Can Genes and Gut Microbes Predict Who Responds to Semaglutide for Liver Disease?

In the phase 3 ESSENCE trial, 62.9 percent of patients receiving weekly semaglutide 2.4 mg achieved resolution of steatohepatitis without worsening of fibrosis, compared with 34.3 percent on placebo. A further 37.1 percent of treated patients missed the endpoint entirely. Clinicians currently have no validated way to predict, before the first injection, which patient will land in which group. A new review published in Frontiers in Endocrinology argues that this gap can be closed with biology measured before treatment, and it assembles the candidate evidence into a framework with unusually strict rules about what that evidence is actually allowed to do.

The review, led by Guojian Zhu and Wenlong Yang of Nanchang University, is a narrative synthesis rather than a new study, and its central claim is deliberately modest. Host genotype and gut microbial features, the authors argue, are the two measurable domains that could eventually forecast who benefits from GLP-1-based therapy in metabolic dysfunction-associated steatohepatitis, or MASH. But neither pillar has been tested for these drugs. The genotype evidence comes from dietary weight loss rather than pharmacotherapy, and the microbial evidence comes from a dietary prebiotic. The framework therefore grades every finding by how mature the evidence behind it is, and it assigns both of its own pretreatment pillars to the lowest tier, where they may generate hypotheses but may not gate a prescribing decision.

The mechanistic starting point explains why these two domains were chosen. Hepatocytes carry almost no GLP-1 receptor, so these drugs cannot act on the liver directly. Their hepatic benefit arrives instead through two relays: systemic weight loss, which lowers circulating free fatty acids and improves insulin sensitivity, and signaling along the gut-liver axis, where microbial communities transform bile acids and produce short-chain fatty acids that act on receptors such as FXR and TGR5. Because both relays can be influenced from outside the liver, the review argues, they can in principle be profiled from outside the liver too. Host genotype is fixed at birth and assayable from blood or saliva; the gut microbiome is dynamic but assessable before therapy. Both sit on the pathway through which the drug reaches the liver, which makes them candidate modifiers of how efficiently any given amount of weight loss is converted into hepatic benefit.

The genetic pillar rests on a compact set of variants. PNPLA3 rs738409, TM6SF2 E167K, and a protective splice variant in HSD17B13 have been replicated across cohorts and ancestries as determinants of steatosis, fibrosis progression, and liver-related events. The most direct response data, however, come from a retrospective single-center cohort of 140 Japanese patients with biopsy-proven fatty liver who received a year of dietary consultation. In carriers of the PNPLA3 risk allele, the change in liver stiffness tracked weight change, while in wild-type patients it did not. Carriers also lost weight less readily, at 2.9, 1.3, and 0.2 percent across the CC, CG, and GG genotypes. The review is careful about what this establishes: the analysis compared correlations within strata without a formal genotype-by-weight-change interaction test, and no genotype-stratified GLP-1 trial has demonstrated the effect for these drugs.

The microbial pillar rests on a stronger but structurally different result. In a randomized placebo-controlled trial of resistant starch in MASLD, baseline microbiota was identified as the dominant determinant of response heterogeneity, and a model integrating baseline microbial and clinical features reached an area under the curve of 0.74, with a 95 percent confidence interval of 0.66 to 0.81, in independent multicenter external validation. High Prevotella abundance marked low responders, with coculture and transfer experiments supporting a causal role for the genus. But the authors stress that this is a transfer between two causal architectures, not a short step along a shared pathway. Resistant starch is a substrate delivered to the community, so a community unable to use it is a mechanism of non-response. A GLP-1 receptor agonist acts on host receptors, and the community sits downstream of the drug target. There is no equivalent notion of a community being unable to use the drug.

The framework’s tier system is what gives it teeth. Tier 1 admits findings replicated in at least two independent human cohorts or tested within a single randomized trial, on a hepatic histological or validated outcome, and such findings may carry weight in a stratification decision. Tier 2 admits contested or single-study human evidence, which may enter as an input with an explicit confidence estimate but may not gate a decision. Tier 3 covers extrapolation across drug class, endpoint, or population, which carries no weight at all. Strikingly, no item in the top tier concerns prediction of response from genotype or from baseline microbiota, because no such finding currently meets the criteria. The tier 1 items are the histological efficacy of incretin therapy itself, the dependence of that efficacy on the magnitude of weight loss, the replication of the three risk loci for disease progression, cardiovascular benefit from semaglutide, and the genotype-restricted AZD2693 trials, which enrolled only PNPLA3 148M homozygotes and produced an unambiguous pharmacological effect in them.

Several findings complicate the picture in ways the review treats as load-bearing rather than incidental. The commonly cited fourth locus, MBOAT7 rs641738, was associated with advanced fibrosis only under a recessive model and only in adults of European descent, placing it in tier 2. A signal that diabetes status modifies drug-specific fibrosis response, in which semaglutide improved fibrosis among patients without diabetes but not among those with diabetes while resmetirom showed the reverse, rests on across-trial subgroup comparison without a formal interaction test, and larger syntheses disagree with the reading. On the microbial side, the interventional record is sobering: the largest synbiotic trial altered the fecal microbiome and moved neither liver fat nor validated fibrosis scores, and three randomized trials of fecal microbiota transplantation reported no improvement in hepatic fat or insulin sensitivity. Metformin exposure further confounds the microbial signature, producing the same direction of change in Escherichia and Shigella abundance that is attributed to disease itself.

The review also insists that hepatic and cardiovascular benefits be predicted separately, because the two do not track together. In the SELECT trial of 17,604 patients with established cardiovascular disease and overweight or obesity but without diabetes, semaglutide reduced major adverse cardiovascular events by 20 percent, and the benefit did not order by baseline fibrosis risk as measured by the Fibrosis-4 score. Meanwhile, Mendelian randomization using PNPLA3 I148M as an instrument found that genetically determined liver fat, which conferred an odds ratio of 3.28 for cirrhosis, showed no association with ischemic heart disease across 279,013 individuals, with an odds ratio of 0.98 per allele. The variant anchoring the genetic axis of the framework is therefore the one this analysis finds uninformative about coronary risk. A framework calibrated only for the liver, the authors argue, would be answering the lesser question, since cardiovascular disease rather than hepatic decompensation is the dominant cause of death in this population.

What the framework permits at the bedside today is limited, and one negative rule takes precedence over every positive inference. Predicted hepatic non-response is not, by itself, sufficient grounds for withholding or discontinuing a GLP-1 receptor agonist, because the cardiovascular benefit established in SELECT was never conditioned on hepatic status at all; the trial required no diagnosis of MASLD. The authors specify refutation conditions for each axis, including genotype-stratified reanalysis of completed incretin trials and baseline microbial profiling of stored samples, none of which requires new patient recruitment. Until those analyses are done, the framework specifies what precision therapy in MASH would require rather than what it delivers. The immediate value, the review concludes, lies in study design: completed trials with stored biospecimens can be interrogated now, and trials beginning now can prespecify the stratification that retrospective analysis can only approximate.

Subject of Research: An evidence-tiered framework for predicting response to GLP-1-based therapy in MASLD/MASH using host genotype and gut microbiome features

Article Title: Predicting response to GLP-1–based therapy in MASLD/MASH: an evidence-tiered, endpoint-specific framework built on host genotype and gut microbial features

Article References: Zhu, G., Huang, Y., Huang, Y., & Yang, W. (2026). Predicting response to GLP-1–based therapy in MASLD/MASH: an evidence-tiered, endpoint-specific framework built on host genotype and gut microbial features. Frontiers in Endocrinology, 17, 1962068. https://doi.org/10.3389/fendo.2026.1962068

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.1962068

Keywords: GLP-1 receptor agonist, semaglutide, MASH, MASLD, PNPLA3, gut microbiome, precision medicine, liver fibrosis, ESSENCE trial, cardiovascular risk, genetic risk score, response prediction