Proteomic Signature of Acute Liver Failure: From Discovery and Verification in a Pig Model to Confirmation in Humans

Abstract

Acute liver failure (ALF) is a fatal condition hallmarked by rapid development. The present study aimed to describe the dynamic alterations of serum proteins associated with ALF development, and to seek for novel biomarkers of ALF. Miniature pigs (n = 38) were employed to establish ALF models by infusing d-galactosamine (GALN, 1.3 g/kg). A total of 1310 serum proteins were compared in pooled serum samples (n = 10) before and 36 h after GALN administration through label-free quantitation (LFQ) based shotgun proteomics. Functional analysis suggested a significant enrichment of ALF-related proteins involved in energy metabolism. Temporal changes of 20 energy metabolism related proteins were investigated in individual pigs (n = 8) via parallel reaction monitoring (PRM) based targeted proteomics. In addition, mitochondrion degeneration and gene expression alteration of aerobic metabolism genes were confirmed in GALN-insulted pig liver. In clinical validation study enrolled 34 ALF patients and 40 healthy controls, fructose-1,6-bisphosphatase 1 (FBP1) showed a prognostic value for short-term survival (30 days) equal to that of the Model of End-stage Liver Disease score (ROC-AUC = 0.778). Survival analysis suggested significantly higher death-related hazard in ALF patients with higher FBP1 levels (>16.89 μg/dL) than in those with lower FBP1 levels (p = 0.002). Additionally, serum retinol binding protein 4 (RBP4) level was found decreased prior to ALT elevation in GALN-insulted pig model. We also confirmed that serum level of RBP4 is significantly lower in ALF patients (p < 0.001) as compared with healthy controls. In summary, this translational study, displayed by multistaged proteomics techniques, unveiled underlying functional changes related to the development of ALF and facilitated the discovery of novel ALF markers.

Acute liver failure (ALF)¹ is characterized by abrupt onset and high mortality with limited therapeutic choice. The heterogeneous etiology of ALF shows regional disparity, with viral infections and drug-induced hepatotoxicity being the leading causes in developing and developed countries, respectively. Other causes include ischemic hepatocellular injury, metabolic disease (Wilson's disease), and indeterminate etiology, which also constitutes a sizable fraction of ALF cases (1). Clinical syndromes of ALF typically deteriorate abruptly, thus survival is critically dependent on timely medical intervention, which in turn underscores the importance of early diagnosis. On the other hand, the outcome of ALF patients varies dramatically, with a significant portion of patients facing ominous prognosis that ultimately requires liver transplantation (LT). Though extracorporeal artificial liver support (ALS) can prolong life while on the waiting list for LT or facilitate spontaneous recovery, the outcome of ALS still relies on early recognition of ALF (1, 2). A significant proportion of ALF patients may miss the chance for LT because of delayed diagnosis (3). Moreover, identification of ALF patients with high mortality risk is of great importance for the objective and efficient allocation of precious medical resources in case LT or/and ALS are required. The most widely used conditions that aid in the diagnosis of ALF is the emergence of hepatic encephalopathy and coagulopathy. However, diagnosis of hepatic encephalopathy is based on descriptive neuropsychological evaluation rather than quantitative indicators, which make diagnosis subjective. In addition, the presence of hepatic encephalopathy reflects severe loss of liver function and predicts poor outcome for ALF patients, therefore this has limited use for early diagnosis. For years, alanine aminotransferase (ALT) has been deemed as a sensitive marker for liver injury. Nevertheless, extrahepatic expression and isoenzyme activities of ALT have been recorded in muscle and celiac disease (15). On the other hand, liver biopsy pathology is considered reliable to ascertain ALF, but it may cause massive hemorrhage in ALF patients with coagulopathy. Therefore, this invasive and inconvenient approach is not recommended in clinical practice.

Given the heterogeneous etiological background and short clinical course of ALF, it is difficult to detect the early development of ALF based on clinical samples. In contrast, animal models serve as ideal surrogates to investigate the pathogenesis of ALF. We previously developed a porcine model of d-galactosamine (GALN)-induced ALF that develops both hepatic encephalopathy (dysphoria, ataxia, and coma) and coagulopathy (constantly rising international normalized ratio, i.e. INR) (2), both are prominent ALF complications. We propose that leveraging the well-defined pathological course of ALF in a pig model will facilitate the development of ALF-related markers that may eventually be brought into clinical use.

The objectives of this study were (1) to identify potential ALF-associated protein signatures by comparing the serum proteomic profiles between healthy and GALN-challenged miniature pigs; (2) to describe the temporal evolution of potential protein marker levels over the course of ALF progression in our miniature pig model; and (3) to validate potential protein markers in ALF patients. Our study eventually identified 20 ALF-associated proteins that were related to energy metabolism, among which RBP4 and FBP1 were highlighted as promising biomarkers for early detection and short-term prognosis of ALF.

EXPERIMENTAL PROCEDURES

Experimental Design

Thirty-eight pigs were used to build the ALF model, among which ten pigs were used for exploratory proteomics study, eight were used for longitudinal profiling of candidate markers, twenty for expression verification on liver tissue level and histological investigations. For exploratory proteomics study, pooled samples from 10 pigs at P0 and P36 were examined by label-free quantitation (LFQ) and data-independent acquisition (DIA). Technical triplicates were performed in DIA experiment. Functional annotation was applied to screen the proteins for parallel reaction monitoring (PRM). Serum samples from 8 pigs collected sequentially from 5 time points during ALF development were analyzed by PRM to monitor the dynamic changes of the candidate proteins. Terminal validation was performed by ELISA of the 34 ALF patients and 40 healthy individuals (Table I). The procedures to generate reliable markers were detailed in Table II. The overall design of this translational study was illustrated in Fig. 1.

Table I. Clinical characteristics of patients.

	Healthy (n = 40)	ALF (n = 34)
Age, years	44.2 (15.4)	47 (16)
Sex, male (female)	11 (29)	11 (20)
Death (30 days)	N/A	16
ALT	26.52 (16.9)	605.87 (595.8)***
AST	N/A	559.06 (608.2)
ALP	N/A	162.76 (79.6)
Albumin	48.7 (3.0)	32.9 (5.7)***
Globulin	25.7 (4.7)	24.5 (8.1)*
Albumin/Globulin	2.01 (0.3)	1.52 (0.6)***
PT	N/A	28.28 (10.0)
INR	N/A	2.39 (0.8)
TB	13.18 (5.8)	379.73 (101.5)***
Creatinine	75.2 (16.8)	54.74 (14.6)
MELD score	N/A	22.32 (5.9)
RBP4	13.28 (0.4)	2.99 (0.5)***
FBP1	16.2 (8.2)	32.6 (39.1)**

*p < 0.05,

**p < 0.01,

***p < 0.001 vs. healthy participants. Data are presented as means (standard error of the mean). ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; DILI, drug-induced liver injury; MELD, Model for End-stage Liver Disease; TB, total bilirubin; RBP4, retinol binding protein 4; FBP1, fructose-1,6-bisphosphatase 1.

Table II. Steps in refining the 1310 proteins down to the 2 markers.

Protein numbers	Refinement	Method or criteria
1310	Unique Protein ID identified from HAP and LAP fraction combined	I. at least two unique peptides.
		AND
		II. protein level FDR<0.1%.
325	Differential expressed protein based on LFQ analysis	I. >2 fold-changes based on LFQ quantitation.
		OR
		II. absent in one condition while the LFQ intensity exceed E10 in the other condition.
137	Verification by DIA quantitation	I. DIA quantitative results in agree with LFQ results.
		AND
		II. Reliable spectrum and chromatogram that can be used for targeted PRM study.
29	Functional analysis	Energy and carbohydrate metabolism
20	PRM	Reliable quantification as previously described in the supplementary method.
9	Tissue specificity	Hepatic specific as described in the method section.
2	ELISA validation	Successfully validated in human study by ELISA.

Fig. 1.

Design of workflow for ALF-related serum marker development. Pooled samples from normal (P0, n = 10) and ALF (P36, n = 10) were divided into high-abundance species (HAP) and low-abundance species (LAP) for proteomic comparison using label-free quantitation (LFQ) and DIA quantitation verification. Individual serum samples from external verification groups (n = 8) were assayed by targeted proteomic technique (PRM) to assess temporal alterations from P0-P60 of potential markers derived from LFQ-DIA quantitation and functional annotation. Terminal candidates (FBP1, RBP4) were validated by ELISA on the serum of the healthy (n = 40) and ALF patients (n = 34). CPLL, combinatorial peptide ligand library; DDA, data-dependent acquisition; DIA, data-independent acquisition; IHC, immunohistochemistry; TEM, transmission electron microscope.

Porcine Model of ALF

Thirty-eight Bama miniature male pigs (Pig Breeding Centre of Taihe Biotechnology, Jiangsu, China), weighing 18–23 kg, were used in this experiment. All pigs were fed with standard diet and housed in an air-conditioned room (20–25 °C) with a 12-hour light and dark cycle. The protocol of establishing GALN-induced ALF was described previously (2). Briefly, 8-hour fasted pigs were injected with intramuscular ketamine (4 mg/kg; Fujian Gutian Pharmaceutical Co., Ltd., Fujian, China) for sedation and then catheterized with 6.5-F dual-lumen catheter (Baihe Biotechnology, Guangdong, China) through the external jugular vein. Continuous anesthesia was sustained throughout the operation with propofol infused at 2.5 mg/kg via the auricular vein. GALN (1.3g/kg) was administered via catheter for 36 h postoperation, and the end of treatment was marked as time 0. Blood samples (n = 18) were taken through the catheter every 12 h after GALN administration, and the corresponding samples were labeled P0, P12, P24, P36, P48, and P60. Twenty additional pigs were sacrificed at P0, P12, P24, P36, and P60 (n = 4 for each point) post-GALN injection to harvest liver tissues. Death was defined by respiratory arrest and undetectable blood pressure. All protocols had received prior approval of the Animal Care Ethics Committee of Zhejiang Academy of Traditional Chinese Medicine.

Label-free Proteomics for Biomarker Discovery

Pooled serum samples from 10 pigs at P0 and P36 were used in a large-scale label-free comparative proteomic study, from which ALF-related markers were further verified in individual samples at all 5 consecutive time points subsequently from 8 pigs. Before proteomic detection, serum samples were first fractionated by combinatorial peptide ligand library (CPLL) into high-abundance proteins (HAP) and low-abundance proteins (LAP) according to the protocol described previously with several modifications (4, 5). Briefly, a ProteoMiner column (Bio-Rad, Hercules, CA) was loaded with 200 μl of serum sample mixed with 20 μl of PBS and SDS at a final concentration of 0.1%, followed by vortexing for 2 h to obtain the flow-through fraction as HAP. After washing the column three times using 200 μl of PBS supplemented with 0.1% SDS, three consecutive loading cycles, each with 200 μl of loading volume, were performed to process a total of 600 μl of pooled serum for study. The eluent was collected in 40 μl of 4% SDS with 25 mm dithiothreitol (DTT) as LAP. Both HAP and LAP fractions were digested into peptides using the filter-aided sample preparation (FASP) method (6). Briefly, 200 μg of proteins quantified by bicinchoninic acid protein assay were reduced using 25 mm DTT and transferred to a 30 kDa spin filter (Millipore, Temecula, CA). After washing the samples twice using 200 μl of 50 mm triethyl ammonium bicarbonate (TEAB), protein was alkylated by 30 mm iodoacetamide for 30 min in the dark. Samples were washed by 200 μl TEAB (50 mm) twice. Trypsin (Promega Corporation, Madison, WI) was added in 200 μl of 50 mm TEAB with an enzyme-to-protein ratio of 1:50. Peptides were collected after incubation at 37 °C for 14 h. Tryptic peptides were then subjected to high-throughput label-free shotgun proteomic comparisons, employing high-pH reverse phase liquid chromatography (LC) peptide fractionation and low-pH reverse phase nanoLC separation hyphened with MS/MS analyses on a Qualdrople-Orbitrap mass spectrometer (Q Exactive™ Hybrid, ThermoFisher, MA) operated under data-dependent acquisition mode. Data were processed by MaxQuant with LFQ quantitation module. Technical details and selection criteria for significantly changed proteins can be found in supplemental Material. To verify the quantitative results of the differentially expressed proteins selected based on label-free proteomic analyses, and to further guide the peptide pick for the PRM targeted assay, digested peptides from pooled serum (P0 and P36) without prefractionation were subjected to DIA analyses. Briefly, peptides were separated by the same nanoLC-MS method except using the MS/MS scan range of 500–900 Th at 200 m/z. For data analyses, a spectral library was generated by e.g. Skyline (version 3.1.0.7382) based on e.g. MaxQuant search results. Technical details of instrument parameters and data processing can be found in the supplemental Material.

Functional Annotation

DAVID (https://david.ncifcrf.gov/) and STRING (http://www.string-db.org/) were used for Gene Ontology (GO) annotations and functional protein association network analysis. Protein-protein interaction network was visualized by Cytoscape. Liver-specificity was defined by the target hepatic expression levels two times higher than in any nonliver organ as annotated by ProteomicsDB, PaxDb databases.

Temporal Verification of Biomarkers by Targeted Proteomics

PRM, a targeted proteomic technology, was employed to monitor the temporal alteration of 20 ALF-associated protein candidates using serum samples collected from 8 pigs at 0, 12, 24, 36, 48, 60 h post GALN injection. The list of peptides targeted by PRM acquisition can be found in supplemental Table S2. All MS parameters and PRM data analysis procedures are described in the supplemental Materials. For this verification study, individual samples (200 μl) were also processed by CPLL except only one loading cycle was used.

Validation of Expression Level of FBP1 and RBP4 in Liver by Immunohistochemistry and qPCR

Liver tissues were cut into cubes (1-cm square), embedded in paraffin, and then subjected to hematoxylin and eosin (HE) staining and immunohistochemical analysis. A specific antibody against fructose-1,6-bisphosphatase 1 (FBP1) (Abcam, Cambridge, UK) was used in dilution of 1:150 to determine FBP1 expression in the liver. Total RNA was extracted from the liver tissue samples and reverse transcribed into cDNA. From each sample, 10 ng of cDNA was amplified with each primer in RT2 Profiler 96-well PCR array plates (Qiagen, Valencia, Spain) using the ABI 7500 real-time PCR System (Applied Biosystems, Foster City, CA). Gene list and primer sequence can be found in supplemental Table S3. The median cycle threshold and the fold change at each time point post GANL treatment relative to P0 for each gene were calculated.

Transmission Electron Microscope (TEM) Evaluation of Mitochondrial Morphology

Liver tissues from P0 and P36 were minced to obtain grain-sized particles (<5 mm), soaked in 2.5% glutaraldehyde, postfixed in 1% osmium tetroxide, dehydrated in graded alcohols, and embedded in Epon. Appropriately sized particles (200–400 Å) were placed on nickel grids and examined using a digital electron microscope (JEM-1400; JEOL Ltd., Tokyo, Japan). Mitochondrial quantitation from TEM was obtained by Image J (http://rsb.info.nih.gov/ij/).

Clinical Study

ALF is clinically defined by the presence of hepatic encephalopathy and coagulopathy with INR >1.5 within 12 weeks of disease onset without preexisting liver disease (1, 7–10). Based on this criterion, the study enrolled 34 patients between January 2012 and January 2015 at the First Affiliated Hospital of Zhejiang University. All patients were followed up for 30 days, and those admitted to LT were excluded. The control group enrolled 40 healthy individuals from the Health Examination Centre of the First Affiliated Hospital of Zhejiang University. The study protocol was approved by the Ethics Committee of the First Affiliated Hospital, School of Medicine, Zhejiang University. Written informed consent was obtained from the patients or their representatives. Both serum and plasma samples were collected from all study participants. Plasma was used for prothrombin time (PT) and INR assessment by the clinical laboratory of the First Affiliated Hospital, Zhejiang University. Serum was used for biochemical analyses performed using an automated biochemical analyzer (Abbott Aeroset, Abbott Laboratories, Chicago, IL) and frozen at −80 °C until proteomic analysis. The severity of ALF was evaluated by Model of End-stage Liver Disease (MELD) score, which was calculated by 3.78 × ln(serum bilirubin) + 11.2 × ln(INR) + 9.57 × ln(serum creatinine) + 6.4. The possibility of drug-induced liver injury (DILI) was assessed using the Roussel Uclaf Causality Assessment Method (RUCAM), as described elsewhere (11).

Validation of Serum RBP4 and FBP1 Levels in ALF Patients

ELISA was employed to evaluate the serum level of FBP1 (LifeSpan BioSciences, Seattle, WA) and RBP4 (Abcam, Cambridge, UK) according to the manufacturer instruction on the coated 96-well plate.

Statistical Analysis

The results presented are expressed as mean value±standard error of the mean (S.E.). Comparison between different groups was performed using student's t-test. The receiver operating characteristic (ROC) curve analysis was used to assess the predictive value of promising biomarkers. All analyses were performed using SPSS Statistics version 19 (SPSS, Inc., Chicago, IL). Differences were considered significant for p values <0.05, or as otherwise indicated.

RESULTS

d-galactosamine Induced ALF Model

Reduced appetite was observed in all pigs at around 24 h after GALN injection. Yellow urine and slight vesania were noted at around 36 h, followed by somnolence and coma. On average, death occurred at 60 ± 2 h after injection. Judging by the clinical parameters, a gradual deterioration of liver function was noted (Fig. 2A). All parameters demonstrated a sustained elevation starting from P12. In addition to parameters indicative of altered liver function, creatinine levels as a kidney function indicator rose moderately accompanying liver failure. Pathology indicated a clear and intact structure of the hepatic sinusoid and lobule in normal livers, which was devastated at P36. Significant hemorrhage, infiltration of inflammatory cells, presence of symbolic apoptosis bodies and vacuolar hepatocytes can be observed at P36, and all-together indicated the onset of ALF. At P60, hepatocytes were fully vacuolated and pseudolobules were formed (Fig. 2B). Taking all data into account, ALF appeared to be established by 36 h after GALN administration.

Fig. 2.

Dynamic biochemical and histological changes during the ALF progress in pig model. A, Hematological and biochemical parameters changes. Significant changes versus P0 were noted by *p < 0.05, **p < 0.01, ***p < 0.001. ALT: Alanine aminotransferase. AST: Aspartate aminotransferase. INR: International Normalized Ratio. Data are means ± S.E. B, Histological images (HE, ×5 and ×10 shown in the lower right) of pig liver tissue from 0, 36, 60 h post GALN insult. Apoptosis bodies were indicated with black arrow.

Identification of ALF-associated Potential Markers in Pig Model

To capture the overall changes in the pig circulating proteome related to ALF onset, serum samples of P0 and P36 post GALN injection were subjected to a large-scale label-free shotgun proteomic comparison. Serum samples were pooled from 10 pigs, and fractionated by CPLL prior to two-dimensional (2D) LC-MSMS analyses. The large-scale label-free comparative proteomics resulted in identification of 1147 and 490 proteins in the HAP and LAP fraction, respectively, using a 1% FDR threshold. After combining the data, a total of 1310 proteins were identified (Fig. 3A). Subsequently, 325 significantly changed proteins were quantified by DIA, analyzed by Skyline, and identified 137 reliable candidates for the next stage selection. The whole selection procedure was detailed in Table II.

Fig. 3.

Label-free quantitation results. A, Protein identification in HAP and LAP fractions divided from pig serum proteome via CPLL approach. B, Selection of differentially expressed proteins by comparing pooled samples collected immediately after treatment (P0) and 36 h after GALN treatment (P36). Proteins were categorized into significantly and insignificantly changed ones. Proteomics data was analyzed by MaxQuant. Protein quantity were estimated by LFQ method and then log2 transformed. C, In silico interactome analysis of 325 significant changed proteins associated with ALF onset. Color shades (red for high, green for low, gray for data not available) represent their relative abundance when comparing ALF versus normal as quantified via LFQ method. D, The zoom-in view of the red square from (C) highlighting protein network with high connectivity. Proteins marked by hexagon were involved in energy metabolism or mitochondrial related.

Temporal Alteration of 20 Energy Metabolism-related Markers During ALF

Among the 137 proteins derived from DIA results, 29 proteins were energy and carbohydrate metabolism-related and listed for PRM (Fig. 4A). Serum samples collected at 5 time points (P0, P12, P24, P36, P60) post-GALN insult from 8 additional pigs were used in this external verification study to portray the longitudinal changes of these candidates during the ALF pathogenesis. The targeted proteomic study eventually provided reliable quantitation data of 20 proteins, of which 19 shown continuous increasing trends (Fig. 4B) while RBP4 (mentioned below) shown decreasing levels during ALF onset. Significant proportion of the 19 increased proteins were associated with glycolysis (pyruvate kinase, PKLR; lactate dehydrogenase A, LDHA; phosphoglycerate kinase 1, PGK1; glyceraldehyde-3-phosphate, GAPDH), gluconeogenesis (fructose-1,6-bisphosphatase 1, FBP1; phosphoenolpyruvate carboxykinase 2, PCK2), and glycogenolysis (amylo-alpha-1,6-glucosidase, AGL; enolase superfamily member 1, ENOSF1; phosphoglucomutase 1, PGM1). According to the liver specific criterion mentioned before, FBP1, Glutathione S-Transferase Omega 1 (GSTO1), PGM1, carbamoyl-phosphate synthase 1 (CPS1), PKLR, 4-hydroxyphenylpyruvate dioxygenase (HPD), AGL, Phosphorylase, Glycogen, Liver (PYGL), and PCK2 were considered as liver specific. Among these 9 liver specific genes, FBP1 exhibited the highest intensity. Retinol binding protein 4 (RBP4) was the only protein that constantly decreased from the start. Hence, we appointed FBP1 and RBP4 as potential markers for further validation.

Fig. 4.

Energy metabolism alteration in the pig ALF models. A, Serum level changes of energy related proteins based on DIA quantitation. Color depth from green to red indicates the intensity detected by DIA increasing from 0 to 10⁸. B, Increased serum levels of 19 ALF-related candidates protein (except RPB4) detected by PRM. Intensities were log10 transformed by MSstats (described in Supplementary Materials). C, Real-time PCR results of energy metabolism related genes in liver during ALF onset. D, Gene mRNA expression level of mitochondrial specific genes (COX1, COX2) in liver during ALF onset. E, TEM images of mitochondrial morphology changes before and during ALF onset. Compared with the normal hepatocytes, enormous lipid droplets formation and mitochondria cristae depletion were noted in ALF hepatocytes. L, lipid droplets; M, mitochondrion. F, Autophagosome and mitophagosome were indicated by black and red arrow, respectively.

Liver Expression of Energy Metabolism Related Genes During ALF

Among the 20 proteins successfully detected from PRM, two terminal candidates (FBP1 and RBP4) and 3 anaerobic metabolism related proteins (PGM1, PCK2, and LDHA) were chosen to determine their liver mRNA expression alterations. We also extended the investigation to an additional 3 aerobic metabolism genes, enoyl-CoA delta isomerase 1 (ECI1), fumarate hydratase 1 (FH1) and citrate synthase (CS). Results showed gene expression level of RBP4 decreased at P12 whereas others increased at the first 12 h after GALN administration, but subsequently declined (Fig. 4C).

Mitochondrial Morphological Changes During ALF

The proteomic studies indicated a substantial alteration in the energy metabolism that suggested functional changes in the mitochondrion during ALF progression. To test this hypothesis, we measured the mRNA expression of two key mitochondria-specific genes, cytochrome c oxidase 1 and 2 (COX1, COX2), and evaluated the mitochondrial morphology via TEM. As illustrated in Fig. 4D, expression level of both COX1 and COX2 significantly decreased in response to GALN impact at P12 and remain low throughout the following time points. Coordinately, TEM of hepatocytes revealed the degenerative morphology of mitochondria with cristae membrane depletion and matrix swelling at P36. Quantitative analysis also suggested fewer numbers of mitochondria (0.12 ± 0.02 per μm²) with increased length (1.28 ± 0.29 μm) and area (0.028 ± 0.0062 μm²) in ALF versus in normal hepatocytes (0.88 ± 0.1 per μm², 1.07 ± 0.34 μm, and 0.016 ± 0.0049 μm², respectively) (Fig. 4E). All evidence suggested a significant loss of mitochondrial function during ALF development. Interestingly, autophagosomes emerged in the hepatocytes at P36 enclosed with impaired mitochondrion, organelles, or other cytoplasmic contents (Fig. 4F).

Patient Characteristics

All ALF patients experienced rapid deterioration of liver function as indicated by significantly increased INR and liver injury indicators (ALT and total bilirubin (TB)) as well as significant loss of hepatic synthesis capacity as indicated by decreased levels of albumin and albumin/globulin (Table. 1). Nearly half of the ALF patients died within 30 days after hospitalization. Regarding the etiology, we carefully excluded all cases with indication of chronic alcoholic liver disease, steatohepatitis, or chronic viral hepatic infection, and focused on those without chronic liver injury. This selection criterion resulted in a study cohort of ALF patients liver disease mainly caused by drug-induced liver injury (DILI), and therefore were assessed by the RUCAM score system. The cohorts included patients that took Chinese herbal medicine (n = 19), body building supplements (n = 3), antituberculotic, psychotropics, and chemotherapeutic drugs (n = 9). The RUCAM assessment indicated 10 and 16 cases with probable DILI (RUCAM score 6–8) and possible DILI (RUCAM score 3–5), respectively. Patients without a definite medical history or other known causes were considered to have indeterminate etiology (supplementary Table S4).

Validation of FBP1 and RBP4 as ALF Markers in Clinical Samples

The serum levels of FBP1, as assessed via PRM, increased consistently starting from 12 h after GALN administration, whereas the corresponding mRNA levels increased only at P12 (Fig. 5A–5B). Immunohistochemistry shown FBP1 expression in the pig liver was significantly higher at ALF onset than the normal (Fig. 5C). The higher serum level of FBP1 in ALF patients was further confirmed increased 2-fold by ELISA assay (p < 0.05) (Fig. 5D). Additionally, serum FBP1 also showed promising prognostic value for ALF, as its levels in nonsurvivors were higher than the ALF-survivors (p = 0.064) (Fig. 5E). ROC analysis suggested that FBP1 has comparable prognostic value (AUC = 0.778) of short time LT-free mortality (30-day) as MELD (Fig. 5F). At the optimal cut-off value (16.89 μg/dL), FBP1 can discriminate ALF survivors from nonsurvivors with sensitivity of 0.80 and specificity of 0.69. The cumulative survival function indicated that patients with a serum concentration of FBP1 higher than 16.89 μg/dL had a higher death related hazard (HR = 9.89, p = 0,002) than those patients with lower FBP1 level (Fig. 5G).

Fig. 5.

Characterization of FBP1. A, Serum level of FBP1 in pig ALF model determined via PRM. B, Gene expression level of FBP1 in porcine liver. C, Immunohistochemical images of FBP1 indicating its higher expression in the ALF liver compared with that in the normal of porcine liver. D, Serum concentration of FBP1 in ALF patients compared with the healthy (*p < 0.05) through ELISA. E, Comparison of serum concentration of FBP1 between the survivor and non-survivor among the ALF patients (p = 0.064). F, Comparison of ROC curves of FBP1 and MELD to predict 30-day survival of ALF patients. G, Comparison of cumulative survival function of ALF subgroups (p = 0.002) divided by FBP1 cutoff (16.89 μg/dl) derived from ROC analysis.

Distinct from other markers targeted in the PRM study, the RBP4 showed a decline since the first 12 h after GALN administration, which was also confirmed at the mRNA level in the tissue. ELISA assay also confirmed that the serum levels of RBP4 in ALF patients were 4 times lower compared with that in healthy controls (p < 0.01) (Fig. 6).

Fig. 6.

Characterization of RBP4. A, Time-dependent change of RBP4 level in the porcine serum was revealed by PRM as indicated on the left axis, and was compared with that of ALT, which was indicated by the right axis. B, Dynamic mRNA expression changes of RBP4 in pig liver. C, Serum concentration of RBP4 in ALF patients and in healthy individuals, evaluated by ELISA. ***p < 0.001.

DISCUSSION

ALF is a life-threatening condition hallmarked by abrupt onset and severe deterioration of liver function. Therefore, early diagnosis is of great importance for timely treatment and reduction of ALF-related mortality. Furthermore, it is worth developing biomarkers that can help classify ALF patients, on its onset, into subgroups that differ in the extent of liver damage and mortality risk and hence guide personalized treatment and how aggressive the treatment should be. However, these tasks remain challenging because of the heterogeneous background of patients, which makes it difficult to characterize the clinical or molecular changes relevant to ALF development particularly at the early stage. Animal models, such as rodents, dogs, and pigs have been adapted to model ALF pathogenesis, with desirable controllability and homogeneity, to circumvent limitations of clinical studies (12). Large animals such as pigs allow easier sequential sampling and are more appropriate for the assessment of treatment intended for humans. Both hepatotoxic drugs and surgery can be used to induce ALF in animal models (12). However, models built on surgical procedures (hepatectomy and devascularization) are dependent on the expertise of the surgeons, which compromises model reproducibility. In our study, we applied the GALN-induced ALF on the miniature pig model with a survival time of 60 ± 2h and biochemical parameters that displayed good reproducibility of disease onset (2). Furthermore, our model developed both hematological and neurological complications that closely resembled the clinical manifestations of ALF patients. Therefore, we conjectured that liver injury markers developed in this model can be translated into clinical use.

In this study, a multistage discovery-verification proteomic strategy was used to develop potential ALF-related biomarkers using a unique pig model. First, the portrait of a comprehensive view of proteomic alteration in response to GALN insults, and selection of individual marker candidates was accomplished by a high-throughput proteomic workflow based on LFQ semiquantitative strategy. The quantitative results of differentially expressed proteins were confirmed with DIA analysis. Following this exploratory stage, potential biomarkers were verified on a longitudinal scheme using a targeted PRM technique. This approach greatly facilitated the accuration and highly multiplexed quantitation of potential markers developed in the pig model as most of them lack well-characterized, specific antibodies for immunoassay based measurement.

Functional analysis unveiled overrepresentation of energy metabolism-related proteins. It is worthy to mention that because of the limited curation level of porcine protein database, uncharacterized proteins were excluded from the downstream verification process. Though they can be targeted upon future updated annotations, we decided to focus on those with clear functional roles. Using PRM we further verified that serum levels of 19 proteins, mostly involved with the glycolysis and glycogenolysis pathways, followed a parallel increasing trend with the progression of ALF starting at P12 (Fig. 4B). It is highly probable that the elevated serum level of these markers was the direct result of upregulation of gene expression, as shown by FBP1, PGM1, and LDHA, in the liver at P12 in response to a higher energy demand caused by hepatoxic challenge (Fig. 4C). At P24 and thereafter, decreased gene expression was observed that may reflect the impaired gene expression commonly occurring during liver failure. However, the high expression rate at P12 may lead to accumulation of proteins in liver at ALF onset (Fig. 5C) and the increasing level in serum can partly be attributed to increased membrane permeability.

The enrichment of energy metabolism-related proteins changed in our ALF model suggesting its crucial role during the ALF pathogenesis as previous studies displayed (13–18). Furthermore, our PRM results show the upregulation of anaerobic metabolism-related proteins. We therefore propose that there was an energy metabolism switch during ALF from mitochondria dominated aerobic metabolism to anaerobic metabolism. To test this hypothesis, we investigated the impact of GALN on mitochondrial-specific genes COX1 and COX2, both known as the core components of the mitochondrion (complex IV), which serves as the terminal enzyme of the mitochondrial respiratory chain (19). We found that the mRNA levels of both genes in the liver diminished continuously upon GALN administration (Fig. 4D). It was shown that along with the progress of ALF, mitochondrial injury induced mitochondrial permeability transition and oxidative phosphorylation uncoupling, which lead to ATP shortage and apoptosis (17). Electron microscopy showed both a decreasing quantity and morphologic deterioration of the mitochondria at P36 (Fig. 4F). These injured mitochondria may be selectively removed by autophagosomes as a stabilizing mechanism called mitophagy (20, 21), which can be seen at P36 (Fig. 4G). However, further studies are needed to demonstrate the switch of ATP source.

In addition to the significant cellular impact of GANL shown by proteomics, our results also lead to the development of two promising ALF-associated biomarkers: FBP1 and RBP4. FBP1 catalyzes the rate-limiting hydrolysis of fructose-1, 6-bisphosphate to fructose-6-phosphate in the glycerol gluconeogenesis pathway. The upregulation of FBP1 may indicate the high level of glucose metabolism, which protects the liver from the injury caused by the hepatocellular apoptosis (22). Here in our GALN-induced ALF model, the serum level of FBP1 rose since P12 continuously whereas COX 1 and COX 2 decreased, followed by the mitochondrial injury shown in TEM and the hepatocellular apoptosis and necrosis in HE staining at P36. Based on this result, we hypothesized that FBP1 is triggered through its role of protection and energy supply before apoptosis begins. Clinical validation of ALF patients in our study displayed a 2-fold increase of serum FBP1 in ALF versus the normal (p < 0.05). Furthermore, higher levels of FBP1 also correlated with higher risk of ALF-related death. This promising prognostic value of FBP1 may be improved by a combination with other indicators and needs further research.

RBP4 is synthesized in the liver and adipose tissue, and is the sole circulating carrier for retinol (vitamin A). Recent reports also suggest an important role of RBP4 in insulin resistance and glucose metabolism (18, 24–27). However, it has been confirmed that adipose tissue has a limited contribution to the serum pool of RBP4, i.e. the serum level of RBP4 reflects the liver protein synthesis function (24). Despite that increasing level of circulating RBP4 was reported as the result of virus-induced steatosis in patients with chronic hepatitis C (CHC) (28), subsequent research has shown that serum RBP4 displayed a significant decrease in line with the severity of liver injury based on histological grading in CHC, probably ascribed to impaired protein synthesis (25, 29). Consistent with the later study, serum RBP4 levels experienced a sharp decrease (Fig. 6A) accompanied by massive hepatic cell death and reduced synthesis rate (Fig. 6B). It is worth mentioning that serum RBP4 declined sharply within the first 12 h whereas another common index of hepatic product, albumin, remained unchanged until 60 h (data not shown). This different reaction time may be explained by the different half-life times: for RBP4, it is ∼12 h, much shorter than that of ALB (18–20 days)(15, 16, 18). Moreover, RBP4 levels showed changes much earlier than ALT, the most widely used indicator of liver injury. RBP4 decreased (p < 0.01) 2-fold at P12 whereas ALT increased nearly 1.5-fold (p < 0.05). PCR results also confirmed this significant decrease of liver synthesis level at P12. All these findings suggested the possibility of RBP4 as an early indicator of liver damage. Our clinical tests also showed a significant decrease of serum RBP4 on comparing ALF patients with healthy individuals (p < 0.001), implying substantial liver malfunction (Fig. 6C). However, the clinical diagnostic value of RBP4 will be the focus of a future prospective study to see if low levels of RBP4 can predict ALF events in advance. Despite the interesting results provided by our data, this study also has limitations. One major limitation of our study is that only ALF patients caused by DILI (RUCAM > 6) and indeterminate causes were enrolled (supplementary Table. S4). Particularly, half the patient cohort had a history of herbals medicine use, which has increasingly drawn attention as previous studies conducted through the Drug-Induced Liver Injury Network have shown increased morbidity caused by herbal agent use (30, 31). However, we argue that regardless of etiology, the presumed underlying pathophysiology during ALF onset, including loss of mitochondrial function and impaired protein synthesis, should be the same. Therefore, further investigations are warranted to assess diagnostic and prognostic values of our markers in cohorts with alcohol or hepatovirus-related liver injury. Another limitation of our study is the relatively small size of clinical study cohorts, in particular pre-ALF patients, to capture the fluctuation of our markers before ALF onset. To confirm the practical value of our markers and clarify the underlying mechanism of ALF, further studies are warranted, involving a larger cohort size with varying etiologies.

Overall, our proteomic findings confirmed that there was a substantial modulation of energy metabolism during ALF in porcine models. Targeted proteomics results revealed the increasing serum levels of energy metabolism-related proteins, especially those involved in anaerobic metabolism, which is in line with the extensive mitochondrial injury during the ALF onset. Furthermore, two promising ALF biomarkers among these energy metabolism-related proteins were developed: FBP1 may serve as a short-term prognosis indicator for ALF, with higher serum level of FBP1 correlated with higher ALF-related mortality in human studies, whereas RBP4 values signaled liver injury at an early stage, as its decreasing level was observed before ALF elevation in the pig ALF model, and its significant decreasing level in ALF patients was confirmed.

DATA AVAILABILITY

All mass spectrometric raw files are uploaded to iProX reservoir (http://www.iprox.org/, project ID: IPX0000785000).