Serum proteomics of adults with acute liver failure provides mechanistic insights and attractive prognostic biomarkers

Abstract

Background & aims: Acute liver failure (ALF) is defined as rapid onset coagulopathy and encephalopathy in patients without a prior history of liver disease. We performed untargeted and targeted serum proteomics to delineate processes occurring in adult patients with ALF and to identify potential biomarkers.

Methods: Sera of 319 adult patients with ALF (∼50% acetaminophen [APAP]-related cases) were randomly selected from admission samples of the multicenter USA Acute Liver Failure Study Group consortium and subdivided into discovery/validation cohorts. They were analyzed using untargeted proteomics with mass spectroscopy and a serum cytokine profiling and compared with 30 healthy controls. The primary clinical outcome was 21-day transplant-free survival. Single-cell RNAseq data mapped biomarkers to cells of origin; functional enrichment analysis provided mechanistic insights. Novel prognostic scores were compared with the model for end-stage liver disease and ALFSG prognostic index scores.

Results: In the discovery cohort, 117 proteins differed between patients with ALF and healthy controls. There were 167 proteins associated with APAP-related ALF, with the majority being hepatocyte-derived. Three hepatocellular proteins (ALDOB, CAT, and PIGR) robustly and reproducibly discriminated APAP from non-APAP cases (AUROCs ∼0.9). In the discovery cohort, 37 proteins were related to 21-day outcome. The key processes associated with survival were acute-phase response and hepatocyte nuclear factor 1α signaling. SERPINA1 and LRG1 were the best individual discriminators of 21-day transplant-free survival in both cohorts. Two models of blood-based proteomic biomarkers outperformed the model for end-stage liver disease and ALFSG prognostic index and were reproduced in the validation cohort (AUROCs 0.83-0.86) for 21-day transplant-free survival.

Conclusions: Proteomics and cytokine profiling identified new, reproducible biomarkers associated with APAP etiology and 21-day outcome. These biomarkers may improve prognostication and understanding of the etiopathogenesis of ALF but need to be independently validated.

Impact and implications: Acute liver failure (ALF) is a sudden, and severe condition associated with high fatality. More sensitive and specific prognostic scores are urgently needed to facilitate decision-making regarding liver transplantation in patients with ALF. Our proteomic analysis uncovered marked differences between acetaminophen and non-acetaminophen-related ALF. The identification of routinely measurable biomarkers that are associated with 21-day transplant-free survival and the derivation of novel prognostic scores may facilitate clinical management as well as decisions for/against liver transplantation. Further studies are needed to quantify less abundant proteins. Although we used two cohorts, our findings still need to be independently and prospectively validated.

Keywords: ALF subtyping; Acetaminophen; Acute liver injury; Proteomic profiling.

Graphical abstract

Fig. 1

Serum proteomic alterations in patients with ALF vs. healthy controls. (A) The first two dimensions of the principal component analysis highlight marked differences in proteomic signatures of patients with ALF (discovery cohort) and healthy controls. (B) The corresponding volcano plot depicts the 187 differentially abundant proteins (98 elevated and 89 diminished in ALF, Bayesian linear regression). A log₂ fold-change >0 indicates proteins elevated in patients with ALF vs. healthy controls (discovery cohort). (C) Chord diagram assigning the proteins altered between patients with ALF and healthy controls to publicly available liver single-cell RNAseq data (discovery cohort). Up/down refers to features increased/decreased in patients with ALF. (D) Serum levels of characteristic proteins that differ between patients with ALF and healthy controls, displayed via dot plots. Significance levels are as follows: ∗p <0.05, ∗∗p <0.01, ∗∗∗p <0.001 (Bayesian linear regression). (E) Volcano plot visualizing the results of the differential abundance analysis of a cytokine profiling comparing serum levels between patients with ALF and healthy controls (discovery cohort). The dotted horizontal lines in (B) and (E) depict an FDR <0.05. ALDOB, aldolase B; ALF, acute liver failure; APOC1, apolipoprotein C–I; APOC3, apolipoprotein C-III; FAH, fumarylacetoacetase, FDR, false discovery rate; NA, not available; not sig, features not reaching statistical significance in differential abundance analysis (p <0.05); vWF, von Willebrand factor.

Fig. 2

Serum proteomic alterations in patients with APAP vs. non-APAP ALF. (A) The first two dimensions of the principal component analysis highlight profound differences in proteomic signatures of patients with APAP and non-APAP ALF (discovery cohort). (B) The corresponding volcano plot depicts 167 differentially abundant proteins (83 elevated and 84 diminished in APAP; Bayesian linear regression). A log₂ fold-change >0 indicates proteins elevated in APAP vs. non-APAP (discovery cohort). (C) Chord diagram representing the mapping of features altered between patients with APAP and non-APAP ALF to publicly available liver single-cell RNAseq data (discovery cohort). Up/down refers to features increased/decreased in patients with APAP ALF. (D/E) Serum levels of the top three discriminating features (ALDOB, CAT, and PIGR) are depicted via dot plots for both the discovery (D) and the validation (E) cohort. Significance levels are indicated as follows: ∗p <0.05, ∗∗p <0.01, ∗∗∗p <0.001 (Bayesian linear regression). (F) Volcano plot representing the results of the differential abundance analysis of the cytokine profiling between patients with APAP and non-APAP ALF (discovery cohort). Dotted horizontal lines in panels (B) and (F) depict FDR <0.05 (Bayesian linear regression). (G/H): Discriminative performances for the three proteins were assessed in univariable logistic regression models, results are depicted via receiver operating curves for both the discovery (G) and validation (H) cohort. ALDOB, aldolase B; ALF, acute liver failure; APAP, acetaminophen; APOC1, apolipoprotein C–I; APOC3, apolipoprotein C-III; FAH, fumarylacetoacetase; FDR, false discovery rate; NA, not available; not sig, features not reaching statistical significance in differential abundance analysis (p <0.05); vWF, von Willebrand factor.

Fig. 3

Serum proteomic alterations in patients with ALF with vs. without spontaneous survival. (A/B) The first two dimensions of the principal component analysis (A) show no obvious separation between patients passing away or receiving liver transplantation within 21 days after admission (non-SpS) and spontaneous survivors (SpS). A 39-protein signature (12 elevated and 27 diminished in SpS) discriminated both groups as seen in the corresponding volcano plot (B) (FDR <0.05, Bayesian linear regression). A log₂ fold-change >0 indicates proteins elevated in non-SpS vs. SpS (discovery cohort). (C) The signature of altered proteins was mapped to publicly available liver single-cell RNAseq data (discovery cohort). (D/E) SERPINA1 and LRG1 are robustly associated with 21-day outcome in both discovery (D) and validation (E) cohort. Significance levels are indicated as follows: ∗p <0.05, ∗∗p <0.01, ∗∗∗p <0.001 (Bayesian linear regression). (F/G) Ingenuity Pathway Core Analysis (IPA) depicts 10 pathways (F) and upstream regulator networks (G) that were most significantly altered in the 21-day outcome dataset (full results, see Tables S9 and S10) (Fisher’s exact test) (discovery cohort). (H) Results of the cytokine profiling comparing non-SpS and SpS are depicted in a volcano plot; six features were differentially abundant (FDR <0.05, Bayesian linear regression) (discovery cohort). 21-day outcome, spontaneous survival vs. liver transplantation or death during the first 21 days post study admission; FDR, false discovery rate; hep, intr, hepatocellular intracellular features; hep, secr, hepatocellular secreted features; LRG1: leucine-rich alpha-2-glycoprotein; MELD: model for end-stage liver disease; NA, not available; not sig, features not reaching statistical significance in differential abundance analysis (p <0.05); SERPINA1: alpha1-antitrypsin.

Fig. 4

Ability of selected proteins and models to predict 21-day outcome in patients with ALF. (A/B) Areas under receiver operating curves (AUROCs) delineate the ability of the top three biomarkers to predict ALF outcome in the discovery (A) and validation (B) cohort (univariable logistic regression). (C/D) AUROCs for best-performing five-feature models (model 1 comprising SERPINA1, IL6, EGF, ATRN, and serum bilirubin; model 2 comprising SERPINA1, INR, the need for mechanical ventilation, EGF and serum bilirubin) are compared to the ALFSG-PI, the model for end-stage liver disease (MELD) and the King’s College Criteria (KCC) in the discovery (C) and validation (D) cohort (multivariable logistic regression). ALFSG-PI: ALFSG prognostic index; ATRN: attractin; EGF, epidermal growth factor; INR: international normalized ratio; SERPINA1: alpha1-antitrypsin.

Fig. 5

Proteins associated with spontaneous survival among patients with APAP-induced ALF. (A/B) Volcano plots depicting differential abundance analysis results in patients with APAP in the discovery (A) and validation cohort (B) (Bayesian linear regression). A log₂ fold-change >0 indicates proteins elevated in non-SpS vs. SpS. (C/D) Dot plots visualizing protein levels of features associated with 21-day outcome (SERPINA1, LRG1, AGT) in APAP cases (discovery: C, validation: D). Significance levels are indicated as follows: ∗p <0.05, ∗∗p <0.01, ∗∗∗p <0.001 (Bayesian linear regression). (E/F) Predictive performances as determined via logistic regression (discovery: E, validation: F) of SERPINA1, LRG1, AGT, and MELD in patients with APAP. AGT, angiotensinogen; LRG1, leucine-rich alpha-2-glycoprotein; MELD, model for end-stage liver disease; not sig, features not reaching statistical significance in differential abundance analysis (p <0.05); SERPINA1, alpha1-antitrypsin.