Bridging metabolism and immuno-inflammation: a novel framework to characterize dilated cardiomyopathy subtype

Abstract

Background

The heterogeneous subtypes in dilated cardiomyopathy (DCM) are poorly characterized, thus posing challenges to risk stratification. This study aimed to establish a DCM subtype framework based on metabolic and immunoinflammatory factors.

Methods

DCM subtypes were identified using unsupervised clustering based on the expression patterns of metabolism-related genes in the left ventricular myocardium of 89 DCM patients. By comparing metabolic pathways, clinical characteristics, immune cell infiltration, inflammatory responses, and immunotherapy efficacy between the subtypes, key metabolic genes were identified through correlation analysis and validated at both bulk and single-cell levels. The alterations in gene expression were verified using the DCM mouse model. Molecular docking was performed to assess the binding affinity between the target protein and potential therapeutic small molecules.

Results

Two subtypes were identified; subtypes 1 and 2 were characterized by increased amino acid metabolism and decreased glucose and energy-related metabolisms, respectively. Subtype 2 displayed worse left ventricular structure and function, higher levels of immune and inflammatory activity, and a more favorable response to immunotherapy. The integrative analysis identified DHRS7C as a key regulator of glucose/energy metabolism; its expression was inversely correlated with left ventricular impairment. The DCM mice showed downregulated DHRS7C expression, which positively correlated with cardiac dysfunction. Additionally, molecular docking identified 17beta-estradiol as a potential therapeutic agent targeting DHRS7C.

Conclusions

This study suggested two heterogeneous DCM subtypes with different metabolic and immunoinflammatory profiles. Furthermore, DHRS7C was inversely correlated with DCM indices and could be targeted by 17beta-estradiol.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-07165-z.

Introduction

Dilated cardiomyopathy (DCM) is triggered by a complex interplay of intrinsic genetic susceptibility and/or various external environmental factors; most of them progress silently and asymptomatically over an extended period, eventually causing ventricular dilation and impaired contractile function [1, 2]. The current pharmacological and non-pharmacological treatments for heart failure have greatly improved the prognosis and progression of DCM; however, these therapies generally lack specificity and targeted action. This results in varying degrees of benefit across different patients; some of whom still have a 10-years mortality rate exceeding 40% [3]. Moreover, the DCM subtypes should also be precisely characterized, which will allow for risk stratification and the development of personalized therapies tailored to specific patient subgroups, thus maximizing treatment efficacy.

Advances in various “omics” technologies, such as transcriptomics and metabolomics, have led to the uncovering of the complex and multifactorial nature of the DCM onset and progression; in particular, these technologies have highlighted the increasingly significant roles of metabolic and immunoinflammatory factors in DCM [4–6]. It has been reported that DCM patients have a shift in the metabolic substrate preference from fatty acids to glucose, accompanied by the accumulation of reactive oxygen species (ROS) and mitochondrial dysfunction [7]. Furthermore, emerging evidence highlights immunoinflammatory mechanisms as equally pivotal pathophysiological regulatory factors, offering novel avenues for targeted therapeutic development [2]. Nevertheless, the inherent spatiotemporal dynamics of immune regulatory networks as well as the complexity of their multilayered molecular interactions pose significant challenges to the clinical translation of precision therapies. These challenges are further compounded by the intricate cross-talks between immune and metabolic pathways, which create a bidirectional regulatory loop, further complicating mechanistic dissection and therapeutic intervention [2, 8]. Therefore, incorporating the metabolic and immunoinflammatory factors might aid in the characterization of DCM subtypes.

The current study attempted to establish robust molecular subtypes based on gene expression patterns in the myocardium of DCM patients. Moreover, the study also explored the heterogeneity of DCM from multiple perspectives, such as metabolic characteristics, the intensity of immunoinflammatory response activity, clinical features, and immunotherapy responses. Based on the molecular signatures, the key genes associated with the characteristics of these subtypes were further investigated and verified in the DCM mouse model. Overall, the classification of DCM subtypes based on metabolism and immuno-inflammation provides a novel basis for the risk stratification of DCM.

Materials and methods

Data sources

A total of 2,752 metabolism-related genes were obtained from previous studies, which were cross-referenced with the metabolic pathway profiles from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (Supplementary Table S1) [9]. Three datasets (GSE29819 [10], GSE21610 [11], and GSE17800 [12]) were downloaded from the Gene Expression Omnibus (GEO) database and integrated using the “ComBat” function from the R software package “sva”, resulting in a meta dataset containing 89 DCM samples and 22 control samples. The GSE17800 and GSE19303 [13] datasets recorded the detailed clinical information of the same patient cohort; the patients received immunoadsorption with subsequent immunoglobulin substitution at baseline and follow-up. The clinical information of the two datasets included gender, age, body mass index (BMI), left ventricular ejection fraction (LVEF), left ventricular internal diameter at diastole (LVIDd), and cardiac immune cells (CD3⁺ T lymphocytes and/or CD68⁺ macrophages). The GSE19303 dataset included 33 DCM patients, who underwent immunoadsorption therapy. This dataset al.so provided complete transcriptomics and clinical information before and after treatment; this information was used for the evaluation of subsequent clinical indicators. The GSE135055 [14] dataset served as a validation set to assess the stability of clinical features between different subtypes. Furthermore, a single-cell dataset GSE121893 [15], containing 2 healthy controls and 4 DCM patients, was used for the subsequent validation and assessment of the key genes. For more detailed information about these datasets, please refer to the original studies and the appendices (Supplementary Table S2).

Unsupervised cluster analysis

The R software package “ConsensuClusterPlus” [16] was used to perform two rounds of unsupervised clustering on the DCM patients to identify the heterogeneous groups. The first clustering was performed based on the expression patterns of 2,752 metabolism-related genes, dividing the patients into clusters 1 and 2. The second clustering was conducted based on the 18 differentially expressed metabolism-related genes between the two clusters, resulting in subtypes 1 and 2, which served as the basis for subsequent analysis. The rationality of the clustering results was assessed comprehensively based on four indicators: consensus matrix, cumulative distribution function (CDF) plots, delta area plots, and cluster consensus analysis. The heterogeneity of sample clustering and the differences in metabolism-associated gene expression patterns between clusters were displayed separately using principal component analysis (PCA) and heatmaps.

Differential expression analysis

Differentially expressed genes (DEGs, absolute fold change > 2, adjusted P-value < 0.05) were identified using the R software package “limma”.

Gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA)

GSEA was conducted using the R software package “clusterProfiler” to evaluate coordinated regulation of gene ontology (GO) biological processes [17]. Significance thresholds were set at |normalized enrichment score (NES)| >1 with adjusted P-value < 0.05. The top enriched terms (the ten most highly upregulated and downregulated pathways per subtype) were visualized using the R software package “ggplot2”.

We selected 84 gene sets related to metabolic pathways from the KEGG database and used the “GSVA” package to assess the variability of these gene sets across all samples in an unsupervised manner [18]. In order to explain the phenotypic heterogeneity, the metabolic DEGs were compared between the DCM subtypes and control samples, with the adjusted P-value < 0.05.

Gene feature classifier

The extreme gradient boosting (XGBoost) model is a boosting-based gradient tree model that excels in handling large datasets and complex models, effectively preventing overfitting and enhancing generalization ability [19]. The study first constructed a sparse matrix based on 17 metabolic DEGs between DCM subtypes as classification features. The R package “caret” was used to optimize the key parameters, such as the number of iterations (nrounds), learning rate (eta), maximum tree depth (max_depth), minimum child weight (min_child_weight), and split point threshold (gamma). The optimal parameter combination was determined using a grid search. Additionally, a 5-fold adaptive resampling was combined for model training. The classification of 15 patients with heart functions graded III to IV from the GSE135055 dataset was predicted using the trained XGBoost model, followed by an assessment of the differences in ejection fraction between the predicted subtypes.

Immunoinflammatory landscapes

The ESTIMATE deconvolution method was used to analyze the differences in stromal and immune scores between the DCM subtypes [20]. The gene set related to the inflammatory response (HALLMARK_INFLAMMATORY_RESPONSE) was obtained from the Molecular Signatures Database. The single-sample GSEA (ssGSEA) method was used to quantify the inflammatory response scores for each patient. In addition, the infiltration scores of 28 immune cells in each DCM sample were calculated using ssGSEA. Spearman correlation analysis was used to explore correlations between the key genes and 28 immune cells.

Enrichment analysis

In order to better understand the biological functions of the key gene DHRS7C, the genes that were significantly and positively correlated with DHRS7C in the transcriptome of subtype 2 (correlation > 0.7 and false discovery rate [FDR] < 0.05) were selected. GO and KEGG enrichment analyses were performed on these genes.

Single-cell RNA-seq analysis

The GSE121893 single-cell RNA-seq (scRNA-seq) dataset was processed through a standardized “Seurat” workflow, implementing stringent quality control (genes expressed in ≥ 5 cells, 500–7,500 features/cell, and mitochondrial gene ratio < 15%) [21]. Batch effects among the 6 samples were removed using the R software package “harmony”. The top ten highly specific genes expressed in each cell cluster were identified using the R software package “genesorteR”, followed by annotating the cell identity of the clusters in combination with existing literature [15, 22]. The metabolic activity of DHRS7C-positive and negative cardiomyocytes derived from DCM samples was compared using the R software package “scMetabolism” [23]. Based on classical ligand-receptor pairs in the scRNA-seq data, the R software package “CellChat” was used to infer communication between cardiomyocytes and immune cells [24].

DCM mouse model

Male BALB/c mice (6–8 weeks old, weighing about 20 g) were purchased from the Hubei Province Experimental Animal Research Center of China and housed in the specific pathogen-free barrier facility in the Experimental Animal Center, Tongji Medical College, Huazhong University of Science and Technology. The mice were randomly divided into the control and DCM groups. Cardiac myosin heavy chain α (MyHC-α) peptide (AC-RSLKLMATLFSTYASADR-OH) was emulsified with complete Freund’s adjuvant (CFA, Sigma-Aldrich, USA) to obtain a 1-mg/mL working solution. The control group mice were administered with only the CFA. On days 0, 7, and 28, the DCM group mice were subcutaneously injected with 100-µL working solution at the axillary and inguinal regions [25]. All the mice were euthanized on day 45, and their serum samples and heart tissues were harvested for subsequent experiments.

Echocardiography

Echocardiographic examinations were performed by an experienced investigator using the VisualSonics Vevo 2100 system in a blinded manner. The mice were anesthetized via inhalation of 3% isoflurane for induction, followed by maintaining at 1.0–1.5% isoflurane to keep them under stable anesthesia. The cardiac functional parameters, including LVEF, fractional shortening (FS), and left ventricular internal diameter at systole/diastole (LVIDs/LVIDd), were assessed in the parasternal long-axis view of the left ventricle using M-mode echocardiography. All the parameters were measured three times for each mouse, and their average was calculated to ensure data reproducibility.

Enzyme-linked immunosorbent assay (ELISA)

The serum levels of brain natriuretic peptide (BNP), cardiac troponin T (cTnT) and 17β-estradiol were quantified using their respective commercially available ELISA kits (BNP: KE1275; cTnT: KE1753, Immunoway, USA; 17β-estradiol: E-OSEL-M0008, Elabscience, China) following the manufacturer’s protocols. Circulating anti-MyHC autoantibody levels were identified using ELISA as described previously [26].

Histological analysis

Cardiac tissue specimens were fixed overnight in 4% paraformaldehyde and subsequently embedded in paraffin. For inflammatory assessment, the cardiac tissue sections were stained with hematoxylin and eosin (HE) and systematically examined under ×200 magnification. Five randomly selected fields per sample were scored according to percentages of inflammatory cell infiltration area using the following criteria: 0 (no infiltration), 1 (< 25%), 2 (25–50%), 3 (50–75%), and 4 (>75%) [27]. The final cardiac pathological score was identified by calculating the mean score from all the fields. The tissue sections were stained with Masson’s trichrome to evaluate myocardial collagen deposition. The quantitative analysis of the fibrotic areas was performed by calculating the percentage of blue-stained collagen fibers relative to the total tissue area using ImageJ software. Immunohistochemical staining was performed using an anti-DHRS7C antibody (26020-1-AP, Proteintech, China) to evaluate the protein expression levels, followed by a semi-quantitative analysis of staining intensity using ImageJ software. For the localization assessment, immunofluorescence staining of DHRS7C was conducted on parallel sections, with cardiomyocytes counterstained using an anti-cTnI antibody (21652-1-AP, Proteintech, China) and nuclei visualized by DAPI counterstaining.

Quantitative real-time polymerase chain reaction (qRT-PCR)

The total RNA was extracted from cardiac tissue using TRIzol reagent and reverse-transcribed into cDNA using the PrimeScript RT kit. The qRT-PCR reactions were performed using the SYBR Premix Ex Taq kit (TaKaRa Biotechnology, Japan) on an ABI PRISM 7900 Sequence Detector System (Applied Biosystems, USA). The PCR reaction conditions were as follows: denaturation at 95 °C for 2 min, followed by 40 cycles of denaturation at 95 °C for 5 s and annealing and amplification at 60 °C for 30 s. The expression levels of DHRS7C were normalized to that of β-actin, and the relative expression levels were calculated using the 2^–ΔΔCt method. The primer sequences were as follows: Dhrs7c (Forward: AGTACGGAGGAAGAAGCAAGAGG, Reverse: GCCACCACAGCGAAGAAGAAC) and β-actin (Forward: TATGCTCTCCCTCACGCCATCC, Reverse: GTCACGCACGATTTCCCTCTCAG).

Western blotting

Total proteins were extracted from mouse cardiac tissues using RIPA lysis buffer (Roche Diagnostics, Switzerland) supplemented with 1% protease inhibitors (MedChemExpress, USA). The total proteins were separated using 10% SDS-PAGE electrophoresis and subsequently transferred onto PVDF membranes. The membranes were blocked with 5% skim milk for 1 h at room temperature, followed by overnight incubation at 4 °C with the primary antibodies: anti-DHRS7C (26020-1-AP, Proteintech, China) and anti-β-actin (66009-1-Ig, Proteintech, China). After washing, the membranes were incubated with HRP-conjugated secondary antibodies (ANT020, AntGene, China) for 1 h at room temperature. The protein signals were finally visualized using an enhanced chemiluminescence detection system.

Identification of potential therapeutic compounds and molecular Docking

The small molecules that potentially target DHRS7C were screened using the Connectivity Map (CMap) platform (https://clue.io/). The small molecule with the highest score was then subjected to molecular docking with the target protein. The structures of the small molecule and protein were obtained from the PubChem (https://pubchem.ncbi.nlm.nih.gov/) and UniProt (https://www.uniprot.org/) databases, respectively. A blinded protein-ligand docking was performed using the CB-Dock2 server to predict the binding sites and conformations of the protein and small molecule [28]. The affinity of the structure-based docking poses was assessed based on the “Vina score”, with lower scores indicating a higher degree of binding affinity. The conformation with the most stable binding was selected for further visualization.

Statistical analyses

For the comparison of continuous variables between groups, the Wilcoxon and Student’s t-tests were employed where appropriate. Paired tests were used to compare the data before and after immunoadsorption treatment in the same patient, while all other comparisons were performed in an unpaired format. Group comparisons of categorical variables were performed using chi-square tests unless conditions required Fisher’s exact test (total sample size < 40 or any expected cell count < 1). The correlation coefficients were calculated using Spearman’s method. Statistical significance was denoted as ns for P ≥ 0.05, * for P < 0.05, ** for P < 0.01, *** for P < 0.001, and **** for P < 0.0001. All the data analyses were performed using the R software (version 4.3.1) and GraphPad Prism (version 9.5).

Results

Identification of metabolic heterogeneity in DCM

After correcting for batch effects across three datasets (Supplementary Fig. S1), an integrated dataset, consisting of 89 DCM patients and 22 healthy controls, was compiled. An unsupervised consensus clustering was applied to the DCM samples based on the expression profiles of all metabolic genes. The analysis outcomes, as depicted in the consensus matrix, CDF plots, delta area plot, and cluster consensus (Fig. 1A–D), suggested that the two-cluster partition (cluster 1, n = 61; cluster 2, n = 28) was more robust than the other configurations (Supplementary Fig. S2A, B). PCA further validated the significant differences in distribution between the two clusters were further (Fig. 1E). The differential expression of 2,752 metabolism-related genes between the clusters is shown in Fig. 1F.

Fig. 1.

Consensus clustering of DCM based on expression patterns of 2752 metabolism-related genes. A Consensus matrix for DCM samples with k = 2. B Cumulative distribution function curve for consensus clustering with k ranging from 2 to 5. C Delta area plot depicting the relative change in the area under the CDF between consecutive k-values. D Consensus clustering consistency scores for each cluster when k = 2. E PCA illustrating the distribution of samples within each cluster when k = 2. F Expression of the 2752 metabolism-related genes in two clusters. DCM, dilated cardiomyopathy; CDF, cumulative distribution function; PCA, principal component analysis

Differential analysis revealed 18 metabolic DEGs between the two clusters (Fig. 2A–B); the genes showed lower expression in cluster 2 compared to cluster 1 (Fig. 2B). Based on these 18 genes, a comparable clustering method was applied to robustly classify the 89 DCM samples into two subtypes (subtype 1, n = 60; subtype 2, n = 29) (Fig. 2C–F; Supplementary Fig. S3A, B). Only one sample changed groups in the two clustering iterations (Supplementary Table S3), while the remaining samples remained in the same groups in both clustering iterations. PCA revealed that the second clustering exhibited more pronounced differences in sample distribution and internal consistency of the subgroups as compared to the first one (Figs. 1D and E and 2F, G). The heatmap showed consistent expression patterns of the 18 metabolic DEGs across both clustering iterations (Fig. 2H). Overall, these results suggested that two metabolically heterogeneous populations existed within DCM. Given the superior performance of the second clustering, these two groups were used as the final subtypes for subsequent analyses.

Fig. 2.

Second consensus clustering of DCM samples. A Volcano plot of differentially expressed genes between cluster 1 and cluster 2. Red triangles indicate upregulation, blue inverted triangles indicate downregulation, and gray circles represent non-significant changes. B Heatmap of differentially expressed metabolism-related genes between cluster 1 and cluster 2, with red indicating upregulation and blue indicating downregulation. C Consensus matrix for DCM samples with k = 2. D Consensus clustering CDF curve for k = 2 to 5. E Delta area plot illustrating the relative change in the area under the CDF between adjacent k values. F Consensus clustering consistency scores for each cluster when k = 2. G PCA of sample distribution within each cluster when k = 2. H Expression of 18 metabolism-related genes between the two clusters. DCM, dilated cardiomyopathy; CDF, cumulative distribution function; PCA, principal component analysis

Metabolic reprogramming in DCM subtypes

In order to investigate the intrinsic molecular signatures distinguishing DCM subtypes, the DEGs in subtypes 1 and 2 compared to the healthy group were analyzed. The results showed that subtype 2 had a greater number of DEGs than subtype 1 (Fig. 3A and C; Supplementary Tables S4 and S5). GSEA revealed 92 and 276 enriched biological processes in the subtypes 1 and 2 respectively, suggesting subtype 2 as the pathway-rich phenotype (Supplementary Tables S6 and S7). The subtype 1 exhibited the most significant upregulation in “extracellular structure organization” (Supplementary Fig. S4A), whereas the subtype 2 demonstrated predominant activation of the biological processes associated with transforming growth factor-beta (TGF-β) response (Supplementary Fig. S4B). The top ten downregulated genes in both subtypes showed high overlap, indicating a common impairment in ATP biosynthesis in the myocardium of DCM patients (Fig. 3B and D). GSVA of 84 metabolic pathways showed that subtype 1 was characterized by a relative upregulation of amino acid metabolism pathways (Fig. 3E; Supplementary Table S8), while subtype 2 was characterized by a relative downregulation of carbohydrate and energy metabolism pathways (Fig. 3F; Supplementary Table S9). Overall, these results indicated that the DCM patients commonly exhibited defects in myocardial bioenergetic synthesis, and the metabolic heterogeneity was manifested as two distinct metabolic reprogramming phenotypes.

Fig. 3.

Metabolic characteristics of DCM subtypes. A Volcano plot of differentially expressed genes for subtype 1. Red triangles denote upregulated genes, blue inverted triangles denote downregulated genes, and grey circles indicate no significant change. B Top ten downregulated biological processes in subtype (1) C Volcano plot of differentially expressed genes for subtype (2) D Top ten downregulated biological processes in subtype 2. E Upregulated and downregulated metabolic pathways in subtype 1. F Upregulated and downregulated metabolic pathways in subtype 2

Clinical characteristics of DCM subtypes

The GSE17800 dataset contains the baseline clinical parameters of 40 patients, recorded before the initiation of immunoadsorption therapy; 28 and 12 patients were classified into subtypes 1 and 2, respectively (Supplementary Table S10). The subtype 2 demonstrated significantly impaired left ventricular systolic function and ventricular structural remodeling. Specifically, subtype 2 exhibited significantly reduced LVEF (P = 7e-04, Fig. 4A) and increased LVIDd (P = 5.4e-03, Fig. 4B) as compared to subtype 1. The proportion of inflammatory cell infiltration (P = 0.78), BMI (P = 0.53), age (P = 0.56), and gender distribution (P = 0.94) showed no statistically significant differences between the subtypes (Fig. 4C–F).

Fig. 4.

Clinical features of DCM subtypes. A–F LVEF, LVIDd, Inflammatory cell counts per square millimeter, BMI, Age, Gender. DCM, dilated cardiomyopathy; LVEF, left ventricular ejection fraction; LVIDd, left ventricular internal diameter in diastole; BMI, body mass index

Validation of clinical characteristics in DCM subtypes

The metabolic DEGs between the two subtypes were used as classification features in the supervised XGBoost classification model (Supplementary Fig. S5A, B). The model was thoroughly trained by altering various parameter combinations to elucidate the two subtypes of 89 DCM patients divided by consensus clustering in the integrated dataset (Supplementary Table S11). The optimal hyperparameter configuration (eta = 0.1, max_depth = 10, gamma = 0, colsample_bytree = 0.8, min_child_weight = 1, subsample = 0.8, and nrounds = 50) achieved exceptional model performance, showing 99.7% accuracy and near-perfect agreement (Kappa = 0.993, Supplementary Fig. S5C–D and Supplementary Table S11). The importance of each feature was illustrated in a beeswarm plot (Supplementary Fig. S5E). The well-trained XGBoost model was applied to the external dataset GSE135055, where 10 and 5 cases were assigned to subtypes 1 and 2, respectively (Supplementary Table S12). Similar to the clinical characteristics of subtypes in the 89 DCM patients, subtype 2 in GSE135055 exhibited a lower LVEF (P = 0.074, Supplementary Fig. 5F), and no difference was observed in other clinical characteristics, including age and gender (Supplementary Table S12).

Immunoinflammatory landscapes in DCM subtypes

The ESTIMATE analysis revealed that subtype 2 exhibited significantly higher stromal (P = 6.7e-04), immune (P = 0.026), and inflammatory response (p = 4.8e-03) scores as compared to subtype 1 (Fig. 5A–C). Immune cell infiltration analysis revealed 16 differentially enriched immune cell populations between the subtypes (Fig. 5D). Strikingly, the patients in subtype 2 demonstrated enrichment of 15 immune cells, while Th17 cells showed preferential enrichment in subtype 1 (Fig. 5D).

Fig. 5.

Immunoinflammatory landscapes in DCM subtypes. A–C Stromal score, Immune score and Inflammatory response score for subtypes of DCM. D Abundance of 28 immune cell infiltration in two subtypes. (ns indicates no significance, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). E–F Changes in LVEF, LVIDd, and inflammatory cell counts before and after immunoadsorption treatment in patients of subtype 1 (n = 24) and subtype 2 (n = 9). G–I Improvement of LVEF, LVIDd, and inflammatory cell counts after immunotherapy in the two subtypes. J Immune treatment response in the two subtypes. DCM, dilated cardiomyopathy; LVEF, left ventricular ejection fraction; LVIDd, left ventricular internal diameter in diastole; BL, baseline; FU, follow-up; R, responders; NR non-responders

Furthermore, the changes in LVEF, LVIDd, and cardiac immune cell counts were also evaluated before and after immunoadsorption treatment in 33 DCM patients (subtype 1, n = 24; subtype 2, n = 9). Overall, the paired analyses demonstrated that both subtypes exhibited a significant increase in LVEF (subtype 1, P = 2.8e-04; subtype 2, P = 0.015) and a significant decrease in LVIDd (subtype 1, P = 1.4e-03; subtype 2, P = 0.029) six months after treatment (Fig. 5E–F). The number of cardiac immune cells/mm² decreased in subtype 2 (P = 0.042), while that in subtype 1 showed no significant change (P = 0.89, Fig. 5E–F). The magnitude of changes in LVEF and LVIDd between the two subtypes showed no significant changes (Fig. 5G–H; Supplementary Table S13). However, the reduction in immune cell counts was more pronounced in subtype 2 as compared to subtype 1 (P = 0.029, Fig. 5I; Supplementary Table S13).

According to the original research report, the DCM patients who showed improvement in LVEF (≥ 20 relative and ≥ 5% absolute) after 6 months of immunoadsorption therapy were considered responders, while those who did not meet this definition were classified as non-responders [12]. In subtypes 1 and 2, 54.2% (13 of 24) and 66.7% (6 of 9) were categorized as responders, respectively (P = 0.698, Fig. 5J; Supplementary Table S13). The insignificant difference in immunotherapy response might be due to limited statistical power. While point estimates suggest a clinically relevant trend favoring subtype 2, validation in larger cohorts is essential.

Overall, these results suggested that the immunoinflammatory factors played a significant role in the process of DCM, and the immunoinflammatory activity in subtype 2 was more pronounced than that in subtype 1. To some extent, the significant improvement in immunoinflammatory markers after immunotherapy in subtype 2 was parallel to its better response to immunotherapy.

Key genes determining the intrinsic characteristics of subtype 2

Given that subtype 2 may represent a distinct branch within the DCM characterized by poorer left ventricular structure and function, identifying its biomarkers might facilitate personalized interventions to improve prognosis. Through a systematic bioinformatic screening of the metabolic genes delineating subtype 2 from both the healthy control and subtype 1, DHRS7C was identified as a candidate DEG (Fig. 6A). Quantitative analysis revealed progressively reduced expression levels of DHRS7C across groups, with subtype 2 demonstrating significantly lower expression levels as compared to both control and subtype 1 (Fig. 6B). Further clinical correlation studies revealed that DHRS7C was significantly positively correlated with LVEF (R = 0.32, P = 0.042) and negatively correlated with LVIDd (R = − 0.37, P = 0.019, Fig. 6C). Among the patients undergoing immunotherapy, the DHRS7C expression level was significantly elevated in responders (P = 0.049), whereas it remained unchanged in non-responders (P = 0.86, Fig. 6D).

Fig. 6.

Clinical correlation and biological functions of DHRS7C. A Venn diagram of metabolism-related genes differentially expressed in subtype 2 compared to controls and subtype 1. B Expression of DHRS7C across three groups (**p < 0.01, ***p < 0.001, ****p < 0.0001). C Correlation between DHRS7C and LVEF and LVIDd. D Changes in DHRS7C expression before and after immunotherapy in responders (n = 19) and non-responders (n = 14). E Correlation of DHRS7C with 28 immune cell types. F Genes positively correlated with DHRS7C expression. G GO and KEGG enrichment analysis. MRG, metabolism-related genes; LVEF, left ventricular ejection fraction; LVIDd, left ventricular internal diameter in diastole; Cor, correlation; BL, baseline; FU, follow-up; R, responders; NR non-responders; GO, Gene Ontology; KEGG Kyoto Encyclopedia of Genes and Genomes; BP, biological processes; CC, cellular compartment; MF, molecular functions

The correlation analysis of DHRS7C with 28 types of immune cells indicated that DHRS7C was significantly positively correlated with central memory CD8 T cells, Th17 cells, neutrophils, gamma delta T cells, and MDSCs and negatively correlated with central memory CD4 T cells and CD56^bright natural killer cells (Fig. 6E). Moreover, GO and KEGG enrichment analyses on DHRS7C and its positively correlated genes were performed (Supplementary Table S14). The results showed that the biological functions of DHRS7C were closely associated with carbohydrate and energy metabolism (Fig. 6F–G; Supplementary Table S15); this aligned with the metabolic characteristics of subtype 2. Collectively, these results suggested that DHRS7C might serve as a potential biomarker for DCM subtype 2, with its expression paralleling the immune metabolic features and clinical phenotypes of this subtype.

Functional characterization of DHRS7C at the single-cell level

Following quality control, dimensionality reduction, and clustering of single-cell data, a total of 16 cellular clusters were identified (Fig. 7A). Based on the marker genes reported in the literature (Supplementary Fig. S6), nine distinct cell types were annotated, including cardiomyocytes, endothelial cells, endocardial cells, fibroblasts, lymphatic endothelial cells, macrophages, pericytes, smooth muscle cells, and T cells (Fig. 7B). DHRS7C was highly expressed in cardiomyocytes (Fig. 7C), and its expression was significantly reduced in DCM cardiomyocytes as compared to the normal group (P = 2.34e-05, Fig. 7D). Based on their expression levels, cardiomyocytes were categorized into DHRS7C-positive and DHRS7C-negative groups, and the activity of pathways related to glycolysis and energy metabolism was assessed in both groups. The results indicated that these two metabolic pathways were highly active in DHRS7C-positive cardiomyocyte populations (Fig. 7E). Additionally, comparing the interactions between the two groups of cells and immune cells revealed that DHRS7C-positive cardiomyocytes exhibited more ligand-receptor pairs with macrophages and T cells (Fig. 7F; Supplementary Fig. S7).

Fig. 7.

Characterization of DHRS7C at the single-cell level. A UMAP plot depicting the distribution of 16 cellular clusters. B Manual annotation resulting in the identification of 9 distinct cell types. C Specific expression of DHRS7C in cardiomyocytes. D Comparison of DHRS7C expression levels in cardiomyocytes between normal controls and patients with DCM. E Metabolic pathway analysis of DHRS7C in cardiomyocytes. F Communication between immune cells and DHRS7C-labeled cardiomyocytes. CM, cardiomyocyte; EC, endothelial cell; EndoC, endocardial cell; FB, fibroblast; LEC, lymphatic endothelial cell; MP, macrophage; SMC, smooth muscle cell; N, negative; P, positive

Verification of DHRS7C in DCM mice model

As compared to the control group, the peptide-immunized mice exhibited enlarged cardiac morphology and elevated heart weight-to-body weight ratios (Fig. 8A, Supplementary Fig. S8A). The echocardiographic analysis demonstrated that the DCM group mice exhibited significantly reduced LVEF and FS, along with left ventricular dilation (increased LVIDs and LVIDd) as compared to the control group. These functional impairments were accompanied by elevated serum levels of BNP and cTnT (Fig. 8B, C). Histopathological analysis showed increased inflammatory cell infiltration and extensive myocardial fibrosis in the heart tissues of the DCM group mice (Fig. 8D). Notably, DHRS7C expression was significantly downregulated in the DCM group (Fig. 8E). Immunohistochemical staining and Western blot analysis confirmed reduced DHRS7C protein abundance in the myocardium of DCM mice (Fig. 8D and F). Immunofluorescence analysis demonstrated the DHRS7C was expressed in cardiomyocytes in the control group mice, while the DCM mice exhibited attenuated fluorescence intensity distribution areas in their heart tissues (Fig. 8D). Furthermore, exploratory analyses in the DCM mice revealed a similar negative association between DHRS7C expression and cardiac structural/functional indices (Supplementary Fig. S8B). In addition, the circulating anti-MyHC autoantibody levels in DCM mice were also measured. As compared to the control group, the circulating anti-MyHC autoantibodies were significantly elevated in DCM mice (Supplementary Fig. S8C). Furthermore, correlation analysis indicated that cardiac mRNA expression of DHRS7C was negatively associated with circulating anti-MyHC antibodies in DCM mice (Supplementary Fig. S8C).

Fig. 8.

Validation of DHRS7C in DCM model and molecular docking. A Representative gross cardiac morphology from PBS-treated control (left) and MyHC-α peptide-induced DCM (right) groups. B Comparative M-mode echocardiographic images of left ventricular function in control (top) and DCM (bottom) mice. C Comparisons of cardiac functional parameters: EF (%), FS (%), LVIDs (mm), LVIDd (mm) and serum BNP and cTnT (pg/ml). D Histopathological evaluation: Top panel shows representative images of HE staining, Masson’s trichrome staining, immunohistochemical staining for DHRS7C, and DHRS7C immunofluorescence in control vs. DCM hearts. Bottom panel displays semi-quantitative analyses of inflammatory scores, fibrotic area percentage, IHC optical density. E Relative mRNA expression levels of DHRS7C in myocardial tissues. F Western blot semi-quantification of DHRS7C protein abundance in control and DCM groups. G The best conformation of blind docking between 17-β-estradiol and the protein encoded by DHRS7C. All data presented as mean ± SD (n = 6 per group). Scale bars: 100 μm (HE/Masson/IHC/immunofluorescence). *p < 0.05, **p < 0.01, ***p < 0.001, **** p < 0.0001. EF, ejection fraction; FS, fractional shortening; LVIDs/d, left ventricular internal diameter in systole/diastole; HE, hematoxylin and eosin; IHC, immunohistochemical

The CMap platform was used to investigate the small molecules that could target DHRS7C. Based on the screening parameters, 17β-estradiol was identified as the best candidate small molecule with the maximal prediction confidence scores across screening parameters (Supplementary Table S16). Of note, 17β-estradiol levels were significantly lower in the DCM model compared to controls (Supplementary Fig. S8D). Molecular docking demonstrated that 17β-estradiol could form multiple hydrogen bonds with DHRS7C with a calculated binding affinity of − 7.5 kcal/mol (Fig. 8G, Supplementary Table S17), indicating a strong spontaneous binding propensity.

Discussion

In this study, consensus clustering analysis of metabolism-related gene expression profiles from 89 DCM patients identified two robust subtypes. The two subtypes maintained comparable clinical manifestations; however, they demonstrated marked divergence across multiple pathophysiological dimensions, including metabolic pattern configurations, severity of left ventricular structural and functional impairment, immunoinflammatory activation status, and differential responsiveness to immunotherapeutic interventions. Additionally, the study found that the expression levels of DHRS7C were significantly negatively correlated with the deterioration of left ventricular structure and function in DCM patients. Furthermore, 17β-estradiol was identified as a potential therapeutic candidate for balancing myocardial energy metabolism and immuno-inflammation by targeting DHRS7C.

Normal cardiomyocytes uptake metabolic substrates from the bloodstream and generate ATP via oxidative phosphorylation in the mitochondrial respiratory chain to meet the high-energy demands of contractile function [29]. A disruption in the bioenergetics of cardiomyocyte mitochondria leads to reduced energy supply and increased ROS production, thereby facilitating the onset of DCM and potentially culminating in end-stage heart failure [7]. The current study indicated that despite the DEGs between the two DCM subtypes and healthy individuals, both subtypes 1 and 2 universally exhibited mitochondrial dysfunction, manifested by impairments in the electron transport chain and uncoupling of oxidative phosphorylation. Under sustained oxidative stress, the heart, already compromised in terms of energy supply, might ultimately succumb to failure [30]. It is well established that DCM patients exhibit metabolic remodeling, such as alterations in metabolic demands and substrate preference [31, 32]. Notably, the two DCM subtypes displayed distinct metabolic patterns. Subtype 1 showed active amino acid metabolism, such as taurine, hypotaurine, phenylalanine, tyrosine, and β-alanine metabolisms. In DCM pathogenesis, taurine is involved in contractile function and redox homeostasis [33], and the imbalance of phenylalanine-tyrosine can induce cardiotoxicity predisposition to ventricular remodeling [34], while β-alanine-mediated antioxidant synthesis can cause ischemic protection [35]. Thus, given the pivotal role of amino acids in myocardial metabolism, management of subtype 1 – a subtype with a larger patient population and relatively milder symptoms – should focus on conventional immunomodulatory and energy metabolism regulation therapies (e.g., SGLT2 inhibitors), while actively investigating targeted amino acid modulation strategies [36]. Subtype 2 was characterized by a significant downregulation of glucose and energy metabolism-related pathways, particularly glycolysis/gluconeogenesis and oxidative phosphorylation [31]. A significant metabolic shift in a failing heart is the transition of the primary energy source from fatty acid β-oxidation to glycolysis [37]. Glycolysis only provides about 5% of ATP under healthy conditions, which is remarkably less than the 95% provided by mitochondrial oxidative phosphorylation [29]; however, the compensatory increase in glycolysis is crucial for maintaining cardiac contractile function under pathological conditions [38]. A decrease in compensatory glycolysis can cause an extreme energy deficit dilemma for the heart [30]. Beyond these subtype-specific metabolic patterns, the current study also observed widespread alterations in fundamental metabolic pathways common to both DCM subtypes. These dysregulated metabolic networks collectively contributed to their distinct phenotypic manifestations.

In terms of clinical features, the results showed that subtype 2 exhibited worse left ventricular structure and function as compared to subtype 1, demonstrating a larger LVIDd and a lower LVEF. The differences in the metabolic profiles of the two subtypes might be a potential underlying cause for the differences in their clinical features. The predictions using the trained XGBoost model on the GSE135055 dataset (n = 15) showed reduced LVEF in subtype 2; however, the decrease did not reach statistical significance, which might be due to limited statistical power from the small cohort size.

Studies have reported a sophisticated regulatory network of immune and inflammatory responses in DCM progression [39]. The current study also revealed that the two DCM subtypes presented with divergent immune and inflammatory profiles. It was observed that subtype 2 stood out in terms of immune, stromal, and inflammatory response scores. Additionally, subtype 2 was associated with a higher prevalence of immune cells, encompassing both innate immune components, such as natural killer cells and macrophages, as well as diverse subsets of T and B cells that participate in adaptive immunity. These disparities in inflammatory and immune dysregulation might be fundamental to the phenotypic heterogeneities observed in DCM and act as modulating factors affecting the divergent outcomes between the subtypes. The data analysis from the follow-up of immunoadsorption therapy for both subtypes demonstrated that this treatment could significantly ameliorate the left ventricular structure and function in DCM patients. Importantly, subtype 2 exhibited a more pronounced and substantial decrease in the proportion of inflammatory cell infiltration following immunotherapy. Although not reaching statistical significance under small-sample constraints, subtype 2 demonstrated a numerically higher overall response rate to immunoadsorption. When considered alongside its quantifiable inflammatory resolution, these observations suggested this subtype may derive clinically meaningful benefits from immunoadsorption. Currently, immunoadsorption therapy targeting cardiac autoantibodies has shown therapeutic effects in numerous clinical trials [40, 41]. Circulating autoantibodies play a dual bridging role in myocardial energy metabolism and immune inflammatory response. The evidence of immunoadsorption treatment in the DCM patients subtype 2 further confirms the significance of the DCM clustering based on energy metabolism and inflammatory regulation.

The correlation analysis of gene expression levels and clinical features suggested DHRS7C as a potential biomarker for subtype stratification. DHRS7C is a member of the short-chain dehydrogenase/reductase family, located in the sarcoplasmic reticulum, and specifically expressed in myocardium, skeletal muscle, and skin [42]. A previous study showed that DHRS7C was downregulated in response to adrenergic stimulation and heart failure [42]; the current study results revealed consistent results. The DCM patients exhibited decreased DHRS7C expression, which was significantly correlated with LVEF and LVIDd. Moreover, poorer left ventricular structure and function corresponded to its lower expression levels. Enrichment analysis indicated that the biological functions of DHRS7C were primarily related to glucose and energy metabolism. Its overexpression in skeletal muscle can enhance glucose uptake and glycogen storage, thereby improving muscle performance [43]. Among patients undergoing immunoadsorption, the DHRS7C expression significantly increased from baseline to six months post-treatment in responders. Given that immunoadsorption therapy primarily targets autoantibodies in patients, the exploratory analysis in the current study showed an inverse correlation between DHRS7C expression and circulating anti-MyHC antibody, suggesting potential regulatory interplay between DHRS7C and pathogenic autoimmunity in DCM progression. While screening for compounds targeting DHRS7C, 17β-estradiol emerged as a lead compound with optimal pharmacodynamic properties. Moreover, molecular docking confirmed a high affinity between DHRS7C and 17β-estradiol. Our observation of significantly reduced serum 17β-estradiol in male DCM mice aligns with previous findings in human males, where increased free estradiol has a protective effect on heart failure risk among men [44]. Mechanically, studies have shown that 17β-estradiol can prevent cardiomyocyte apoptosis, inhibit ROS-induced cardiac injury, and mitigate cardiac hypertrophy and fibrosis [45]. Therefore, 17β-estradiol might represent a therapeutic option for improving the prognosis of DCM patients, although its molecular mechanisms warrant further investigation [46].

This study provided new insights into DCM subtypes. However, certain limitations need to be addressed. First, an integrative framework incorporating etiology, comorbidities, and multi-omics data (proteomic, metabolomic, and imaging profiles) should be established to delineate inter-subtype distinctions and validate classification reproducibility across diverse cohorts [47]. Second, developing adaptive machine learning algorithms capable of processing multidimensional patient data is essential to achieve precise subtyping with minimal feature requirements, thereby bridging the gap toward clinical translation for personalized therapeutic decision-making [48]. Third, due to the lack of animal models corresponding to the subtypes, this study only compared the differences in DHRS7C expression between controls and DCM models. Further experiments are needed to elucidate the molecular mechanisms of DHRS7C in DCM and verify the existence and significance of its biological binding with 17β-estradiol. Overcoming these limitations will undoubtedly help in achieving a more precise stratification of DCM.

Conclusions

In summary, this study revealed the existence of two subtypes within DCM, which were characterized by distinct metabolic patterns and immunoinflammatory regulatory networks, along with heterogeneous clinical features. Notably, DHRS7C exhibited an inverse correlation with cardiac functional indices in DCM, while 17β-estradiol emerged as a promising candidate for drug repurposing. The subtype classification approach proposed in this study might aid in stratifying risk among DCM and developing precise treatment strategies.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

DCM

Dilated cardiomyopathy

ROS

Reactive oxygen species

KEGG

Kyoto encyclopedia of genes and genomes

GEO

Gene expression omnibus

BMI

Body mass index

LVEF

Left ventricular ejection fraction

LVIDd

Left ventricular internal diameter at diastole

CDF

Cumulative distribution function

PCA

Principal component analysis

GSEA

Gene set enrichment analysis

GSVA

Gene set variation analysis

GO

Gene ontology

XGBoost

eXtreme gradient boosting

ssGSEA

Single-sample gene set enrichment analysis

FDR

False discovery rate

scRNA-seq

Single-cell RNA sequencing

MyHC

Myosin heavy chain

qRT-PCR

Quantitative real-time polymerase chain reaction

Cmap

Connectivity map

Contributions

Jing Yuan: Conceptualization, Supervision, Writing—Review & Editing, Funding acquisition; Cheng Yu: Conceptualization, Investigation, Formal analysis, Writing—Original Draft; Changhu Liu: Conceptualization, Investigation, Formal analysis, Writing—Original Draft; Bingjun Liu, Jiaying Feng and Yichen Deng: Validation, Data curation. All authors read and approved the final manuscript.

Funding

This study was funded by the Noncommunicable Chronic Diseases-National Science and Technology Major Project (NO:2023ZD0504600) and National Natural Science Foundation of China (82470368 to J.Y.).

Data availability

Datasets used in this study are available in Gene Expression Omnibus under GSE29819, GSE21610, GSE17800, GSE19303, GSE135055, GSE121893.

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interest

All authors declare no disclosure of interest for this contribution.