Biomarkers of Alcoholic Hepatitis

Abstract
Our overarching goal is to develop minimally invasive approaches to better predict outcome and novel
mechanisms in alcoholic hepatitis (AH). AH is characterized by acute hepatic decompensation and multiple
organ failure. Although supportive care for AH has improved, short-term mortality has largely remained
unchanged (30-40%) for decades. Effective approaches to predict risk hamper the treatment of AH. The hepatic
extracellular matrix (ECM) responds dynamically to organ injury and ECM turnover increases; we propose to
take advantage of this to develop new biomarkers for AH. The peptidome, low molecular weight peptides in
biologic fluids, includes not only synthesized peptides, but fragments of degraded proteins (i.e., ‘degradome’).
We hypothesize that the ECM degradome in plasma will yield new biomarkers to predict outcome and
mechanisms in AH. We will test this hypothesis via the following Specific Aims: 1). To identify key changes
in the peptidome as predictive biomarkers of outcome in AH. Unbiased peptidomics and multivariate analyses
will identify degradomic features independently linked to prognosis. Protease activity that could produce
significantly changed peptides will be predicted using Proteasix. We will also determine the mechanistic role of
ECM turnover in the in parallel established models of alcohol-induced liver injury. 2) To develop probabilistic
graphical models to predict outcome in AH. Whereas we expect the results of Aim 1 to establish that the
peptidome profile in patients correlates with overall outcome, biomarkers alone are often insufficient to accurately
predict individual patient outcome. We will therefore employ machine learning methods like probabilistic
graphical models (PGMs) over mixed data types to integrate peptidomic and individual patient clinical data, into
a single probabilistic graphical framework. The resulting graphs will then be used to infer causal interactions
between variables, select informative biomarkers that will more specifically predict the outcome, and gain new
mechanistic insight into the biology of AH (hypothesis generation). 3) To validate the use of the peptidome as
a predictive tool for determining outcome in AH. Using a large prospectively-designed patient cohort with
established outcomes, we will test the ability of the algorithms and biomarkers generated in this study to predict
outcome. The successful completion of the proposed work will produce significant results at various levels: (1)
Biomarker discovery: we will identify biomarkers and conditional biomarkers for AH prognosis. (2) Mechanistic
understanding of AH: our models will generate hypotheses about the interactions between variables at different
scales (molecular, individual) that will provide insights on the proteins that are involved in AH. (3) Algorithm
development: through this project we will extend our mixed data graph learning algorithms to include censored
variables (i.e., survival data). As a result of the above, this project is likely to yield novel diagnostic tools for AH
that may also translate to other liver diseases.

Grant Number: 5R01AA028436-05
NIH Institute/Center: NIH
Principal Investigator: Gavin Arteel