Abstract
Background/Aims
MicroRNA (miRNA) isoforms (isomiRs) broaden the regulatory landscape of canonical miRNAs, but their role in metabolic dysfunction-associated steatotic liver disease (MASLD) remains unknown. We aimed to characterize the hepatic isomiR landscape in MASLD and define their association with disease activity and fibrosis.
Methods
Small RNA (sRNA) sequencing was performed on liver biopsies from 79 patients across the histological spectrum of MASLD. IsomiRs were annotated and quantified. Their association to disease activity and fibrosis score was assessed by differential expression, ordinal regression, and machine learning. Parallel mRNA sequencing and pathway enrichment were used to map isomiR–mRNA interactions and regulatory networks, which were validated against an independent dataset.
Results
MiRNAs accounted for 75% of sRNAs in liver tissue, of which 67% were isomiRs. Across MASLD severity, 173 isomiRs correlated with disease activity and 58 with fibrosis stage. Key findings included a miR-122 isomiR uniquely targeting INSIG1 (cholesterol metabolism) and a miR-21 isomiR targeting PPARA and HMGCS2 (lipid and fibrosis pathways). Integration with mRNA data revealed 33 dysregulated pathways, including PPAR signaling, insulin resistance, and TGF-β response. Several novel isomiRs from miR-26b, let-7c, and miR-32 families were also linked to lipid metabolism and fibrosis progression.
Conclusions
IsomiRs represent the majority of hepatic miRNAs and uncover novel regulatory networks masked by canonical miRNA analysis. These findings provide new insights into the molecular heterogeneity of MASLD, highlight candidate pathways driving disease progression, and identify potential biomarkers and therapeutic targets for precision hepatology.
Keywords: MicroRNA, IsomiR, MASLD, Fibrosis
Graphical Abstract
INTRODUCTION
Metabolic dysfunction-associated steatotic liver disease (MASLD) is a leading cause of liver-related mortality and increases the risk of cardiac and cancer related deaths, contributing to the burden of non-communicable diseases [1-3]. MASLD includes two main histological phenotypes: metabolic dysfunction-associated steatotic liver (MASL) and metabolic dysfunction-associated steatohepatitis (MASH), the latter being more active and more likely to progress to cirrhosis and end-stage liver disease.
MASLD is a heterogeneous disorder driven by complex gene-environment interactions, contributing to variable clinical and histological responses to therapy, even with promising drugs such as GLP-1, FGF-21, and THR-beta receptor agonists [4,5]. Understanding this heterogeneity is essential for advancing precision therapeutics. One source of variability lies in differences in gene expression across individuals with similar histological severity, influencing downstream pathways in metabolism, inflammation, and fibrosis [6,7].
MicroRNAs (miRNAs) regulate this process by targeting messenger RNA (mRNA), leading to pleiotropic translational repression [8]. Nearly two decades ago, we reported the first miRNA profile of MASLD (previously termed NAFLD) [9]. Since then, multiple miRNAs have been linked to liver disease through their role in lipid metabolism (e.g., miR-122 and miR-34a), inflammation (e.g., miR-22, miR-122), fibrosis (e.g., miR-10b, miR-132, miR-21) and oncogenesis (e.g., miR-22, miR-221, miR-122, miR-132) [10-16]. These findings have led to clinical trials of anti-miR-22-5p (RES-010, Resalis Therapeutics) and anti-miR-132-3p therapeutics (Regulus Therapeutics) [17,18].
Recently, attention has shifted to isomiRs, isoforms of canonical miRNAs with 5’ and 3’ shifts or single nucleotide polymorphisms that can alter regulatory function [19]. Prior studies examined isomiRs in lipid metabolism and cardiovascular risk in MASLD and cirrhosis [20-22], none have examined their associations across the histological spectrum of MASLD. Therefore, we hypothesized that isomiRs can play a role in MASLD biology and provide additional insights on regulatory targets involved in disease. To address this we (i) defined the spectrum of isomiRs across MASLD histology, (ii) linked specific isomiRs to disease activity and fibrosis score, and (iii) predicted their mRNA targets and pathways to uncover regulatory networks underlying MASLD heterogeneity.
MATERIALS AND METHODS
Patient cohort, liver samples, and histological assessment
Patients with MASLD attending clinics at Virginia Commonwealth University Health System provided consent for the use of liver tissue in translational studies. Biopsies were stained with H&E or Masson trichrome. Steatosis, hepatocellular ballooning, lobular inflammation, and fibrosis were scored (Supplementary Table 1). NAFLD activity score (NAS) was calculated as the sum of steatosis, lobular inflammation, and ballooning.
RNA extraction
Total RNA was extracted from 10 mg of frozen tissue using the miRNeasy Tissue/Cells Advanced Kit. RNA integrity (RIN) was measured using LabChip GX Touch RNA Assay. Concentrations were quantified with the Quant-IT RNA Assay Kit.
Small RNA (sRNA) sequencing
Libraries were generated from 300 ng of total RNA using the NextFlex Small RNA-Seq Kit v3 (Revvity Health Sciences Inc., Boston, MA, USA) with 18 PCR cycles. Fragments of 140–200nt were size-selected using 3% agarose gel. Library quality was assessed using the NGS 3K DNA Assay on a LabChip GX Touch, and concentrations were measured with the Quant-IT dsDNA Assay Kit (Invitrogen, Waltham, MA, USA). Libraries were pooled to 1.6 nM and sequenced on a NovaSeq 6000. Data is available under BioProject PRJNA1242911.
sRNA mapping and isomiR annotation
Adaptor trimming was performed with a Regex-based algorithm (Supplementary Material). PCR duplicate levels were estimated using read count/Unique Molecular Index ratios. Trimmed reads were mapped in two steps: (i) alignment to 2,632 canonical miRNA sequences, with 10nt extensions at the 5’ and 3’ ends, using Bowtie I, allowing two mismatches; additional mismatches were permitted at the 5′ and 3′ ends, with penalties assigned for templated additions or deletions (1), non-templated additions (2), and swaps (5); (ii) unmapped reads were aligned to 17–95nt tiled human genome (hg38) using RefSeq, piRBase, and miRbase annotations. Multimaps outside of miRNAs were prioritized to piRNA, then tRNA, rRNA, protein coding genes, and other non-coding RNAs (ncRNAs). Expression was calculated as reads per million (RPM), with miRNA family levels defined by the sum of sRNAs mapping to the family locus. Outliers were identified using log2RPM values (Supplementary Table 2).
mRNA library preparation and sequencing
Libraries were prepared from 300 ng of total RNA with the NextFlex Rapid Directional RNA Seq Kit. Library quality was assessed using the NGS 3K DNA assay on a LabChip GX Touch. Concentrations were measured using the Quant-IT dsDNA Assay Kit. Libraries were pooled to 1.6 nM and sequenced on a NovaSeq 6000. Reads were trimmed (Trimmomatic), mapped (HiSat2), and quantified (HTSeq-count).
Differential expression and ordinal regression analyses
Differential expression was performed in R using DEseq2 [23], adjusting for RIN as a covariate. Ordinal regression of expression against NAS and fibrosis scores was performed using the ‘ordinal’ R package, using the proportional odds model. Cumulative link logit models were fit for each sRNA or gene as described by Hoang et al. [7], with fold changes reported as differences in mean log2RPM between the top and bottom two score categories. Significance thresholds were set at FDR<0.05 for sRNA analyses and FDR<0.1 for mRNA analyses. The higher threshold for mRNA was chosen to allow for the inclusion of a greater number of potential miRNA regulatory targets, which would otherwise be limited at FDR<0.05. Results were filtered for false positives using two independent sources: a secondary RNA-seq dataset [7] and knowledge graph–based information, described below. Results are provided in Supplementary Tables 2 and 6.
Machine learning (ML) models
Models were developed with 10-fold cross-validation. Within each fold, the top 10,000 sRNAs were preselected based on differential expression. Features were further reduced using elastic net regression (glmnet, α=0.5). A support vector machine (SVM) with linear kernel was trained on the reduced feature set and evaluated on the test set using receiver operator characteristic area under the curve (AUROC). Results were combined across folds with the ROCR package. Optimal feature sets were defined as the smallest models achieving peak AUROC. Based on this approach we selected up to 27 features per fold for NAS high vs. low, 40 for fibrosis high vs. low, and 66 for MASL vs. MASH (Supplementary Tables 3, 4).
Gene set enrichment analysis
Enrichment analysis of differentially expressed genes and TargetScan-predicted targets was performed using clusterProfiler and ReactomePA on GO Biological Process, Kyoto Encyclopedia of Genes and Genomes (KEGG), and Reactome databases. Pathways containing 3–300 genes were tested. Significance was defined as FDR<0.05.
Artificial intelligence (AI)-generated knowledge graph for miRNA-disease-gene-pathway associations
To construct a knowledge graph of miRNA–disease–gene–pathway associations, multiple large language models were applied to PubMed and other repositories (Supplementary Tables 5, 8, 9). These models provided three association lists: miRNA-to-Disease, Disease-to-Genes, and Disease-to-Pathways. Associations were extracted for (i) miRNAs and liver disease (“liver,” “steatohepatitis”), (ii) diseases and genes (“fibrosis,” “cirrhosis”), and (iii) genes/pathways and MASLD severity (“NAFLD,” “liver fibrosis”). Results from different LLMs were cross-validated for consistency (Supplementary Material).
Independent mRNA expression dataset
Our gene expression results were compared to those of an independent dataset previously published [7]. In this study, also conducted at Virginia Commonwealth University, liver biopsies were obtained from 72 MASLD patients and 6 histologically normal age- and weight-matched controls. Histological assessments, RNA extraction/sequencing, and expression/ordinal regression analysis were performed following protocols similar to those employed here.
RESULTS
MiRNAs are the prominent sRNA type in liver
We analyzed 79 liver biopsy samples from patients with MASLD: 19 with MASL, 47 with MASH, 11 with MASLD-related cirrhosis, and 2 with MASLD of unclassified histology. The cohort spanned a range of NAS and fibrosis score (Fig. 1A-D, Supplementary Table 1). Because NAS decreases in cirrhosis [24,25], values from cirrhotic patients were excluded from NAS-based analysis (Fig. 1C).
Figure 1.
Data overview. (A) Study design. (B–D) Distribution of the number of samples for each key clinical information: Diagnosis (B), NAS score (Ballooning + Steatosis + Lobular Inflammation) (C), and fibrosis score (D). The colors indicate the type of MASLD diagnosis. NAS scores for cirrhosis patients were excluded from analysis because extensive fibrosis makes the scoring of ballooning, inflammation, and steatosis inaccurate. (E) Summary table of patient clinical characteristics across the three comparative groups used in the study. MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; MASLD, metabolic dysfunction-associated steatotic liver disease; mRNA, messenger RNA; NAS, metabolic dysfunction-associated steatotic liver disease activity score; sRNA, small RNA.
sRNA sequencing yielded an average of 14% of reads in the 17-26nt range, consistent with small, non-coding regulatory RNAs (Supplementary Fig. 1A). Principal component analysis of normalized expression (log2RPM) showed limited separation by diagnosis (Supplementary Fig. 1B). In contrast, pairwise correlation analysis of the sample metadata revealed that RIN strongly influenced data structure and was therefore included as a covariate in subsequent analysis (Supplementary Fig. 2). We did not observe demographic bias across the clinical groups in our patient cohort (Fig. 1E, Supplementary Fig. 3).
Across all samples, sequencing identified 102,943 unique sRNA transcripts mapping to seven distinct classes: miRNA, piRNA, tRNA-derived fragments (tRFs), rRNA-derived fragments (rRFs), mRNA-derived fragments (mRFs), other ncRNA, and intron/intergenic regions (Supplementary Table 2). The majority mapped to miRNA (74.6%) or piRNA loci (11.3%), with lengths primarily 20-24nt (Fig. 2A). We further categorized the miRNAs into 758 families. The 50 highest expressed families accounted for approximately 90% of the total miRNA reads. As expected, the hepatocyte-specific miR-122-5p family dominated (41.2% of the miRNA reads), far exceeding the next most abundant family, miR-143-3p (5.6%, Fig. 2B and Supplementary Fig. 4).
Figure 2.
sRNA expression in liver biopsy samples. (A) Distribution of sRNAs size and class across samples. (B) Proportion of miRNA expression by families (top) and by miRNA sequence type (bottom). (C) The expression breakdown of the isomiRs and canonical miRNAs within the 50 most highly expressed miRNA families. The black areas correspond to low expressed isomiRs. miRNA, microRNA; mRF, mRNA fragment; ncRNA, non-coding RNA; piRNA, PIWI-interacting RNA; rRF, rRNA fragment; sRNA, small RNA; tRF, tRNA fragment.
The majority of miRNA reads corresponded to isomiRs with diverse variation of templated and non-templated 5’ and 3’ ends, underscoring their prevalence in MASLD (Fig. 2B). The abundance of isomiRs varied substantially across families; in miR-192-5p, miR-125a-5p, and miR-30e-5p, they accounted for more than 93% of all family reads, leaving only a minor fraction as canonical sequences (Fig. 2C).
IsomiRs reshape the landscape of liver-expressed small ncRNAs
IsomiRs arise from alternative RNase III cleavage by DROSHA or DICER which create Templated Additions and Deletions on the 5’ or 3’ ends (referred to as TADs), nucleotidyl transferases that append Non-Templated Additions (referred to as NTA), or Single Nucleotide Polymorphisms (SNPs) (Fig. 3A).
Figure 3.
MicroRNA (miRNA) isoform landscape. (A) Presentation of the five types of isomiR modifications. (B) Expression levels of the 20 most abundant isomiRs within the two most highly expressed miRNA families (miR-122-5p, left; miR-143-3p, right). (C) Proportion of each type of isomiR modifications (addition, deletion, or swap), within the two most highly expressed miRNA families (miR-122-5p, left; miR-143-3p, right). (D) Summary of isomiR modifications occurrence across the 50 most highly expressed miRNA families. Each line is a heatmap representation of the proportion as presented on panel C, for each corresponding miRNA family.
For the two most abundant families, miR-122-5p and miR-143-3p, the canonical sequence was only the second most expressed isomiR. The dominant isomiR in both families carried one 3’ NTA (Fig. 3B). Across these families, isomiR modifications were enriched at the 3′ end, particularly NTAs and TADs (Fig. 3C). The 5′ end was highly conserved, with internal swaps being rare and almost absent in the seed region (nt 2–8, <0.3% of transcripts) [8]. This pattern was consistent across the 50 most abundant families (Fig. 3D).
Seed-altering isomiRs occurred at low frequency overall (12% of all miRNAs), but several families including miR-192-5p, miR-199a-3p, miR-101-3p, miR-10a-5p, miR-126-3p, miR-122-3p and miR-140-3p, displayed above-average levels of 5’ TADs (Fig. 3D). In addition, 5′ NTAs (mostly adenosine additions) were observed in 0.5–5% of reads per family (Supplementary Fig. 5A), a modification known to affect RNA Induced Silencing Complex (RISC) loading [26].
The 3′ end showed greater variability from the canonical terminal nucleotides (e.g., miR-125a-5p, miR-30e-5p, miR-122-3p; Fig. 3D). The 3′ end influences binding stability within RISC [27,28], cellular localization [29], and stability/decay [30]. On average, 18.6% of reads carried 3′ NTAs, reaching as high as 57.4% for miR-143-3p. Adenosine and uridine additions predominated, consistent with distinct stability and localization outcomes (Supplementary Fig. 5B).
Diversity of isomiRs across MASLD histology
We next investigated whether miRNA isomiR profiles are associated with histological severity. Two complementary approaches were applied. First, differential expression analysis comparing (i) MASH versus MASL diagnosis, (ii) NAS high (5-6) versus low (0-4) and (iii) fibrosis score high (3-4) versus low (0-2). Second, ordinal regression modeling to identify isomiRs correlated with NAS or fibrosis stage. RIN was included as a covariate. Differential gene expression analysis using dichotomized groups identified: 299 differentially expressed (DE) sRNAs in MASH vs. MASL, 1,824 DE sRNAs in high vs. low NAS, and 7,818 DE sRNAs in high vs. low fibrosis score (BH-adjusted P<0.05). Ordinal regression identified 99 and 961 sRNAs correlated to NAS and fibrosis score respectively (BH-adjusted P<0.05, Fig. 4A, Supplementary Table 2).
Figure 4.
Differential expression analysis. (A) DE sRNAs identified across the three comparator groups and two ordinal regression analyses (BH P-adj <0.05). (B) Breakdown of significant sRNA by type from the ordinal regression analysis with fibrosis. (C) DE isomiRs within the 50 most highly expressed miRNA families from the ordinal regression analysis with fibrosis. (D) Directionality and significance of isomiRs (with average log2RPM >2) within two miRNA families, left: miR-122-5p, right: miR-21-5p. The canonical miRNA sequence is highlighted in bold. Blue: down-regulated, red: up-regulated. DE, differential expression; MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; miRNA, microRNA; mRF, mRNA fragment; NAS, metabolic dysfunction-associated steatotic liver disease activity score; ncRNA, non-coding RNA; piRNA, PIWI-interacting RNA; rRF, rRNA fragment; sRNA, small RNA; tRF, tRNA fragment.
Overall, the modest number of DE sRNAs for MASH vs. MASL may reflect the limited ability of histology-based diagnosis to accurately capture underlying disease biology [31,32]. In contrast, progression based on Fibrosis score was associated with extensive isomiR remodeling, consistent with a strong role of RNA interference in fibrogenesis [11-15,33,34].
Class distribution analysis revealed that dysregulated sRNAs were enriched for miRNAs (Fig. 4B, Supplementary Fig. 6A). Within the 50 most abundant families, several displayed stage-specific patterns (Fig. 4C, Supplementary Fig. 6B). For example: miR-122-5p isomiRs: downregulated with advancing fibrosis; miR-143-3p, miR-21-5p, miR-146b-5p isomiRs: upregulated with fibrosis; miR-26a-5p, miR-26b-5p, miR-92a-3p, let-7c-5p isomiRs: downregulated with high NAS.
We observed a consistent trend of up or down regulation across isomiRs families (Fig. 4D and Supplementary Fig. 7), suggesting an upstream change of expression at the pre-miR level. Notably, within the miR-122-5p family, nearly all isomiRs were downregulated with fibrosis except for a single significantly upregulated isoform (Fig. 4D). This variant shared the canonical 5′ seed but carried a distinct 3′ terminus, potentially altering stability and localization. Overall, isomiR profiles showed extensive variability across MASLD histology, with consistent dysregulation linked to advanced fibrosis.
ML implicated isomiRs in MASLD progression
We applied ML to further identify “important” sRNAs defined by their ability to predict pathology (Fig. 5A). Elastic net regression was used for feature selection, followed by linear SVM classification with 10-fold cross-validation. MASL vs. MASH classification had poor performance AUROC=0.605 (one-sided T-test P=0.151, Supplementary Fig. 8) consistent with limited DE sRNAs. NAS high vs. low: AUROC=0.748 (one-sided T-test P=0.034, Fig. 5B), based on 23 sRNAs including isomiRs from miR-122-5p, miR-143-3p, miR-21-5p, and miR-26a-5p. Fibrosis score high vs. low: AUROC=0.792 (one-sided T-test P=0.00017, Fig. 5C), based on 7 sRNAs from miR-26a-5p, let-7b-5p, miR-103-3p, and miR-146b-5p (Supplementary Tables 3, 4).
Figure 5.
Machine learning (ML) assisted sRNA selection. (A) Method overview. (B, C) Predictive outcomes with confusion matrix (left) and ROC curve (right) for high/low NAS scores (B) and high/low fibrosis scores (C). AUROC, receiver operator characteristic area under the curve; CI, confidence interval; NAS, metabolic dysfunction-associated steatotic liver disease activity score; sRNA, small RNA.
Correlation analysis (|r|>0.75) was used to expand ML-selected features to alternative sRNAs with similar expression profiles, yielding extended predictive panels of 13,151 (NAS) and 311 (fibrosis) candidates. After filtering for sequencing quality and expression (log₂RPM≥2), 3,948 NAS and 268 fibrosis-associated sRNAs remained (Supplementary Tables 3, 4).
Intersection of DE/ordinal regression results with ML-expanded panels produced final lists of 271 NAS-associated and 72 fibrosis-associated sRNAs, including 173 and 58 isomiRs from 41 and 17 miRNA families, respectively (Fig. 6). Notably, confirmed associations included miR-122 with MASH severity and miR-21 with Fibrosis (Fig. 6C, 6D, Supplementary Table 5). Novel associations included isomiRs from miR-26b, let-7c, miR-30c, miR-191, and miR-320a families not previously linked to liver disease.
Figure 6.
Identification of miRNA isoforms with potential disease involvement. (A, B) Upset plot showing overlap between differential expression analysis and machine learning (ML) outcome in relation to high NAS (A) or fibrosis scores (B). (C, D) Summary of the isomiRs identified through both differential expression analysis and ML as well as an estimated number of publications, identified through an AI-generated knowledge graph, associating each miRNA family to MASLD severity (C) or fibrosis (D). MiR families are ranked left to right in decreasing order of expression. Only isomiRs from the top 50 most highly expressed miRNA families are shown in panel (C). AI, artificial intelligence; MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; MASLD, metabolic dysfunction-associated steatotic liver disease; miRNA, microRNA; NAS, metabolic dysfunction-associated steatotic liver disease activity score; sRNA, small RNA.
IsomiR-mediated regulatory networks in MASLD
To identify downstream effects of isomiR dysregulation, we performed mRNA sequencing on 77 liver samples. Differential expression and ordinal regression identified 1,256 DE genes (BH-adjusted P<0.1, Supplementary Fig. 9A, Supplementary Table 6). Inter-study comparisons with an independent MASLD dataset [7] demonstrated strong correlation of fold-changes (MASH: r=0.33, fibrosis: r=0.44, both P<10-10, Supplementary Fig. 9B, 9C), supporting robustness. DE genes overlapped significantly with pathways known in MASLD and fibrosis, including repression of glycolysis (adj. P=7×10-7), fatty acid metabolism (adj. P=0.01), and activation of peroxisome proliferator-activated receptor (PPAR) pathway (adj. P=0.02), focal adhesion (P=5×10-6), and collagen fibril organization (P=6×10-11) (Supplementary Fig. 9D, Supplementary Table 7). Many are consistent with dysregulated pathways through comparative analysis further supporting the robustness and reliability of our results (Supplementary Fig. 9D, Supplementary Table 7).
Using the mRNA sequencing data, we selected 33 pathways, 15 for MASH and 27 for fibrosis, either dysregulated or associated with liver disease or fibrosis from our inhouse AI-generated knowledge graph dataset (Supplementary Table 8, Fig. 7A, 7B). For each DE sRNA selected through ML, the 2-8 seed sequence was used to obtain custom or canonical target lists from TargetScan [35,36] that were either (i) differentially expressed and involved in these 33 pathways (Fig. 7A, 7B), (ii) reported as implicated in MASLD/fibrosis using an AI generated knowledge graph (30 for NAS and 25 for fibrosis, Supplementary Table 9), or (iii) predictive genes in our previous independent study [7]. From those, regulatory networks were generated for each sRNAs (Supplementary Table 10), confirming already proposed mechanisms of action, such as INSIG1 regulation by miR-363 [37], or SCD regulation by miR-27a [38]. Importantly, novel regulatory mechanisms were also identified in relation to the disease, involving non-canonical seed targeting.
Figure 7.
Predicted isomiR regulatory role in MASLD. (A, B) Overview of selected sRNA regulated pathways involved in MASH severity (A) and advanced fibrosis (B). Purple corresponds to the number of targeted genes (based on Target-Scan) in the pathway, and stars indicates significant target enrichment (BH P-adj: <0.1: *, <0.01: **, <0.001: ***). Underlined seeds indicate canonical sequences. (C) Expression of the ML-selected differentially expressed isomiR from miR-122-5p family (5’ templated deletion) is associated with MASH severity across diagnosis, fibrosis scores and NAS. For fibrosis and NAS, log2FC indicates log2-FoldChange for high over low scores, and ‘padj’ indicates the BH-adjusted P-value of the ordinal regression method. (D) AI-predicted regulatory network involved in MASH severity of the same miR-122-5p isomiR using Target Scan (TS) as target predictor. Stars indicate genes that are also dysregulated in the same direction in the Hoang et al. study (BH P-adj <0.1). (E, F) Same as (C, D) for an isomiR of miR-21-5p (5’ non-templated A addition) associated with fibrosis. AI, artificial intelligence; DE, differential expression; MASH, metabolic dysfunction-associated steatohepatitis; MASLD, metabolic dysfunction-associated steatotic liver disease; NAS, metabolic dysfunction-associated steatotic liver disease activity score; sRNA, small RNA.
Examples include a miR-122-5p 5′-deletion isomiR, which was negatively correlated with NAS and was predicted to target 1,157 genes including INSIG1 and CDKN1A (that are up-regulated in MASH) (Fig. 7C, 7D). Canonical miR-122-5p had 52 predicted targets, none linked to driving MASLD, NAS or fibrosis pathways (Supplementary Fig. 10A, 10B). A miR-21-5p +5′A isomiR, upregulated in fibrosis, is predicted to target PIK3R1 (focal adhesion/insulin resistance), PPARA (PPAR signaling, TGF-β response), and HMGCS2 (cholesterol metabolism), all of which have previously been implicated in liver disease (Fig. 7E, 7F) [39-41]. Canonical miR-21 lacked these predicted targets (Supplementary Fig. 10C, 10D). MiR-26b-5p isomiR: repressed with high NAS; targets ACSL4, encoding a long-chain acyl-CoA ligase upregulated with MASH (Supplementary Fig. 11A, 11B). miR-320a-2p isomiR: upregulated with fibrosis; targets RXRA, a PPARA co-regulator with anti-fibrotic activity [42] (Supplementary Fig. 11C, 11D).
Target gene expression patterns were validated in the independent MASLD dataset (e.g., CDKN1A, PPARA, HMGCS2, ACSL4, RXRA; indicated in Fig. 7D–7F, Supplementary Fig. 10, 11). These data reveal that isomiRs frequently diverge from their canonical counterparts in both abundance and regulatory networks, enabling novel mechanisms of gene regulation in MASLD progression.
DISCUSSION
This study confirmed that miRNAs are the dominant sRNA type expressed in liver tissue across the histological spectrum of MASLD, with 67% occurring as non-canonical isomiRs. The extent and type of modification varied across miRNA families and patients, and several isomiRs, particularly those involving non-canonical seed changes, were linked to novel targets and disease severity. These findings have several implications for understanding MASLD biolo-gy and clinical heterogeneity.
IsomiRs as major contributors to miRNA complexity in MASLD
We showed that isomiRs comprise a substantial amount of liver miRNAs and are enriched within relatively few miRNA families. Some isomiRs had 5’ alterations that potentially created distinct regulatory profiles. For example, non-canonical seed variants were predicted to target novel mRNA such as CDKN1A, ACSL4, and RXRA, expanding known biology beyond canonical interactions. In contrast, 3’ modifications, more frequently observed among differentially expressed isomiRs, may influence stability and localization, consistent with prior reports [19,27,28]. While our findings suggest novel regulatory roles for isomiRs, the specific target(s) require additional validation using reporters, and gain or loss of function studies. Future, prospective studies using proteomic or metabolomic approaches on larger patient populations that include healthy controls could validate the role of these isomiRs in disease etiology and progression.
Novel insights into disease heterogeneity and target biology
Many of the predicted isomiR-target genes overlapped with established MASLD pathways, including PPARA and PI3KR1 in focal adhesion and fibrosis [13,40]. We also predicted new targets tied to lipid metabolism and nuclear receptor signaling, offering a potential explanation for the variability in gene expression across patients. These findings suggest that isomiRs may act as molecular modulators of disease that could be targeted with oligonucleotide therapies. However, it is important to note that the proposed regulatory networks rely heavily on in silico target prediction algorithms that might not contain the fully accurate targets of each isomiRs.
Association between fibrosis and sRNA remodeling
The number of isomiRs associated with fibrosis severity greatly exceeded those linked to disease activity, underscoring sRNA remodeling as a potential dominant regulator of fibrosis. Interestingly, miR-122-5p isomiRs declined with progression, while miR-143-3p and miR-21-5p increased. Circulating miR-122 is known to rise in parallel with declining hepatic levels, consistent with export or release during injury [43]. Notably, one miR-122-5p isomiR increased in advanced fibrosis despite others decreasing, carrying 3′ modifications that may alter stability or localization. This heterogeneity highlights the need to consider isomiRs individually rather than as a single pooled species.
Clinical implications and future directions
In MASH, the underlying metabolic perturbations, inflammatory responses, and fibrogenesis are all active energy-requiring processes. Even under normal circumstances, this perivenular zone has a low oxygen tension, and the increased energy requirements in MASLD may further induce tissue hypoxemia and modulate the injury and fibrogenic response. For instance, MASH disease severity has been associated with sleep apnea which causes hypoxemia [44]. The miR-21 family has been implicated in hypoxemia and oncogenesis [45,46]. In the current study, a strong relationship between miR-21-5p and fibrosis severity was noted. This raises the possibility for miR-21 to be one potential link relating the greater incidence of hepatocellular cancer in MASH with advanced fibrosis compared to early-stage disease and provides direction for future research. Several circulating miRNAs have been evaluated as biomarkers reflective of disease severity in MASLD such as miR-34a, miR-21 and miR-122 [47]. The identification of isomiRs associated with different stages of liver disease provides valuable insights into underlying pathogenic mechanisms and offers potential avenues for personalized therapeutic interventions.
Abbreviations
- ACSL4
achaete-scute homolog 4
- AUROC
receiver operator characteristic area under the curve
- CDKN1A
cyclin-dependent kinase inhibitor 1A
- DE
differential expression
- DGE
differential gene expression
- HMGCS2
mitochondrial 3-hydroxy-3-methylglutaryl-CoA synthase 2
- INSIG
insulin-induced gene 1
- LLMs
large language models
- MASH
metabolic dysfunction-associated steatohepatitis
- MASL
metabolic dysfunction-associated steatotic liver
- MASLD
metabolic dysfunction-associated steatotic liver disease
- miRNA
microRNA
- mRFs
mRNA fragments
- mRNA
messenger RNA
- NAS
metabolic dysfunction-associated steatotic liver disease (previously NAFLD) activity score
- ncRNA
non-coding RNA (here other than miRNA
- NTA
non-templated addition
- PIK3R1
phosphoinositide-3-kinase regulatory subunit 1
- piRNA
PIWI-interacting RNA
- PPAR
peroxisome proliferator-activated receptor
- rRFs
rRNA fragments
- RXRA
retinoid X receptor alpha
- SNP
single nucleotide polymorphism
- sRNA
small RNA
- TAD
templated addition or deletion
- TGFb
transforming growth factor-beta
- tRFs
tRNA fragments
Study Highlights
• This study provides a comprehensive profiling of hepatic microRNA isoforms (isomiRs) in MASLD. IsomiRs comprised two-thirds of hepatic miRNAs and showed strong associations with disease activity and fibrosis progression. Integration of mRNA sequencing and machine learning revealed novel regulatory networks and unique targets, including INSIG1 (miR-122 isomiR) and PPARA/HMGCS2 (miR-21 isomiR), not captured by canonical miRNAs. These findings establish isomiRs as major contributors to MASLD heterogeneity and as potential sources of new biomarkers and therapeutic targets.
Footnotes
Authors’ contributions
CB, SAH, DJH, NCF, DWS, and AJS contributed to the conceptualization of the study. The methodology was developed by CB, SAH, MRL, MAS, NCF, and DWS. The investigation was carried out by CB, GW, FM, JA, MRL, ZZ, BS, MSS, and AA. Data visualization was handled by CB, SAH, and MRL. NCF, DWS, and AJS were responsible for funding, project administration, and supervision. The manuscript was prepared by CB, AJS, SAH, GW, MRL, NCF, and DWS.
Acknowledgements
The work is funded through intramural grants from Virginia Commonwealth University and the NIH NIDDKD 5R01DK129564-03. This study complied with ethical regulations.
Conflicts of Interest
CB, GW, JA, MRL, ZZ, BS, MAS, DJH, NCF, and DWS are employees and shareholders of Gatehouse Bio, and DWS and NCF serve on the Board of Directors. AJS has stock options in Tiziana, Inversago, Rivus, NorthSea, Durect. He has served as a consultant to Novo Nordisk, Eli Lilly, Boehringer Ingelhiem, Inventiva, Gilead, Takeda, LG Chem, Hanmi, Corcept, Surrozen, Poxel, 89 Bio, Boston Pharmaceuticals, Regeneron, Merck, Alnylam, Aligos, Akero, Myovant, Salix, Avant Sante, NorthSea Pharma, Madrigal, Path AI, Histoindex, Astra Zeneca, Abbvie, Zydus. His institution has received grants from Intercept, Novo Nordisk, Boehringer Ingelhiem, Eli Lilly, Merck, Takeda, Salix, Inventiva, Gilead, Akero, Hanmi, Histoindex, 89Bio. He receives royalties from Wolter Kluwers (UptoDate) and Elsevier. All other authors declare that they have no competing interests.
SUPPLEMENTARY MATERIAL
Supplementary material is available at Clinical and Molecular Hepatology website (http://www.e-cmh.org).
sRNA sequencing output. (A) Sequencing depth and size selection for sRNA sequencing. (B) Principal component analysis of the sRNA sequencing data using the 500 most variable features. No clear clusters or outliers were identified. MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; sRNA, small RNA.
Pairwise meta data covariate comparison. This analysis presents pairwise correlations across samples, patient covariates, and the principal components from the PCA shown in Supplementary Figure 1. The goal was to identify potential confounding factors. The absence of significant correlations between clinical endpoints and variables such as age, sex, or body mass index (BMI) indicates a well-balanced patient cohort. However, RNA integrity number (RIN) demonstrated a notable influence on the data structure (PC1 and PC2) and will therefore be included as a covariate in subsequent statistical models. Factor variables are indicated with a star. Continuous to continuous, factor to continuous, and factor to factor correlations values were calculated by Spearman’s correlation method, Kruskal–Wallis based Cramer’s V, and Pearson’s Chi2 based Cramer’s V, respectively. Benjamini-Hochberg FDR correction was applied globally across test types. PCA, principal component analysis.
Distribution of the baseline demographic characteristics across comparative groups in the MASLD patient’s cohort. BMI, body mass index; MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; MASLD, metabolic dysfunction-associated steatotic liver disease; NAS, metabolic dysfunction-associated steatotic liver disease activity score.
Distribution of the absolute expression of the 50 most highly expressed miRNA families. The black areas correspond to numerous lowly expressed isomiRs. miRNA, microRNA.
Type and occurrence of non-templated additions (NTA) at the 5’ (A) or 3’ (B) ends of the isomiRs for each of the 50 most highly expressed miRNA families. miRNA, microRNA.
(A) Breakdown of the number of significant sRNA (BH P-adj <0.05) by type across four comparative methods (MASH vs. MASL DEA, NAS high vs. low DEA, fibrosis high vs. low DEA, and NAS ordinal regression analysis). (B) Number of significant isomiRs within each of the 50 most highly expressed miRNA families across the same four comparative methods. DEA, differential expression analysis; MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; miRNA, microRNA; mRF, mRNA fragment; NAS, metabolic dysfunction-associated steatotic liver disease activity score; ncRNA, non-coding RNA; piRNA, PIWI-interacting RNA; rRF, rRNA fragment; sRNA, small RNA; tRF, tRNA fragment.
Differential expression analysis using normalized quantifications per family (relative abundance). Sequence, directionality, and significance of isomiRs (with average log2RPM >2) were provided for isomiRs within two miRNA families, left: miR-122-5p, right: miR-21-5p. The canonical miRNA sequence is highlighted in bold. Blue: down-regulated, red: up-regulated. miRNA, microRNA.
Machine learning predictive outcome for MASH vs. MASL with confusion matrix (left) and ROC curve (right). MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; AUROC, receiver operator characteristic area under the curve.
mRNA differential expression analysis. (A) Number of significant genes in each of the three comparator groups and two ordinal regression analyses (BH P-adj <0.1). (B, C) Interstudy comparison of significant genes associated with MASH (C), or liver fibrosis (D). Both cases showed a highly significant positive correlation between the log2FoldChange across studies. (D) Interstudy comparison of gene set enrichment analysis. Significantly differentially expressed genes across each comparative method were considered with the following threshold: BH P-adjust <0.1 from our study, and <0.05 from Hoang et al. study. We used KEGG (K), Reactome (R), and GO: Biological Process (G) repositories. NAS ordinal regression did not generate enough significant genes for successful enrichment analysis. DEA, differential expression analysis; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MASH, metabolic dysfunction-associated steatohepatitis; mRNA, messenger RNA; MASL, metabolic dysfunction-associated steatotic liver; NAS, metabolic dysfunction-associated steatotic liver disease activity score.
MicroRNA regulatory role in MASLD. (A) Expression of an ML-selected differentially expressed isomiR from miR-122-5p family (canonical seed) associated with MASH severity across diagnosis, fibrosis scores and NAS. For fibrosis and NAS, log2FC indicates log2-FoldChange for high over low scores, and ‘padj’ indicates the BH-adjusted P-value of the ordinal regression method. (B) AI-informed putative regulatory network involved in MASH severity of the same miR-122-5p isomiR using Target Scan (TS) as target predictor. Stars indicate genes that are also significantly dysregulated in the same direction in the Hoang et al. study (BH P-adjusted <0.1). (C, D) Same as (A, B) for an isomiR of miR-21-5p (canonical seed) associated to advanced fibrosis. AI, artificial intelligence; MASH, metabolic dysfunction-associated steatohepatitis; MASLD, metabolic dysfunction-associated steatotic liver disease; ML, machine learning; NAS, metabolic dysfunction-associated steatotic liver disease activity score.
MicroRNA regulatory role in MASLD. (A) Expression of an ML-selected differentially expressed isomiR from miR-26b-5p family (one 5’ deletion) associated with MASH severity from across diagnosis, fibrosis scores and NAS. For fibrosis and NAS, log2FC indicates log2-FoldChange for high over low scores, and ‘padj’ indicates the BH-adjusted P-value of the ordinal regression method. (B) AI-informed putative regulatory network involved in MASH severity of the same miR-26b-5p isomiR using Target Scan (TS) as target predictor. Stars indicate genes that are also significantly dysregulated in the same direction in the Hoang et al. study (BH P-adjusted <0.1). (C, D) Same as (A, B) for an isomiR of miR-320a-5p (canonical seed) associated with advanced fibrosis. AI, artificial intelligence; MASH, metabolic dysfunction-associated steatohepatitis; MASLD, metabolic dysfunction-associated steatotic liver disease; ML, machine learning; NAS, metabolic dysfunction-associated steatotic liver disease activity score.
MASLD patents overview (number in bracket are standard deviation)
List of sRNA with annotation and differential analyses results (after filtering for sequencing quality and minimum log2 RPM expression ≥2.0)
A. List of predictive sRNA for ML model related to MASH severity. B. Extended list of predictive sRNA for ML model related to MASH severity through correlation analysis
A. List of predictive sRNA for ML model related to fibrosis. B. Extended list of predictive sRNA for ML model related to fibrosis through correlation analysis
AI generated knowledge graph representing miRNA to disease associations
Gene differential expression analyses results (significant padj >0.1 in at least one DEA)
Pathway enrichment analyses
AI generated knowledge graph representing disease to pathway associations
AI generated knowledge graph representing disease to gene associations
List of targeted genes for the 182 seed sRNA selected in either NAS of fibrosis progression and relation with the disease
REFERENCES
- 1.Lekakis V, Papatheodoridis GV. Natural history of metabolic dysfunction-associated steatotic liver disease. Eur J Intern Med. 2024;122:3–10. doi: 10.1016/j.ejim.2023.11.005. [DOI] [PubMed] [Google Scholar]
- 2.Eskridge W, Cryer DR, Schattenberg JM, Gastaldelli A, Malhi H, Allen AM, et al. Metabolic dysfunction-associated steatotic liver disease and metabolic dysfunction-associated steatohepatitis: the patient and physician perspective. J Clin Med. 2023;12:6216. doi: 10.3390/jcm12196216. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Ferenc K, Jarmakiewicz-Czaja S, Sokal-Dembowska A, Stasik K, Filip R. Common denominator of MASLD and some non-communicable diseases. Curr Issues Mol Biol. 2024;46:6690–6709. doi: 10.3390/cimb46070399. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Petta S, Targher G, Romeo S, Pajvani UB, Zheng MH, Aghemo A, et al. The first MASH drug therapy on the horizon: current perspectives of resmetirom. Liver Int. 2024;44:1526–1536. doi: 10.1111/liv.15930. [DOI] [PubMed] [Google Scholar]
- 5.Abushamat LA, Shah PA, Eckel RH, Harrison SA, Barb D. The emerging role of glucagon-like peptide-1 receptor agonists for the treatment of metabolic dysfunction-associated steatohepatitis. Clin Gastroenterol Hepatol. 2024;22:1565–1574. doi: 10.1016/j.cgh.2024.01.032. [DOI] [PubMed] [Google Scholar]
- 6.Govaere O, Cockell S, Tiniakos D, Queen R, Younes R, Vacca M, et al. Transcriptomic profiling across the nonalcoholic fatty liver disease spectrum reveals gene signatures for steatohepatitis and fibrosis. Sci Transl Med. 2020;12:eaba4448. doi: 10.1126/scitranslmed.aba4448. [DOI] [PubMed] [Google Scholar]
- 7.Hoang SA, Oseini A, Feaver RE, Cole BK, Asgharpour A, Vincent R, et al. Gene expression predicts histological severity and reveals distinct molecular profiles of nonalcoholic fatty liver disease. Sci Rep. 2019;9:12541. doi: 10.1038/s41598-019-48746-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Lewis BP, Shih IH, Jones-Rhoades MW, Bartel DP, Burge CB. Prediction of mammalian microRNA targets. Cell. 2003;115:787–798. doi: 10.1016/s0092-8674(03)01018-3. [DOI] [PubMed] [Google Scholar]
- 9.Cheung O, Puri P, Eicken C, Contos MJ, Mirshahi F, Maher JW, et al. Nonalcoholic steatohepatitis is associated with altered hepatic MicroRNA expression. Hepatology. 2008;48:1810–1820. doi: 10.1002/hep.22569. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Ma N, Tan J, Chen Y, Yang L, Li M, He Y, et al. MicroRNAs in metabolic dysfunction-associated diseases: pathogenesis and therapeutic opportunities. FASEB J. 2024;38:e70038. doi: 10.1096/fj.202401464R. [DOI] [PubMed] [Google Scholar]
- 11.Bandiera S, Pfeffer S, Baumert TF, Zeisel MB. MiR-122--a key factor and therapeutic target in liver disease. J Hepatol. 2015;62:448–457. doi: 10.1016/j.jhep.2014.10.004. [DOI] [PubMed] [Google Scholar]
- 12.Hu Y, Peng X, Du G, Zhang Z, Zhai Y, Xiong X, et al. MicroRNA-122-5p inhibition improves inflammation and oxidative stress damage in dietary-induced non-alcoholic fatty liver disease through targeting FOXO3. Front Physiol. 2022;13:803445. doi: 10.3389/fphys.2022.803445. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Loyer X, Paradis V, Hénique C, Vion AC, Colnot N, Guerin CL, et al. Liver microRNA-21 is overexpressed in non-alcoholic steatohepatitis and contributes to the disease in experimental models by inhibiting PPARα expression. Gut. 2016;65:1882–1894. doi: 10.1136/gutjnl-2014-308883. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Zhang T, Yang Z, Kusumanchi P, Han S, Liangpunsakul S. Critical role of microRNA-21 in the pathogenesis of liver diseases. Front Med (Lausanne) 2020;7:7. doi: 10.3389/fmed.2020.00007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Salah A, Karimi MH, Sajedianfard J, Nazifi S, Yaghobi R. Expression pattern of microRNA-21 during the liver ischemia/reperfusion. Iran J Allergy Asthma Immunol. 2021;20:88–97. doi: 10.18502/ijaai.v20i1.5415. [DOI] [PubMed] [Google Scholar]
- 16.Carpi S, Daniele S, de Almeida JFM, Gabbia D. Recent advances in miRNA-based therapy for MASLD/MASH and MASH-associated HCC. Int J Mol Sci. 2024;25:12229. doi: 10.3390/ijms252212229. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Papazyan R, Kinberger G, Wang D, Lang G, Gogas K, Wright T, et al. LBP-40-development of oligonucleotide-based miR-132 antagonists for the treatment of NASH. J Hepatol. 2019;70:e160–e161. [Google Scholar]
- 18.Panella R, Zanderigo F, Morandini F, Federico D, Vicentini E, Andreetta F, et al. Assessment of immunostimulatory responses to the antimiR-22 oligonucleotide compound RES-010 in human peripheral blood mononuclear cells. Front Pharmacol. 2023;14:1125654. doi: 10.3389/fphar.2023.1125654. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Tomasello L, Distefano R, Nigita G, Croce CM. The microRNA family gets wider: the isomiRs classification and role. Front Cell Dev Biol. 2021;9:668648. doi: 10.3389/fcell.2021.668648. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Wojcicka A, Swierniak M, Kornasiewicz O, Gierlikowski W, Maciag M, Kolanowska M, et al. Next generation sequencing reveals microRNA isoforms in liver cirrhosis and hepatocellular carcinoma. Int J Biochem Cell Biol. 2014;53:208–217. doi: 10.1016/j.biocel.2014.05.020. [DOI] [PubMed] [Google Scholar]
- 21.Makarenkov N, Yoel U, Haim Y, Pincu Y, Bhandarkar NS, Shalev A, et al. Circulating isomiRs may be superior biomarkers compared to their corresponding miRNAs: a pilot biomarker study of using isomiR-Ome to detect coronary calcium-based cardiovascular risk in patients with NAFLD. Int J Mol Sci. 2024;25:890. doi: 10.3390/ijms25020890. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Vickers KC, Sethupathy P, Baran-Gale J, Remaley AT. Complexity of microRNA function and the role of isomiRs in lipid homeostasis. J Lipid Res. 2013;54:1182–1191. doi: 10.1194/jlr.R034801. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15:550. doi: 10.1186/s13059-014-0550-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Ekstedt M, Franzén LE, Mathiesen UL, Kechagias S. Low clinical relevance of the nonalcoholic fatty liver disease activity score (NAS) in predicting fibrosis progression. Scand J Gastroenterol. 2012;47:108–115. doi: 10.3109/00365521.2011.634024. [DOI] [PubMed] [Google Scholar]
- 25.Sanyal AJ, Harrison SA, Ratziu V, Abdelmalek MF, Diehl AM, Caldwell S, et al. The natural history of advanced fibrosis due to nonalcoholic steatohepatitis: data from the Simtuzumab trials. Hepatology. 2019;70:1913–1927. doi: 10.1002/hep.30664. [DOI] [PubMed] [Google Scholar]
- 26.Nakanishi K. Anatomy of four human Argonaute proteins. Nucleic Acids Res. 2022;50:6618–6638. doi: 10.1093/nar/gkac519. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.McGeary SE, Bisaria N, Pham TM, Wang PY, Bartel DP. MicroRNA 3’-compensatory pairing occurs through two binding modes, with affinity shaped by nucleotide identity and position. Elife. 2022;11:e69803. doi: 10.7554/eLife.69803. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Kosek DM, Banijamali E, Becker W, Petzold K, Andersson ER. Efficient 3’-pairing renders microRNA targeting less sensitive to mRNA seed accessibility. Nucleic Acids Res. 2023;51:11162–11177. doi: 10.1093/nar/gkad795. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Khudayberdiev SA, Zampa F, Rajman M, Schratt G. A comprehensive characterization of the nuclear microRNA repertoire of post-mitotic neurons. Front Mol Neurosci. 2013;6:43. doi: 10.3389/fnmol.2013.00043. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Han J, Mendell JT. MicroRNA turnover: a tale of tailing, trimming, and targets. Trends Biochem Sci. 2023;48:26–39. doi: 10.1016/j.tibs.2022.06.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Brunt EM, Kleiner DE, Wilson LA, Belt P, Neuschwander-Tetri BA. Nonalcoholic fatty liver disease (NAFLD) activity score and the histopathologic diagnosis in NAFLD: distinct clinicopathologic meanings. Hepatology. 2011;53:810–820. doi: 10.1002/hep.24127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Filozof CM, Lackner C, Romero-Gómez M, Imperial JC, Mc-Gee R, Dimick-Santos L, et al. Best practices in liver biopsy histologic assessment for nonalcoholic steatohepatitis clinical trials: expert opinion. GastroHep. 2022;2022:1–11. [Google Scholar]
- 33.Pirola CJ, Fernández Gianotti T, Castaño GO, Mallardi P, San Martino J, Mora Gonzalez Lopez Ledesma M, et al. Circulating microRNA signature in non-alcoholic fatty liver disease: from serum non-coding RNAs to liver histology and disease pathogenesis. Gut. 2015;64:800–812. doi: 10.1136/gutjnl-2014-306996. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Xu X, Kriegel AJ, Jiao X, Liu H, Bai X, Olson J, et al. MiR-21 in ischemia/reperfusion injury: a double-edged sword? Physiol Genomics. 2014;46:789–797. doi: 10.1152/physiolgenomics.00020.2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Agarwal V, Bell GW, Nam JW, Bartel DP. Predicting effective microRNA target sites in mammalian mRNAs. Elife. 2015;4:e05005. doi: 10.7554/eLife.05005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Lewis BP, Burge CB, Bartel DP. Conserved seed pairing, often flanked by adenosines, indicates that thousands of human genes are microRNA targets. Cell. 2005;120:15–20. doi: 10.1016/j.cell.2004.12.035. [DOI] [PubMed] [Google Scholar]
- 37.Wang L, Jia G, Fu R, Liang J, Xue W, Zheng J, et al. Hepatic miR-363 promotes nonalcoholic fatty liver disease by suppressing INSIG1. J Nutr Biochem. 2024;134:109717. doi: 10.1016/j.jnutbio.2024.109717. [DOI] [PubMed] [Google Scholar]
- 38.Zhang M, Sun W, Zhou M, Tang Y. MicroRNA-27a regulates hepatic lipid metabolism and alleviates NAFLD via repressing FAS and SCD1. Sci Rep. 2017;7:14493. doi: 10.1038/s41598-017-15141-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Mungamuri SK, Gupta YK, Rao PN, Ravishankar B. Evolution of non-alcoholic fatty liver disease to liver cancer: insights from genome-wide association studies. Gene Expr. 2023;22:47–61. [Google Scholar]
- 40.Hu P, Li K, Peng X, Kan Y, Li H, Zhu Y, et al. Nuclear receptor PPARα as a therapeutic target in diseases associated with lipid metabolism disorders. Nutrients. 2023;15:4772. doi: 10.3390/nu15224772. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Wang YH, Liu CL, Chiu WC, Twu YC, Liao YJ. HMGCS2 mediates ketone production and regulates the proliferation and metastasis of hepatocellular carcinoma. Cancers (Basel) 2019;11:1876. doi: 10.3390/cancers11121876. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Li B, Cai SY, Boyer JL. The role of the retinoid receptor, RAR/RXR heterodimer, in liver physiology. Biochim Biophys Acta Mol Basis Dis. 2021;1867:166085. doi: 10.1016/j.bbadis.2021.166085. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Colaianni F, Zelli V, Compagnoni C, Miscione MS, Rossi M, Vecchiotti D, et al. Role of circulating microRNAs in liver disease and HCC: focus on miR-122. Genes (Basel) 2024;15:1313. doi: 10.3390/genes15101313. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Sheng W, Ji G, Zhang L. Management of non-alcoholic fatty liver disease patients with sleep apnea syndrome. World J Gastroenterol. 2022;28:6099–6108. doi: 10.3748/wjg.v28.i43.6099. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Sarkar J, Gou D, Turaka P, Viktorova E, Ramchandran R, Raj JU, et al. MicroRNA-21 plays a role in hypoxia-mediated pulmonary artery smooth muscle cell proliferation and migration. Am J Physiol Lung Cell Mol Physiol. 2010;299:L861–L871. doi: 10.1152/ajplung.00201.2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Rhim J, Baek W, Seo Y, Kim JH. From molecular mechanisms to therapeutics: understanding microRNA-21 in cancer. Cells. 2022;11:2791. doi: 10.3390/cells11182791. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Gjorgjieva M, Sobolewski C, Dolicka D, Correia de Sousa M, Foti M. MiRNAs and NAFLD: from pathophysiology to therapy. Gut. 2019;68:2065–2079. doi: 10.1136/gutjnl-2018-318146. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
sRNA sequencing output. (A) Sequencing depth and size selection for sRNA sequencing. (B) Principal component analysis of the sRNA sequencing data using the 500 most variable features. No clear clusters or outliers were identified. MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; sRNA, small RNA.
Pairwise meta data covariate comparison. This analysis presents pairwise correlations across samples, patient covariates, and the principal components from the PCA shown in Supplementary Figure 1. The goal was to identify potential confounding factors. The absence of significant correlations between clinical endpoints and variables such as age, sex, or body mass index (BMI) indicates a well-balanced patient cohort. However, RNA integrity number (RIN) demonstrated a notable influence on the data structure (PC1 and PC2) and will therefore be included as a covariate in subsequent statistical models. Factor variables are indicated with a star. Continuous to continuous, factor to continuous, and factor to factor correlations values were calculated by Spearman’s correlation method, Kruskal–Wallis based Cramer’s V, and Pearson’s Chi2 based Cramer’s V, respectively. Benjamini-Hochberg FDR correction was applied globally across test types. PCA, principal component analysis.
Distribution of the baseline demographic characteristics across comparative groups in the MASLD patient’s cohort. BMI, body mass index; MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; MASLD, metabolic dysfunction-associated steatotic liver disease; NAS, metabolic dysfunction-associated steatotic liver disease activity score.
Distribution of the absolute expression of the 50 most highly expressed miRNA families. The black areas correspond to numerous lowly expressed isomiRs. miRNA, microRNA.
Type and occurrence of non-templated additions (NTA) at the 5’ (A) or 3’ (B) ends of the isomiRs for each of the 50 most highly expressed miRNA families. miRNA, microRNA.
(A) Breakdown of the number of significant sRNA (BH P-adj <0.05) by type across four comparative methods (MASH vs. MASL DEA, NAS high vs. low DEA, fibrosis high vs. low DEA, and NAS ordinal regression analysis). (B) Number of significant isomiRs within each of the 50 most highly expressed miRNA families across the same four comparative methods. DEA, differential expression analysis; MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; miRNA, microRNA; mRF, mRNA fragment; NAS, metabolic dysfunction-associated steatotic liver disease activity score; ncRNA, non-coding RNA; piRNA, PIWI-interacting RNA; rRF, rRNA fragment; sRNA, small RNA; tRF, tRNA fragment.
Differential expression analysis using normalized quantifications per family (relative abundance). Sequence, directionality, and significance of isomiRs (with average log2RPM >2) were provided for isomiRs within two miRNA families, left: miR-122-5p, right: miR-21-5p. The canonical miRNA sequence is highlighted in bold. Blue: down-regulated, red: up-regulated. miRNA, microRNA.
Machine learning predictive outcome for MASH vs. MASL with confusion matrix (left) and ROC curve (right). MASH, metabolic dysfunction-associated steatohepatitis; MASL, metabolic dysfunction-associated steatotic liver; AUROC, receiver operator characteristic area under the curve.
mRNA differential expression analysis. (A) Number of significant genes in each of the three comparator groups and two ordinal regression analyses (BH P-adj <0.1). (B, C) Interstudy comparison of significant genes associated with MASH (C), or liver fibrosis (D). Both cases showed a highly significant positive correlation between the log2FoldChange across studies. (D) Interstudy comparison of gene set enrichment analysis. Significantly differentially expressed genes across each comparative method were considered with the following threshold: BH P-adjust <0.1 from our study, and <0.05 from Hoang et al. study. We used KEGG (K), Reactome (R), and GO: Biological Process (G) repositories. NAS ordinal regression did not generate enough significant genes for successful enrichment analysis. DEA, differential expression analysis; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MASH, metabolic dysfunction-associated steatohepatitis; mRNA, messenger RNA; MASL, metabolic dysfunction-associated steatotic liver; NAS, metabolic dysfunction-associated steatotic liver disease activity score.
MicroRNA regulatory role in MASLD. (A) Expression of an ML-selected differentially expressed isomiR from miR-122-5p family (canonical seed) associated with MASH severity across diagnosis, fibrosis scores and NAS. For fibrosis and NAS, log2FC indicates log2-FoldChange for high over low scores, and ‘padj’ indicates the BH-adjusted P-value of the ordinal regression method. (B) AI-informed putative regulatory network involved in MASH severity of the same miR-122-5p isomiR using Target Scan (TS) as target predictor. Stars indicate genes that are also significantly dysregulated in the same direction in the Hoang et al. study (BH P-adjusted <0.1). (C, D) Same as (A, B) for an isomiR of miR-21-5p (canonical seed) associated to advanced fibrosis. AI, artificial intelligence; MASH, metabolic dysfunction-associated steatohepatitis; MASLD, metabolic dysfunction-associated steatotic liver disease; ML, machine learning; NAS, metabolic dysfunction-associated steatotic liver disease activity score.
MicroRNA regulatory role in MASLD. (A) Expression of an ML-selected differentially expressed isomiR from miR-26b-5p family (one 5’ deletion) associated with MASH severity from across diagnosis, fibrosis scores and NAS. For fibrosis and NAS, log2FC indicates log2-FoldChange for high over low scores, and ‘padj’ indicates the BH-adjusted P-value of the ordinal regression method. (B) AI-informed putative regulatory network involved in MASH severity of the same miR-26b-5p isomiR using Target Scan (TS) as target predictor. Stars indicate genes that are also significantly dysregulated in the same direction in the Hoang et al. study (BH P-adjusted <0.1). (C, D) Same as (A, B) for an isomiR of miR-320a-5p (canonical seed) associated with advanced fibrosis. AI, artificial intelligence; MASH, metabolic dysfunction-associated steatohepatitis; MASLD, metabolic dysfunction-associated steatotic liver disease; ML, machine learning; NAS, metabolic dysfunction-associated steatotic liver disease activity score.
MASLD patents overview (number in bracket are standard deviation)
List of sRNA with annotation and differential analyses results (after filtering for sequencing quality and minimum log2 RPM expression ≥2.0)
A. List of predictive sRNA for ML model related to MASH severity. B. Extended list of predictive sRNA for ML model related to MASH severity through correlation analysis
A. List of predictive sRNA for ML model related to fibrosis. B. Extended list of predictive sRNA for ML model related to fibrosis through correlation analysis
AI generated knowledge graph representing miRNA to disease associations
Gene differential expression analyses results (significant padj >0.1 in at least one DEA)
Pathway enrichment analyses
AI generated knowledge graph representing disease to pathway associations
AI generated knowledge graph representing disease to gene associations
List of targeted genes for the 182 seed sRNA selected in either NAS of fibrosis progression and relation with the disease