2.1 Data acquisition

In this investigation, human heart scRNA-seq data were obtained from the GSE247468 dataset in the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). This dataset comprised samples from 2 healthy donors and 4 HF. The training dataset was generated from GSE57338 in the GEO database, which contained RNA expression patterns of 313 people with and without HF. For validation, three distinct approaches were integrated: (1) Cardiac tissue sequencing from TAC-induced HF mouse models; (2) The GSE42955 dataset, which included 29 human left ventricular samples (12 with dilated cardiomyopathy, 12 with ischemic cardiomyopathy, and 5 controls); (3) The GSE76701 dataset, which comprised 8 human heart specimens (4 from failing hearts and 4 from non-failing donors). Additionally, 332 LRGs curated from previous research [12] were incorporated into the analysis. Although this LRG set was initially identified in a hepatocellular carcinoma (HCC) context, the fundamental biochemical pathway of lactylation—from lactate conversion to lactyl-CoA and subsequent modification by core writers (e.g., EP300) and erasers (e.g., HDACs)—is highly conserved across various eukaryotic cell types and diseases, including cardiovascular pathologies. Due to the current lack of a well-curated, HF-specific lactylome database, this comprehensive gene set was used to evaluate lactylation-associated transcriptomic signatures.

2.2 ScRNA-seq dataset analysis

Raw scRNA-seq data were systematically analyzed using the Seurat R package [13]. Following the conversion of raw data into Seurat objects, stringent quality control criteria were applied based on data distribution characteristics and empirical thresholds (nFeature_RNA: 200−5000; nCount_RNA: 200−30,000; mitochondrial gene percentage pMT ≤ 30%; hemoglobin gene percentage pHB ≤ 5%). Low-quality cells were filtered out using the subset function. The retained high-quality cellular data underwent LogNormalize standardization, followed by identification of highly variable genes (HVGs). Principal component analysis (PCA) was employed for data dimensionality reduction, feature integration, and further dimension reduction. Automated cell clustering was achieved using the “FindNeighbors” and “FindClusters” algorithms, and cell type annotation was performed based on well-established marker genes documented in previous studies [14].

2.3 Evaluation of inferred lactylation activity

To evaluate the lactylation-related gene set enrichment score at single-cell resolution, this study adapted the methodological framework of SCENIC’s AUCell for transcription factor activity assessment, integrating five well-established computational algorithms. Following standardized protocols, lactylation-related gene set enrichment scores were calculated using four individual methods: AUCell [15], UCell [16], singscore [17], and AddModuleScore [18]. Their sum was defined as a composite score (Score1). Cells were stratified into high- and low-activity groups based on the median lactylation-related gene set enrichment score across the population. Subsequent Spearman correlation analysis identified genes significantly associated with lactylation activity, while differentially expressed genes (DEGs) upregulated in the high-activity group were detected using Seurat’s “FindMarkers” function [19]. The intersecting genes from correlation analysis and DEGs were subjected to downstream functional exploration.

2.4 Enrichment analysis

To comprehensively examine biological pathways enriched in high-lactylation activity clusters, multi-level enrichment analysis was performed on filtered candidate gene sets using the Gene Ontology (GO) and Kyoto Encyclopaedia of Genes and Genomes (KEGG) databases. The “clusterProfiler” R package [20] was employed for functional annotation across three ontological domains: Biological Process (BP), Molecular Function (MF), and Cellular Component (CC), while identifying significantly enriched metabolic and signaling pathways (adjusted p-value <0.05). This integrated analytical approach elucidated the core biological functions of candidate genes and delineated key regulatory networks associated with elevated lactylation activity.

2.5 Machine learning algorithms for feature gene selection

A multi-model machine learning strategy was employed to systematically identify key regulators associated with lactylation activity for in-depth analysis of pre-filtered genes. Four feature selection algorithms were integrated: eXtreme Gradient Boosting (XGBoost) [21], Boruta algorithm [22], Random Forest [23], and Least Absolute Shrinkage and Selection Operator (LASSO) regression [24]. Gene importance rankings were independently evaluated using each algorithm, and consensus hub genes were identified through stringent cross-algorithm concordance criteria. To elucidate the molecular mechanisms underlying HF, boxplot analysis was conducted to compare expression profiles of these genes between HF and control groups, with statistical significance assessed using a three-tiered annotation system (***: p ≤ 0.001; **: 0.001 < p ≤ 0.01; *: 0.01 < p ≤ 0.05).

2.6 Evaluation of feature gene expression and diagnostic potential in validation cohorts

Cross-species validation was conducted using multi-platform datasets to validate the expression patterns and diagnostic potential of the optimal feature genes in HF tissues. For human cardiac samples, transcriptomic data from three independent HF cohorts (GSE57338, GSE42955, GSE76701) were analyzed via the Wilcoxon rank-sum test to assess expression differences of optimal feature genes between HF and control groups. Additionally, cardiac sequencing data from TAC-induced murine HF models were included for cross-species validation. Receiver Operating Characteristic (ROC) curves were generated across these four datasets to evaluate diagnostic ability, with Area Under the Curve (AUC) values measuring predictive efficacy (AUC > 0.7 indicating strong diagnostic capacity) [25].

2.7 GSEA and TF-miRNA

To elucidate the regulatory mechanisms of HMGN2, VIM, and NUCKS1 in HF, TF-miRNA interaction networks were predicted using the KnockTF database and the MultiMiR package [26]. Single-gene GSEA enrichment analysis was performed on HMGN2 through Reactome and KEGG databases [27].

2.8 Trajectory analysis

Cells were dichotomously classified to construct cellular differentiation trajectories based on the presence/absence of optimal feature gene expression. The scaled unique molecular identifier (UMI) count matrix extracted from Seurat subsets was used as input for trajectory inference using the Monocle 2 algorithm with default parameters [28].

2.9 Interactions between intercellular communication

CellChat [29] was employed to analyze intercellular communication dynamics, using the default CellChatDB ligand-receptor database. Cell type-specific interaction modules were identified by systematically detecting overexpressed ligands or receptors in cellular subsets and evaluating their ligand-receptor interaction strengths.

2.10 TAC-induced HF model

This study utilized TAC in 8–10-week-old male C57BL/6 mice (Nantong University Laboratory Animal Center) to minimize the cardiovascular protective effects of estrogen and hormonal fluctuations, randomly assigned to sham or TAC groups. Mice were anesthetized with ketamine (100 mg/kg) and xylazine (2.5 mg/kg) via intraperitoneal injection. A left anterolateral thoracotomy was performed to expose the aortic arch under visual guidance. A 27-gauge needle-guided 6–0 silk suture was ligated around the aortic arch to create TAC, followed by needle withdrawal and wound closure with 4–0 sutures. Sham controls underwent identical surgical procedures without aortic ligation [30]. Postoperative mice were maintained on a temperature-controlled heating pad until full recovery before returning to standard housing. Terminal euthanasia was performed via cervical dislocation under deep anesthesia. All animal experiments were strictly performed in accordance with the guidelines for the care and use of laboratory animals and were approved by the Institutional Animal Care and Use Committee (IACUC) of Nantong University (Approval No. P20251021-005).

2.11 RNA-seq sequencing

For RNA-seq analysis of TAC-modeled mice, myocardial tissues from control and 8-week post-TAC groups were dissected and snap-frozen in liquid nitrogen. Total RNA was isolated using TRIzol reagent (Invitrogen) following the manufacturer’s protocols, with strict adherence to RNA integrity and purity standards. RNA concentration and purity (OD260/280 and OD260/230 ratios) were quantified via NanoDrop spectrophotometry, and integrity was verified by 1.5% agarose gel electrophoresis. Ribosomal RNA was depleted using the Ribo-Zero Gold rRNA Removal Kit (Illumina) to enrich mRNA. First-strand cDNA synthesis used random hexamer primers, followed by second-strand synthesis with buffer, dNTPs, RNase H, and DNA Polymerase I. Double-stranded cDNA was purified via magnetic bead-based systems and eluted in EB buffer. Purified cDNA underwent end repair, A-tailing, and Illumina adapter ligation. UNG enzyme treatment degraded uracil-containing second strands. Size selection (200–500 bp) via gel electrophoresis preceded PCR amplification and sequencing on an Illumina Novaseq 6000. The entire experimental workflow was rigorously conducted under stringent quality control measures to ensure high-quality sequencing data with demonstrated reliability.

2.12 Evaluation of lactylation activity in HF model

Following the establishment of a PE-induced HF model, H9C2 cardiomyocytes were harvested for subsequent analysis. Total RNA was extracted from cell pellets using TRIzol reagent according to the manufacturer’s protocol, with RNA concentration and purity verified spectrophotometrically. cDNA was synthesized from 1 μg of total RNA using a PrimeScript RT reagent kit with gDNA Eraser. After sealing the plate with the prepared samples, amplification was performed on a real-time PCR detection system. The standard cycling conditions included an initial denaturation at 95°C for 30 s, followed by 40 cycles of denaturation at 95°C for 5 s and annealing/extension at 60°C for 30 s. First, the mRNA expression levels of established cardiac hypertrophy markers atrial natriuretic peptide (ANP), brain natriuretic peptide (BNP), and β-myosin heavy chain (β-MHC) were quantified to validate successful model establishment. The model group exhibited significantly elevated expression of these markers, confirming successful induction of the HF phenotype. Subsequently, the expression level of lactate dehydrogenase A (LDHA) [31], a key enzyme regulating lactate production and lactylation modification, was assessed. Using TB Green Premix Ex Taq II, all reactions were performed in triplicate. The relative expression of target genes, normalized to GAPDH, was analyzed using the 2 − ΔΔCt method.

The primer sequences used were as follows:

1. LDHA forward, 5’- TAATGAAGGACTTGGCGGATGA-3’ and reverse, 5’- ACCAGCTTGGAGTTCGCAGTTA-3’;
2. GAPDH forward, 5’- CCTCGTCCCGTAGACAAAATG-3’ and reverse, 5’- TGAGGTCAATGAAGGGGTCGT-3’.
3. β-MHC forward, 5’- CCCAGAAACAAGTGAAGAGCCT-3’ and reverse, 5’- GTTCCACGATGGCGATGTTC-3’.
4. ANP forward, 5’- CTTCTTCCTCGTCTTGGCCTTT-3’ and reverse, 5’- TCCAGGTGGTCTAGCAGGTTCT-3’.
5. BNP forward, 5’- CTTTATCTGTCACCGCTGGGAG-3’ and reverse, 5’- ACAACTTCAGTGCGTTACAGCC-3’.

2.13 Echocardiography

Mice were anesthetized with 2% isoflurane. Chest hair was depilated, and ultrasound coupling gel was applied to the anterior thoracic region. Cardiac parameters (IVSd, LVIDd, LVPWd, IVSs, LVIDs, LVPWs, ejection fraction (EF), and fractional shortening (FS)) were acquired in the parasternal long-axis view using a Vevo 2100 high-resolution imaging system equipped with an FMS-250 probe (13–24 MHz), with values averaged over 4–5 consecutive cardiac cycles [32].

2.14 Immunostaining and histological analysis

For comprehensive histological analysis of cardiac tissues, samples were prepared using both paraffin-embedded and frozen-section methodologies. Sections of 5 μm thickness obtained from paraffin-embedded tissues were stained with hematoxylin and eosin (H&E) to assess general morphology and with Masson’s trichrome to evaluate collagen content and fibrosis. To specifically examine cardiomyocyte hypertrophy, frozen sections were permeabilized with 0.1% Triton X-100, blocked in 5% bovine serum albumin (BSA), and incubated with FITC-conjugated wheat germ agglutinin (WGA) (1:200) for 30 minutes at 37 °C. Imaging was performed using an epifluorescence microscope.

2.15 Western blotting

Protein lysates were prepared from both cardiac tissues of TAC-induced mice and PE-stimulated H9c2 cardiomyocytes. For the in vitro model, H9c2 cells were cultured for 48 h, synchronized by serum starvation for 24 h in high-glucose basal medium supplemented with 0.1% 5-bromodeoxyuridine (Brdu) (0.1 mmol/L), and subsequently treated for 48 h as follows: The control group received the synchronization medium (100 µl). The model group was treated with the same medium containing 100 µM PE. Proteins were separated by SDS-PAGE, transferred to a membrane, and probed with specific primary antibodies. To ensure equal loading, membranes were incubated with an anti-α-tubulin antibody (Servicebio, China). The band intensity of the protein of interest, HMGN2 (ABCAM, UK), was quantified using image analysis software and normalized to α-tubulin. The results are expressed as the relative integrated density value compared to the control group. All primary antibody incubations were performed overnight at 4°C [33].

2.16 Statistical analysis

Statistical analyses were performed using R 4.1.3 (R Foundation) and GraphPad Prism 9.0 (GraphPad Software, USA). Continuous data are expressed as mean ± standard error of the mean (SEM). Group differences for continuous variables were evaluated with either the Wilcoxon test or t-test for two-group comparisons, while one-way ANOVA was applied for multi-group analyses. Intervariable associations were examined via Spearman’s correlation coefficients. All significance thresholds used two-tailed tests with p < 0.05 denoting statistical significance. For all in vitro and in vivo experiments, data are presented as the mean ± standard error of the mean (SEM) of independent biological replicates. Specifically, n = 5 per group for bulk RNA-seq; n = 3 per group for echocardiography, histological staining, qPCR, and Western blotting experiments.

2.17 Ethical statement

All animal experiments were strictly performed in accordance with the guidelines for the care and use of laboratory animals and were approved by the Institutional Animal Care and Use Committee (IACUC) of Nantong University (Approval No. P20251021-005).

3.1 Schematic of the experimental workflow

A workflow diagram for this study is shown in Fig 1.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Schematic of the experimental workflow.

https://doi.org/10.1371/journal.pone.0357971.g001

3.2 ScRNA-seq uncovers cellular heterogeneity and dynamic remodeling in HF

To evaluate the quality of scRNA-seq data and select high-quality cells for HF analysis, mitochondrial (pMT) and hemoglobin (pHB) gene percentages were calculated using the PercentageFeatureSet function. Then, four quality metrics (nCount_RNA [RNA counts], nFeature_RNA [gene features], pMT, and pHB) were visualized through VlnPlot-generated violin plots. Cell filtering was performed using the subset function with empirically determined thresholds: nFeature_RNA (200–5000), nCount_RNA (200–30000), pMT ≤ 30%, and pHB ≤ 5%, based on data distribution patterns (Fig 2A). Data normalization was performed using LogNormalize with a scale factor of 10,000. 2,000 HVGs were selected for scaling, and PCA was carried out for preliminary data exploration. Initial PCA revealed significant batch effects across sample groups. The integration of the Harmony algorithm, with orig.ident designated as the batch variable, effectively removed technical variations, as evidenced by the post-correction two-dimensional (2D) embeddings (harmony_1/harmony_2 coordinates) that confirmed effective batch effect mitigation (Fig 2B). To investigate the cellular heterogeneity and sample distribution patterns associated with HF, cell clustering was conducted using the Seurat algorithm, resulting in the identification of 34 distinct cell clusters (numbered 0–33) (Fig 2C). Subsequent UMAP visualization provided a clear depiction of both cluster relationships and cellular distributions across different samples (HF1–4, Control1–2) (Fig 2D). Through manual annotation based on canonical marker genes, nine cell types were identified: B cells, cardiomyocytes, endothelial cells, fibroblasts, macrophages, mast cells, neurons, SMC, and T/NK cells (Fig 2E). The accuracy of this annotation was further substantiated by dot plot visualization of marker genes, which showcased both the statistical significance and expression levels of cell type-specific genes (Fig 2F). To investigate the dynamic changes in cellular composition during HF progression, stacked bar plots were generated to compare cell type proportions between the Control and HF groups via scRNA-seq analysis. The analytical results demonstrated a notable expansion of fibroblasts (pink-shaded segment) and a depletion of cardiomyocytes (teal-colored segment) in the HF cohort compared to the Control cohort, with macrophages and mast cells showing distinct redistribution patterns (Fig 2G). Furthermore, gene expression visualization experiments were carried out using single-cell transcriptomic data to analyze the subpopulation-specific expression patterns of marker genes. Cellular distributions were visualized through UMAP dimensionality reduction, and representative marker genes for each major cell type were displayed in heatmaps with a blue (low)-to green/yellow (high) expression gradient (Fig 2H).

Download:

• PNG
  larger image
• TIFF
  original image

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.

https://doi.org/10.1371/journal.pone.0357971.g002

3.3 Analyzing cell-type-specific lactylation signatures and their functional implications in HF by scRNA-seq

To decipher the lactylation-associated gene expression patterns across different cardiac cell types and their pathological relevance to HF, we conducted a systematic evaluation of nine cell populations (including B cells, macrophages, and fibroblasts) using five computational algorithms (AUCell, UCell, singscore, Addmodulescore, and Score1 [summation of first four]) based on single-cell transcriptomic data. The results indicated elevated lactylation signature scores in SMCs, fibroblasts, and T/NK cells, suggesting their involvement in HF progression through lactylation pathways. In contrast, neurons showed limited association, providing novel evidence for cell-type-specific metabolic regulation in HF (Fig 3A). To investigate the distributional differences in lactylation-related gene set enrichment scores between the HF and Control groups, we visualized single-cell transcriptomes using UMAP-based dimensionality reduction. This visualization confirmed elevated lactylation activity in fibroblasts as a key pathological feature (Fig 3B). Using predefined lactylation-associated gene sets, we conducted pathway activity scoring (Score1) on single-cell data, followed by UMAP dimensionality reduction for cellular distribution visualization. The results demonstrated that the cluster with high lactylation activity was predominantly composed of fibroblasts and SMCs (Fig 3C). To decipher the disparities in lactylation-associated pathways across cell types during HF, we compared the Score1 lactylation signature scores between the HF and Control groups using the Wilcoxon rank-sum test, both globally and across individual cell types. The results demonstrated a significantly elevated median Score1 in HF versus Control (p < 2.2e-16). These findings indicate that lactylation activity is selectively enhanced in HF, particularly in SMCs, fibroblasts, and immune-related cells (B and mast cells). This suggests that lactylation may contribute to the pathogenesis of HF by modulating vascular remodeling, fibrosis, and inflammatory responses (Fig 3D and 3E). To decipher the regulatory features of lactylation pathway-associated genes in HF single-cell data, we conducted Spearman correlation analysis to quantify associations between gene expression and pathway scores (Score1), followed by functional classification based on correlation coefficients (cor) and gene ranking. Correlation analysis identified 100 genes significantly associated with lactylation modifications (Fig 3F and 3G). Screening of DEGs in HF single-cell data based on lactylation pathway scores revealed 562 upregulated genes between high- and low-expression groups (Fig 3H). Venn diagram analysis was used to assess the overlap between the upregulated gene set (562 genes) and the lactylation-related pathway-associated gene set (Top 100, ranked by correlation). The analysis revealed that 25 genes were common to both sets (Fig 3I). To investigate the biological functions of HF-associated lactylation genes, GO and KEGG enrichment analyses were performed on the 25 overlapping genes. KEGG analysis significantly enriched these genes in nucleocytoplasmic transport, cytoskeleton organization in muscle cells, and amyotrophic lateral sclerosis pathways (S1 Fig). GO analysis revealed that CCs were substantially correlated with focal adhesion, MFs were associated with cadherin binding, and BPs primarily involved ribosome biogenesis and ribosomal subunit export (S1 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

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.

https://doi.org/10.1371/journal.pone.0357971.g003

3.4 Machine learning-based identification of core feature genes for HF

Four machine learning algorithms were used to identify key feature genes among the 25 candidate genes. LASSO regression identified 16 key genes with the lowest binomial deviance (Fig 4A). The Boruta algorithm identified 19 critical genes after 500 iterations (Fig 4B and 4C). The random forest algorithm prioritized 10 significant genes (Fig 4D and 4E). XGBoost analysis detected 24 candidate genes (Fig 4F). Seven core feature genes (HMGN2, CNN3, CACYBP, NUCKS1, H2AFZ, VIM, and PFN1) were identified by combining the results of all four algorithms (Fig 4G).

Download:

• PNG
  larger image
• TIFF
  original image

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).

https://doi.org/10.1371/journal.pone.0357971.g004

3.5 HMGN2 as an optimal feature gene revealed by immune interactions and cross-species validation

To explore the crosstalk between key genes and the immune microenvironment, ssGSEA-based immune infiltration analysis was performed. The correlations between seven key genes (HMGN2, CNN3, CACYBP, NUCKS1, H2AFZ, VIM, PFN1) and 28 immune cell types were visualized using bubble plots and heatmaps. The bubble plot revealed significant positive correlations between all seven genes and specific immune subclusters. The heatmap further confirmed the statistical significance of these associations, with all marked as *** (p ≤ 0.001). For instance, HMGN2 showed the strongest correlation with plasmacytoid dendritic cells and activated dendritic cells, while CNN3 was highly associated with Type 1 T helper cells. VIM presented a strong correlation with the majority of immune cells (Fig 5A and 5B). To investigate the molecular mechanisms underlying HF and identify potential therapeutic targets, we compared the expression levels of the seven key genes between HF and control groups using box plot analysis, with statistical significance assessed using significance markers. The results showed significant upregulation of HMGN2, NUCKS1, and VIM in the HF group. Based on their expression levels and functional relevance, these three genes were prioritized as optimal hub genes, with subsequent analyses focusing on their regulatory networks and disease associations (Fig 5C). To validate these findings, we evaluated the three candidate biomarkers at the bulk RNA-seq level. In the GSE57338 training cohort, HMGN2 achieved an AUC of 0.956 in HF patients, demonstrating exceptional sensitivity and specificity for distinguishing HF from controls. In contrast, NUCKS1 (AUC = 0.694) and VIM (AUC = 0.583) showed lower diagnostic value (Fig 5D). In the independent validation set GSE76701, differential expression analysis (box plots) and ROC curve analysis confirmed significant upregulation of HMGN2 in HF (AUC = 1.0), consistent with previous findings. NUCKS1 showed marginal upregulation (AUC = 1.0), while VIM displayed no statistical significance (AUC = 0.562) (Fig 5E and 5F). Analysis of the GSE42955 validation cohort using box plots and ROC curves indicated upregulation of HMGN2 (AUC = 0.883), whereas NUCKS1 (AUC = 0.617) and VIM (AUC = 0.583) performed near random chance levels (Fig 5G and 5H). We established a TAC-induced murine model to identify cross-species conserved core genes in HF with clinical translatability and performed RNA sequencing on cardiac tissues for final validation. In murine HF models, HMGN2 and VIM showed AUCs of 0.8 and 0.88, respectively, while NUCKS1 achieved an AUC of 0.56 (Fig 5I and 5J). HMGN2 was ultimately prioritized as the core gene based on its consistent upregulation across datasets, cross-dataset stability in human cohorts, and conserved diagnostic performance in cross-species validations.

Download:

• PNG
  larger image
• TIFF
  original image

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.

https://doi.org/10.1371/journal.pone.0357971.g005

3.6 Results of single-gene GSEA and TF-miRNA network

To systematically investigate the multi-dimensional regulatory mechanisms of HMGN2 as a core gene in HF, we conducted single-gene GSEA using the Reactome and KEGG databases. GSEA revealed significant enrichment of HMGN2-high samples in Reactome pathways, including MHC class I antigen presentation, FcεRI-mediated NF-KB activation, and IL-1 family signaling. These findings suggest that HMGN2 may be involved in HF progression through immune regulation and pro-inflammatory responses. Additionally, KEGG analysis demonstrated prominent enrichment of proteasome, spliceosome, and ubiquitin-mediated protein degradation pathways, indicating that HMGN2 contributes to HF progression through dysregulation of protein homeostasis and defects in transcriptional control (Fig 6A and 6B). These findings suggest that HMGN2 serves as a potential regulatory hub in HF pathology, potentially operating through dual mechanisms: modulation of immune-inflammatory signaling and disruption of protein metabolic homeostasis. To dissect the regulatory mechanisms of HMGN2, VIM, and NUCKS1 in HF, we predicted and constructed their interaction networks with transcription factors (TFs) and miRNAs using KnockTF and MultiMiR (Fig 6C).

Download:

• PNG
  larger image
• TIFF
  original image

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.

https://doi.org/10.1371/journal.pone.0357971.g006

3.7 Results of trajectory analysis and cell interactions

To investigate single-cell expression heterogeneity of HMGN2 in the HF microenvironment, its distribution across cell types was analyzed using integrated UMAP dimensionality reduction, violin plots, and stacked density plots. UMAP clustering revealed that HMGN2 expression was predominantly concentrated within fibroblast and SMC populations (Fig 7A). In alignment, violin plots and ridge density analyses validated high levels of HMGN2 expression in fibroblasts/SMCs, in contrast to minimal expression in mast cells and cardiomyocytes (Fig 7B and 7C). Fibroblasts and SMCs were stratified into HMGN2 + / − Fibroblasts and HMGN2 + / − SMCs. Trajectory analysis was performed using Monocle2 to investigate transcriptional heterogeneity in fibroblasts and SMCs. During quasitemporal ordering, HMGN2 + fibroblasts and HMGN2 + SMCs exhibited synchronized proportional expansion. Pseudotemporal analysis revealed distinct relative expression patterns of HMGN2, with divergent trends along pseudotime between Control and HF groups (Fig 7D–7F). Cellular communication analysis revealed differences in both the number and strength of interactions between HMGN2 + /HMGN2 − SMCs, fibroblasts, and other cell types (Fig 7G). Compared to HMGN2 − fibroblasts, HMGN2 + fibroblasts exhibited stronger outgoing and incoming interaction strengths, suggesting their heightened participation in cellular communication. Similarly, HMGN2 + SMCs displayed higher activity in both efferent and afferent interactions compared to HMGN2 − SMCs (S2 Fig). Ligand-receptor interaction analysis further elucidated the connectivity between HMGN2 + /HMGN2 − SMCs, fibroblasts, and other cell types. HMGN2 + SMCs interacted with other cells through receptor-ligand pairs such as MIF- (CD74 + CXCR4), PTN-NCL, and PGF-PEGFR1. Similarly, HMGN2 + fibroblasts communicated via PTN-NCL, MIF- (CD74 + CD44), and CXCL12-CXCR4 pairs (Fig 7H). Heatmap analysis highlighted elevated communication probabilities in HMGN2 + fibroblasts/SMCs, alongside complex signaling networks involving diverse cell types. HMGN2 + fibroblasts exhibited heightened activity in pathways, including IGF, PROS, MK, and MIF, whereas incoming signals such as Periostin and TGF-β were predominantly expressed in HMGN2 − fibroblasts. Additionally, HMGN2 + SMCs displayed enhanced activity in ANGPT, CSF, and VEGF pathways, with incoming signals like PROS and PDGF preferentially expressed in HMGN2 + SMCs. These signaling molecules play crucial roles in regulating cellular behaviors, participating in tissue repair, ECM remodeling, and immune modulation (S3 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

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).

https://doi.org/10.1371/journal.pone.0357971.g007

3.8 Experimental validation of optimal feature genes in H9C2 cells and TAC mouse model

It is worth noting that while our scRNA-seq analysis primarily highlighted the involvement of fibroblasts and smooth muscle cells, primary cardiac fibroblasts are prone to spontaneous activation and phenotypic drift in vitro. Therefore, as an initial proof-of-concept, we utilized the stable H9c2 cardiomyocyte cell line for our preliminary in vitro validation.To elucidate the role of HMGN2 in HF pathogenesis, H9C2 cardiomyocytes were treated with PE (100 µM) and 0.1% Brdu in high-glucose basal medium for 48 h to simulate HF conditions, while control cells received Brdu alone. Total RNA was isolated, reverse-transcribed, and analyzed via quantitative PCR (qPCR). PE treatment significantly increased the mRNA expression of ANP, BNP, and β-MHC (Fig 8A–8C), confirming successful HF modeling. Moreover, LDHA mRNA was markedly upregulated, indicating elevated lactylation activity (Fig 8D). Western blotting demonstrated enhanced HMGN2 protein levels in PE-stimulated cells compared to controls (Fig 8E). To further assess the functional relevance of HMGN2 in vivo, a TAC-induced HF mouse model was utilized. Echocardiographic analyses revealed substantial impairment of cardiac contractility (Fig 8F), characterized by reduced EF (EF%) and FS (FS%) (Fig 8G and 8H). H&E, WGA, and Masson staining validated the presence of hypertrophy, fibrosis, and architectural disruption (Fig 8I–8K). Strikingly, HMGN2 expression was broadly elevated in TAC hearts (Fig 8L). Together, these results demonstrate that HMGN2 expression is upregulated concurrently with markers of hypertrophy and potential lactate metabolism (LDHA), supporting its positive association with HF progression, although further investigation is required to establish causal relationships.

Download:

• PNG
  larger image
• TIFF
  original image

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.

https://doi.org/10.1371/journal.pone.0357971.g008