Exploring hub genes related to adipocytokines in keloids: a combined analysis integrating single-cell, Mendelian randomization and bulk transcriptome data with experimental verification

Figures

• 

  FIGURE 1. Identification and functional profiling of adipocytokine-related genes in keloid tissues reveals dysregulated signaling networks. (A) Genome-wide differential expression analysis reveals extensive transcriptomic alterations in keloids. The volcano plot displays 818 significantly dysregulated genes (311 upregulated, red; 507 downregulated, blue) in keloid versus normal skin tissues from the GSE113619 dataset (18 keloid, 14 control). (B) Heatmap of differentially expressed genes. The top color bar annotation: pink represents Keloid samples, blue represents control samples. In the heatmap, the y-axis represents genes, with red indicating highly expressed genes and blue indicating lowly expressed genes. (C) Venn diagram of candidate genes. Intersection analysis pinpoints 19 candidate adipocytokine-related genes dysregulated in keloids. (D) GO enrichment analysis highlights biological processes, molecular functions, and cellular components associated with the 19 candidate genes, emphasizing their roles in cellular response to peptides and related pathways. (E) KEGG pathway analysis identifies significant enrichment in pathways such as “proteoglycans in lipid and atherosclerosis”,…

• 

  FIGURE 2. Integrative machine learning identifies PIK3R3 and ANGPTL5 as hub genes for keloids. (A) SVM-RFE plot shows classification accuracy (y-axis) as a function of the number of included features (x-axis), with the peak accuracy indicating the optimal feature set. (B) Corresponding error rate plot for SVM-RFE highlights the minimal error achieved with the selected feature subset. (C) Boruta feature selection analysis identifies eight genes as important predictors, with confirmed (green), tentative (yellow), and rejected (red) features displayed. (D) XGBoost analysis ranks gene importance, with the top eight genes contributing most to model performance. (E) The intersection of genes selected by SVM-RFE (13 genes), Boruta (8 genes), and XGBoost (8 genes) yielded a consensus set of seven genes: GAS6, IGF1, GSK3B, PIK3R3, IL1B, ANGPTL5, and OBSCN. (F) Validation in two cohorts (GSE113619 and GSE158395) confirmed that PIK3R3 is significantly upregulated and ANGPTL5 is significantly downregulated in keloids compared to normal skin (p < 0.05), establishing them as the final hub genes. (G) GeneMANIA functional association network illustrates interactions between the two hub genes and 20…

• 

  FIGURE 3. Mendelian randomization identifies PIK3R3 as a causal risk factor for keloids. (A) Forest plot compares OR and 95% CI from five Mendelian randomization methods, showing consistent positive association of PIK3R3 with keloid risk using the IVW method as primary reference (OR = 1.156, 95% CI = 1.054–1.268, p = 0.0020). (B) Forest plot displays individual IV effect sizes for PIK3R3. (C) Scatter plot of SNP-exposure vs. SNP-outcome associations demonstrates positive slope for PIK3R3. (D) Funnel plot assesses symmetry of IV estimates; even distribution of SNPs around the IVW line indicates absence of directional pleiotropy and validates MR model assumptions. (E) Leave-one-out sensitivity analysis forest plot confirms that no single SNP disproportionately drives the causal estimate, reinforcing the robustness of the PIK3R3–keloid association.

• 

  FIGURE 4. Multi-layer regulatory networks and potential therapeutic targets for PIK3R3 and ANGPTL5 in keloids. (A) mRNA-miRNA interaction network shows 97 miRNAs targeting PIK3R3 and 3 targeting ANGPTL5, including key pairs PIK3R3-hsa-miR-363-3p and ANGPTL5-hsa-miR-376a-2-5p (green: miRNAs; orange: hub genes). (B) mRNA-TF regulatory network predicts 30 TFs regulating PIK3R3 and 5 regulating ANGPTL5, such as PIK3R3-ZNF263 and ANGPTL5-CHD1 (green: TFs; orange: hub genes). (C) TFBS analysis identifies 8 binding sites in PIK3R3 and 14 in ANGPTL5 promoters; MAFF and IRF1 are common to both, suggesting shared transcriptional regulation (green: TFBS; orange: hub genes). (D) Drug-gene network reveals 47 candidate drugs for PIK3R3 (e.g., apigenin) and 2 for ANGPTL5 (e.g., silica), highlighting potential therapeutic options (green: drugs; orange: hub genes).

• 

  FIGURE 5. GSEA and immune infiltration analyses identify keloid-associated pathways and altered immune cell profiles. (A) GSEA for PIK3R3 reveals enrichment in 261 pathways including the Biocarta Cytokine Pathway, implicating inflammatory signaling in keloids. (B) GSEA for ANGPTL5 shows 337 enriched pathways such as the URS Adipocyte Differentiation DN set, linking it to adipocyte-related processes. (C) Immune infiltration profiling in keloid (n = 18) vs. normal (n = 14) samples highlights activated dendritic cells as highly abundant in both groups. (D) Macrophages and regulatory T cells differ significantly between keloid and normal tissues (p < 0.05), indicating their role in keloid immune dysregulation. (E) Correlation analysis identifies a strong negative association between macrophages and regulatory T cells (cor = −0.69, p = 2.44 × 10−5), suggesting immune network coordination in keloids.

• 

  FIGURE 6. Single-cell RNA sequencing identifies seven major cell populations in keloid and normal scar tissues. (A) Pre-quality control metrics (nFeature, nCount, percent. mt) across samples from GSE163973 (3 keloid, 3 normal scar samples). (B) Post-quality control metrics after filtering cells with 200 < nFeature <4,000, nCount <10000, and percent. mt < 15%. (C) Highly variable gene plot: red dots denote the top 3,000 variable genes used for downstream analysis. (D) PCA plot shows separation of cell populations based on transcriptional variation. (E) Scree plot indicates the first 40 principal components capture the majority of biological variation. The x-axis represents the number of principal components, and the y-axis represents the standard deviation of principal component scores. (F) Unsupervised clustering identifies 18 initial cell clusters (resolution = 0.4). (G) UMAP visualization of seven annotated cell types: vascular endothelial cells, smooth muscle cells, fibroblasts, keratinocytes, lymphatic endothelial cells, Schwann cells, and melanocytes. (H) Cell type abundance varies between keloid and normal individuals, with VEC enriched in keloid and FIB enriched in normal tissue.…

• 

  FIGURE 7. Fibroblasts are identified as key cells expressing hub genes in keloid pathogenesis. (A) Comparison of cell cluster proportions between keloid (red) and normal (blue) samples identifies fibroblasts and vascular endothelial cells as differentially abundant cell populations (p < 0.05). (B) Functional enrichment analysis of differentially abundant cells reveals 1,725 enriched pathways, including amine oxidase reactions. (C) UMAP plots show PIK3R3 distributed in both FIB and VEC, while ANGPTL5 expression is primarily confined to FIB (color intensity reflects expression level). (D) Expression profiles of ANGPTL5 (upper) and PIK3R3 (lower) across annotated cell clusters confirm high expression of both hub genes in fibroblasts. (E) Aggregate expression analysis indicates PIK3R3 is significantly upregulated and ANGPTL5 is downregulated in keloid compared to normal cells across all cell types (p < 0.05). (F) Subtype-specific analysis confirms ANGPTL5 expression is significantly altered in FIB, while PIK3R3 expression differs significantly in VEC between keloid and normal tissues.

• 

  FIGURE 8. Single-cell analysis reveals fibroblast heterogeneity, differentiation trajectories, intercellular communication, and metabolic activity in keloids. (A) UMAP plot of fibroblasts subclustered into three subtypes (FIB#1, FIB#2, FIB#4) from keloid and normal samples. (B) Marker expression bubble plot confirms subtype identity: FIB#1 expresses pro-inflammatory genes, FIB#2 mesenchymal markers, and FIB#4 secretory papillary fibroblast markers. (C) Pseudotime trajectory orders fibroblasts from early (blue) to late (red) differentiation states. (D) Distribution of fibroblast subtypes along pseudotime shows FIB#1 and FIB#2 predominant early, and FIB#4 enriched late. (E) Differentiation state map highlights three distinct developmental stages. (F) Dynamic expression of PIK3R3 and ANGPTL5 declines over pseudotime, with ANGPTL5 higher in early subtypes. (G) Branch-point analysis identifies gene clusters associated with distinct differentiation fates. (H–K) Cell-cell interaction networks show strong fibroblast-Schwann cell communication in keloids, and fibroblast-melanocyte interaction in normal tissue. Bubble plots highlight CD99−CD99 as a key ligand-receptor pair reduced in normal tissue.…

• 

  FIGURE 9. Experimental validation confirms PIK3R3 upregulation in keloids but reveals discordant ANGPTL5 expression in clinical samples. (A) Transcriptomic analysis shows PIK3R3 is significantly upregulated in keloid tissues compared to normal skin in the training dataset (GSE113619, p < 0.05). (B) Transcriptomic analysis shows ANGPTL5 is significantly downregulated in keloid tissues compared to normal skin in the training dataset (GSE113619, p < 0.05). (C) RT-qPCR validation in five paired clinical samples confirms PIK3R3 upregulation in keloid tissues (p < 0.05), consistent with bioinformatic predictions. (D) RT-qPCR analysis shows ANGPTL5 is upregulated in clinical keloid samples (p < 0.05), contrary to transcriptomic predictions, potentially due to limited sample size or microenvironmental variability.