‐7.3) | 0.18 |
| Serum albumin (g/dL) | 2.5 (1.7‐3.6) | 2.2 (1.7‐2.9) | 0.13 | 2.3 (1.8‐3.0) | 2.1 (1.6‐3.2) | 0.14 |
| INR | 2.0 (1.5‐4.0) | 2.0 (1.74‐3.0) | 0.72 | 1.9 (1.4‐4.2) | 2.0 (1.74‐3.0) | 0.44 |
| Hb (g/dL) | 9.7 (6.8‐14.8) | 9.8 (7.4‐11.6) | 0.83 | 8.7 (6.5‐13.8) | 10.2 (6.4‐12.6) | 0.32 |
| TLC (cells/μL) | 12.0 (4.0‐31.9) | 15.2 (7.9‐33) | 0.01 | 11.9 (4.2‐32.9) | 15.9 (8.0‐34) | 0.03 |
| Neutrophils (%) | 78 (46‐90) | 81 (67‐90) | 0.71 | 80 (45‐86) | 83 (65‐89) | 0.13 |
| Platelet count (cells/μL) | 134 (45‐379) | 218 (28‐410) | 0.04 | 140 (48‐349) | 227 (30‐398) | 0.02 |
| Urea (mg/dL) | 21 (4‐85) | 31 (7‐71) | 0.62 | 20 (4‐82) | 34 (8‐79) | 0.20 |
| Creatinine (mg/dL) | 0.5 (0.09‐1.3) | 0.5 (0.02‐1.1) | 0.83 | 0.45 (0.07‐1.4) | 0.5 (0.03‐1.3) | 0.32 |
| eGFR (mL/minute/1.73m2) | 105 (98‐113) | 101 (95‐107) | 0.15 | 106 (102‐110) | 101 (95‐106) | 0.19 |
| Serum sodium (mEq/L) | 131 (115‐142) | 130 (118‐137) | 0.29 | 132 (115‐140) | 129 (117‐138) | 0.89 |
| Serum potassium (mEq/L) | 4.1 (3.0‐5.6) | 4.4 (3.3‐5.5) | 0.42 | 4.0 (3.4‐5.7) | 4.6 (3.0‐5.5) | 0.20 |
| Serum TNFα (pg/mL) | 11.8 (0.5‐718.0) | 8.7 (04‐270.0) | 0.58 | 12.5 (0.3‐700.0) | 7.5 (0.4‐670.0) | 0.76 |
| HVPG (mm Hg) n = 46 | 19 (10‐29) | 20 (15‐31) | 0.95 | 19 (11‐32) | 20 (16‐34) | 0.51 |
| CP score | 12 (12‐16) | 12 (10‐12) | 0.32 | 12 (12‐16) | 11 (10‐12) | 0.24 |
| MELD score | 25 (18‐32) | 25 (16‐32) | 0.94 | 24 (19‐33) | 26 (16‐30) | 0.09 |
| MELDNa | 28 (18‐38) | 30 (16‐40) | 0.83 | 29 (19‐39) | 32 (16‐44) | 0.30 |
| GAH score | 8 (10‐14) | 8.5(7‐9) | 0.12 | 9(09‐14) | 9.5 (8‐10) | 0.12 |
| MDF | 72 (33‐157) | 75 (56‐145) | 0.73 | 71 (32‐150) | 73 (56‐149) | 0.29 |
| Lille score | 0.1 (0.04‐0.4) | 0.7 (0.5‐0.9) | 0.00 | 0.1 (0.04‐0.4) | 0.8(0.5‐0.9) | 0.00 |
|
90‐day mortality (No./Total number [%]) |
2/52 (3.8) | 4/8 (50) | 0.01 | 6/60 (10) | 14/20 (70) | 0.01 |
Unless specified, values are medians (range).
Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; CP, Child‐Pugh score; eGFR, estimated glomerular filtration rate; GAH, Glasgow alcoholic hepatitis score; Hb, hemoglobin levels; HVPG, hepatic vein pressure gradient; INR, international normalized ratio; TLC, total leukocyte count; TNFα, tumor necrosis factor α.
CHARACTERIZATION OF THE URINE METABOLOME IN THE DISCOVERY COHORT
In this untargeted urine metabolome profiling approach, 4,472 features were detected in positive and negative electrospray ionization conditions. We were able to annotate and validate a total of 212 (∼5%) features from the negative and positive ionization modes (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). We considered a feature only if it matched any two of the following validation criteria: (a) m/z matching structure, (b) retention times matching standard metabolite, (c) tandem (MS/MS) matching standard metabolite, (d) interpretation of MS‐MS spectrum, or (e) interpretation of MS spectrum (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). Functional annotation of metabolites identified diverse subclasses (e.g., alkaloid derivatives, amino acid derivatives, benzyl alcohols, primary and secondary bile acids, fatty acids and derivatives, steroids, sugar alcohols), each being enriched with more than five distinct metabolites (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full).
BASELINE URINE METABOLOME ROBUSTLY DISTINGUISHES NRs IN THE DISCOVERY COHORT
Partial least square discriminating analysis clearly separated NRs from Rs (Fig. 1A) and was validated by 100 permutation tests (Fig. 1B; principal component analyses in http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). A total of 29/212 (13.6%) urinary metabolites with variable importance on projection scores >1 were identified (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). The up‐regulated metabolites in NRs were linked to the energy metabolism/trichloroacetic acid cycle; D‐glutamine/D‐glutamate metabolism; alanine, aspartate, and glutamate metabolism; lysine biosynthesis; and vitamin B6 metabolism (P < 0.05; pathway impact >0.05)22 (see http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full; Fig. 1C). The down‐regulated metabolites were significantly enriched in beta‐alanine metabolism and phenylalanine metabolism (Fig. 1D). In the cohort, nine metabolites (4.2%) significantly differed between NRs and Rs before and after MSTUS normalization and fulfilled each of the following criteria: change >1.5‐fold; P < 0.05; variable importance on projection >1; and Benjamini–Hochberg q correction <0.05). Accordingly, these nine metabolites were considered to be the most reliable metabolites. Of these, seven were increased in NRs, including acetyl‐L‐carnitine, octanoylcarnitine, decanoylcarnitine, decenedioic acid, alpha‐ketoglutaric acid, histidylproline diketopiperazine, and Gly‐Ala‐Pro‐Thr (tetra peptide), and two were decreased: glycerol‐3‐phosphate and N‐acetylneuraminic acid (Fig. 1E; Table 2; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full).
Figure 1.
Baseline urine metabolome in severe alcoholic hepatitis patients (responders and nonresponders) according to their 7‐day response to corticosteroid therapy. (A) PLS‐DA plot documenting clear differences between NRs (n = 8) and Rs (n = 52). (B) Internal cross‐validation plot (Q2) for baseline urine metabolites of NRs versus Rs. (C) Up‐regulated metabolite pathway enrichment (bubble plot analysis) based on the HMDB database in NRs. (D) Down‐regulated pathway enrichment (bubble plot analysis) based on the HMDB database in NRs. (E) Key mitochondrial metabolites significantly altered in NRs versus Rs after normalization (***P < 0.001, **P < 0.01,*P < 0.05). Abbreviations: CoA, coenzyme A; Comp, Component; Do, day zero (baseline); HMDB, Human Metabolome Database; PLS‐DA, partial least square discriminating analysis plot; t, matrix consisting of n row vectors; TCA, trichloroacetic acid. Data is represented as Mean and SD for the metabolites.
Table 2.
Nine Metabolites That Differentiate Responders From Nonresponders After a Stringent Selection Process Operated in the Discovery Cohorta
| Metabolites | Label | RT (Minutes) | m/z | Attribution | Subclass | Before MSTUS Normalization | After MSTUS Normalization | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Median R | Median NR | NR/R | P value | VIP | Median R | Median NR | NR/R | P value | VIP | ||||||
| Acetyl‐L‐carnitine | M98 | 1.3 | 204.12 | [(M+H)]+ | Fatty acid esters | 7.62E+07 | 4.44E+09 | 58.25 | 0.00 | 1.38 | 1.49E+08 | 1.93E+09 | 12.97 | 0.00 | 3.00 |
| Gln Ala Pro Thr (tetra peptide) | M191 | 5.78 | 416.21 | [(M+H)]+ | Amino acids, peptides, and analogues | 1.52E+07 | 5.95E+08 | 39.19 | 0.00 | 1.22 | 1.94E+07 | 1.56E+08 | 8.02 | 0.02 | 2.30 |
| Octanoylcarnitine | M159 | 7.92 | 288.22 | [(M+H)]+ | Fatty acid esters | 1.24E+07 | 3.31E+08 | 26.64 | 0.00 | 1.19 | 2.87E+07 | 1.32E+08 | 4.58 | 0.00 | 2.20 |
| Decanoylcarnitine | M171 | 9.02 | 316.25 | [(M+H)]+ | Fatty acid esters | 1.40E+07 | 2.68E+08 | 19.17 | 0.00 | 1.18 | 1.97E+07 | 7.45E+07 | 3.78 | 0.02 | 1.80 |
| Decenedioic Acid | M90 | 7.35 | 201.11 | [(M+H)]+ | Fatty acids and conjugates | 1.34E+07 | 2.19E+08 | 16.36 | 0.00 | 1.02 | 2.75E+07 | 8.80E+07 | 3.20 | 0.04 | 1.80 |
| C12H15O2N4/histidylproline diketopiperazine | M136 | 6.15 | 247.12 | [(M–H)]– | Piperazino piperidines | 5.63E+04 | 6.78E+05 | 12.03 | 0.00 | 1.44 | 8.45E+04 | 2.53E+05 | 2.99 | 0.05 | 1.70 |
| Alpha‐ketoglutaric acid | M26 | 1.1 | 145.01 | [M–H]– | Gamma keto acid and derivatives | 3.49E+08 | 3.14E+09 | 9.01 | 0.00 | 1.20 | 6.11E+08 | 1.10E+09 | 1.80 | 0.01 | 1.10 |
| Glycerol‐3‐phosphate | M45 | 0.89 | 171.01 | HMDB02520 | Glycerophosphates | 1.77E+05 | 2.27E+05 | 1.28 | 0.02 | 1.00 | 3.09E+05 | 1.52E+05 | 0.49 | 0.02 | 1.00 |
| N‐Acetylneuraminic acid | M169 | 0.91 | 308.1 | [(M–H)]– | Sugar acid and derivatives | 4.48E+05 | 7.15E+05 | 1.60 | 0.03 | 1.02 | 8.26E+05 | 3.59E+05 | 0.43 | 0.03 | 1.30 |
To be selected, metabolites should be significantly different between NRs and Rs before and after MSTUS normalization and after univariate and multivariate analysis, exhibit a VIP score of >1, have a fold‐change > (for up‐regulated) or < (for down‐regulated) 1.5, with P < 0.05.
Abbreviations: HMDB, Human Metabolome Database; M+H, Mono isotopic ion state in positive mode; M−H, Mono isotopic ion state in negative mode; RT, Retention Time; VIP, variable important in projection.
BASELINE URINE METABOLITES CORRELATE WITH OUTCOMES IN THE DISCOVERY COHORT
In the discovery cohort, higher levels of acetyl‐L‐carnitine, octanoylcarnitine, and alpha‐ketoglutaric acid corresponded to a higher risk of nonresponse to corticosteroids (Table 3). In addition, there was a positive correlation between the levels of each of these three metabolites and each of the severity scores (MELD, MDF, Child‐Pugh) and the 1‐month mortality rate (Table 3; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). Because acetyl‐L‐carnitine was associated with the highest values of AUROC (Table 3) and odds ratio for predicting NRs in the discovery cohort, we validated these results using enzyme‐linked immunosorbent assay (ELISA) measurement of acetyl‐L‐carnitine in the validation cohort. In the discovery cohort, we found that acetyl‐L‐carnitine levels assessed by MS significantly correlated with levels measured using the ELISA technique (regression coefficients (r s) = 0.838; P < 0.001; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). Levels of acetyl‐L‐carnitine (ELISA) were significantly increased in NRs in both the discovery and validation cohorts (Fig. 2A).
Table 3.
Assessment of Ability of Top Nine Metabolites to Predict the Response to Corticosteroids and Correlation of These Metabolites With Mortality and Severity Scores in the Discovery Cohort
| AUROC Curves | OR for Nonresponse Determination | Predictive Values | Spearman Correlation Coefficient | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Test Result Variable(s) | Area | Sig | 95% CI | OR | P value | PPV | NPV | Mortality | MELD score | MDF | CP score | |
| LB | UB | |||||||||||
| Acetyl‐L‐carnitine | 0.909 | 0.000 | 0.816 | 0.99 | 15.3 (7.4‐29) | 0.0001 | 100 | 90 | 0.458** | 0.340** | 0.3179** | 0.302** |
| Octanoylcarnitine | 0.868 | 0.001 | 0.743 | 0.992 | 5 (1.9‐30) | 0.0010 | 87 | 60 | 0.412** | 0.532** | 0.732** | 0.283** |
| Alpha ketoglutaric acid | 0.798 | 0.007 | 0.68 | 0.916 | 4 (1.6‐36) | 0.0010 | 92 | 70 | 0.482** | 0.416** | 0.351** | 0.390** |
| Decanoylcarnitine | 0.767 | 0.016 | 0.616 | 0.918 | 3.4 (1.6‐26) | 0.0210 | 85 | 67 | 0.148 | 0.017 | 0.094 | 0.064 |
| GlnAlaProThr (tetra‐peptide) | 0.72 | 0.047 | 0.523 | 0.917 | 3.0 (1.3‐17.8) | 0.0410 | 75 | 70 | 0.171 | 0.144 | 0.146 | 0.251 |
| Decenedioic acid | 0.742 | 0.029 | 0.526 | 0.958 | 1.6 (1.2‐35) | 0.0450 | 85 | 80 | 0.134 | 0.061 | ‐0.013 | 0.045 |
| C12h15o2n4/Histidylproline diketopiperazine | 0.663 | 0.139 | 0.431 | 0.896 | 1.2 (1.1‐15) | 0.0310 | 86 | 65 | 0.218 | –0.321* | –0.334** | –0.308* |
| Glycerol 3 phosphate | 0.227 | 0.014 | 0.109 | 0.345 | 2.0 (1.5‐50) | 0.0040 | 85 | 75 | –0.455** | –0.483** | –0.455** | –0.506** |
| N‐Acetylneuraminic acid | 0.244 | 0.021 | 0.109 | 0.379 | 3.6 (2.0‐135) | 0.0170 | 80 | 90 | –0.351** | –0.351** | –0.321* | –0.318* |
* = P < 0.05, ** = P < 0.01, and *** = P < 0.001.
Abbreviations: CI, confidence interval; CP, Child‐Pugh; LB, lower bound; NPV, negative predictive value; OR, odds ratio; PPV, positive predictive value; Sig, Significance at P < 0.05; UB, upper bound.
Figure 2.
Performance evaluation of baseline predictors of nonresponse and mortality. (A) Acetyl‐L‐carnitine measurements in urine (ELISA) in 8 NRs (3,350 ng/mL) and 2 Rs (936 ng/mL) in the discovery cohort and validated (3,293 ng/mL) in 20 NRs and 60 Rs (1,566 ng/mL; ***P < 0.001). (B) Cox proportional analysis of Acetyl‐L‐carnitine in comparison to other clinical factors. Hazard ratio of Acetyl‐L‐carnitine was significantly higher than any other clinical factors compared in multivariate analysis. (C) AUROC was significantly higher with acetyl‐L‐carnitine than with CTP, MELD, MDF, or TLC for predicting nonresponse. (D) Kaplan–Meier curve analysis documented differences between nonsurvivors and survivors based on the cut‐off point of acetyl‐L‐carnitine (2,500 ng/mL) in urine samples of patients with SAH. Abbreviations: ALCAR, acetyl‐L‐carnitine; b, standardize; CI, confidence interval; CTP, Child‐Turcotte‐Pugh; HR, hazard ratio; Sig, Significance at P < 0.05; TLC, thin‐layer chromatography.
URINE ACETYL‐L‐CARNITINE LEVELS CORRELATE WITH OUTCOME
In the discovery cohort, higher acetyl‐L‐carnitine levels and higher MDF were independent predictors of death (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). In the entire cohort (discovery plus validation), higher acetyl‐L‐carnitine levels, higher total leukocyte count, and higher MDF were independent predictors of death (Fig. 2B). Interestingly, in the entire cohort, acetyl‐L‐carnitine documented the highest AUROC of 0.96 (95% confidence interval, 0.89‐0.96) for predicting a response to corticosteroids when compared to other clinical factors (Fig. 2C). Further, based on the AUROC of acetyl‐L‐carnitine of 0.96 and a likelihood ratio of 5.6, a cutoff for the prediction of nonresponse was determined at 2,500 ng/mL and was used to assess survival. In the entire cohort, survival was significantly lower among patients with acetyl‐L‐carnitine levels above 2,500 ng/mL than among those with levels below 2,500 ng/mL (log‐rank test <0.01; Fig. 2D).
BASELINE HEPATIC TRANSCRIPTOME CORROBORATES WITH URINARY METABOLOME SIGNATURES
We assessed whether changes in baseline urine metabolome in patients with SAH were linked to alterations in basal hepatic gene expression. We used results of hepatic and PBMC transcriptomics of 32 patients with SAH before any treatment.25 At baseline, there were 1,662 differentially expressed genes between liver and PBMCs. Of these, 1,340 were overexpressed in the liver (liver‐specific genes) and 322 underexpressed in the liver (immune‐cell‐related genes; see Patients and Methods).25 Of the 1,340 liver‐specific genes, 403 were differentially expressed between NRs and Rs (Fig. 3A). Among these, very few (12 genes, “cluster 1”) had higher expression in NRs than Rs, while the remaining 391 genes (“cluster 2”) had lower expression in NRs than Rs (Fig. 3A; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). Functional annotation of cluster 1 did not show any significant feature; in contrast, genes in cluster 2 were related to protein synthesis and tissue homeostasis (Fig. 3B; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). Of the 322 immune‐cell‐related genes, 118 genes differentially expressed between NRs and Rs. Of these, 89 had higher expression in NRs than Rs (“cluster 3”), while the remaining 29 genes (“cluster 4”) had lower expression in NRs than Rs (Fig. 3A; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). Genes in cluster 3 were related to G‐protein‐coupled receptor (GPCR) signaling and activity; genes in cluster 4 were related to positive immune cell regulation and cell adhesion (Fig. 3B; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full). The results for both transcriptomics and metabolomics were available in 16 Rs and 8 NRs. In these patients, we tested the association between the two “omics” and found that gene cluster intensities significantly regressed against 54 metabolites (Table 4) irrespective of their response status. Sixteen metabolites regressed against cluster 1 intensities, 15 metabolites against cluster 2 intensities, 15 metabolites against cluster 3 intensities, and 15 metabolites against cluster 4 intensities. Some metabolites (e.g., decanoylcarnitine) regressed against different gene clusters. Top metabolites predicting poor outcome (see Table 3) were among metabolites that regressed against clusters exhibiting a “prominence” of NRs over Rs. Counterintuitively, the regression of acetyl‐L‐carnitine, octanoylcarnitine, and decanoylcarnitine levels against cluster 1 intensities exhibited negative regression coefficients (i.e., negative β values for unstandardized coefficients; see Table 4). Accordingly, we examined the influence of being Rs or NRs on the direction of the association between metabolites and cluster 1 intensity (using Spearman correlation). In Rs, there was a significant negative correlation of each metabolite with cluster 1 intensity (r s values were –0.75, –0.64, and –0.70, for acetyl‐L‐carnitine, octanoylcarnitine, and decanoylcarnitine, respectively). In contrast, in NRs, metabolites either were correlated positively (r s was 0.98 for acetyl‐L‐carnitine) or did not correlate with cluster intensity (r s value was similarly 0.07 with octanoylcarnitine and decanoylcarnitine). The differences in the direction of the association between metabolites and cluster 1 intensity in NR versus R may explain the low value of regression (Table 4) and correlation (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full) coefficients observed in the whole group. Together, these results suggest that the counterintuitive negative correlation of metabolites versus cluster 1 intensity found in the whole population may reflect the negative correlation observed in Rs. For the whole patient group, the coefficient values (i.e., standardized coefficient beta for regression in Table 4; r s in http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full) were relatively low, suggesting that the strength of the association between metabolite levels and cluster 1 intensity was weak; hence, results should be interpreted with caution. More interestingly, cluster 3 intensities positively correlated with decanoylcarnitine and Gln‐Ala‐Pro‐Thr (tetra peptide) levels (Table 4; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full).
Figure 3.
Baseline hepatic transcriptome in patients with SAH (NRs and Rs). (A) The 1,340 genes that were up‐regulated at baseline in liver compared to corresponding PBMCs were analyzed for their expression in NRs versus Rs. We found 403 DEGs between NRs and Rs: 12 genes were up‐regulated in NRs (cluster 1); 391 genes were down‐regulated in NRs (cluster 2). The 322 genes that were down‐regulated in liver versus PBMCs were also analyzed for their expression in NRs versus Rs. There were 118 DEGs between NRs and Rs: 89 genes up‐regulated in NRs (cluster 3); 29 genes down‐regulated in NRs (cluster 4). Average intensities for gene clusters 1‐4 are provided for NRs versus Rs. (B) GSEA of genes included in clusters 2‐4. Cluster 1 did not show any significant enrichment. Abbreviations: DEG, differentially expressed gene; FDR, false discovery rate; GSEA, gene set enrichment analysis.
Table 4.
Stepwise Linear Regression of Urine Metabolites Against Gene Cluster Intensities in Patients With SAH
| Model | Metabolite ID | Metabolite ID | Unstandardized Coefficients | Standardized Coefficient | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| B | 95% CI of B (Lower Bound) | 95% CI of B (Upper Bound) | Beta | Sig. | HMDB | KEGG | Pathway | Biofunctions | |||
| Cluster 1 | M1 | Dihydro‐5‐methyl‐2(3h)‐furanone | 5.748 | 5.747 | 5.749 | 0.661 | 0.000 | HMDB33840 | ‐ | NA | Nutrient |
| M16 | 3‐Methyl‐2‐oxovaleric acid | –1.015E‐05 | 0.000 | 0.000 | 0.000 | 0.001 | HMDB00491 | C03465 | Valine, leucine, and isoleucine degradation |
Cell signaling Essential amino acid Fuel and energy storage Fuel or energy source Membrane integrity/stability |
|
| M18 | N‐acetylputrescine | 1.474 | 1.473 | 1.475 | 0.356 | 0.000 | HMDB02064 | C02714 | Arginine and proline metabolism | Endogenous or microbial | |
| M29 | Creatinine | 0.000 | 0.000 | 0.000 | 0.000 | 0.004 | HMDB00562 | C00791 | Arginine and proline metabolism | NA | |
| M32 | L‐histidine | 0.001 | 0.0001 | 0.01 | 0.001 | 0.000 | HMDB00177 | C00135 | Beta‐alanine metabolism |
Component of histidine metabolism Component of nitrogen metabolism |
|
| M49 | Iso valeric acid isomer | 1.547 | 1.546 | 1.548 | 0.202 | 0.000 | HMDB40529 | ‐ | NA | Fuel and energy storage | |
| M57 | 1‐Methyluric acid | ‐0.790 | –0.785 | –0.795 | –0.165 | 0.000 | HMDB03099 | C16359 | Caffeine metabolism | Waste products | |
| M83 | 2‐Methylhippuric acid | ‐0.928 | –0.927 | –0.929 | –0.103 | 0.000 | HMDB11723 | C01586 | Phenylalanine metabolism | NA | |
| M116 | 3‐Hydroxysebacic acid | 0.056 | 0.055 | 0.057 | 0.092 | 0.000 | HMDB00424 | ‐ | NA | NA | |
| M134 | C11 H20 O4 N2/Glutamine derivative | 0.079 | 0.077 | 0.080 | 0.053 | 0.000 | ‐ | ‐ | NA | NA | |
| M141 | N‐acetylgalactosamine | 0.021 | 0.020 | 0.022 | 0.003 | 0.000 | HMDB00212 | C01074 | Amoebiasis |
Component of glutamate metabolism Component of keratan sulfate biosynthesis Component of N‐glycan biosynthesis |
|
| M98 | Acetyl‐L‐carnitine | –0.207 | –0.206 | –0.300 | –0.186 | 0.000 | HMDB00201 | C02571 | Mitochondrial beta‐oxidation of short chain saturated fatty acids | Lipid catabolism, fatty acid transport, energy production | |
| M159 | Octanoylcarnitine | –0.137 | –0.136 | –0.138 | –0.158 | 0.000 | HMDB00791 | C02838 | Mitochondrial beta‐oxidation of short chain saturated fatty acids | Lipid catabolism, fatty acid transport, energy production | |
| M171 | Decanoylcarnitine | –0.014 | –0.0130 | –0.0150 | –0.027 | 0.000 | HMDB00651 | ‐ | Mitochondrial beta‐oxidation of short chain saturated fatty acids | Lipid catabolism, fatty acid transport, energy production | |
| M174 | Valproic acid glucuronide | –2.867 | –2.865 | –2.869 | –0.323 | 0.000 | HMDB00901 | C03033 | Valproic acid metabolism pathway | Fuel or energy source | |
| M211 | Glycochenodeoxycholic acid 3‐glucuronide or isomer | 1.155 | 1.153 | 1.156 | 0.574 | 0.000 | HMDB02579 | C03033 | Bile secretion | Fuel and energy storage | |
| Cluster 2 | M7 | 3‐Hydroxy‐3‐methylbutyric acid (3‐hydroxyisovaleric acid) | –0.323 | –0.321 | –0.324 | –0.342 | 0.000 | HMDB00754 | ‐ | NA | NA |
| M8 | 2‐Hydroxy‐3‐methylbutyric acid | 0.055 | 0.052 | 0.057 | 0.052 | 0.000 | HMDB00407 | ‐ | NA | Fuel or energy source | |
| M10 | Nicotinic acid | 0.275 | 0.274 | 0.277 | 0.141 | 0.000 | HMDB01488 | C00253 | Nicotinate and nicotinamide metabolism | Essential vitamins | |
| M41 | 4‐Pyridoxolactone | 0.453 | 0.452 | 0.455 | 0.243 | 0.000 | HMDB03454 | C00971 | Vitamin B6 metabolism | NA | |
| M75 | Methylhippuric acid | –0.001 | –0.0001 | –0.01 | 0.000 | 0.009 | HMDB00859 | ‐ | NA | NA | |
| M108 | Hexose | 0.032 | 0.031 | 0.033 | 0.017 | 0.000 | HMDB12326 | C15923 | Ascorbate and aldarate metabolism | NA | |
| M129 | Tiglylcarnitine/2‐ethylacrylylcarnitine | 1.714 | 1.712 | 1.715 | 0.935 | 0.000 | HMDB02366 | ‐ | NA | Lipid catabolism, fatty acid transport, energy production | |
| M130 | N‐Acetyl‐Dl‐tryptophan | 0.007 | 0.006 | 0.008 | 0.009 | 0.001 | HMDB13713 | ‐ | NA | NA | |
| M146 | Glu‐Leu | 1.042 | 1.041 | 1.043 | 0.323 | 0.000 | HMDB28823 | ‐ | NA | NA | |
| M153 | Isovalerylglucuronide | –0.415 | –0.413 | –0.416 | –0.118 | 0.000 | HMDB02091 | C03033 | Pentose and glucuronate interconversions | Waste products | |
| M154 | 4‐Hydroxypheny lacetylglutamine | –0.235 | –0.234 | –0.236 | –0.284 | 0.000 | HMDB06061 | C05595 | Tyrosine metabolism | NA | |
| M195 | Phe Try Asp | –0.047 | –0.045 | –0.048 | –0.066 | 0.000 | ‐ | ‐ | NA | NA | |
| M202 | Androsterone glucuronide | 0.000 | 0.000 | 0.000 | 0.000 | 0.001 | HMDB02829 | C11135 | Steroid hormone biosynthesis | Waste products | |
| M210 | Glycochenodeoxycholic acid 3‐glucuronide or isomer | –0.541 | –0.540 | –0.542 | –0.480 | 0.000 | HMDB02579 | C03033 | Bile secretion | Waste products | |
| M212 | Glycochenodeoxycholic acid 3‐glucuronide or isomer | 0.000 | 0.000 | 0.000 | –0.044 | 0.000 | HMDB02579 | C03033 | Bile secretion | Waste products | |
| Cluster 3 | M14 | D‐1‐Piperidine‐2‐carboxylic acid | –0.279 | –0.278 | –0.280 | –0.071 | 0.000 | HMDB01084 | C04092 | Lysine degradation | Protein synthesis, amino acid biosynthesis |
| M20 | Isoleucine/Leucine | 1.330 | 1.329 | 1.332 | 0.127 | 0.000 | HMDB00172 | C00407 | Biosynthesis of secondary metabolites | Component of valine, leucine, and isoleucine biosynthesis | |
| M31 | 2,5‐Dihydroxybenzoic acid | 5.453 | 5.452 | 5.454 | 0.781 | 0.000 | HMDB00152 | C00628 | Tyrosine metabolism | NA | |
| M52 | Acetyl‐(Leu/Ile) | –0.476 | –0.475 | –0.478 | –0.036 | 0.000 | ‐ | ‐ | NA | NA | |
| M56 | 8‐Hydroxy‐7‐methylguanine | –0.889 | –0.887 | –0.890 | –0.184 | 0.000 | HMDB06037 | ‐ | NA | NA | |
| M86 | C9h10n2o3/Pyridylacetylglycine | 2.768 | 2.766 | 2.769 | 0.348 | 0.000 | HMDB59723 | ‐ | NA | NA | |
| M112 | C6h14o6/Mannitol or isomers | 4.511 | 4.509 | 4.512 | 0.561 | 0.000 | HMDB00765 | C00392 | Phosphotransferase system | ||
| M117 | Pantothenic acid | –0.677 | –0.676 | –0.678 | –0.196 | 0.000 | HMDB00210 | C00864 | beta‐Alanine metabolism | Component of pantothenate and CoA biosynthesis | |
| M135 | Isovalerylcarnitine | –0.171 | –0.170 | –0.172 | –0.098 | 0.000 | HMDB00688 | ‐ | NA | Lipid catabolism, Fatty acid transport, Energy production | |
| M171 | Decanoylcarnitine | 0.107 | 0.106 | 0.108 | 0.214 | 0.000 | HMDB00651 | ‐ | NA | Lipid catabolism, fatty acid transport, energy production | |
| M177 | Galactosylhydroxylysine | 0.744 | 0.743 | 0.745 | 0.321 | 0.000 | HMDB00600 | C05547 | NA | NA | |
| M191 | Gln Ala Pro Thr (tetra peptide) | 0.091 | 0.090 | 0.092 | 0.375 | 0.000 | ‐ | ‐ | NA | NA | |
| M201 | Glycocholic acid | 3.035 | 3.034 | 3.036 | 0.680 | 0.000 | HMDB00138 | C01921 | Primary bile acid biosynthesis | Fuel and energy storage | |
| M206 | Glycochenodeoxycholate‐3‐sulfate or isomers | 0.280 | 0.279 | 0.282 | 0.918 | 0.004 | HMDB02497 | ‐ | NA | Fat solubilization and waste products | |
| M212 | Glycochenodeoxycholic acid 3‐glucuronide or isomer | 0.001 | 0.0001 | 0.01 | 0.242 | 0.001 | HMDB02579 | C03033 | NA | Fuel and energy storage | |
| Cluster 4 | M18 | N‐Acetylputrescine | –0.134 | –0.132 | –0.135 | –0.194 | 0.000 | HMDB02064 | C02714 | Arginine and proline metabolism | NA |
| M22 | 4‐Hydroxybenzoic acid | –0.117 | –0.116 | –0.118 | –0.173 | 0.000 | HMDB00500 | C00156 | Ubiquinone biosynthesis | NA | |
| M34 | 3‐Methylcrotonyl glycine | 0.026 | 0.025 | 0.027 | 0.029 | 0.000 | HMDB00459 | ‐ | NA | NA | |
| M40 | L‐Fucose | 0.242 | 0.240 | 0.243 | 0.215 | 0.000 | HMDB00174 | C01019 | Fructose and mannose metabolism | Component of fructose and mannose metabolism | |
| M56 | 8‐Hydroxy‐7‐methylguanine | –6.775E‐06 | 0.000 | 0.000 | 0.000 | 0.001 | HMDB06037 | ‐ | NA | NA | |
| M95 | Dl‐Tryptophan | 0.000 | 0.000 | 0.000 | 0.001 | 0.002 | HMDB30396 | ‐ | NA | NA | |
| M112 | C6h14o6/Mannitol or isomers | –1.883 | –1.882 | –1.884 | –1.388 | 0.000 | HMDB00765 | C00392 | Fructose and mannose metabolism | Fuel and energy storage | |
| M129 | Tiglylcarnitine/2‐ethylacrylylcarnitine | 0.217 | 0.215 | 0.218 | .287 | 0.000 | HMDB02366 | ‐ | NA | Lipid catabolism, fatty acid transport, energy production | |
| M138 | 3‐Hydroxydodecanedioic acid | 0.032 | 0.031 | 0.033 | .431 | 0.000 | HMDB00413 | ‐ | Beta oxidation of FFA | Fuel and energy storage | |
| M139 | Norepinephrine sulfate | –0.169 | –0.168 | –0.170 | –.306 | 0.000 | HMDB02062 | ‐ | NA | Waste products | |
| M158 | Ophthalmic acid | –0.018 | –0.017 | –0.019 | –.012 | 0.000 | HMDB05765 | ‐ | NA | NA | |
| M172 | Acylcarnitine of a dicarboxylic acid (C8h14o4) | 0.005 | 0.004 | 0.006 | .028 | 0.000 | ‐ | ‐ | NA | NA | |
| M173 | C12 H15 O10/Glucuronide of a dicarboxylic acid | 5.551 | 5.550 | 5.552 | 1.641 | 0.000 | ‐ | ‐ | NA | NA | |
| M182 | C17h24o5n2/Carnitine ester of C10h11n03/2‐methyl hippuric acid‐carnitin | 0.008 | 0.007 | 0.009 | 0.013 | 0.000 | ‐ | ‐ | NA | NA | |
| M190 | Glucuronide of C14h23o2 | 0.053 | 0.052 | 0.054 | 0.226 | 0.000 | ‐ | ‐ | NA | NA | |
For more information on the strategy used for identification of gene clusters and on the composition of these clusters, see Patients and Methods and Results sections; Fig. 3; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full.
Abbreviations: CI, confidence interval; CoA, coenzyme A; FFA, free fatty acid; HMDB, Human Metabolome Database; KEGG, Kyoto Encyclopedia of Genes and Genomes; NA, not applicable; Sig, Significance at P < 0.05.
Discussion
Noninvasive and easy to access methods for early identification of corticosteroid nonresponse or mortality for patients with SAH are not available. To address this issue, we investigated 140 patients with SAH who were divided into two cohorts; the first (discovery cohort) enrolled 60 patients (13% NRs) and the second (for validation) enrolled 80 patients (25% NRs). We have no clear explanation for finding that the proportion of NRs was lower in our discovery cohort than in our validation cohort. The only difference between the two cohorts was related to the period of enrollment; patients in the first cohort were enrolled during 2013, and those of the validation cohort were enrolled in 2014 and 2015. The response to steroid in our population of SAH was slightly higher with fewer NRs compared to Western countries (35%).4, 29 The reasons for these differences between countries are unclear but may involve differences in genetic and environmental factors between Indian and Western patients that contribute to differences in the response to corticosteroids. Further studies are needed.
We investigated baseline urinary metabolites (i.e., before corticosteroid therapy) in the discovery cohort. Using a high‐resolution MS‐based, untargeted, metabolomics approach, we could annotate 212 metabolites that were enriched in energy metabolism pathways, bile acid biosynthesis, amino acid biosynthesis, and others. Our novel observations demonstrated that baseline urinary metabolome can be used to identify patients with SAH who are unlikely to respond to corticosteroid therapy or die within a month.
Our technique for urine metabolome analysis was carefully designed. Although the preparation of urine samples for analysis is simple and the concentration of many metabolites is amplified by bladder storage, the biological interpretation of data is complicated by a variation in diuresis from subject to subject. Various normalization methods have been used and published to address this issue, including the traditional use of urinary creatinine concentration, osmolality,30, 31 total useful MS signal,30 and specific gravity19, 32 as well as a combination of creatinine concentration and normalization of the MS signal20 and the determination of the total concentration of chemically labeled metabolites by using liquid chromatography‐ultraviolet.33 However, many studies do not use normalization procedures, and there is still no consensus on this point.34 We employed a MSTUS normalization strategy,19, 31 which uses the total intensity of metabolites that are common to all samples and which is easy to implement and was found to perform better than creatinine normalization.20 For the selection of metabolites of interest, we chose to take into account metabolites that had concentration differences between Rs and NRs that were statistically significant with or without MSTUS normalization in order to improve result reliability.
In our discovery cohort, the baseline urine excretion of acetyl‐L‐carnitine, octanoylcarnitine, decanoylcarnitine, decenedioic acid, and alpha‐ketoglutaric acid was significantly higher among NRs. This suggests a marked derangement of energy biosynthesis and beta‐oxidation of fatty acids in NRs, consistent with results showing that SAH is associated with an alteration in the trichloroacetic acid cycle and beta‐oxidation of fatty acids.35
Because urine acetyl‐L‐carnitine levels measured with MS significantly correlated with levels measured with an ELISA technique in our discovery cohort, we used this technique in a validation cohort of 80 patients. In the validation cohort, higher levels of acetyl‐L‐carnitine significantly predicted the nonresponse to steroids and mortality. These findings were confirmed when results obtained in the discovery and validation cohorts were pooled.
Integration of data sets obtained with high‐throughput omics approaches can provide new insights into the pathophysiology of liver diseases. In this study, we explored the hypothesis that changes in baseline urine metabolome in patients with SAH could be associated with alterations in basal hepatic gene expression. For this, we identified four hepatic gene clusters that differentiated NRs from Rs at baseline. Two clusters were composed of liver‐specific genes, and the other two included genes related to immune cell functions. Using stepwise linear regression, we found that 54 metabolites significantly regressed against gene clusters, suggesting a link between alterations in gene expression within the liver and changes in urine metabolome composition. It is noteworthy that metabolites found to strongly predict poor outcome were among metabolites that were associated with intensity of clusters, including genes overexpressed in NRs. Cluster 1, which was up‐regulated in NRs, negatively regressed with acetyl‐L‐carnitine, octanoylcarnitine, and decanoylcarnitine (stepwise regression; Table 4). This surprising negative regression may be related to the fact that the correlation was strongly negative in Rs but was either positive or nonsignificant in NRs. Of note for the whole group of patients, the values of coefficients (i.e., standardized coefficient beta for regression [Table 4]; r s for correlation [http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full]) were relatively low, suggesting that the strength of the association between metabolite levels and cluster 1 intensity was weak; hence, results should be interpreted with caution. More interestingly, cluster 3 (which is up‐regulated in NRs) positively correlated with decanoylcarnitine and Gln‐Ala‐Pro‐Thr (tetra peptide). Genes included in cluster 3 were related to GPCR signaling activity consistent with enhanced GPCR signals in liver of NRs. Cells exhibiting increased GPCR signaling could be immune cells (infiltrating and/or resident) or progenitors.36 Our finding that metabolites correlated with cluster 3 suggests that these metabolites are markers of crucial pathophysiologic mechanisms that develop in the liver of NRs.
Our study has a limitation of being monocentric. Future multicentric studies should be performed to integrate data sets obtained with metabolomics and transcriptomics in large series of patients with SAH.
To conclude, baseline urine metabolome clearly discriminates corticosteroid Rs from NRs. In particular, baseline acetyl‐L‐carnitine can be used as a marker for early assessment of corticosteroid nonresponse and clinical outcome. Our study affirms that integration of metabolomics and liver transcriptomics substantially improves understanding the pathophysiology of alcoholic hepatitis.
Supporting information
Additional Supporting Information may be found at http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full.
Supporting Information Figure 1
Supporting Information Figure 2
Supporting Information Figure 3
Supporting Information 1
Potential conflict of interest: Nothing to report.
S.K.S. and R.M. were supported by the Indo‐French Center for Promotion of Advanced Research (CEFIPRA/IFCPAR no. 4903‐3C).
REFERENCES
- 1. Kim W, Kim DJ. Severe alcoholic hepatitis‐current concepts, diagnosis and treatment options. World J Hepatol 2014;6:688‐695. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Basra G, Basra S, Parupudi S. Symptoms and signs of acute alcoholic hepatitis. World J Hepatol 2011;3:118‐120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. European Association for the Study of Liver . EASL clinical practical guidelines: management of alcoholic liver disease. J Hepatol 2012;57:399‐420. [DOI] [PubMed] [Google Scholar]
- 4. Forrest E, Mellor J, Stanton L, Bowers M, Ryder P, Austin A, et al. Steroids or pentoxifylline for alcoholic hepatitis (STOPAH): study protocol for a randomised controlled trial. Trials 2013;14:262. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Kok VC, Horng JT, Huang HK, Chao TM, Hong YF. Regular inhaled corticosteroids in adult‐onset asthma and the risk for future cancer: a population‐based cohort study with proper person‐time analysis. Ther Clin Risk Manag 2015;11:489‐499. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Louvet A, Naveau S, Abdelnour M, Ramond MJ, Diaz E, Fartoux L, et al. The Lille model: a new tool for therapeutic strategy in patients with severe alcoholic hepatitis treated with steroids. Hepatology 2007;45:1348‐1354. [DOI] [PubMed] [Google Scholar]
- 7. Mathurin P, O'Grady J, Carithers RL, Phillips M, Louvet A, Mendenhall CL, et al. Corticosteroids improve short‐term survival in patients with severe alcoholic hepatitis: meta‐analysis of individual patient data. Gut 2011;60:255‐260. [DOI] [PubMed] [Google Scholar]
- 8. Zhang N, Truong‐Tran QA, Tancowny B, Harris KE, Schleimer RP. Glucocorticoids enhance or spare innate immunity: effects in airway epithelium are mediated by CCAAT/enhancer binding proteins. J Immunol 2007;179:578‐589. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Louvet A. Prednisolone vs. pentoxifylline for severe alcoholic hepatitis. J Hepatol 2014;61:723‐724. [DOI] [PubMed] [Google Scholar]
- 10. Kwon HJ, Won YS, Park O, Feng D, Gao B. Opposing effects of prednisolone treatment on T/NKT cell‐ and hepatotoxin‐mediated hepatitis in mice. Hepatology 2014;59:1094‐1106. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Louvet A, Wartel F, Castel H, Dharancy S, Hollebecque A, Canva‐Delcambre V, et al. Infection in patients with severe alcoholic hepatitis treated with steroids: early response to therapy is the key factor. Gastroenterology 2009;137:541‐548. [DOI] [PubMed] [Google Scholar]
- 12. Masson S, Emmerson I, Henderson E, Fletcher EH, Burt AD, Day CP, et al. Clinical but not histological factors predict long‐term prognosis in patients with histologically advanced non‐decompensated alcoholic liver disease. Liver Int 2014;34:235‐242. [DOI] [PubMed] [Google Scholar]
- 13. Parker R, Armstrong MJ, Corbett C, Rowe IA, Houlihan DD. Systematic review: pentoxifylline for the treatment of severe alcoholic hepatitis. Aliment Pharmacol Ther 2013;37:845‐854. [DOI] [PubMed] [Google Scholar]
- 14. Dhanda AD, Lee RW, Collins PL, McCune CA. Molecular targets in the treatment of alcoholic hepatitis. World J Gastroenterol 2012;18:5504‐5513. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Struck‐Lewicka W, Kordalewska M, Bujak R, Yumba Mpanga A, Markuszewski M, Jacyna J, et al. Urine metabolic fingerprinting using LC‐MS and GC‐MS reveals metabolite changes in prostate cancer:a pilot study. J Pharm Biomed Anal 2015;111:351‐361. [DOI] [PubMed] [Google Scholar]
- 16. Stojiljkovic N, Paris A, Garcia P, Popot MA, Bonnaire Y, Tabet JC, et al. Evaluation of horse urine sample preparation methods for metabolomics using LC coupled to HRMS. Bioanalysis 2014;6:785‐803. [DOI] [PubMed] [Google Scholar]
- 17. Boudah S, Olivier MF, Aros‐Calt S, Oliveira L, Fenaille F, Tabet JC, et al. Annotation of the human serum metabolome by coupling three liquid chromatography methods to high‐resolution mass spectrometry. J Chromatogr B Analyt Technol Biomed Life Sci 2014;966:34‐47. [DOI] [PubMed] [Google Scholar]
- 18. Kalafateli M, Wickham F, Burniston M, Cholongitas E, Theocharidou E, Garcovich M, et al. Development and validation of a mathematical equation to estimate glomerular filtration rate in cirrhosis: the royal free hospital cirrhosis glomerular filtration rate. Hepatology 2017;65:582‐591. [DOI] [PubMed] [Google Scholar]
- 19. Jacob CC, Dervilly‐Pinel G, Deceuninck Y, Gicquiau A, Chevillon P, Bonneau M, et al. Urinary signature of pig carcasses with boar taint by liquid chromatography‐high‐resolution mass spectrometry. Food Addit Contam Part A Chem Anal Control Expo Risk Assess 2017;34:218‐227. [DOI] [PubMed] [Google Scholar]
- 20. Chen Y, Shen G, Zhang R, He J, Zhang Y, Xu J, et al. Combination of injection volume calibration by creatinine and MS signals' normalization to overcome urine variability in LC‐MS‐based metabolomics studies. Anal Chem 2013;85:7659‐7665. [DOI] [PubMed] [Google Scholar]
- 21. Xia J, Mandal R, Sinelnikov IV, Broadhurst D, Wishart DS. MetaboAnalyst 2.0‐‐a comprehensive server for metabolomic data analysis. Nucleic Acids Res 2012;40(Web Server issue):W127‐W133. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Xia J, Sinelnikov IV, Han B, Wishart DS. MetaboAnalyst 3.0‐‐making metabolomics more meaningful. Nucleic Acids Res 2015;43(W1):W251‐W257. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958;53:457‐481. [Google Scholar]
- 24. Cox DR. Regression models and life‐tables. J R Stat Soc 1972;34:187‐220. [Google Scholar]
- 25. Sharma S, Maras JS, Das S, Hussain S, Mishra AK, Shasthry SM, et al. Pre‐therapy liver transcriptome landscape in Indian and French patients with severe alcoholic hepatitis and steroid responsiveness. Sci Rep 2017;7:6816. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Li S, Sullivan NL, Rouphael N, Yu T, Banton S, Maddur MS, et al. Metabolic phenotypes of response to vaccination in humans. Cell 2017;169:862‐877. e817. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge‐based approach for interpreting genome‐wide expression profiles. Proc Natl Acad Sci U S A 2005;102:15545‐15550. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Kanehisa M, Furumichi M, Tanabe M, Sato Y, Morishima K. KEGG: new perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res 2017;45(D1):D353‐D361. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Thursz MR, Richardson P, Allison M, Austin A, Bowers M, Day CP, et al.; STOPAH Trial . Prednisolone or pentoxifylline for alcoholic hepatitis. N Engl J Med 2015;372:1619‐1628. [DOI] [PubMed] [Google Scholar]
- 30. Warrack BM, Hnatyshyn S, Ott KH, Reily MD, Sanders M, Zhang H, et al. Normalization strategies for metabonomic analysis of urine samples. J Chromatogr B Analyt Technol Biomed Life Sci 2009;877:547‐552. [DOI] [PubMed] [Google Scholar]
- 31. Thevenot EA, Roux A, Xu Y, Ezan E, Junot C. Analysis of the human adult urinary metabolome variations with age, body mass index, and gender by implementing a comprehensive workflow for univariate and OPLS statistical analyses. J Proteome Res 2015;14:3322‐3335. [DOI] [PubMed] [Google Scholar]
- 32. Edmands WM, Ferrari P, Scalbert A. Normalization to specific gravity prior to analysis improves information recovery from high resolution mass spectrometry metabolomic profiles of human urine. Anal Chem 2014;86:10925‐10931. [DOI] [PubMed] [Google Scholar]
- 33. Wu Y, Li L. Determination of total concentration of chemically labeled metabolites as a means of metabolome sample normalization and sample loading optimization in mass spectrometry‐based metabolomics. Anal Chem 2012;84:10723‐10731. [DOI] [PubMed] [Google Scholar]
- 34. Zhang T, Watson DG. A short review of applications of liquid chromatography mass spectrometry based metabolomics techniques to the analysis of human urine. Analyst 2015;140:2907‐2915. [DOI] [PubMed] [Google Scholar]
- 35. Rachakonda V, Gabbert C, Raina A, Bell LN, Cooper S, Malik S, et al. Serum metabolomic profiling in acute alcoholic hepatitis identifies multiple dysregulated pathways. PLoS One 2014;9:e113860. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Druey KM. Regulation of G‐protein‐coupled signaling pathways in allergic inflammation. Immunol Res 2009;43:62‐76. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Additional Supporting Information may be found at http://onlinelibrary.wiley.com/doi/10.1002/hep4.1176/full.
Supporting Information Figure 1
Supporting Information Figure 2
Supporting Information Figure 3
Supporting Information 1