Clinical application and immune infiltration landscape of stemness‐related genes in heart failure

Abstract

Background

Heart failure (HF) is the leading cause of morbidity and mortality worldwide. Stemness refers to the self‐renewal and differentiation ability of cells. However, little is known about the heart's stemness properties. Thus, the current study aims to identify putative stemness‐related biomarkers to construct a viable prediction model of HF and characterize the immune infiltration features of HF.

Methods

HF datasets from the Gene Expression Omnibus (GEO) database were adopted as the training and validation cohorts while stemness‐related genes were obtained from GeneCards and previously published papers. Feature selection was performed using two machine learning algorithms. Nomogram models were then constructed to predict HF risk based on the selected key genes. Moreover, the biological functions of the key genes were evaluated using Gene Ontology (GO) and Kyoto Encyclopedia of Genes Genomes (KEGG) pathway analyses, and gene set variation analysis (GSVA) and enrichment analysis (GSEA) were performed between the high‐ and low‐risk groups. The immune infiltration landscape in HF was investigated, and the interaction network of key genes was analysed to predict potential targets and molecular mechanisms.

Results

Seven key genes, namely SMOC2, LUM, FNDC1, SCUBE2, CD163, BLM and S1PR3, were included in the proposed nomogram. This nomogram showed good predictive performance for HF diagnosis in the training and validation sets. GO and KEGG analyses revealed that the key genes were primarily associated with ageing, inflammatory processes and DNA oxidation. GSEA and GSVA identified various inflammatory and immune signalling pathways that were enriched between the high‐ and low‐risk groups. The infiltration of 15 immune cell subsets suggests that adaptive immunity has an important role in HF.

Conclusions

Our study identified a clinically significant stemness‐related signature for predicting HF risk, with the potential to improve early disease diagnosis, optimize risk stratification and provide new strategies for treating patients with HF.

Introduction

Heart failure (HF) is a complex clinical disease and a leading cause of hospitalization and death worldwide. During the course of heart disease or ageing, damaged cardiomyocytes are unable to renew and regenerate, triggering cardiac remodelling and HF. ¹ The current treatment for HF is suboptimal; symptomatic treatment via devices and medications primarily slows disease progression rather than targets the underlying myocardial pathophysiology. Meanwhile, recent studies have confirmed that foetal and neonatal cardiomyocytes can proliferate and repair injured tissues; in contrast, mature mammalian hearts do not compensate for the enormous loss of viable cardiomyocytes due to the low cellular turnover rate after injury. ² The injured myocardial tissue undergoes repair in an immune inflammatory response with massive fibroblast infiltration, resulting in fibrous scar formation and, eventually, HF. Hence, understanding the biological basis of HF is crucial for developing more effective treatments. Furthermore, the advent of genetic testing holds promise for achieving early HF diagnosis by identifying potential diagnostic markers and providing insights into the mechanisms of HF development.

The heart is the first organ to form during embryonic development. ³ Interestingly, the possible involvement of cardiac stem cells in HF is at the centre of intense debate. Accumulating evidence indicates that pathological conditions negatively impact the capacity of cardiac progenitors for cardiomyogenic differentiation, which may explain the progressive exhaustion of local cardiac progenitors and provide insights into the stemness of the heart in healthy and pathological circumstances. ⁴ Hence, one of the main objectives of cardiovascular research is to investigate alterations in cardiomyocyte plasticity to promote cardiac regeneration and repair. Currently, cardiac regeneration is primarily achieved using tissue engineering techniques, stem cell treatment, direct cell reprogramming and natural cardiomyocyte renewal ability to repair damaged hearts. ⁵ Over the past decade, research has revealed that adult heart progenitor cells (hCPCs) express stem cell‐related markers, such as TP‐binding cassette superfamily G member 2 (ABCG2), c‐Kit (CD117), the LIM‐homeodomain transcription factor Islet‐1 (ISL1) and stem cell antigen‐1 (SCA1); these cells express stemness transcription factors (TFs) in an undifferentiated state. ⁶ , ⁷ For patients with myocardial infarction (MI) and HF, regenerative therapies based on resident hCPCs offer a promising alternative to conventional interventions. Additionally, induced pluripotent stem cells and resident cardiac stem cells (CSCs) have been used for heart regeneration and repair in clinical trials. ⁸ However, the effect of stemness‐related genes (SRGs) on the onset and progression of HF remains unclear.

The heart is immunologically active, with inflammation and other immune responses playing vital roles in HF progression and pathogenesis. Indeed, multiple inflammatory pathways contribute to cardiac inflammation, including the tumour necrosis factor (TNF)/nuclear factor kappa‐B pathway and pattern recognition receptors expressed by macrophages. ⁹ The activation of pro‐inflammatory T lymphocytes and accumulation of pro‐inflammatory macrophages in the heart is associated with a poor prognosis in HF. ¹⁰ However, how variations in SRG expression affect cardiac immunoreactivity and inflammation as HF progresses remains unknown.

In this study, we combine machine learning and bioinformatics techniques to identify important genes linked to stemness changes in HF, advancing the current understanding of cardiac cell stemness (Figure 1). Additionally, we construct and verify a stemness‐related genetic diagnostic nomogram to estimate the risk of HF and explore the effects of the relevant stemness genes on HF onset and progression. Additional analyses of the relationship between SRGs and immune cell infiltration will yield novel insights for the early diagnosis and management of HF.

Figure 1.

Workflow employed in this study. DEGs, differentially expressed genes; SRGs, stemness‐related genes; SRDEGs, stemness‐related differentially expressed genes.

Materials and methods

Data acquisition and pre‐processing

Gene expression profile data (GSE141910, ¹¹ GSE116250 ¹² and GSE46224 ¹³ datasets) was downloaded from the Gene Expression Omnibus (GEO) database ¹⁴ using the R Package ‘GEOquery’ ¹⁵ (Table 1). In the GSE46224 dataset, only RNA‐Seq data sampled from patients with HF before left ventricular assist device implantation and non‐failing donors were included in the subsequent analyses. A total of 24 samples were included: 16 patients with HF and 8 controls. Based on the annotation data in the GEO Platform (GPL), the probes were converted into gene symbols. The GSE116250 and GSE46224 datasets were then merged, and batch effects were removed using the R package ‘sva’ ¹⁶ to get an integrated GEO dataset (GEO‐Combined), including 66 with HF and 22 controls. The GSE141910 and GEO‐combined datasets were then standardized and normalized using the R package ‘limma’ ¹⁷ and included in the subsequent analyses as the training and validation sets, respectively.

Table 1.

Heart failure dataset information list.

	GSE141910	GSE116250	GSE46224
Platform	GPL16791	GPL16791	GPL11154
Species	Homo sapiens	Homo sapiens	Homo sapiens
Tissue	Left ventricular free‐wall tissue	Left ventricle tissue	Left ventricle tissue
HFs	200	50	16
controls	166	14	8

Note: HFs, samples in the HF group; controls, samples in the control group.

Abbreviation: HF, heart failure.

The GeneCards database ¹⁸ (http://www.genecards.org/) is a comprehensive and authoritative compendium of annotative information for human genes. A total of 2303 SRGs were obtained by searching GeneCards using ‘Stemness’ as the keyword and retaining only the protein‐coding genes. In addition, 4419 SRGs were identified in published literature from PubMed. ¹⁹ Finally, 5861 SRGs were identified after merging and deduplication (Table S1).

Differential gene expression analysis

Gene expression analysis was carried out on the HF datasets GSE141910 and GEO‐combined separately using the ‘limma’ ¹⁷ package to obtain differentially expressed genes (DEGs) between different groups (HF/control). The criteria used to assess significance were P.adj < 0.05 and |logFC| > 1. Subsequently, stemness‐related DEGs (SRDEGs) were identified by taking the intersection of the DEGs from the two HF datasets and SRGs. The results were processed using the R package ‘ggplot2’ to generate volcano plots and the R package ‘pheatmap’ to generate heat maps.

Screening key genes and least absolute shrinkage and selection operator risk model construction

Using the concept of integrated learning, the random forest (RF) ²⁰ algorithm combines several decision trees. To select the optimal genes, the ‘RandomForest’ ²¹ package was used to construct models based on SRDEG expression in the GSE141910 dataset using the following parameters: set.seed = 234 and ntree = 1000. A higher node purity indicated the presence of fewer impurities (smaller Gini coefficient). Subsequently, the results of the specific analyses were filtered using the criterion IncNodePurity > 2.

Next, least absolute shrinkage and selection operator (LASSO) regression analysis ²² was performed based on the SRDEGs included in the RF using the R package ‘glmnet’ ²³ with set.seed = 500 as the parameter to select key genes for subsequent analyses. To avoid overfitting, a run period of 200 was selected. The LASSO risk score was calculated as follows:

$$\mathit{\text{RiskScore}}=\sum _{\mathit{i}}\mathit{\text{Coefficient}}\left({\mathit{\text{gene}}}_{\mathrm{i}}\right)*\mathit{\text{mRNA Expression}}\left({\mathit{\text{gene}}}_{\mathrm{i}}\right)$$

To obtain the risk score, the expression of key genes from the GEO‐combined and GSE141910 datasets was plugged into the formula, and the samples were divided into high‐ and low‐risk groups using the median risk score of the HF group in the GSE141910 and GEO‐combined datasets.

Functional enrichment analysis

Gene ontology (GO) ²⁴ analysis is commonly employed to perform extensive functional enrichment investigations related to biological processes (BPs), cellular components (CC) and molecular functions (MF). Similarly, the Kyoto Encyclopedia of Genes and Genomes (KEGG) ²⁵ is a popular database for storing data on diseases, medications, biological pathways and genomes. The R package ‘clusterProfiler’ ²⁶ was used with the statistically significant screening criteria of P < 0.05 and false discovery rate (FDR,q.value) <0.05 for GO and KEGG analyses.

Diagnostic performance and validation of the risk signature

Nomograms were established ²⁷ to predict the diagnostic efficacy of key genes for HF using the R package ‘rms’. To evaluate the clinical usefulness of the predictive models, a decision curve analysis (DCA) was performed ²⁸ using the R package ‘ggDCA’. The discrimination accuracy of the prediction model was described using receiver operating characteristic (ROC) analysis, ²⁹ yielding the ROC curve and the area under the ROC curve (AUC) to evaluate the predictive ability of the models.

Validation of the expression differences and functional similarity analysis of the key genes

The Mann–Whitney U‐test (Wilcoxon rank‐sum test) was used to compare the expression of key genes between different subgroups (HF/control). Next, the ‘pROC’ package plotted the ROC curves of key genes across the HF and control groups, and the AUC was calculated to evaluate the diagnostic efficacy of key gene expression in patients with HF.

Semantic comparison of GO annotations provides a quantitative method for calculating the similarity between genes and genomes, representing an important foundation for many bioinformatics analysis tools. Herein, the GO semantic similarity which is defined as the geometric mean of their semantic similarities in MF, CC and MF aspect of GO of key genes was calculated using the ‘GOSemSim’ package. ³⁰ The findings were visualized by ‘ggplot’ package.

Gene set enrichment analysis and gene set variation analysis

Gene set enrichment analysis (GSEA) ³¹ and gene set variation analysis (GSVA) ³² were run in the GSE141910 dataset to compare the high‐ and low‐risk groups. GSEA was carried out using the ‘clusterProfiler’ package with the following parameters: minSize = 10, maxSize = 500 and nperm = 10 000. GSEA was performed using the signalling pathway dataset ‘c2.cp.all.v2022.1.Hs.symbols.gmt’ from the Molecular Signatures Database (MSigDB) as the background. ³³ The screening threshold was P < 0. 05 and q.adj < 0.05. GSVA was performed using the gene set ‘h.all.v7.4. symbols.gmt’ as the background and P.adj < 0.05 as the substantial enrichment threshold.

Immune infiltration analysis based on the high‐ and low‐risk groups

The single‐sample gene set enrichment analysis (ssGSEA) ³⁴ algorithm was employed to compute the infiltration abundance of 28 immune cell subsets in the high‐ and low‐risk groups of HF patient samples. The Mann–Whitney U‐test was then used to examine the degree of infiltration of the immune cell subsets. Pearson's correlation was computed for the various immune cells. Additionally, the relationships between immune cells and key genes in high‐ and low‐risk groups were evaluated by combining the gene expression matrix of the GSE141910 dataset.

Interaction network analysis of key genes

The ENCORI (Starbase V3.0) ³⁵ was used to predict miRNAs that interact with important genes. mRNA–miRNA interaction pairings were screened based on a PancancerNum > 2 and visualized using Cytoscape. The CHIPBase database ³⁶ (version 3.0) was used to identify the TFs associated with key genes. The screening criteria included adding the number of samples detected (upstream and downstream) and setting a threshold >4. Furthermore, the ENCORI database predicted the RNA‐binding proteins (RBP) that interact with important genes using ClusterNum > 3 as the screening criterion. A reference count of >1 was used as a screening criterion to identify possible medications or small molecule chemicals that interact with key genes in the Comparative Toxicogenomics Database (CTD). ³⁷

Spatial protein structure of key genes

Alphafold (https://www.alphafold.ebi.ac.uk/) was the first to propose that protein structures can be predicted with atomic precision based on computational methods in the absence of homologous templates and that 98.5% of the predicted structures cover known human proteins and homologous proteins of other organisms. ³⁸ Thus, in this study, the protein structures of key genes were predicted using the AlphaFold website.

Statistical analysis

Statistical analyses were performed using R version 4.2.2. Continuous data were expressed as mean ± standard error of the mean (SEM). The Kruskal–Wallis test was used to compare three or more groups, while the Wilcoxon rank‐sum test compared two groups. Results were subjected to Spearman's correlation analysis to assess relationships between molecules unless otherwise specified. All statistical P values were two‐tailed, with P < 0.05 regarded as statistically significant.

Results

A total of 266 HF samples and 188 control samples from three databases were used in our study (see Table 1), and each one was verified by combining databases and intersection mutual authentication to increase the accuracy and consistency of the study.

Data pre‐processing

The batch effects were removed from two HF datasets (GSE116250 and GSE46224) before performing integration analyses to produce the integrated GEO dataset (GEO‐combined). The datasets before and after batch‐effect removal were compared using box and principal component analysis plots (Figure 2). The outcomes demonstrated that the batch effects were substantially eliminated in the GEO‐combined dataset.

Figure 2.

Data pre‐processing. Boxplot of the integrated GEO dataset (GEO‐combined) before (A) and after (B) batch‐effect removal. Principal component analysis plots of the GEO‐combined dataset before (C) and after (D) eliminating the batch effects. GEO, Gene Expression Omnibus.

Identification of SRDEGs

Volcano plots were used to display the results of the difference analyses. A total of 1074 DEGs were identified in GSE141910, of which 742 were up‐regulated, and 332 were down‐regulated (Figure 3A). Additionally, 172 DEGs were obtained from the GEO‐combined dataset, of which 105 were up‐regulated and 67 were down‐regulated (Figure 3B).

Figure 3.

Identification of stemness‐related differentially expressed genes (SRDEGs). DEGs in the GSE141910 (A) and GEO‐combined (B) datasets are displayed using a volcano plot. (C) Venn diagram showing the SRDEGs. Hierarchical cluster heat map of the SRDEGs between different groups (HF/Control) in the GSE141910 (D) and GEO‐combined (E) datasets. DEGs, differentially expressed genes; GEO, Gene Expression Omnibus; SRDEGs, stemness‐related differentially expressed genes.

The DEGs were then intersected with the SRGs to obtain 52 SRDEGs (Table S2); the results were plotted in a Venn diagram (Figure 3C). Next, the differential expression of 52 SRDEGs was examined between groups (HF/control) in the GSE141910 and GEO‐combined datasets; the expression levels were displayed using a heat map (Figure 3D,E), identifying 52 SRDEGs in the GSE141910 and GEO‐combined datasets across subgroups (HF/control).

Screening key genes and model construction

To construct an SRDEG‐based signature to predict HF occurrence, the expressions of the 52 SRDEGs in the GSE141910 dataset were analysed using the RF algorithm to select feature variables, and 8 SRDEGs were screened (Figure 4A,B). Based on these eight SRDEGs, LASSO regression analyses were performed to further identify seven SRDEGs (SMOC2, LUM, FNDC1, SCUBE2, CD163, BLM and S1PR3) as the key genes for model construction (Figure 4C,D). The LASSO regression analysis of the seven key genes was performed using a forest plot (Figure 4E). The RiskSore was calculated as follows:

$$\mathit{\text{RiskScore}}=\sum _{\mathrm{i}}\mathit{\text{Coefficient}}\left({\mathit{\text{gene}}}_{\mathrm{i}}\right)*\mathit{\text{mRNA Expression}}\left({\mathit{\text{gene}}}_{\mathrm{i}}\right)$$

Figure 4.

Two algorithms were used for feature selection. (A) Model training error plot for the RF algorithm. (B) RF model showing the SRDEGs (top 30 in descending order of IncNodePurity). (C) LASSO regression analysis of the seven key genes. (D) Identification of the best penalization coefficient in the LASSO regression, lambda (λ). (E) Forest plot of the LASSO regression model of the seven key genes. LASSO, least absolute shrinkage and selection operator; RF, random forest.

Samples from the GSE141910 and GEO‐combined datasets were classified into the high‐ and low‐risk subgroups based on the median scores of the HF groups.

Diagnostic performance of the key genes and validation of the risk model

A diagnostic nomogram model of HF in the GSE141910 dataset was constructed based on the seven key genes (Figure 5A). The results showed that the expression of FNDC1 and S1PR3 contributed more significantly to HF diagnosis than the other genes. In contrast, the expression of SCUBE2 contributed significantly less to HF diagnosis than the other genes. The net benefit of the nomogram was significantly positive (Figure 5C), indicating that it was valid for clinical applications. Additionally, the ROC curves indicated that this nomogram had an excellent discriminatory ability for HF (Figure 5E, AUC = 0.947).

Figure 5.

Nomogram model predicting HF risk based on seven key genes. Nomograms to estimate the risk of HF for the training set GSE141910 (A) and validation set GEO‐combined (B). Decision curve analysis curves of the nomogram in the GSE141910 (C) and GEO‐combined (D) datasets. ROC curves in the GSE141910 (E) and GEO‐Combined (F) datasets. HF, Heart failure; GEO, Gene Expression Omnibus.

Similarly, the nomogram based on the seven key genes in the GEO‐combined dataset (validation set) indicated that the expression of FNDC1 contributed more significantly to HF diagnosis than the other genes whereas CD163 and BLM expression contributed significantly less to HF diagnosis (Figure 5B). According to the DCA (Figure 5D), the net benefit of the nomogram in determining HF risk was significantly positive, indicating that it was valid for clinical applications. Similarly, the risk prediction nomogram model in the GEO‐combined validation dataset presented good discriminatory power for HF (Figure 5F, AUC = 0.951).

Differential expression validation and functional similarity analysis of the key genes

According to the differential expression analysis results, the expression of all seven key genes was highly significant across various subgroups (HF/control) in the GSE141910 (Figure 6A) and GEO‐combined (Figure 6B) datasets. Four genes were up‐regulated (SMOC2, LUM, FNDC1 and SCUBE2), and three were down‐regulated (CD163, BLM and S1PR3) in the HF samples compared to the controls.

Figure 6.

Differential expression validation and functional similarity analysis of the key genes. Analysis of the differential expression of key genes between various subgroups (HF/control) in the GSE141910 (A) and GEO‐combined (B) datasets. (C–F) ROC curves of the key genes in the GSE141910 dataset. (G–J) Receiver operating characteristic curves of the key genes in the GEO‐combined dataset. (K) Functional similarity analysis of the key genes. *P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant. HF, Heart failure; GEO, Gene Expression Omnibus.

The ROC curves of the seven key genes in the GEO‐combined and GSE141910 datasets were used to determine the diagnostic role of each gene in each dataset (Figure 6C–J). In the GSE141910 dataset, the expression levels of SMOC2 (AUC = 0.988, Figure 6C), LUM (AUC = 0.952, Figure 6C), FNDC1 (AUC = 0.988, Figure 6D), SCUBE2 (AUC = 0.967, Figure 6D), CD163 (AUC = 0.971, Figure 6E), BLM (AUC = 0.959, Figure 6E) and S1PR3 (AUC = 0.947, Figure 6F) exhibited high accuracy for the diagnosis of both subgroups (HF/control). Similarly, in the GEO‐combined dataset, the expression levels of SMOC2 (AUC = 0.956, Figure 6G), LUM (AUC = 0.966, Figure 6G), FNDC1 (AUC = 0.965, Figure 6H), SCUBE2 (AUC = 0.948, Figure 6H), BLM (AUC = 0.942, Figure 6I) and S1PR3 (AUC = 0.951, Figure 6J) were highly accurate for the diagnosis of both subgroups (HF/control), while CD163 (AUC = 0.815, Figure 6I) expression was moderately accurate for the diagnosis of both subgroups (HF/control).

The functional similarities between the seven key genes were arranged in descending order and visualized in Figure 6K, revealing that SCUBE2 had the highest functional similarity with the other key genes.

Functional enrichment analysis of key genes

The relationships between the BPs, CCs, MFs and biological pathways of the seven key genes (SMOC2, LUM, FNDC1, SCUBE2, CD163, BLM and S1PR3) were further explored using GO and KEGG enrichment analyses (Tables S3 and S4). According to the GO enrichment results (Figure 7), these genes were primarily enriched in BPs, including telomeric loop disassembly, extracellular matrix (ECM) organization, regulation of TGF‐β1 (TGFβ1) production, negative regulation of endothelial cell (EC) differentiation and cellular response to hydroxyurea. The enriched CCs included fibrillar collagen trimers, banded collagen fibrils, lateral elements, collagen trimer complexes and collagen‐containing ECMs. Finally, the enriched MFs associated with the key genes included oxidized DNA binding, annealing activity, 3′–5′ DNA helicase activity and bioactive lipid receptor activity.

Figure 7.

Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis of key genes. Bar plot graph (A), bubble graph (B) and network diagrams (C–F) of the GO and KEGG enrichment analysis results.

KEGG enrichment analysis revealed that the key genes were significantly associated with homologous recombination, Fanconi anaemia and the Hedgehog signalling pathway (Figure 8).

Figure 8.

Pathway maps of the KEGG enrichment analysis. (A) Homologous recombination (hsa03440) pathway. (B) Fanconi anaemia pathway (hsa03460). (C) Hedgehog signalling pathway (hsa04340).

GSEA and GSVA of the high‐ and low‐risk groups

GSEA and GSVA were performed to more accurately explore the differences in the pathways and BPs activated by the differential expression of the key genes. The GSEA results identified oxidative phosphorylation, influenza infection and selenoamino acid metabolism as the significantly enriched pathways (Figure 9). In summary, the significant enrichment of oxidative phosphorylation pathways between the high‐ and low‐risk groups suggests that cardiac metabolism plays an important role in HF development.

Figure 9.

Gene set enrichment analysis (GSEA) of all genes between the high‐ and low‐risk groups in the GSE141910 dataset. (A) Four main biological features in the GSEA results. (B) Oxidative phosphorylation, (C) influenza infection, (D) oxidative phosphorylation and (E) selenoamino acid metabolism.

The GSVA results (Figure 10) detected 20 significantly enriched pathways: UV response up, UV response down, reactive oxygen species, protein secretion, pancreatic beta cells, P53, oxidative phosphorylation, NOTCH signalling myogenesis, MYC targets V2, MYC targets V1, mitotic spindle, KRAS signalling up, IL6–JAK–STAT3 signalling, Hedgehog signalling, G2M checkpoint, DNA repair, cholesterol homeostasis, bile acid metabolism and androgen response. Hence, in addition to cardiac metabolism‐related pathways, cell proliferation, differentiation and apoptosis are also associated with HF risk.

Figure 10.

Gene set variation analysis (GSVA) of all genes between high‐ and low‐risk groups in GSE14190. GSVA results presented in a hierarchical cluster heatmap (A) and boxplot (B). *P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant.

ssGSEA of immune infiltration based on the high‐/low‐risk groups in the GSE141910 dataset

ssGSEA was used to examine the differences in immune cell infiltration between the high‐ and low‐risk groups in the GSE141910 dataset. Significant variations in the myocardial infiltration of 15 immune cell types were found between the high‐ and low‐risk groups (Figure 11A). Specifically, CD8⁺ T cells, natural killer (NK) CD56^bright cells, NK CD56^dim cells, central memory CD4⁺ T cells, myeloid‐derived suppressor cells, monocytes and plasmacytoid dendritic cells were more abundant in the high‐risk group, whereas CD4⁺ T cells, effector memory CD4⁺ T cells, immature B cells, memory B cells, NK cells, NK⁺ T cells, type 17 T helper cells (Th17) and type 2 T helper cells (Th2) were more abundant in the low‐risk group. Further analysis of these immune cell types revealed a positive correlation between infiltration abundance and the 15 immune cell types (Figure 11B). The strongest synergistic effect was observed between NK T cells and CD4⁺ T cells, memory B cells and CD4⁺ T cells, Th2 and memory B cells and effector memory CD4⁺ T cells and Th2 cells. In contrast, there were strong competitive effects among NK CD56^dim cells, memory B cells, Th2 cells and effector memory CD4⁺ T cells.

Figure 11.

Immune cell infiltration landscape between the high‐ and low‐ risk groups in GSE141910. (A) Immune cell distribution. (B) Immune cell correlation heatmap. (C) Correlations between immune cells and key genes. *P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant.

The relationships between the 15 immune cells and seven key genes (SMOC2, LUM, FNDC1, SCUBE2, CD163, BLM and S1PR3) were also evaluated in patient samples from the GSE141910 dataset. There was a substantially positive association between the abundance of the 15 immune cells and the expression of most key genes (Figure 11C). In addition, a negative correlation was evident between NK CD56^dim cells and LUM, and monocytes and S1PR3.

Key gene interaction network analysis

The interactions between the seven key genes (SMOC2, LUM, FNDC1, SCUBE2, CD163, BLM and S1PR3) and other molecules were further explored. The mRNA–miRNA interaction network comprised four mRNAs (BLM, LUM, S1PR3 and SMOC2) and 77 miRNAs, constituting 86 mRNA–miRNA interaction pairs (Figure 12A, Table S5). The mRNA–TF interaction network comprised 59 TFs, constituting 104 mRNA–TF interaction pairs (Figure 12B, Table S6). Additionally, 5 mRNAs (BLM, FNDC1, S1PR3, SCUBE2 and SMOC2) and 60 RBP molecules constituted 72 mRNA–RBP interaction pairs (Figure 12C, Table S7). In the final analysis, our mRNA–drug interaction network included 14 drug molecules, constituting 36 mRNA–drug interaction pairs (Figure 12D, Table S8).

Figure 12.

Interaction network analysis of the key genes. (A) mRNA–miRNA interaction network. (B) mRNA–TF interaction network. (C) mRNA–RBP interaction network. (D) mRNA–drug interaction network.

Spatial protein structures of the key genes

The protein structures of the seven key genes (SMOC2, LUM, FNDC1, SCUBE2, CD163, BLM and S1PR3) were analysed using the AlphaFold website (Figure 13A–13F). SCUBE2 and CD163 have the same protein structure (Figure 13D).

Figure 13.

Spatial protein structures of the key genes.

Discussion

HF is a significant healthcare problem worldwide and is the terminal stage of many cardiovascular diseases. Cellular senescence is defined as a stable, growth‐arrested end‐state for cells. Similar to other age‐related diseases, cardiac senescence is characterized by a loss of tissue homeostasis and a decline in organ function, along with a progressive accumulation of senescent stem cells with reduced proliferative and regenerative potential, which occurs in the pathophysiological changes of HF associated with myocardial remodelling and cardiac dysfunction. ³⁹ The determination of cardiomyogenic lineages is tightly controlled by epigenetic mechanisms, consistent with previous studies on the cellular plasticity of mammalian cardiomyocytes. Hence, exploring the epigenomic changes in cardiomyocytes will provide important insights regarding cardiac stemness, enabling the development of therapies to induce heart regeneration and prevent HF.

In this study, seven candidate SRGs were screened for optimal diagnostic potential: SMOC2, LUM, FNDC1, SCUBE2, CD163, BLM and S1PR3 (Figure 14). Based on these key genes, a new diagnostic nomogram was constructed for HF risk assessment, exhibiting high accuracy and robustness, implying strong relevance with HF that may improve HF diagnosis. GO results showed that these key genes were primarily related to the ECM‐associated CCs, telomeric loop disassembly and TGFβ1 production‐associated BPs, DNA damage and repair response and G‐quadruplex DNA binding‐associated MFs. Moreover, homologous recombination, Fanconi anaemia and the Hedgehog signalling pathways were enriched.

Figure 14.

Cartoon representation of the major findings of the study. SRGs, stemness‐related genes.

SMOC2 (SPARC‐related modular calcium binding 2) is a matricellular protein that accumulates in the ECM during ageing, leading to impaired muscle stem cell function and regeneration. ⁴⁰ The proteins encoded by SMOC2 are involved in wound healing, matrix assembly and remodelling, which are associated with inflammatory damage and fibrosis in HF. ⁴¹ , ⁴² Additionally, SMOC2 knockdown partially abrogates cardiac functional impairment and fibrosis in vivo, ⁴³ suppressing the development of HF by altering TGF‐β1/Smad3 signalling‐mediated autophagy. ⁴⁴ These findings suggest that SMOC2 is a potential therapeutic target for treating HF.

Sphingosine‐1‐phosphate (S1P) is a circulating bioactive lipid metabolite involved in various cellular responses mediated by the S1P receptor (S1PR). Recently, S1P has received increasing attention owing to its cytoprotective effects. In the cardiovascular system, S1P protects the heart from ischaemia/reperfusion (I/R) damage by increasing cardiomyocyte survival and decreasing the size of infarct areas in isolated hearts. ⁴⁵ S1P accumulates in cardiac tissues in HF and is associated with cardiac remodelling in patients with HF. ⁴⁶ Moreover, plasma S1P levels correlate with left ventricular function and the clinical symptoms of HF. ⁴⁷ Although the precise mechanisms underlying the intricate alterations in S1P in HF are unknown, S1P may represent a promising candidate for further exploring HF diagnosis and treatment.

LUM (Lumican) binds to fibrillar collagen and regulates collagen fibrogenesis, implying that it plays a vital role in cardiac remodelling after LV pressure overload. LUM mRNA and protein levels are elevated in the hearts of patients and mice with end‐stage dilated cardiomyopathy (DCM) and HF whereas LUM deficiency attenuates collagen cross‐linking and contributes to decreased survival, increased systolic dysfunction and left ventricular dilatation, all of which are key processes in HF progression. ⁴⁸ Additionally, LUM deficiency exacerbates isoproterenol‐induced HF by upregulating the TGF‐β and MMP signalling pathways, implying that LUM is a novel therapeutic target for cardiac fibrosis. ⁴⁹ LUM is also vital for inflammation, which may have important implications in the HF process. Thus, further studies on the effects of LUM may provide new therapeutic targets or biomarkers for patients with HF.

CD163 is a hallmark of alternatively activated macrophages; it is cleaved from the surface of macrophages and becomes soluble CD163 (sCD163). ⁵⁰ Recent cardiovascular research has suggested that CD163 and sCD163 represent potential biomarkers of inflammatory activation in various chronic inflammatory conditions affecting the cardiovascular system, including atherosclerosis, myocarditis, atrial fibrillation, MI and HF. ⁵¹ , ⁵² , ⁵³ , ⁵⁴ Additionally, CD163 plasma levels become elevated before adverse clinical events and are an independent predictor of clinical outcomes in HF populations; however, its precise role and potential prognostic properties in HF are unclear. Thus, mechanistic studies are required to provide deeper insights into the biological roles of CD163. ⁵⁵

SCUBE2 (Signal Peptide, CUB Domain and EGF Like Domain Containing 2) is highly expressed in vascular ECs and highly vascularized tissues. ⁵⁶ SCUBE2 might participate in the development of atherosclerotic plaques and serve as a potential therapeutic target for coronary artery disease. Previous studies have shown that FNDC1 (Fibronectin Type III Domain Containing 1) is up‐regulated in cardiomyocytes in response to ischaemia or hypoxia and may cause cardiomyocyte death under inflammatory or hypoxic conditions. ⁵⁷ , ⁵⁸ In addition, FNDC1 is expressed in ECs and participates in vascular endothelial growth factor‐induced EC angiogenesis. ⁵⁹ , ⁶⁰ Bloom syndrome helicase (BLM) is a member of the RecQ family that participates in the DNA damage response (DDR) and DNA metabolism with important functions in maintaining genome integrity. The molecular effects of BLM on DNA replication, DDR and telomere maintenance have been described; however, the underlying mechanisms remain largely unknown. ⁶¹ , ⁶² , ⁶³ In summary, few studies have described the significance of these three key genes in HF, warranting additional studies to characterize their roles and potential significance in HF.

In the current study, GSEA and GSVA were performed for all genes in the high‐ and low‐risk groups. In particular, the oxidative phosphorylation and reactive oxygen species pathways were widely enriched in patients at high risk of HF, consistent with a recent study showing that cardiac metabolism and oxidative stress contribute to the onset and development of HF. ⁶⁴ More importantly, cell proliferation, differentiation, senescence and apoptosis‐related pathways were also enriched in the high‐risk group, these included the Notch, the P53, MYC targets V1 and V2, DNA repair and myogenesis pathways. Among these, the Notch signalling pathway has been confirmed to be involved in cardiac development, guiding cardiomyocytes and vascular differentiation and is a crucial therapeutic target for promoting angiogenesis and myocardial repair after MI. ⁶⁵ The Notch pathway is intact in the adult CSC niche, which could influence the proliferation and differentiation of progenitor cells, highlighting the reparative potential of cardiac progenitor cells in treating heart disorders. ⁶⁶ , ⁶⁷ Furthermore, Notch receptors and ligands have been found in experimental and clinical HF cardiac tissues and appear to be associated with regulating ECM remodelling and inflammation. ⁶⁸ Therefore, we hypothesized that the Notch signalling pathway is involved in HF pathogenesis.

p53 is a tumour suppressor protein with a crucial role in DNA repair and apoptosis. In the cardiovascular system, cardiomyocyte apoptosis, which is mediated in part through the p53 signalling pathway, plays an important role in pathological remodelling after MI and in HF progression. ⁶⁹ , ⁷⁰ Additionally, p53 is a key regulator of inflammatory processes in age‐related diseases and can trigger cellular dysfunction in most HF cases by activating inflammation. ⁷¹ , ⁷² Multiple studies have revealed that Myc expression is markedly low in the adult myocardium under normal physiological conditions but is rapidly elevated in response to nearly all hypertrophic stimuli, such as hypertrophic cardiomyopathy and HF. ⁷³ , ⁷⁴ Notably, Myc inhibition is emerging as an attractive paradigm in the fight against cardiomyopathy and HF. ⁷⁵ Taken together, these pathways are strongly linked to HF development and may serve as a target for HF diagnosis and treatment.

Recently, the involvement of the immune system in cardiac remodelling has become evident. It is generally accepted that the heart is immunoactive and that persistent inflammation and immune activation contribute to HF pathogenesis. Based on ssGSEA analysis, patients with HF displayed high infiltration of various lymphocytes, indicating the role of adaptive immunity in HF. Both innate and adaptive immunity function in an orderly manner, with the specificity of the immune response depending on the initial stimulus that triggered cardiac inflammation. Generally, neutrophils are the first to be recruited, accompanied by monocytes/macrophages, which secrete various pro‐inflammatory cytokines, such as tumour necrosis factor‐alpha (TNF‐α), interferon‐gamma (IFN‐γ), interleukin (IL)‐6 and IL‐1, which trigger lymphocyte‐mediated adaptive immunity. During this process, the adaptive immune system makes an important contribution to the chronic phase of cardiac remodelling through the reactivation and persistence of the inflammatory cascade. ⁷⁶

Th1, Th2, T‐regulatory (Treg) cells and Th17 are subsets of CD4⁺ T cells. Multiple animal models have demonstrated that the balance of these immune cells is critical to the pathophysiology of HF. ⁷⁷ , ⁷⁸ Indeed, atherosclerosis‐induced MI is a major cause of HF. Current studies suggest that Th1 and Th17 cells contribute to atherosclerosis whereas Th2 and Tregs are protective. The restoration of CD4⁺ T cells in immunocompromised animals significantly increases lesion size, similar to the enhanced T cell recruitment observed in atherosclerotic plaques. ⁷⁹ In addition, CD8⁺ T lymphocytes are recruited and activated in ischaemic myocardial tissue, triggering cardiomyocyte death and unfavourable cardiac remodelling. ⁸⁰ , ⁸¹

However, adult lymphocyte‐deficient mice lack myocardial regenerative capacity despite a causal relationship between adult T cells and the loss of neonatal myocardial regenerative potential. ⁸² This aligns with the results of another study showing that the specific ablation of CD4⁺ T cells restores myocardial regenerative capacity in P8 mice but not in adult mice. ⁸³ These findings highlight the complexity of the interplay between cardiac regeneration potential and T‐cell competence development.

Conclusions

This study demonstrates the potential diagnostic utility of SRGs in HF. Moreover, the stemness‐related biomarker‐based nomogram provides a unique diagnostic and prognostic tool for HF. These findings have the potential to improve risk stratification and contribute to future research on the molecular mechanisms, diagnosis and treatment of HF. However, this study has certain limitations. First, the datasets were obtained from the GEO database, and various confounding factors such as age, sex, clinical complications and pharmacological treatment may have influenced the results. Moreover, further experimental data are required to explore the functions and molecular mechanisms of the key genes identified in this study. Finally, to confirm the clinical feasibility of our results, a large prospective investigation is needed. Additional studies are also needed to demonstrate whether genetic susceptibility to HF can be identified from a stemness perspective.

Conflict of interest statement

The authors declare no competing interests.

Funding

This work was supported by the Natural Science Foundation of Gansu Province (No: 22JR5RA665).

Supporting information

Table S1 Stemness related genes list.

Table S2 SRDEGs list.

Table S3. GO enrichment Analysis results of key genes.

Table S4. KEGG enrichment Analysis results of key genes.

Table S5. mRNA‐miRNA interaction network nodes.

Table S6. mRNA‐TF interaction network nodes.

Table S7. mRNA‐RBP interaction network nodes.

Table S8. mRNA‐drugs interaction network nodes.

Yan, W. , Li, Y. , Wang, G. , Huang, Y. , and Xie, P. (2025) Clinical application and immune infiltration landscape of stemness‐related genes in heart failure. ESC Heart Failure, 12: 250–270. 10.1002/ehf2.15055.

Data availability statement

The datasets generated and/or analysed during the current study are available in Gene Expression Omnibus DataSets (https://www.ncbi.nlm.nih.gov/gds). All data generated or analysed during this study are included in this published article and its supplementary information files.