# A gene interaction network-based molecular taxonomy of dilated cardiomyopathy highlighting immune and inflammatory heterogeneity
Researcher: Ge Zhang
Chinese name: 张格
ORCID identity anchor: https://orcid.org/0000-0002-3116-3246
Canonical researcher: https://drgezhang.com/#person
Journal: Clinical and Experimental Medicine
Year: 2026
Type: Article
DOI: 10.1007/s10238-026-02130-6
Canonical page: https://drgezhang.com/papers/doi-10-1007-s10238-026-02130-6.html
Markdown record: https://drgezhang.com/papers/doi-10-1007-s10238-026-02130-6.md
## Full Authors
1. Youyang Zheng
2. Fangfang Wan
3. Shulun Ma
4. Haijie Chen
5. Leqi Wang
6. Yuanrui Deng
7. Ge Zhang
8. Lei Song
9. Mo Zhang
## Research & review context
### Research fields
- Cardiomyopathy molecular classification
- Cardiovascular systems biology
### Review topics
- Myocardial transcriptomic heterogeneity
- Network-based patient stratification
### Scientific role
Retrospective human expression-network subtyping with cross-dataset assignment, inferred immune profiles and a selected cell-model expression check.
### Evidence position
Computational and exploratory translational evidence; treatment-response expression similarity does not demonstrate subtype-guided therapeutic benefit.
### Related concepts
- Edge perturbation
- Immune-signature inference
- Immunoadsorption response hypotheses
### Possible citation contexts
- Methods: interaction-level features for molecular DCM classification across bulk myocardial expression datasets.
- Background: immune and metabolic heterogeneity in DCM, with exploratory treatment-response matching explicitly separated from clinical efficacy.
Possible literature-review contexts, not evidence of existing citations. Read the findings and Evidence Scope before citing the original article.
## Evidence Scale
**DCM interaction-network taxonomy: myocardial transcriptomic subtypes and exploratory immunoadsorption-response matching**
A reanalysis of myocardial expression datasets identifies three dilated-cardiomyopathy network subtypes, characterizes their immune and metabolic associations, and explores expression similarity to immunoadsorption responders. Cell-line qPCR checks selected EPS1-associated genes; it does not validate subtype-guided treatment.
| Evidence | Value |
| --- | --- |
| article-reported dataset scale | 245 DCM and 174 control myocardial samples across four datasets; not a verified total of unique individuals |
| discovery dataset | GSE5406: 86 DCM and 16 controls |
| validation datasets | GSE57338: 82 DCM/136 controls; GSE116250: 37 DCM/14 controls; GSE19303: 40 DCM/8 controls |
| mapped discovery interactome | 5,593 genes/nodes and 107,063 interactions/edges |
| selected clustering features | 27,937 edges |
| molecular subtypes | 3: EPS1, EPS2 and EPS3 |
| WGCNA analyzed subset | 74 DCM samples after outlier exclusion |
| cell-model expression check | 8 genes in AC16 cells treated with 10 micromolar angiotensin II for 48 hours |
### Counting note
Counts are those in the article Table 1 and Results; transcriptomic samples, network edges, analyzed subsets and cultured cells are different units. No combined unique patient total is asserted. The 74-sample WGCNA subset is not another cohort. GSE19303 contributes to subtype assignment and the exploratory treatment comparison, so that comparison is not independent prospective treatment validation. Cell-replicate numbers are not inferred.
## Research Question
Can deviations in myocardial gene-interaction patterns distinguish reproducible molecular subtypes of dilated cardiomyopathy, and how do those subtypes relate to inferred immune profiles and expression patterns in an immunoadsorption-treated dataset?
## Author Evidence Summary
The study reanalyzed four public myocardial transcriptomic datasets, using expression-rank differences between connected genes to characterize network perturbations. Discovery was performed in GSE5406 and subtype assignments were evaluated in three other datasets. The analysis identified EPS1, EPS2 and EPS3, with different enrichment patterns involving energy metabolism, humoral immunity, platelet-related pathways and endocytosis. Computational immune-cell signatures also differed among subtypes. EPS1 expression was associated with the responder profile in an existing immunoadsorption/immunoglobulin-substitution dataset. That result is a retrospective expression-matching hypothesis rather than a clinical test of treatment benefit. Angiotensin-II-treated AC16 cells reproduced the direction of selected EPS1-associated expression changes, providing a narrow cell-model check without demonstrating all three human subtypes or their mechanisms.
## Key Findings
### KF1: Four public myocardial datasets supplied discovery and external subtype-assignment data.
Context: Article-reported analysis subsets for DCM network taxonomy; no newly recruited clinical cohort
| Evidence | Value |
| --- | --- |
| discovery subset | GSE5406: 86 DCM and 16 controls |
| external subsets | GSE57338: 82/136; GSE116250: 37/14; GSE19303: 40/8 DCM/control samples |
| reported sample totals | 245 DCM and 174 controls across four datasets; no verified unique-person total |
Source locator: Methods: Data collection and preprocessing (Par6); Results: Cohort characteristics and interactome construction (Par22); Table 1 (Tab1); Figure 2E-G
### KF2: Rank-based perturbation features provided a network representation for subtype discovery.
Context: Expression-rank differences were compared with a control-derived reference before feature selection
| Evidence | Value |
| --- | --- |
| mapped network | 5,593 nodes and 107,063 edges |
| selected features | 27,937 discriminative edges |
| feature-selection rule | intersection of the top 50% of edges for case-control difference and the top 50% for DCM variability |
Source locator: Methods: Quantifying gene interactions and Edge perturbation matrix construction (Par8-Par11); Results: Cohort characteristics and interactome construction (Par22) and Edge perturbation profiling (Par24); Figure 1
### KF3: Three network subtypes were recovered and assigned across the discovery and validation datasets.
Context: Retrospective molecular clustering evaluated with UMAP, silhouette analysis and nearest-template prediction
| Evidence | Value |
| --- | --- |
| selected cluster count | K = 3 |
| EPS1 range | 43.2-52.5% across analyzed datasets |
| EPS2 range | 40.0-44.2% across analyzed datasets |
| EPS3 range | 7.5-13.5% across analyzed datasets |
Source locator: Methods: Consensus clustering for subtype identification and Validation of clustering stability (Par13-Par14); Results: Consensus clustering identifies three stable DCM subtypes (Par25-Par26); Figure 2A-G
### KF4: Coexpression modules and inferred immune patterns differed between the molecular subtypes.
Context: Computational associations in the 74-sample WGCNA subset and gene-expression-based immune profiling; not direct immune-cell counting or causal perturbation
| Evidence | Value |
| --- | --- |
| WGCNA subset and modules | 74 DCM samples after outlier exclusion; 10 modules |
| EPS1-associated module | 540 genes; module-subtype r = 0.84 |
| EPS2-associated module | 513 genes; module-subtype r = 0.74 |
| EPS3-associated module | 87 genes; module-subtype r = 0.78 |
| interpretation | energy/proteostasis, humoral/platelet-related and respiratory-chain/endocytosis enrichment patterns differed; immune composition was inferred by ssGSEA |
Source locator: Methods: Identification of subtype-specific gene modules and Immune cell infiltration analysis (Par15-Par17); Results: Subtype-specific co-expression modules (Par28), Functional annotation of molecular subtypes (Par30-Par32), Distinct immune infiltration patterns (Par34); Figures 3-6A-C
### KF5: EPS1 showed expression similarity to the IA/IgG responder group in an exploratory retrospective analysis.
Context: Submap compared subtype expression with responder/non-responder profiles in GSE19303; this is not a randomized treatment comparison
| Evidence | Value |
| --- | --- |
| response definition in the source cohort | at least 20% relative LVEF improvement together with at least 5 percentage points absolute improvement |
| reported expression-matching result | EPS1 association with responder profiles remained significant after Bonferroni correction; an exact corrected P value is not supplied in the Results prose |
| clinical boundary | no demonstrated benefit of assigning IA/IgG according to EPS subtype |
Source locator: Methods: Prediction of IA/IgG therapy sensitivity (Par18); Results: IA/IgG therapy response prediction (Par36); Figure 6D
### KF6: A cell stress model reproduced the direction of eight selected EPS1-associated gene-expression differences.
Context: qPCR in angiotensin-II-treated AC16 cells checked genes intersecting EPS1-versus-control and DCM-versus-control contrasts, rather than validating all subtype mechanisms
| Evidence | Value |
| --- | --- |
| cell treatment | 10 micromolar angiotensin II for 48 hours |
| higher-expression genes | 3: ASPN, LUM and NPPA |
| lower-expression genes | 5: MYOT, FKBP5, SERPINE1, CD163 and ANKRD2 |
| scope | 8 genes; no cell-replicate count or patient-level diagnostic accuracy is inferred |
Source locator: Methods: Cell culture and treatment (Par20) and Quantitative real-time PCR analysis (Par19); Results: qPCR confirmation of subtype-specific DEGs (Par37); Figure 7; Discussion: qPCR validation of core subtype genes (Par55)
## Study Design & Analytical Framework
### Study profile
- Study Design: Retrospective secondary analysis of public myocardial transcriptomic datasets, with network-based clustering, external subtype assignment, exploratory treatment-response expression matching and a cell-line expression check
- Evidence Type: Human myocardial transcriptomics, Reactome interaction mapping, rank-based edge perturbation, consensus clustering, WGCNA, enrichment, ssGSEA, Submap and AC16-cell qPCR
- Population: DCM and non-failing myocardial samples in four GEO datasets; no new clinical recruitment or randomized subtype-guided treatment trial
- Primary Endpoint: Identification and cross-dataset reproducibility of molecular DCM subtypes
- Secondary Endpoint: Subtype-associated network, pathway and inferred immune features; exploratory similarity to IA/IgG responder profiles; selected gene-expression changes in a cell stress model
- External dataset evaluation: Yes
- Data modalities: myocardial microarray expression, myocardial RNA sequencing, curated gene-interaction network, computed immune signatures, AC16 cardiomyocyte-cell-line qPCR
## What This Study Adds
- Provides an interaction-level view of DCM heterogeneity that complements expression-abundance and conventional clinical classifications.
- Connects externally assigned molecular subtypes with computational pathway and immune-signature differences.
- Supplies a bounded hypothesis for testing subtype-dependent IA/IgG responsiveness in prospective research.
- Adds a cell-line expression-direction check for selected EPS1-associated genes while keeping that evidence distinct from human subtype and clinical-response validation.
## Evidence Scope
### Supports
- Reproducible retrospective expression-network clustering in the analyzed myocardial datasets.
- Subtype-associated coexpression, enrichment and inferred immune-signature differences.
- An exploratory association between EPS1 expression and the previously observed IA/IgG responder profile.
- Concordant expression directions for selected genes in the specified AC16-cell stress model.
### Does Not Establish
- Prospective diagnostic performance, prognosis or generalizability to all clinical DCM populations.
- A causal mechanism for each subtype from pathway enrichment, coexpression or immune deconvolution.
- A validated indication for immunoadsorption, or improved outcomes from subtype-guided treatment.
- Direct immune-cell abundance measurements from ssGSEA scores.
- That the cell experiment reproduces all three human subtypes or validates every inferred pathway.
- A verified count of unique people across public datasets or additional independent cohorts from analysis subsets.
### Limitations
- Public myocardial expression datasets are selected research samples; subtype prevalence and clinical transportability require further study.
- The article reports that longitudinal outcome data were unavailable for prognostic evaluation.
- Coexpression, enrichment and ssGSEA do not experimentally establish subtype mechanisms or immune-cell causation.
- GSE19303 contributes both to subtype evaluation and to the exploratory response analysis; the response claim is not independent prospective validation.
- The cell-line experiment evaluates selected EPS1-associated expression directions, with no basis to infer all subtype mechanisms or treatment effects.
- The source does not provide a clinical decision threshold, calibrated individual treatment-benefit model or randomized comparison of subtype-guided care.
- Supplementary-file links were verified, but the Springer supplementary spreadsheets were not retrieved successfully in this run; no claim depends on uninspected supplementary-only values.
## When This Study Is Useful to Cite
### CU1: What evidence supports interaction-network subtypes of DCM?
Retrospective myocardial-expression subtyping and external subtype assignment; not a prospectively validated clinical diagnostic test.
Evidence: [KF1](#kf1), [KF2](#kf2), [KF3](#kf3)
### CU2: How do molecular DCM subtypes differ biologically?
Coexpression, enrichment and inferred immune signatures differ; causal subtype mechanisms are not established.
Evidence: [KF4](#kf4)
### CU3: Is there evidence linking DCM subtypes with immunoadsorption responsiveness?
EPS1 expression similarity to a historical responder group is an exploratory hypothesis, not proof of treatment efficacy.
Evidence: [KF5](#kf5)
### CU4: What experimental check accompanied the computational taxonomy?
Expression directions of selected EPS1-associated genes were checked in angiotensin-II-treated AC16 cells.
Evidence: [KF6](#kf6)
## What This Study Should Not Be Cited to Claim
- Prospective diagnostic performance, prognosis or generalizability to all clinical DCM populations.
- A causal mechanism for each subtype from pathway enrichment, coexpression or immune deconvolution.
- A validated indication for immunoadsorption, or improved outcomes from subtype-guided treatment.
- Direct immune-cell abundance measurements from ssGSEA scores.
- That the cell experiment reproduces all three human subtypes or validates every inferred pathway.
- A verified count of unique people across public datasets or additional independent cohorts from analysis subsets.
## Evidence Matrix
### KF1: Four public myocardial datasets supplied discovery and external subtype-assignment data.
- Context: Article-reported analysis subsets for DCM network taxonomy; no newly recruited clinical cohort
- Finding: Four public myocardial datasets supplied discovery and external subtype-assignment data. discovery subset: GSE5406: 86 DCM and 16 controls; external subsets: GSE57338: 82/136; GSE116250: 37/14; GSE19303: 40/8 DCM/control samples; reported sample totals: 245 DCM and 174 controls across four datasets; no verified unique-person total
- Evidence Level: Secondary human myocardial transcriptomics
- Scope: Article-reported analysis subsets for DCM network taxonomy; no newly recruited clinical cohort
- Source Locator: Methods: Data collection and preprocessing (Par6); Results: Cohort characteristics and interactome construction (Par22); Table 1 (Tab1); Figure 2E-G
- Evidence refs: [KF1](#kf1)
### KF2: Rank-based perturbation features provided a network representation for subtype discovery.
- Context: Expression-rank differences were compared with a control-derived reference before feature selection
- Finding: Rank-based perturbation features provided a network representation for subtype discovery. mapped network: 5,593 nodes and 107,063 edges; selected features: 27,937 discriminative edges; feature-selection rule: intersection of the top 50% of edges for case-control difference and the top 50% for DCM variability
- Evidence Level: Computational interaction-feature engineering
- Scope: Expression-rank differences were compared with a control-derived reference before feature selection
- Source Locator: Methods: Quantifying gene interactions and Edge perturbation matrix construction (Par8-Par11); Results: Cohort characteristics and interactome construction (Par22) and Edge perturbation profiling (Par24); Figure 1
- Evidence refs: [KF2](#kf2)
### KF3: Three network subtypes were recovered and assigned across the discovery and validation datasets.
- Context: Retrospective molecular clustering evaluated with UMAP, silhouette analysis and nearest-template prediction
- Finding: Three network subtypes were recovered and assigned across the discovery and validation datasets. selected cluster count: K = 3; EPS1 range: 43.2-52.5% across analyzed datasets; EPS2 range: 40.0-44.2% across analyzed datasets; EPS3 range: 7.5-13.5% across analyzed datasets
- Evidence Level: Retrospective clustering and external subtype assignment
- Scope: Retrospective molecular clustering evaluated with UMAP, silhouette analysis and nearest-template prediction
- Source Locator: Methods: Consensus clustering for subtype identification and Validation of clustering stability (Par13-Par14); Results: Consensus clustering identifies three stable DCM subtypes (Par25-Par26); Figure 2A-G
- Evidence refs: [KF3](#kf3)
### KF4: Coexpression modules and inferred immune patterns differed between the molecular subtypes.
- Context: Computational associations in the 74-sample WGCNA subset and gene-expression-based immune profiling; not direct immune-cell counting or causal perturbation
- Finding: Coexpression modules and inferred immune patterns differed between the molecular subtypes. WGCNA subset and modules: 74 DCM samples after outlier exclusion; 10 modules; EPS1-associated module: 540 genes; module-subtype r = 0.84; EPS2-associated module: 513 genes; module-subtype r = 0.74; EPS3-associated module: 87 genes; module-subtype r = 0.78; interpretation: energy/proteostasis, humoral/platelet-related and respiratory-chain/endocytosis enrichment patterns differed; immune composition was inferred by ssGSEA
- Evidence Level: Coexpression, enrichment and inferred immune associations
- Scope: Computational associations in the 74-sample WGCNA subset and gene-expression-based immune profiling; not direct immune-cell counting or causal perturbation
- Source Locator: Methods: Identification of subtype-specific gene modules and Immune cell infiltration analysis (Par15-Par17); Results: Subtype-specific co-expression modules (Par28), Functional annotation of molecular subtypes (Par30-Par32), Distinct immune infiltration patterns (Par34); Figures 3-6A-C
- Evidence refs: [KF4](#kf4)
### KF5: EPS1 showed expression similarity to the IA/IgG responder group in an exploratory retrospective analysis.
- Context: Submap compared subtype expression with responder/non-responder profiles in GSE19303; this is not a randomized treatment comparison
- Finding: EPS1 showed expression similarity to the IA/IgG responder group in an exploratory retrospective analysis. response definition in the source cohort: at least 20% relative LVEF improvement together with at least 5 percentage points absolute improvement; reported expression-matching result: EPS1 association with responder profiles remained significant after Bonferroni correction; an exact corrected P value is not supplied in the Results prose; clinical boundary: no demonstrated benefit of assigning IA/IgG according to EPS subtype
- Evidence Level: Exploratory retrospective expression-profile matching
- Scope: Submap compared subtype expression with responder/non-responder profiles in GSE19303; this is not a randomized treatment comparison
- Source Locator: Methods: Prediction of IA/IgG therapy sensitivity (Par18); Results: IA/IgG therapy response prediction (Par36); Figure 6D
- Evidence refs: [KF5](#kf5)
### KF6: A cell stress model reproduced the direction of eight selected EPS1-associated gene-expression differences.
- Context: qPCR in angiotensin-II-treated AC16 cells checked genes intersecting EPS1-versus-control and DCM-versus-control contrasts, rather than validating all subtype mechanisms
- Finding: A cell stress model reproduced the direction of eight selected EPS1-associated gene-expression differences. cell treatment: 10 micromolar angiotensin II for 48 hours; higher-expression genes: 3: ASPN, LUM and NPPA; lower-expression genes: 5: MYOT, FKBP5, SERPINE1, CD163 and ANKRD2; scope: 8 genes; no cell-replicate count or patient-level diagnostic accuracy is inferred
- Evidence Level: Cell-line qPCR expression-direction check
- Scope: qPCR in angiotensin-II-treated AC16 cells checked genes intersecting EPS1-versus-control and DCM-versus-control contrasts, rather than validating all subtype mechanisms
- Source Locator: Methods: Cell culture and treatment (Par20) and Quantitative real-time PCR analysis (Par19); Results: qPCR confirmation of subtype-specific DEGs (Par37); Figure 7; Discussion: qPCR validation of core subtype genes (Par55)
- Evidence refs: [KF6](#kf6)
## Q&A
### What data underpin the DCM network subtypes?
Four public myocardial expression datasets were analyzed. Table 1 reports 245 DCM and 174 control samples in total. These are sample counts; the evidence page does not present them as a verified unique-patient total.
Evidence: KF1
### How were network perturbations constructed?
The analysis ranked genes within each sample, calculated rank differences for interacting genes and compared those differences with a control-derived reference. Selected perturbation edges were then used for clustering.
Evidence: KF2
### How many molecular subtypes were identified?
The selected solution comprised EPS1, EPS2 and EPS3. Their reported frequencies ranged from 43.2-52.5%, 40.0-44.2% and 7.5-13.5%, respectively, across the analyzed datasets.
Evidence: KF3
### Were the immune differences measured directly in individual cells?
No. Immune-related scores were inferred from bulk gene-expression profiles using ssGSEA. Those associations are useful for generating hypotheses but are not direct cell counts or proof of immune causation.
Evidence: KF4
### Does EPS1 prove that immunoadsorption will work for a patient?
No. EPS1 expression matched the responder profile in a retrospective Submap analysis. A trial of subtype-guided treatment, with prospectively specified outcomes, would be needed to establish clinical benefit.
Evidence: KF5
### What did the AC16-cell experiment validate?
It checked the direction of eight genes selected from overlapping EPS1-versus-control and DCM-versus-control contrasts. Three increased and five decreased after angiotensin-II treatment. It does not validate all three subtypes or their treatment responses.
Evidence: KF6
### Can the subtype frequencies be used as population prevalence estimates?
No. They describe selected public myocardial datasets and the reported analysis procedure. They are not estimates from a population-based DCM sampling frame.
Evidence: KF1, KF3
## Concepts & Entities
### Disease
dilated cardiomyopathy, heart failure, myocardial heterogeneity
### Methods
edge perturbation, gene-interaction network, consensus clustering, nearest-template prediction, WGCNA, ssGSEA, Submap
### Subtypes
EPS1, EPS2, EPS3
### Evidence Boundaries
retrospective transcriptomic subtyping, computational immune inference, exploratory treatment-response matching, AC16-cell expression check
## Related Research
- [Plasma Olink Proteomics Reveals Novel Biomarkers for Prediction and Diagnosis in Dilated Cardiomyopathy with Heart Failure](https://drgezhang.com/papers/olink-dcm.html) — Complementary DCM-HF plasma-protein biomarker study; circulating protein abundance and myocardial interaction-network subtyping measure different biological layers. (DOI: 10.1021/acs.jproteome.4c00522)
- [Differential expression profiles of plasma exosomal microRNAs in dilated cardiomyopathy with chronic heart failure](https://drgezhang.com/papers/doi-10-1111-jcmm-17789.html) — DCM-HF plasma-exosomal microRNA study offering an extracellular-RNA biomarker perspective distinct from myocardial bulk-expression taxonomy. (DOI: 10.1111/jcmm.17789)
## Publication & Provenance
- DOI: https://doi.org/10.1007/s10238-026-02130-6
- Publisher: https://link.springer.com/article/10.1007/s10238-026-02130-6
- Doi Url: https://doi.org/10.1007/s10238-026-02130-6
- PubMed: https://pubmed.ncbi.nlm.nih.gov/41945162/
- Pmc Url: https://pmc.ncbi.nlm.nih.gov/articles/PMC13342204/
- PMCID: PMC13342204
- Version Of Record: Clinical and Experimental Medicine, 2026; published 7 April 2026; DOI 10.1007/s10238-026-02130-6
- License: CC BY-NC-ND 4.0 at article level. This evidence page uses an original factual synthesis; no article abstract, figures, tables or adapted article prose are reproduced.
- License Url: https://creativecommons.org/licenses/by-nc-nd/4.0/
- Author Identity: The version of record lists Ge Zhang at the Department of Cardiology, Tianjian Laboratory of Advanced Biomedical Sciences, Academy of Medical Sciences, First Affiliated Hospital of Zhengzhou University. The article record does not supply an ORCID for this author; no contradictory author identity was found. The same DOI is present in the public works list of ORCID 0000-0002-3116-3246 (put-code 226787148), independently checked 15 October 2026.
- External Datasets: GSE5406; GSE57338; GSE116250; GSE19303. Counts refer to article-reported analyzed subsets, not the full native GEO series or a verified cross-series unique-person total.
- Supporting Material: Version of record links Supplementary Materials 1-4, including two spreadsheets and two TIFF figures. Main-text claims in this page are supported by the retrieved full-text XML and visible article figures/legends; no supplementary-only values are inferred.
- Analysis Boundary: Subtype reproducibility, computational immune/pathway associations, retrospective expression matching and cell-line expression checks are recorded as different evidence layers.
- Evidence Basis: Publisher version of record, PMC/Europe PMC full-text XML, PubMed-indexed author-affiliation metadata and Crossref DOI metadata, checked 14 October 2026.
- Orcid Url: https://orcid.org/0000-0002-3116-3246
## Evidence-page notice
Author-controlled evidence page for this publication. This original evidence-oriented synthesis does not replace the publisher's version of record and is not medical advice.
## Links
- DOI: https://doi.org/10.1007/s10238-026-02130-6
- HTML page: https://drgezhang.com/papers/doi-10-1007-s10238-026-02130-6.html
- Google Scholar query: https://scholar.google.com/scholar?q=%22A%20gene%20interaction%20network-based%20molecular%20taxonomy%20of%20dilated%20cardiomyopathy%20highlighting%20immune%20and%20inflammatory%20heterogeneity%22
- ORCID: https://orcid.org/0000-0002-3116-3246