Considering the Liver as a Cellular and Genetic Ecosystem: A New Way to Read the Hepatic Transcriptome in the Context of Alcohol-Associated Liver Disease by Adapting an Approach from Rainforest Ecology

Gavin E. Arteel, PhD, FAASLD, professor of medicine, Division of Gastroenterology, Hepatology and Nutrition at UPMC and the University of Pittsburgh, is the senior author of a study published in JCI Insight that applies ecological diversity metrics to the hepatic transcriptome in alcohol-associated liver disease. Dr. Arteel also is the associate vice chair, Faculty Development, and associate chief, Basic Science, as well as Pilot and Feasibility Core Director, Pittsburgh Liver Research Center.

The study, “Alpha Diversity Analysis of Hepatic Transcriptome Reveals Distinct Pathways in Alcohol-Associated Hepatitis,” used a novel approach and new mathematical model the team created to categorize how genetic transcription changes in human livers with alcohol-associated hepatitis and alcohol-associated liver disease, compared to healthy livers.

Below, Dr. Arteel explains the thinking behind the work, what it found, and how this approach may have applicability for studying other chronic conditions in different organs.

A summary article of the study and its findings can be found in this article: What Rainforest Ecology Analytical Methods Can Reveal About Alcohol-Associated Liver Disease and Potentially Other Chronic Illnesses.

Q: What is the clinical picture for alcohol-associated hepatitis, and why is it so difficult to address?

A: Alcohol-associated hepatitis is a subacute form of alcohol-associated liver disease. It is a clinical phenomenon where the patient decompensates: the liver stops functioning, they develop jaundice, ascites, other complications of severe hepatic failure. The mortality rate is between 20% and 40% within 90 days. That number has been the same for 60 years. The only U.S. Food and Drug Administration-approved therapy is corticosteroids, which have limited effectiveness, and a lot of current work is focused on predicting, as accurately as we can, who is going to live and who is going to die. That is the primary clinical question. The primary scientific question is why this happens in the first place. If we understand that better, we can target it better.

The other dimension of the problem is scale. Alcohol is the most common toxin voluntarily consumed at toxic levels by humans. In the United States, roughly one in 10 Americans has an alcohol use disorder, which means something on the order of 30 to 35 million people are engaging in behavior patterns that substantially increase their risk of developing liver disease, including alcohol-associated hepatitis. We know who is broadly at risk. What we don't have is any practical means of stratifying those individuals: who among them is approaching the threshold of decompensation? To screen 30 million people with the tools we have now would bankrupt the medical system. So, we are left with a situation where we know who is at risk generally speaking but we don't know who is truly at risk for a very bad outcome, and we don't know why the liver fails when it does, which means we can't effectively treat it once it happens.

Q: Where did the idea to apply ecological diversity analysis to the transcriptome come from?

A: Alcohol-associated hepatitis is now understood, at least in part, as a form of metabolic collapse. The liver has been damaged over such a long period of time that it starts to lose its identity as a liver. Because the liver is constantly trying to regenerate itself from the damage being caused, too few hepatocytes end up doing what hepatocytes are supposed to do, and that is what leads to decompensation. When I was thinking about that, what it reminded me of was ecosystem collapse. When an ecosystem is under sustained severe environmental pressure, species die off. Diversity collapses progressively to a point, and then the system collapses entirely. That is well understood for ecosystems in the natural world. It struck me that what we were seeing in the failing liver was a similar kind of pattern.

There is also a practical connection. A lot of the mathematical tools we already use in RNA sequencing analysis were originally derived from ecosystem analysis. Principal component analysis is one example. We use these tools because when you are looking at the totality of gene expression in an organ, genes don't change one by one. They change in groups, in coordinated responses to stress or other factors. So, the idea of looking at the liver transcriptome as an ecosystem, where each gene is a species and its expression count is the population of that species, was not as large a conceptual leap as it might sound at first. The question was whether we could apply indices of alpha diversity, which ecologists have used for decades to quantify ecosystem health, and use that approach to identify when the liver is approaching a point of systemic collapse.

The Shannon diversity index is probably the most intuitive way to explain what we were measuring. Imagine a forest of 100 trees with 10 species, 10 trees of each. That is a highly diverse forest. Now imagine the same 100 trees with 10 species, but 91 of them are one specie and the other nine species have one tree each. The Shannon index would correctly identify the second forest as far less diverse. That is what we were applying to gene expression: how many different genes are being expressed, and how evenly is that expression distributed in the liver.

Q: What did you find when you applied those metrics to the data?

A: Alpha diversity in the hepatic transcriptome declines in a stepwise pattern across the disease spectrum, from healthy controls to early alcohol-associated steatohepatitis to alcohol-associated hepatitis. Multiple indices show the same trend: Shannon entropy, the Brillouin index, evenness, equitability, all of them fall as disease severity increases. Dominance goes in the opposite direction, meaning that a smaller and smaller subset of genes accounts for a larger and larger share of total transcription activity in severe disease. The transcriptome is becoming less diverse, more dominated by a few highly expressed genes.

When we looked at what was driving that, it was the low-abundance genes dropping off, not changes in the highly expressed genes. The dominant genes barely moved. That is consistent with what happens in ecosystems, where rare species are typically the first to disappear under environmental stress. In the liver, those lower-abundance genes are probably, at least initially, less essential to immediate survival, I would theorize. The organ shuts them off as part of an acute phase response, effectively triaging its resources toward managing the insult. In patients with early subclinical disease, that may actually be an adaptive response. In patients with alcohol-associated hepatitis, we think the response has been too dramatic, too many genes have been lost, and the organ can no longer maintain function. You then end up in liver failure, and that is not a good place to be.

Q: How does the new model your team created, Differential Shannon Diversity (DSD) differ from standard Differentially Expressed Genes (DEG) analysis, and what did it find that DEG missed?

A: Standard DEG analysis looks at genes one at a time. It asks whether a given gene is upregulated or down regulated between conditions, but it doesn't factor in what is happening to the transcriptome as a whole. That has been a recognized limitation for years. Low-abundance genes often don't pass statistical significance in DEG analysis because their variability is weighted against highly expressed genes. The signal gets lost. DSD performs the same gene-by-gene comparison, but it calculates each gene's proportional contribution to total Shannon entropy, essentially its share of the ecosystem, and uses that as the unit of comparison instead of raw expression counts. That means changes are evaluated in the context of what is happening to the whole transcriptome. A gene that increases modestly, in a transcriptome that has lost 30% of its expression overall, is weighted differently than it would be under DEG.

In our analysis, the two methods overlapped substantially, which we expected and is reassuring. But DSD identified genetic pathways that DEG did not flag, including ones related to fatty acid oxidation, cholesterol metabolism, and extracellular matrix degradation. Those findings were not surprising to us biologically in the context of liver disease. They fit the natural history of alcohol-associated liver disease very well. The fact that they did not appear in DEG analysis, but made biological sense once DSD found them, actually validated the approach for us. If DSD had found things that were biologically implausible, that would have been a concern. Finding things that are consistent with what we know about the disease, but that standard analysis was not detecting, tells us the method is working.

Q: What are the translational implications of your team’s work?

A: The biopsy-based approach we used here to get our data is not going to become a clinical screening tool. You cannot biopsy 30 million people, and even if you could, the time required to process and analyze the data would make it grossly impractical for the clinical decisions that need to be made in real time. But what our research can do is help identify the specific genes that may shift at the tipping point between an adaptive physiologic response and total liver collapse. Once you know which genes those are, you can ask what those genes do. The liver is a secretory organ. Its primary function is metabolic, but it also produces essentially all of the blood proteins. If a gene that is critical at that tipping point is involved in something the liver releases into circulation, that could make it a plausible peripheral biomarker, something you could detect in a blood draw.

The goal is to understand the mechanism in enough detail that we can go looking for it in a patient’s blood. That is a more realistic path to a screening tool than trying to biopsy the population at risk, because it is simply too large. And, if something's involved in causing that tipping point, that is also, generally speaking, something that can be targeted therapeutically. Understanding the mechanism and finding something to target are not separate problems.

A final important note I want to point out about our team’s research is that this paradigm is probably not specific to alcohol-associated liver disease. Any organ – kidneys, hearts, intestines – under sustained pathological stress is being asked to do something similar to what the liver does in the face of consistent, long-term alcohol exposure: manage the damage, try to maintain vital functions, triage resources as best as possible. My expectation is that progressive reduction of transcriptional diversity is a feature of chronic disease more broadly, not just liver disease. In fact, we are already applying DSD analysis in other disease contexts to test that hypothesis. If it holds, the method our team has created could have applications well beyond looking at alcohol-associated liver disease. That’s a very exciting prospect, to have an analytic approach that could have far-reaching applicability. We will see.

Study Reference

Chaudhary S, Liu JJ, Liu S, Di M, Beier JI, Bataller R, Argemi J, Benos PV, Arteel GE. Alpha Diversity Analysis of Hepatic Transcriptome Reveals Distinct Pathways in Alcohol-Associated Hepatitis. JCI Insight. 2026. Jan 8: e200727. Online ahead of print.
Note: Paper is open access.