Identification of circadian rhythm-associated key genes in metabolic dysfunction-associated fatty liver disease Authors: Jinbo Luo, Xiuying Ma, Quanyu Wang, Ruyi Zhang, Ling Zhu, Xue Feng, Wenxue Wang, Jiawei Geng Published: 2026 ABSTRACT As of now, current studies have confirmed that circadian rhythm genes (CRGs) can modulate the onset and development of metabolic dysfunction-associated fatty liver disease (MAFLD). Leveraging computational biology methodologies, systematic and thorough investigations into MAFLD and CRGs enhance our capacity to gain deeper insights into the disease’s pathogenic mechanisms. MAFLD-associated datasets (GSE89632 and GSE126848) and 248 circadian rhythm genes genes (CRGs) were analyzed. Principal component analysis (PCA) was performed on the GSE89632 dataset, and samples were divided into training and test sets. Differentially expressed genes (DEGs) from the training set were intersected with key module genes from WGCNA to identify intersection genes. These were then analyzed for expression consistency across datasets, followed by machine learning to identify key genes. Further analyses, including nomogram development, gene set enrichment analysis (GSEA), immune infiltration, and drug prediction, were also conducted. Four outliers were excluded, leaving 17 MAFLD and 22 control samples for analysis. 141 intersection genes were identified, with 36 showing consistent expression trends. Machine learning identified SOCS2, FBXO27, and HMMR as key genes. A nomogram based on these genes predicted MAFLD patient survival. GSEA revealed significant enrichment in pathways like olfactory transduction and peroxisome. Mast cell activation correlated strongly with SOCS2 and FBXO27. SOCS2, FBXO27, and HMMR are identified as key genes in MAFLD, providing new insights into the disease mechanisms.Supplementary InformationThe online version contains supplementary material available at 10.1007/s10238-026-02234-z. Introduction Metabolic dysfunction-associated fatty liver disease (MAFLD) is a type of fatty liver disease diagnosed based on evidence of metabolic dysfunction, and its occurrence and progression are closely associated with insulin resistance and genetic susceptibility, and it is closely associated with insulin resistance (IR) as well as genetic susceptibility [1]. MAFLD progresses along a well-defined pathological continuum, beginning with simple fatty liver, advancing to the inflammatory stage of Metabolic dysfunction-associated steatohepatitis(MASH), and potentially culminating in life-threatening conditions like cirrhosis and liver cancer (HCC) [2]. Furthermore, it is linked to an elevated risk of type 2 diabetes (T2DM), atherosclerotic cardiovascular disease (ASCVD), and certain cancers, including colorectal adenocarcinoma [3]. In the context of the current epidemiological transition characterized by obesity and metabolic syndrome, coupled with significantly accelerated lifestyles, the prevalence of MAFLD has been increasing annually, now affecting approximately 25% of the global population. This condition imposes substantial economic and disease burdens on both patients and society, emerging as a critical worldwide public health challenge [4, 5]. The pathogenesis of MAFLD is highly complex. Current clinical management predominantly relies on lifestyle modification, while specific pharmacological therapeutics remain unavailable. The development of novel treatment strategies therefore represents a pressing unmet medical need [6]. Accumulating research highlights a close link between dysregulated circadian rhythms and the etiology and progression of MAFLD [7]. Unraveling the mechanisms of circadian rhythm disruption and identifying associated biomarkers holds significant promise for developing novel diagnostic and therapeutic strategies against MAFLD. The circadian clock is an endogenous regulatory system that controls physiological and behavioral rhythms, with its core mechanism relying on negative feedback loops to maintain 24-hour oscillatory cycles [8]. In mammals, circadian regulation originates in the central clock located within the suprachiasmatic nucleus (SCN) of the hypothalamus. This master pacemaker receives photic signals transmitted from the retina and generates rhythmic oscillations through a transcription-translation feedback loop composed of core clock genes (CLOCK/BMAL1) and their negative regulators (REV-ERB, PER, and CRY). This molecular mechanism coordinates the synchronization between endogenous biological processes and external environmental cycles [9]. The central rhythmic signals from the SCN are further relayed to peripheral tissues through peripheral clocks (e.g., in the liver, heart, and kidneys). Among these serves as a critical peripheral pacemaker, playing a pivotal role in circadian metabolic regulation and energy homeostasis. Emerging evidence demonstrates that the circadian clock regulates systemic metabolism through nuclear receptors, nutrient-sensing pathways, and rate-limiting enzymes. Disruption of circadian rhythms contributes to dysregulated lipid metabolism, obesity, and metabolic disorders, and is strongly linked to the pathogenesis and progression of MAFLD [10, 11]. However, the precise molecular mechanisms remain incompletely elucidated, underscoring the critical need to identify novel circadian-related molecules and delineate their mechanistic roles. Through an integrated bioinformatics interrogation of public genomic repositories, we sought to delineate pivotal circadian genes underlying MAFLD progression. Collectively, these findings offer a theoretical framework to elucidate MAFLD pathogenesis and devise novel therapeutic interventions. This work thereby provides a crucial advancement in our mechanistic understanding of the disease. Materials and methods Data source For the present research, the microarray dataset GSE89632 (platform: GPL14951) contained 19 liver tissue samples from MAFLD patients and 24 normal control samples. The dataset GSE126848 (platform: GPL18573) included 16 MAFLD liver tissue samples and 14 normal control samples. These datasets were derived from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/gds). The training dataset GSE89632 was generated on the GPL14951 microarray platform. Probe annotation conversion was performed prior to analysis to map probes to official gene symbols. The validation dataset GSE126848 was obtained from the Illumina NextSeq 500 sequencing platform and required no additional probe conversion. These processing steps ensured consistent analytical criteria at the gene level between the two datasets. Unfortunately, detailed demographic and clinical information is not available in datasets retrieved from the GEO database, which is an inherent limitation of public omics data. Besides, 248 CRGs in total were acquired according to existing published literature [12] . This retrospective study enrolled participants who underwent liver biopsy or hepatic cyst resection at the First People’s Hospital of Yunnan Province in 2022. The MAFLD group consisted of patients with pathologically confirmed MAFLD, from whom liver biopsy specimens were collected. The control group included patients undergoing surgical resection for benign hepatic cysts, with all individuals confirmed free of MAFLD and other chronic liver diseases.Clinical infomation of the patients was shown in Supplementary Table S2. Inclusion criteria: (1) age 18–75 years; (2) MAFLD diagnosis established by liver biopsy according to current guidelines; (3) availability of complete clinical and laboratory data, including alanine transaminase (ALT), aspartate transaminase (AST), total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), fasting blood glucose (FBG), and body mass index (BMI). Exclusion criteria: (1) viral hepatitis (HBV/HCV), autoimmune liver disease, drug-induced liver injury, or hereditary liver disorders; (2) liver cirrhosis, hepatocellular carcinoma, or other malignancies; (3) severe cardiovascular, renal, or endocrine diseases; (4) history of alcohol abuse (≥ 20 g/day for men, ≥ 10 g/day for women), pregnancy, lactation, or incomplete clinical data. Differential expression analysis For the purpose of investigating inter-sample similarity, we employed a principal component analysis (PCA) to conduct quality control on the samples within the GSE89632 dataset. Outliers were removed based on the two-dimensional PC1-PC2 space from principal component analysis (PCA). Mahalanobis distance with a threshold corresponding to the 99% quantile of the chi-squared (χ²) distribution was used to identify outliers, which were further confirmed by visual inspection. Following this, the team detected differential gene expression between MAFLD and control samples in the GS E89632 dataset using the limma package (v 3.52.4) [13], and filtering thresholds were defined as |log2FC| > 1 and an adjusted p-value (padj) < 0.05. Differential expression results were presented as a volcano plot and a heatmap, and these two visualization plots were built with the respective R packages: ggplot2 (version 3.3.6) for the volcano plot and pheatmap (version 1.0.12) for the heatmap [14, 15]. Weighted gene co-expression network analysis (WGCNA) Firstly, the abundance of cuproptosis-related pathways was quantified using the ssGSEA algorithm. To elaborate, we calculated an enrichment score for every sample within the GSE89632 dataset, using the predefined gene set consisting of 248 CRGs. Subsequently, the statistical significance of the score difference between the MAFLD and control cohorts was evaluated with a non-parametric Wilcoxon test. Following this, for the construction of co-expression networks, we employed the R-based WGCNA package (version 1.72.1) to conduct an analysis of transcriptomic data from the GSE89632 dataset [16]. After that, To identify potential outliers, all samples in the GSE89632 dataset were subjected to cluster analysis prior to further processing. Subsequently, the soft threshold was identified aiming to ensure that the interactions between genes fit a scale-free distribution as much as possible, a key requirement for reliable co-expression network construction. Additionally, gene modules were detected by means of a hybrid dynamic tree-cutting algorithm, with each module required to contain a minimum of 200 genes. We first assessed gene modules for their correlations with CRGs using Pearson correlation analysis, and subsequently identified the module with the maximum absolute correlation coefficient as the key module. Ultimately, key module genes were chosen with selection thresholds set as |Module Membership (MM)| > 0.8 and |Gene Significance (GS)| > 0.2. Functional enrichment profiling integrated with screening of intersecting genes Moreover, DEGs and key module genes were taken to intersect through the ggvenn package (v 1.2.2) to obtain intersection genes [17]. For exploring the biological roles and pathways of the intersection gene set, our team carried out GO and KEGG enrichment analyses by employing the clusterProfiler R package (v 4.7.1.001), defining the significance threshold as padj < 0.05 [18]. Furthermore, we built a PPI network for intersection genes using the STRING database (http://string.embl.de/) and applied a confidence score cutoff of > 0.4. Screening of key genes To analyze the overlapping genes in the GSE89632 and GSE126848 datasets, we performed the Wilcoxon test. Candidate genes were defined as those showing both a concordant expression direction and significant differential expression across comparative groups. Next, feature genes underwent further refinement via the application of least absolute shrinkage and selection operator (Lasso) regression using the glmnet package, and support vector machine recursive feature elimination (SVM-RFE) through the e1071 package [19, 20]. Taking the intersection of these feature gene subsets (derived from distinct screening approaches) resulted in the final key gene set. Ultimately, the differentiation performance of individual key genes between MAFLD and control samples was assessed via the generation of receiver operating characteristic (ROC) curves with the pROC package (v 1.18.4) [21] .The optimal cutoff value suitable for clinical application was determined based on the Youden index and expressed as standardized expression level. Construction of a nomogram for key genes Leveraging the rms package (v 6.7.0), scores were assigned to key gene expression levels [22], we separately constructed nomograms for predicting the probability of MAFLD diagnosis in the GSE89632 dataset and the GSE126848 dataset, respectively. Moreover, the diagnostic accuracy of the constructed nomogram was evaluated by means of a calibration curve. Additionally, Spearman correlation analysis was carried out on the key genes in the GSE89632 dataset, with the findings visualized using the ggcor package (v 0.9.8.1). GSEA and gene-gene interaction (GGI) analysis of key genes Firstly, Spearman correlations were computed for each key gene against all genes leveraging the psych package (v 2.3.6) [23]. The coefficients derived from this analysis served as the basis for ranking in the GSEA enrichment analysis. We then performed GSEA in R using the clusterProfiler package (v 4.7.1.001) [18], with the c2.cp.kegg.v7.4.symbols.gmt gene set (retrieved from http://www.gsea-msigdb.org/gsea/msigdb) as the reference; significance was determined by an adjusted p-value < 0.05. Moreover, the GeneMANIA (http://genemania.org) was used to construct the GGI network, and top 20 genes interacting with key genes were selected as key nodes to be displayed in the network diagram according to the order of score values from largest to smallest. Immune infiltration analysis The CIBERSORT computational algorithm (v 1.03) [24] and the LM22 immune cell gene set [25] were applied by our team to determine the proportions of 22 immune cell types. Moreover, group differences in the percentages of the 22 immune cell types were evaluated using the Wilcoxon test. For investigating associations between the previously identified key genes and differential immune cells, we performed Spearman correlation analysis via the ggcor package. Development of functional regulatory networks focused on key genes We predicted miRNAs targeting key genes and their interaction pairs using the starBase database (https://rnasysu.com/encori/index.php) and DIANA microT database (https://dianalab.e-ce.uth.gr/microt_webserver), specifically for each respective analysis. The key miRNAs and key interactions were obtained by taking the intersection of the predicted relationship pairs. Subsequently, lncRNAs targeting the key miRNAs were predicted using the starBase database, and key lncRNAs were screened for using a clipExpNum threshold greater than 50. In addition, we identified potential key gene-related drugs by utilizing the Comparative Toxicogenomics Database (CTD; https://ctdbase.org/). Quantitative real-time polymerase chain reaction (qRT-PCR) Subsequently, we validated the expression of key genes via qRT-PCR. A total of 5 paired clinical specimens from The First People’s Hospital of Yunnan Province were analyzed using qRT-PCR, comprising 5 MAFLD and 5 matched normal samples. We first obtained all required ethical approvals from the Ethics Committee of The First People’s Hospital of Yunnan Province. Additionally, informed consent was obtained from each patient participating in this study. Total RNA was extracted from frozen MAFLD and normal tissues via TRIzol reagent. After isolation, equal amounts of mRNA were converted into cDNA through reverse transcription. Quantitative PCR was subsequently carried out using a 2× Universal Blue SYBR Green qPCR Master Mix kit under the following cycling conditions: initial denaturation at 95 °C for 1 min; followed by 40 cycles of denaturation at 95 °C for 20 s, annealing at 55 °C for 20 s, and extension at 72 °C for 30 s. Gene expression levels were ultimately determined via the 2 − ΔΔCT method. We ultimately determined the expression levels of target key genes using the 2 − ΔΔCT method, and all primer sequences used in this experiment are listed in Online Resource 1. Statistical processing of data We performed bioinformatic analyses using the R software. Between-group differences were assessed by the Wilcoxon test, and statistical significance was defined as p < 0.05. Results There were 439 DEGs in GSE89632 In the PCA analysis, four outliers (GSM2385782, GSM2385767, GSM2385731, and GSM2385721) were excluded from the GSE89632 dataset. These outliers were attributed to biological individual heterogeneity or minor technical variation. The remaining 17 control samples and 22 MAFLD samples were used as the training set for subsequent analyses (Fig. 1a.). We found 439 differentially expressed genes (DEGs) overall: among them, 170 genes were noted as significantly upregulated in MAFLD samples, while the remaining 269 exhibited downregulation (Fig. 1b and c). Fig. 1Identification of circadian rhythm-related subtypes in MAFLD. a PCA of circadian gene expression. b Volcano plot of DEGs. c Heatmap of circadian gene expression patterns Identification of circadian rhythm-related subtypes in MAFLD. a PCA of circadian gene expression. b Volcano plot of DEGs. c Heatmap of circadian gene expression patterns There were 400 key module genes in MEyellow module ssGSEA algorithm-based analysis demonstrated a significant difference in CRGs scores among samples, specifically between the MAFLD and control groups. (Fig. 2a). Following cluster analysis, no outlier samples were observed that necessitated elimination (Fig. 2b). Selection of a soft threshold value of 12 resulted in a network conforming to scale-free topology (Fig. 2c). Following this, ten distinct gene modules were identified (Fig. 2d). We designated the MEyellow module as the key module, given that it demonstrated the strongest negative correlation with CRGs scores (r = -0.56, p = 2 × 10⁻⁴) (Fig. 2e). After |MM|>0.8 and |GS|>0.2 as criteria for screening, 400 key module genes were obtained (Fig. 2f). Fig. 2WGCNA identifies MAFLD-associated modules. a Expression of a key circadian gene (CRY1). b Sample clustering and trait association. c Soft-thresholding power selection. d Gene dendrogram and module identification. e Module-trait correlations. f Gene significance-module membership correlation (cor = 0.26, p = 1.1e-32) WGCNA identifies MAFLD-associated modules. a Expression of a key circadian gene (CRY1). b Sample clustering and trait association. c Soft-thresholding power selection. d Gene dendrogram and module identification. e Module-trait correlations. f Gene significance-module membership correlation (cor = 0.26, p = 1.1e-32) The 141 intersection genes were significantly enriched in 339 functional items and 36 signaling pathways By crossing DEGs and key module genes, 141 intersection genes were obtained (Fig. 3a). In GO enrichment analysis, intersection genes significantly enriched 339 functional items (cellular component, CC: 1; molecular function, MF: 3; biological function, BF: 335), among which, fat cell differentiation, multi-organism reproductive process, multi-cellular organism process, which are relevant to MAFLD (Fig. 3b). Additionally, through KEGG pathway enrichment analysis, we found that candidate genes were enriched in 36 signaling pathways—with three key pathways among them (TNF signaling pathway, IL-17 signaling pathway, and osteoclast differentiation signaling pathway) showing close associations with MAFLD. (Fig. 3c). In the PPI network, there were 129 candidate genes that formed 445 interactions at the protein level, such as MAFF-MIDN, ETS2-SOCS2, and so on (Fig. 3d). Fig. 3Functional enrichment analysis of hub genes associated with MAFLD and circadian rhythm. a Venn diagram showing the overlap between differentially expressed genes (DEGs) and genes from key WGCNA modules. b Gene Ontology (GO) enrichment analysis of biological processes for the overlapping genes. c Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. d Protein-protein interaction (PPI) network of the core hub genes Functional enrichment analysis of hub genes associated with MAFLD and circadian rhythm. a Venn diagram showing the overlap between differentially expressed genes (DEGs) and genes from key WGCNA modules. b Gene Ontology (GO) enrichment analysis of biological processes for the overlapping genes. c Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. d Protein-protein interaction (PPI) network of the core hub genes The SOCS2, FBXO27 and HMMR showed excellent predictive performance for MAFLD Following expression profiling of overlapping genes in the GSE89632 and GSE126848 datasets, 36 genes exhibited concordant differential expression patterns and were statistically significant across comparative groups. Thus, we identified these genes as candidate genes for subsequent investigation, with details presented in Fig. 4a. Moreover, 9 (ETS2, FBXO27, FMO1, GRIA3, HMMR, PFKFB3, SOCS2, TK1, TYMS) and 7 feature genes (EPHA2, SOCS2, FBXO27, HMMR, UGT2B11, PELI1, SLITRK3) were selected in Lasso and SVM-RFE analysis, respectively (Fig. 4b and c), which were taken intersection to mined 3 (SOCS2, FBXO27, HMMR) key genes (Fig. 4d). In both the GSE89632 and GSE126848 datasets, key genes exhibited AUC (area under the curve) values exceeding 0.8 in ROC analysis, thus demonstrating strong predictive accuracy (Fig. 4e). The sensitivity, specificity and optimal cutoff value of each key gene all exhibited favorable diagnostic performance in both training and validation datasets (Table S1). Fig. 4Development of a MAFLD diagnostic signature. a Expression patterns (a1: heatmap; a2: boxplot) of candidate genes. b LASSO regression analysis (b1: coefficient profiles; b2: parameter selection). c Feature selection curve. d Overlap of genes selected by LASSO and SVM. e ROC curves of the three-gene signature in training and validation sets Development of a MAFLD diagnostic signature. a Expression patterns (a1: heatmap; a2: boxplot) of candidate genes. b LASSO regression analysis (b1: coefficient profiles; b2: parameter selection). c Feature selection curve. d Overlap of genes selected by LASSO and SVM. e ROC curves of the three-gene signature in training and validation sets The diagnostic performance for the nomogram model of key genes was admirable Utilizing expression profiles of the key genes, we constructed a nomogram for predicting the risk of MAFLD (Fig. 5a). In the GSE89632 dataset, the regression equation of the nomogram was as follows: y = − 59.6928 + (− 8.0389) × SOCS2 + 6.4281 × FBXO27 + 9.4274 × HMMR. The calibration curve showed a slope close to 1, indicating good predictive performance for MAFLD (Fig. 5b). In the validation cohort GSE126848, the risk of disease approached 100% when the total score reached 106 (Fig. 5c). The calibration curve yielded a p-value of 0.694, further supporting its favorable predictive accuracy (Fig. 5d). Meanwhile, correlation analysis among key genes indicated that FBXO27 exhibited the most robust positive correlation with HMMR (r = 0.76, p < 0.01) and the most robust negative correlation with SOCS2 (r = -0.81, p < 0.01) (Fig. 5c). Fig. 5Clinical application of the MAFLD diagnostic signature. a Nomogram for predicting MAFLD risk in the training cohort based on three-gene expression. b Calibration curve of the nomogram in the training cohort, showing agreement between predicted and observed outcomes. c Nomogram for predicting MAFLD risk in the validation cohort based on three-gene expression. d Calibration curve of the nomogram in the validation cohort, showing agreement between predicted and observed outcomes. e Correlation analysis among the three signature genes Clinical application of the MAFLD diagnostic signature. a Nomogram for predicting MAFLD risk in the training cohort based on three-gene expression. b Calibration curve of the nomogram in the training cohort, showing agreement between predicted and observed outcomes. c Nomogram for predicting MAFLD risk in the validation cohort based on three-gene expression. d Calibration curve of the nomogram in the validation cohort, showing agreement between predicted and observed outcomes. e Correlation analysis among the three signature genes The signaling pathways significantly enriched in key genes included peroxisome, olfactory transduction and so on GSEA enrichment analysis was applied to further explore the biological functions of the key genes, with the top 5 significantly enriched signaling pathways for the key genes visualized. Our analysis revealed that key genes all exhibited significant enrichment in pathways including olfactory transduction, cytokine-cytokine receptor interaction, peroxisome, JAK-STAT signaling pathway, and lysosome (Fig. 6a and c). In the GGI network, the key genes formed 300 relationship pairs (such as HMMR-LEPR, SOCS2-SOCS3) with their 20 interacting genes (such as JAK2, CUL5, SC9A1). Further, these genes were all significantly enriched for functions including response to growth hormone, regulation of receptor signaling pathway via JAK-STAT, etc. (Fig. 6d). Fig. 6Functional analysis of signature genes. a GSEA enrichment plot for SOCS2. b GSEA enrichment plot for FBXO27. c GSEA enrichment plot for HMMR. d Protein interaction network and functional annotation of the three signature genes Functional analysis of signature genes. a GSEA enrichment plot for SOCS2. b GSEA enrichment plot for FBXO27. c GSEA enrichment plot for HMMR. d Protein interaction network and functional annotation of the three signature genes The activated mast cells was the immune cells with the highest correlation to key genes Employing the CIBERSORT algorithm, the proportions of infiltration for 22 immune cell types were estimated in MAFLD case samples and control tissues. (Fig. 7a). We conducted a comparative analysis and found a significant reduction in the infiltration abundances of naive B cells, monocytes, activated dendritic cells, activated mast cells, and neutrophils within MAFLD samples. In contrast, an inverse pattern was observed for four other immune cell subtypes: gamma delta T cells, M2 macrophages, resting dendritic cells, and resting mast cells (Fig. 7b). By means of correlation analysis, our findings indicated that SOCS2 exhibited the most robust positive correlation with activated mast cells (r = 0.851, p < 0.001), while FBXO27 showed the most prominent negative correlation with these cells (r = -0.870, p < 0.001) (Fig. 7c). Fig. 7Immune infiltration analysis. a Composition of immune cell infiltration in samples. b Differences in immune cell proportions between Control and MAFLD groups. c Correlation between signature genes and specific immune cell types Immune infiltration analysis. a Composition of immune cell infiltration in samples. b Differences in immune cell proportions between Control and MAFLD groups. c Correlation between signature genes and specific immune cell types The miRNAs, lncRNAs and drugs corresponding to key genes were predicted Based on starBase and DIANA microT databases, 214 and 288 miRNAs were predicted, respectively, and 37 key miRNAs were obtained after their intersection (Fig. 8a). Then, 22 key lncRNAs were predicted to correspond to 31 key miRNAs in the starBase database based on clipExpNum > 50 conditions. The lncRNA-miRNA-mRNA network contains 99 relationship pairs, such as LINC00461-hsa-miR-485-3p-HMMR, AC021078.1-hsa-miR-7-5p-SOCS2, etc. (Fig. 8b). FBXO27 yielded no significant results when predicting drugs in the DsigDB database. Finally, 51 drugs were predicted, of which 11 drugs were predicted by SOCS2 and HMMR, including progesterone, enterolactone, camptothecin, resveratrol, calcitriol, scriptaid, testosterone, trichostatin A, anisomycin, COUMESTROL, and 0175029-0000 (Fig. 8c). Fig. 8Regulatory network and therapeutic prediction. a Overlap of miRNA targets predicted by starbase and microT databases. b ceRNA regulatory network of signature genes. c Potential therapeutic compounds targeting the MAFLD signature pathway Regulatory network and therapeutic prediction. a Overlap of miRNA targets predicted by starbase and microT databases. b ceRNA regulatory network of signature genes. c Potential therapeutic compounds targeting the MAFLD signature pathway The expression of core candidate genes was validated through qRT-PCR In the qRT-PCR validation, both FBXO27 and HMMR were upregulated in MAFLD samples compared to controls, with a statistically significant difference observed for HMMR expression between the groups. SOCS2 was significantly overexpressed in the control group. Their expression trend is consistent with that in GSE89632 and GSE126848. Although FBXO27 was upregulated, the change did not reach statistical significance (p = 0.3166). Given the relatively small sample size (n = 5 per group), these findings need to be further validated in larger independent cohorts (n ≥ 30) (Fig. 9). Fig. 9Experimental validation of signature genes. a FBXO27 expression shows no significant difference (p = 0.3166). b HMMR is significantly upregulated in MAFLD (p = 0.0472). c SOCS2 shows significantly low expression in MAFLD (p = 0.0029). All data normalized to GAPDH Experimental validation of signature genes. a FBXO27 expression shows no significant difference (p = 0.3166). b HMMR is significantly upregulated in MAFLD (p = 0.0472). c SOCS2 shows significantly low expression in MAFLD (p = 0.0029). All data normalized to GAPDH Discussion With the acceleration of modern lifestyles, the prevalence of MAFLD has been rising significantly, while clinically effective pharmacological treatments remain lacking [26].Numerous studies have demonstrated that circadian rhythm alterations contribute significantly to MAFLD progression. Key circadian clock genes, including BMAL1, CLOCK, CRY, and PER, have all been demonstrated to be significantly linked to MAFLD pathogenesis [27].This study employs bioinformatics analysis to identify novel circadian clock genes FBXO27, HMMR, and SOCS2, thereby paving the way for the development of novel interventions for fatty liver disease. Consistent with our experimental validation results, SOCS2 was significantly downregulated in MAFLD hepatic tissues. Although SOCS2 exerts pro-lipogenic and pro-inflammatory effects to facilitate hepatic lipid deposition, its decreased expression may represent a negative feedback regulatory response during disease progression. As a member of the SOCS gene family, Suppressor of Cytokine Signaling 2 (SOCS2) acts as a negative regulator in the JAK-STAT signaling pathway. It adjusts lipid metabolism via the inhibition of growth hormone (GH) signaling and insulin-like growth factor-1 (IGF-1) signaling pathways [28]. SOCS2 binds to the GH receptor (GHR) and promotes JAK2 ubiquitination and degradation, thereby inhibiting STAT5 activation, reducing IGF-1 production, and suppressing lipolysis. Additionally, SOCS2 attenuates the PI3K-AKT pathway by facilitating ubiquitin-mediated degradation of insulin receptor substrates (IRS1/2), which decreases insulin sensitivity and enhances lipid synthesis. Furthermore, SOCS2 indirectly represses SREBP-1c transcriptional activity and PPARα expression, leading to downregulation of fatty acid synthase (FASN, ACC) expression and impaired mitochondrial β-oxidation capacity. SOCS2 also alleviates inflammation-associated lipid accumulation by suppressing TLR4-NF-κB signaling while promoting M2 macrophage polarization. In summary, the mechanisms mediated by SOCS2 drive the development of hepatic lipid deposition—a process that is pivotal for promoting the pathogenesis of fatty liver disease [29].Although there are no reports directly linking SOCS2 to the core circadian regulatory machinery, both immune and metabolic processes—known to be under circadian control—exhibit diurnal oscillations (e.g., rhythmic secretion of inflammatory cytokines). Notably, studies have demonstrated that SOCS3, another member of the SOCS gene family, displays circadian oscillations in baseline expression across brown adipose tissue, liver, and white adipose tissue, indicating direct regulation by the molecular clock. This suggests potential indirect circadian influences on SOCS2-mediated pathways [30]. This study suggests that SOCS2 serves as a key circadian rhythm-associated gene in MAFLD, while the exact underlying mechanisms still warrant additional exploration.Given that SOCS2 belongs to the same SOCS family and shares similar structural domains, it is reasonable to speculate that SOCS2 may also be regulated by circadian rhythms. However, whether and how circadian mechanisms directly regulate hepatic SOCS2 expression remains to be experimentally validated. Hyaluronan-mediated motility receptor (HMMR), a protein characterized by prominent coiled-coil structural domains, functions as a key hyaluronan-binding receptor and critically regulates essential cellular processes including migration, proliferation, and differentiation [31]. Studies have demonstrated that HMMR functions as an oncogene and is closely associated with various cancers [32], including lung adenocarcinoma [33], breast cancer [34], and prostate cancer [35] .HMMR exerts a central regulatory role in MAFLD by activating the NF-κB signaling pathway [36]. Binding of free fatty acids or endotoxins to TLR4/TNFR activates the relevant signaling pathway, which further induces NF-κB activation. This process subsequently promotes upregulated transcription of key pro-inflammatory mediators (e.g., TNF-α and IL-6), thus sustaining persistent chronic hepatic inflammation [37]. Additionally, NF-κB signaling exacerbates insulin resistance by suppressing IRS1/2 phosphorylation while upregulating SREBP-1c to enhance lipogenesis and inhibit PPARα-mediated fatty acid oxidation, ultimately leading to hepatic lipid accumulation [38] . In 2007, the World Health Organization (WHO) classified circadian rhythm disruption as a carcinogenic factor. Studies have confirmed that neoplastic tissues lose responsiveness to physiological regulation and rhythmic entrainment by the central pacemaker, exhibiting circadian rhythms that are desynchronized from normal tissues with periods significantly shorter than 24 h [39]. Following tumor formation, both shortened circadian periods (< 24 h) and constant light-induced rhythm disruption accelerate tumor growth [40]. Based on these observations, we propose that dysregulation of circadian rhythms could facilitate tumor development through modulation of HMMR-regulated cellular migration and proliferation mechanisms. Furthermore, we speculate that HMMR might also influence the pathogenesis of MAFLD through circadian rhythm modulation. F-BXO27 (F-box Protein 27), which may acts as a key member of the F-box protein family, takes part in the ubiquitin-proteasome system (UPS) and facilitates the regulation of various cellular processes including cell cycle control, cell differentiation, and cell survival [41]. Similar to FBXO2 and FBXO6, FBXO27 recognizes glycoproteins modified with high-mannose type N-glycans and can also bind certain glycoproteins bearing complex-type N-glycans [42]. Autophagy is an evolutionarily conserved, genetically programmed catabolic process that promotes cell survival and maintains intracellular homeostasis. Autophagy impairment has been observed in the livers of both human MAFLD/MASH patients and obese murine models, indicating its essential role in fatty liver disease pathogenesis [43, 44]. Notably, FBXO27 may function as a ubiquitinated glycoprotein to modulate autophagy machinery through the SQSTM1-MAP1LC3B pathway, thereby influencing MAFLD progression [45]. Current evidence implies that UPS enzymes can modulate cellular circadian rhythms [46]. Although FBXO27 has not been explicitly reported to participate in circadian regulation, its family member FBXL3 directly controls the stability of the core clock protein CRY [47]. We therefore hypothesize that FBXO27 may indirectly influence circadian rhythms through analogous mechanisms. Our study further demonstrates that FBXO27 may serve as a circadian-related diagnostic biomarker for MAFLD. GSEA analysis revealed that the key genes SOCS2, FBXO27, and HMMR were significantly enriched in the olfactory transduction pathway. Accumulating evidence has demonstrated that multiple olfactory receptors exhibit functional ectopic expression in peripheral metabolic tissues, including the liver, adipose tissue, pancreas, and intestine. These “peripheral olfactory receptors” can be activated by circulating odorants or metabolites and participate in the regulation of lipolysis, fatty acid oxidation, and gluconeogenesis through the cAMP/PKA/CREB signaling cascade. For instance, activation of the ectopically expressed olfactory receptor OR10J5 in human hepatocytes by α-cedrene significantly reduced intracellular triglyceride content via the cAMP-PKA pathway; conversely, OR10J5 silencing increased lipid accumulation, accompanied by upregulated lipogenic genes and downregulated fatty acid oxidation genes [48]. Similarly, Wu et al. reported that the ectopic olfactory receptor Olfr43 in mouse liver regulates systemic energy metabolism and hepatic lipid accumulation via the CREB-HES1-PPARγ signaling axis [49]. In addition, the olfactory system exhibits close bidirectional crosstalk with circadian rhythm regulation. The olfactory bulb has been identified as a circadian oscillator that functions independently of the suprachiasmatic nucleus, and genes involved in the transduction pathway in the olfactory epithelium display robust diurnal oscillations [50, 51] Given that SOCS2, FBXO27, and HMMR were originally identified through circadian rhythm-associated gene analysis, the enrichment in olfactory transduction may reflect a common transcriptional regulatory network linking circadian rhythms to both olfactory signaling and hepatic metabolism, rather than pathway-specific activation in MAFLD. Collectively, the GSEA enrichment of the olfactory transduction pathway suggests a potential mechanism by which ectopic hepatic olfactory receptors contribute to metabolic dysregulation in MAFLD. Although functional evidence for hepatic olfactory receptors in MAFLD/MASH remains limited, this emerging field represents a promising direction for investigating the interplay among circadian rhythms, olfactory signaling, and liver metabolism. Immune infiltration analysis revealed that activated mast cells were significantly decreased, while resting mast cells were correspondingly increased in MAFLD samples. SOCS2 was positively correlated, and FBXO27 was negatively correlated with the abundance of activated mast cells. These findings suggest that the key genes may participate in MAFLD progression, at least in part, by regulating mast cell phenotypes. Mechanistically, SOCS2 acts as a negative regulator of the JAK-STAT pathway. Its downregulation in MAFLD may lead to aberrant JAK-STAT activation [52, 53], thereby affecting mast cell differentiation and degranulation. Recent studies have shown that STAT5 signaling is critical for mast cell survival and activation. Accordingly, the loss of SOCS2-mediated negative regulation may drive mast cells toward an exhausted or dysfunctional state, characterized by a reduced activated population [54].On the other hand, FBXO27 is a component of the SCF (SKP1/CUL1/F-box) ubiquitin ligase complex. It may promote the degradation of glycoproteins or autophagy factors essential for maintaining mast cell homeostasis [42]. Its upregulation may impair mast cell viability or alter the activation threshold, resulting in fewer activated mast cells. We therefore hypothesize that the balance between activated and resting mast cells, rather than their absolute numbers, determines the net effect on liver pathology. Dysregulation of SOCS2 and FBXO27 may disrupt this balance, ultimately promoting MAFLD progression. However, these mechanistic hypotheses remain speculative and require validation through functional experiments. Drug prediction results identified a total of 51 drugs related to SOCS2, FBXO27, and HMMR, among which deferoxamine showed the strongest correlation. Ferroptosis is closely linked to the pathogenesis and progression of MAFLD. In this context, ferroptosis exacerbates hepatocyte inflammation, fibrogenesis, and liver injury [55]. The primary mechanism involves dysregulated iron metabolism, which promotes the Fenton reaction, generating ROS and subsequent lipid peroxidation, ultimately leading to cellular damage or death [56]. Liu et al. [57] demonstrated that ferroptosis triggers inflammatory responses in simple hepatic steatosis and aggravates hepatic lipid peroxidation and apoptosis, thereby driving the transition to MASH. In addition, specific ferroptosis-related genes such as ACSL3, FADS2, and GSS are reported to play pivotal roles in MAFLD progression [58]. Tsurusaki et al. [59] reported that ferroptosis accelerates MASH progression by initiating inflammatory responses in simple hepatic steatosis. Deferoxamine, an iron chelator, mitigates ferroptosis by sequestering iron and inhibiting the Fenton reaction, thereby suppressing lipid peroxidation [60]. Although deferoxamine has not received regulatory approval for the treatment of MAFLD, growing evidence indicates its potential therapeutic benefits. Studies suggest that it may mitigate liver injury and slow disease progression through mechanisms such as reduction of hepatic iron accumulation, attenuation of oxidative stress, and suppression of lipid peroxidation, ultimately improving iron overload-related dysregulation of hepatic lipid metabolism [61]. However, further clinical trials are warranted to comprehensively evaluate its long-term safety and therapeutic efficacy in MAFLD patients. The relatively small sample size used for qRT-PCR validation may limit the statistical power and generalizability of the expression results. Notably, FBXO27 failed to reach statistical significance (p = 0.3166), which does not necessarily indicate a true absence of differential expression in NAFLD. Future validation in larger, independent clinical cohorts is therefore warranted to enhance the reliability and translational potential of these findings. Although SOCS2, FBXO27, and HMMR were identified as circadian rhythm-associated candidate key genes, direct regulatory or physical interactions with core clock components (e.g., CLOCK/BMAL1, PER/CRY) remain experimentally unconfirmed. Further functional investigations are planned to provide direct evidence for whether SOCS2, FBXO27, and HMMR are indeed embedded within the core circadian regulatory network. Conclusions This study identified FBXO27, HMMR, and SOCS2 as novel circadian rhythm-related genes associated with MAFLD through bioinformatics analysis. However, these findings require validation with larger sample sizes and further experimental studies to confirm whether FBXO27, HMMR, and SOCS2 regulate MAFLD pathogenesis through circadian rhythm modulation. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2