Single-cell RNA-seq and machine learning identify HMGN2 as a lactylation-associated hub gene in heart failure

Figures

• 

  Fig 1. Schematic of the experimental workflow.

• 

  Fig 2. HF Cellular Dynamics Uncovered by scRNA-seq. A. Violin plots of quality metrics (nCount_RNA, nFeature_RNA, pMT, pHB) used to filter cells at empirical thresholds. B. PCA projections before and after Harmony integration demonstrate effective batch-effect correction. C. Seurat-based clustering identifies 34 cell clusters. D. UMAP projection visualizes cluster relationships and sample-specific distributions. E. Annotation using canonical markers defines 9 cell types, including fibroblasts and cardiomyocytes. F. Dot plot validates marker gene expression and statistical significance. G. Stacked bar plot shows proportional shifts: fibroblast expansion and cardiomyocyte loss in HF. H. UMAP heatmap displays marker gene expression with a color gradient.

• 

  Fig 3. Cell-specific lactylation in HF. A. Elevated lactylation scores in SMCs, fibroblasts, and T/NK cells link them to HF progression. B. UMAP visualization identifies fibroblasts as a primary cellular hub of inferred lactylation activity in HF. C. Fibroblasts and SMCs constitute the cluster with high lactylation-related gene signature scores. D-E. Global elevation of lactylation in HF (p < 2.2e-16), specifically in SMCs, fibroblasts, and immune cells. F-G. Spearman correlation identifies 100 lactylation-associated genes. H. 562 genes are upregulated in high lactylation cells. I. 25 genes overlap between lactylation-associated and HF-upregulated sets.

• 

  Fig 4. Machine learning identification of core feature genes. A. LASSO regression selected 16 key genes. B-C. The Boruta algorithm identified 19 critical genes. D-E. Random forest prioritized 10 significant genes. F. XGBoost detected 24 candidate genes. G. Integration of all methods revealed 7 core genes (HMGN2, CNN3, CACYBP, NUCKS1, H2AFZ, VIM, PFN1).

• 

  Fig 5. HMGN2 as a conserved HF diagnostic indicator. A-B. All 7 key genes show significant positive correlations with immune cell infiltration (p ≤ 0.001). HMGN2 correlates strongly with dendritic cells. C. HMGN2, NUCKS1, and VIM are upregulated in HF versus controls. D. In cohort GSE57338 (bulk RNA-seq), HMGN2 achieves an AUC of 0.956 for HF diagnosis. E-F. In validation set GSE76701 (bulk RNA-seq), HMGN2 (AUC = 1.0) and NUCKS1 (AUC = 1.0) are upregulated in HF. G-H. In cohort GSE42955 (bulk RNA-seq), HMGN2 remains upregulated (AUC = 0.883). I. In a TAC-induced mouse HF model, HMGN2 (AUC = 0.8) and VIM (AUC = 0.88) show diagnostic value.. J. HMGN2 is prioritized as the core HF gene based on consistent upregulation and cross-species diagnostic performance.

• 

  Fig 6. HMGN2-mediated regulation of immune and proteostatic pathways in HF via GSEA and TF-miRNA networks. (A–B) HMGN2-high samples enrich immune (MHC I, NF-κB, IL-1) and protein degradation pathways (proteasome, spliceosome, ubiquitin). (C) TF-miRNA network depicts HMGN2, VIM, and NUCKS1 interactions with TFs and miRNAs.

• 

  Fig 7. Single-cell heterogeneity of HMGN2 in HF. A. UMAP shows HMGN2 expression concentrated in fibroblasts and SMCs. B-C. Violin and density plots confirm high HMGN2 in fibroblasts/SMCs, low in mast cells and cardiomyocytes. D-F. Trajectory analysis reveals HMGN2 + fibroblasts and SMCs expand synchronously with divergent expression in HF. G. Altered cellular interactions between HMGN2 + / − cells and others. H. HMGN2 + cells communicate via specific ligand-receptor pairs (e.g., MIF-CD74/CXCR4, PTN-NCL).

• 

  Fig 8. HMGN2-induced lactylation and cardiac dysfunction in cellular/animal HF models. A–C. PE-induced upregulation of ANP, BNP, and β-MHC mRNA in H9C2 cells confirms hypertrophic modeling. D. LDHA elevation indicates increased lactylation activity after PE treatment. E. HMGN2 protein is upregulated under PE stimulation. F–H. TAC operation impairs cardiac function, reducing EF% and FS%. I–K. Histology reveals hypertrophy, fibrosis, and structural disruption in TAC hearts. L. HMGN2 is consistently elevated in TAC mice. Data are presented as mean ± SEM. n = 3 independent biological replicates per group for all in vitro and in vivo experiments unless otherwise stated.