Transcriptomics: Five-Gene Signature Flags At-Risk MASH
Deciding which fatty liver patients need treatment still comes down to a biopsy needle. At-risk metabolic dysfunction-associated steatohepatitis (MASH), defined by an activity score of at least 4 with stage 2 or worse fibrosis, is the subgroup that drifts toward cirrhosis, and no blood test separates it cleanly from benign steatosis. A group at Ruijin Hospital went after that gap with liver transcriptomics, asking whether genes governing efferocytosis (the clearance of dying cells by macrophages) could flag at-risk disease.
Key Takeaways
Liver transcriptomics across 300 biopsy samples produced a five-gene efferocytosis signature (CD24, CHI3L1, TREM2, PTGS2, LGR6) separating at-risk MASH from milder fatty liver disease.
The score reached an AUC of 0.866 in training and 0.805 in an independent cohort, which is triage-grade performance rather than diagnostic-grade.
Two markers, TREM2 and CHI3L1, are measurable in plasma, giving a blood-based version of the score with an AUC of 0.736.
Every result comes from public expression datasets plus a mouse model, so nothing has been tested prospectively in one patient cohort.
Liver Transcriptomics Points Toward Efferocytosis
Training used GSE135251, the biopsy-anchored RNA-seq resource assembled by Govaere and colleagues (2020), contributing 206 samples once healthy controls were dropped: 110 at-risk, 96 not-at-risk. Differential testing by limma-voom returned 970 genes, lopsidedly upregulated (854 of them). Intersecting those with 242 curated efferocytosis genes left 17 candidates, enriched around macrophage activation and IL-17 signaling. Stepwise regression, LASSO, and SVM-RFE then ran in parallel, and five genes appeared on all three lists.
Key Findings
Five genes, three methods, one overlap: CD24, CHI3L1, TREM2, PTGS2, and LGR6 were shared by every selection approach and significantly upregulated in at-risk MASH.
Single genes were weak; the combination was not: individual AUCs sat between 0.695 and 0.765, while the five-gene nomogram reached 0.866 in training and held 0.805 in GSE174478.
Mouse expression tracked the human signal: mice on a Western diet for 24 weeks showed higher activity scores, more collagen, and elevated mRNA and protein for all five genes versus 12-week animals.
Only part of the signature reaches blood: across a separate plasma proteomic cohort of 191 patients, TREM2 and CHI3L1 were elevated (AUC 0.693 and 0.691) while CD24 and LGR6 went undetected.
Figure 1. Study workflow. The schematic moves through four stages: screening, where genes differentially expressed in GSE135251 are intersected with efferocytosis-related genes and passed to GO and KEGG enrichment; hub gene selection by stepwise regression, LASSO, and SVM-RFE with mouse validation; nomogram construction, external validation, and plasma biomarker assessment; and function characterization by GSEA, immune infiltration, and single-cell analysis. Adapted from Jiang et al. (2026), Genes.
Turning Gene Expression Profiling Into a Score
Correlation among the five hub genes was low, so collinearity did not inflate the combined model. The number worth anchoring on is external validation in GSE174478 (50 at-risk, 44 not-at-risk): an AUC of 0.805, with risk scores clearly shifted upward in at-risk patients. That sits in the same range as FIB-4 and the NAFLD Fibrosis Score. Reporting the comparison is honest; it also means a tissue panel has to beat cheap clinical indices to earn its place.
Where the Signal Lives in Liver Tissue
CIBERSORT deconvolution showed more plasma cells and M0/M1 macrophages in at-risk livers, with fewer resting NK cells and neutrophils; single-cell RNA-seq agreed. Cell-type attribution matters more than it first appears. If CHI3L1 is largely a neutrophil transcript in liver, bulk expression partly reports infiltration rather than hepatocyte biology.
Matching Tissue RNA to Plasma Protein
The tissue signal and the plasma signal here come from different people, which is the usual compromise when tissue transcriptomics and protein measurement sit in separate labs. Running both layers on matched material closes that gap; the single-injection Omni-MS workflow at Dalton exists partly for this reason, since proteins and metabolites come off the same aliquot instead of a second draw. Teams scoping a multi-omics biomarker discovery program should treat matched-sample design as what makes a score like this testable rather than suggestive.
Frequently Asked Questions
What is transcriptomics used for in biomarker research?
Transcriptomics measures which genes are expressed, and how strongly, in a given tissue. In biomarker work it is the discovery layer: RNA-seq surveys tens of thousands of transcripts and narrows the question to a handful of candidates, which then need protein-level confirmation in a sampleable fluid.
Can transcriptomics identify at-risk MASH without a liver biopsy?
Not yet on its own. This study found candidates in biopsy tissue, then asked whether the matching proteins appear in plasma. Two of five did, and the blood model reached an AUC of 0.736, short of what a biopsy-sparing test needs.
How many samples does an RNA-seq biomarker study need?
This one used 206 training and 94 validation samples, a workable range for a binary phenotype with moderate effect sizes. Signatures built on fewer than 50 samples per group tend not to replicate. Budget for independent validation from the start; that is where most panels fail.
Conclusion
Efferocytosis genes track with histological severity in MASH, and that much looks solid across two public cohorts plus a mouse model. What isn't shown is that the panel adds anything beyond FIB-4 in clinic, since tissue and plasma were never measured in the same patients. The practical read is narrow: TREM2 and CHI3L1 are worth a prospective matched-sample study, and the other three probably aren't.
Related Reading
See how we run these analyses in one lab: Dalton's multi-omics CRO services.
Citation
Jiang, J., Mao, X., Lou, W., Lou, W., Li, Z., Zhang, X., Xie, Q., & Lai, R. (2026). An efferocytosis-associated gene signature for identifying at-risk MASH: Transcriptomic and exploratory plasma biomarker assessment. Genes, 17(8), 948. https://doi.org/10.3390/genes17080948
Note
This blog post summarizes findings from the above-cited research. Figures are adapted from the original publication. For full details, please refer to the source article.
By Seungjun Yeo, CEO at Dalton Bioanalytics. Specializing in multi-omics mass spectrometry for drug discovery and biomarker research.