Identification of Biomarkers for Right Ventricular Dysfunction in Idiopathic Dilated Cardiomyopathy Via Urinary Proteomics and Machine Learning

Abstract

Background

Right ventricular dysfunction (RVD) is a common complication of idiopathic dilated cardiomyopathy linked to poor outcomes. However, reliable noninvasive biomarkers for RVD remain lacking. This study aimed to identify urinary proteomic markers using mass spectrometry and machine learning.

Methods

In this prospective cohort, patients with idiopathic dilated cardiomyopathy were classified by cardiac magnetic resonance imaging into groups with RVD (RV ejection fraction <45%) and without RVD groups. Baseline urine samples were profiled by data‐independent acquisition mass spectrometry. Differentially expressed proteins were identified and selected by least absolute shrinkage and selection operator regression to build a diagnostic model, developed in a training set, and validated in a test set. The primary end point was a composite of cardiovascular death, heart failure rehospitalization, left ventricular assist device implantation, or heart transplantation.

Results

The study enrolled 147 patients with idiopathic dilated cardiomyopathy (64 with RVD, 83 without), with a median follow‐up of 19.3 months. Of 3579 quantified urinary proteins, 46 were differentially expressed between groups. A 3‐protein panel (RARRES1 [retinoic acid receptor responder protein 1], MVB12B [multivesicular body subunit 12B], GSK3A [glycogen synthase kinase 3 alpha]) was identified and showed excellent diagnostic accuracy (training area under the curve 0.946; validation area under the curve0.935), outperforming both NT‐proBNP (N‐terminal pro–brain natriuretic peptide) and tricuspid annular plane systolic excursion. The risk score derived from this panel effectively stratified patients, with the high‐risk group exhibiting significantly worse outcomes than the low‐risk group (hazard ratio, 3.24 [95% CI, 1.56–6.71], P=0.002).

Conclusions

The urinary proteomic panel developed in this study demonstrates diagnostic and prognostic potential for identifying RVD in idiopathic dilated cardiomyopathy, providing a promising noninvasive tool for precise detection and clinical risk stratification.

Nonstandard Abbreviations and Acronyms

GSK3A

glycogen synthase kinase 3 alpha

iDCM

idiopathic dilated cardiomyopathy

MVB12B

multivesicular body subunit 12B

RARRES1

retinoic acid receptor responder protein 1

RVD

right ventricular dysfunction

TAPSE

tricuspid annular plane systolic excursion

Clinical Perspective.

What Is New?

• This study is the first to combine urinary proteomics with cardiac magnetic resonance to identify biomarkers of right ventricular dysfunction in idiopathic dilated cardiomyopathy.

• A 3‐protein urinary panel (RARRES1 [retinoic acid receptor responder protein 1], MVB12B [multivesicular body subunit 12B], GSK3A [glycogen synthase kinase 3 alpha]) not only demonstrated superior diagnostic performance compared with NT‐proBNP (N‐terminal pro–brain natriuretic peptide) and tricuspid annular plane systolic excursion for detecting right ventricular dysfunction but also provided independent prognostic information on adverse clinical outcomes.

What Are the Clinical Implications?

• Urinary proteomic profiling provides a noninvasive method for detecting right ventricular dysfunction with higher diagnostic accuracy than conventional biomarkers, reducing reliance on costly or less accessible imaging modalities.

Idiopathic dilated cardiomyopathy (iDCM) is characterized by left ventricular (LV) systolic dysfunction and chamber enlargement in the absence of identifiable causes. It represents an important cause of heart failure and sudden cardiac death, particularly in younger individuals. ¹ Although recent advances in molecular profiling and imaging‐based phenotyping have enabled more refined classification of iDCM subtypes, ² , ³ , ⁴ the lack of targeted therapies continues to limit outcomes, with 5‐year mortality rates remaining at ∼20% despite optimal medical management. ⁵

Despite affecting 36% to 75% of patients with iDCM, right ventricular dysfunction (RVD) was historically overlooked clinically due to its variable presentation and perceived secondary hemodynamic effects. However, recent advances in longitudinal imaging studies have firmly established RVD as an independent predictor of all‐cause mortality and the need for cardiac transplantation. ⁶ , ⁷ Although cardiac magnetic resonance (CMR) remains the gold standard for RV ejection fraction (EF) quantification, its limited accessibility often delays diagnosis in resource‐constrained settings. Echocardiography, though more widely available, often underdetects RVD due to its limited sensitivity. This diagnostic gap highlights the urgent need for noninvasive, accurate biomarkers to inform risk stratification and guide therapeutic decisions.

Urine proteomics offers a noninvasive and clinically accessible platform for biomarker discovery. Unlike conventional serum or tissue‐based approaches, it enables longitudinal monitoring, earlier detection of disease‐related changes, and improved identification of low‐abundance proteins. Proteins originating from cardiovascular and systemic sources can be excreted in the urine, allowing urinary proteomics to capture systemic pathological processes. ⁸ Although urinary biomarkers have shown promise in animal models of DCM, ⁹ human studies—especially those focusing on RV function—are scarce. This study thus aimed to identify urinary proteomic signatures predictive of RVD in iDCM and to evaluate their prognostic potential.

Methods

Study Design and Population

This was a prospective cohort study conducted at Peking Union Medical College Hospital between June 1, 2021, and July 31, 2024. A total of 376 consecutive patients with heart failure with reduced or mildly reduced EF were screened during hospitalization. Patients were eligible for inclusion if they met the diagnostic criteria for iDCM, defined as LV enlargement—LV end‐diastolic diameter >50 mm in women or >55 mm in men—and reduced systolic function (LVEF <50%), in the absence of identifiable secondary causes. ¹⁰ , ¹¹ , ¹² Exclusion criteria were as follows: (1) ischemic cardiomyopathy due to coronary artery disease; (2) primary valvular disease, (3) persistent or uncontrolled hypertension, (4) prior exposure to cardiotoxic agents (eg, anthracyclines), (5) alternative cardiomyopathies (eg, hypertrophic, arrhythmogenic, restrictive, peripartum cardiomyopathy, tachycardia‐induced cardiomyopathy, or Chagas cardiomyopathy), (6) congenital or structural heart disease, (7) infiltrative or inflammatory myocardial diseases (eg, sarcoidosis, amyloidosis, iron overload, or acute myocarditis), (8) multisystem diseases with potential cardiac involvement (eg, hypereosinophilic syndrome, connective tissue disorders, endomyocardial fibrosis, and certain metabolic or endocrine disorders such as thyroid dysfunction and Fabry disease), (9) familial DCM defined by DCM in at least 1 first‐degree relative, (10) severely impaired renal function (estimated glomerular filtration rate <30 mL/min/1.73 m²), hematuria or proteinuria, (11) inability or unwillingness to undergo CMR imaging, ¹³ and (12) lost to follow‐up.

Of the 376 patients screened, 225 met the inclusion criteria and had available urine samples with regular clinical follow‐up. Among them, 71 were excluded due to lack of CMR imaging. The remaining 154 underwent both CMR and urine sample collection for proteomic analysis. Urine samples were collected on the same morning as the CMR scan, using the midstream portion of the first morning void to minimize diurnal variation and ensure close temporal proximity between biomarker assessment and imaging evaluation. RVD was defined as RVEF <45%, measured by CMR, based on prior studies in patients with iDCM and heart failure. ⁷ , ¹⁴ , ¹⁵ , ¹⁶ After excluding 7 patients due to insufficient urine sample quality, the final analytic cohort included 147 patients. This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Peking Union Medical College Hospital (approval number: ZS‐3533). Written informed consent was obtained from all participants. A schematic overview of patient selection is presented in Figure S1. The data that support the findings of this study are available from the corresponding author on reasonable request.

Follow‐Up and Outcomes

Follow‐up was completed in February 2025 through outpatient visits conducted every 3 to 6 months postdischarge. Clinical events were verified by reviewing medical records and follow‐up communications. The end point was a composite of cardiovascular death, heart failure rehospitalization, LV assist device implantation, and heart transplantation. Survival time was calculated from the baseline assessment to the first occurrence of a primary end point event or the end of the study period.

The complete methodological details, including the echocardiographic and CMR protocol, sample collection and preparation, liquid chromatography–tandem mass spectrometry analysis, database searching, bioinformatics analysis (Gene Ontology/Kyoto Encyclopedia of Genes and Genomes), and the procedures for ELISA validation of candidate urinary biomarkers, are available in the supporting information methods S1 section.

Statistical Analysis

For clinical data, normally distributed variables were presented as mean ± SD, and nonnormally distributed variables were described using median (interquartile range). Normality was assessed using Shapiro–Wilk tests. Categorical data were presented as frequency (percentage). For intergroup comparisons, independent t tests or Mann–Whitney U tests were applied for continuous variables, and chi‐square tests or Fisher’s exact tests were used for categorical variables. Pearson correlation analysis was performed to assess associations between differentially expressed protein levels and selected clinical and imaging parameters. To further identify proteins independently associated with RVD, multivariate logistic regression analysis was subsequently conducted with adjustment for relevant clinical confounders. Kaplan–Meier survival analysis was conducted to compare event‐free survival between groups using the log‐rank test. For risk stratification, a multivariable Cox proportional hazards model was used to generate a composite risk score based on selected variables. Patients were categorized into high‐ and low‐risk groups using the median risk score as the cutoff. Cox regression was further used to assess the association between predictor variables and end point events, adjusting for potential confounders. To evaluate the robustness of the observed associations against potential unmeasured confounding, E‐values were calculated based on the hazard ratios (HR) and 95% CIs from the Cox model. To explore whether the effect of the urinary proteomic risk score on adverse clinical outcomes was mediated by RVD, causal mediation analysis was conducted using the mediation package in R. The risk score was treated as the exposure, RVD as the mediator, and time‐to‐event outcome as the dependent variable. Parametric survival regression models assuming a Weibull distribution were used to estimate the average causal mediation effect, average direct effect, and total effect. Bootstrapping with 1000 simulations was applied to generate 95% CIs and assess statistical significance. The proportion of the total effect mediated through RVD was also calculated.

Urinary proteomics data were analyzed using data‐independent acquisition‐based mass spectrometry. Quality control procedures were performed to ensure data reliability, including the exclusion of samples that did not meet predefined quality criteria. Proteins with more than 50% missing values in both groups were removed to reduce bias. For proteins with <50% missing values, K‐nearest neighbors imputation was applied, while proteins with ≥50% missing values were imputed with half of the minimum detected value. A coefficient of variation threshold of >0.5 was then applied to filter out highly variable proteins, ensuring data stability. The remaining protein intensity values were median‐normalized to correct for systematic variations and log2‐transformed to approximate a normal distribution. To identify differentially expressed proteins, independent 2‐sample t tests were conducted to compare protein expression levels between groups. Multiple testing correction was performed using the Benjamini–Hochberg method to control the false discovery rate at 5%. Proteins were considered statistically significant if they met both the false discovery rate threshold (<0.05) and a predefined fold change cutoff (fold change >1.5 for upregulation; fold change <0.67 for downregulation). For model development, all patients were randomly assigned to the training and validation sets at a 3:1 ratio. Stratified sampling based on age and sex was applied within each subgroup (RVD and non‐RVD) to ensure balanced distributions of these variables across both sets. Candidate proteins were selected in the training set using least absolute shrinkage and selection operator regression with 10‐fold cross‐validation, implemented via the “glmnet” R package. ¹⁷ The optimal regularization parameter (λ) was determined using the minimum mean squared error criterion, with additional assessment based on the 1‐SE rule to enhance model robustness. Proteins identified by least absolute shrinkage and selection operator regression and biological relevance were used to construct a binary logistic regression model for diagnostic performance evaluation. The model’s discriminative ability was quantified using the area under the receiver operating characteristic curve (AUC), computed with the “pROC” R package. Sensitivity, specificity, positive predictive value, and negative predictive value were also estimated in both the training and validation data sets. All statistical analyses were performed using R statistical software (version 4.2.1).

Results

Baseline Characteristics of the Study Cohort

The demographic characteristics of the cohort are summarized in Table 1. A total of 147 patients with iDCM were enrolled, comprising 64 patients with RVD and 83 with non‐RVD. Age and sex distributions were comparable between the 2 groups. Compared with the group without RVD, patients with RVD exhibited lower systolic blood pressure, higher heart rate, and worse functional status, as indicated by a higher proportion of New York Heart Association class III–IV and elevated NT‐proBNP (N‐terminal pro–B‐type natriuretic peptide) levels. CMR revealed that patients with RVD had significantly reduced LVEF and RVEF, accompanied by greater ventricular dilation. Echocardiographic findings further demonstrated more pronounced LV diastolic dysfunction in the RVD group, reflected by elevated early mitral inflow velocity to early diastolic mitral annular velocity ratio (E/e’). Pulmonary artery systolic pressure was also elevated in this group. Further evidence of RV systolic impairment included lower tricuspid annular plane systolic excursion (TAPSE) and a higher prevalence of moderate‐to‐severe tricuspid regurgitation. Regarding pharmacological treatment, patients with RVD were more likely to receive symptomatic therapies such as diuretics and digitalis, whereas the use of guideline‐directed medical therapy for heart failure was comparable between groups.

Table 1.

Baseline Characteristics of Enrolled Patients in This Study

	RVD (N=64)	Non‐RVD (N=83)	P value
Male sex, %	41 (64.1)	43 (51.8)	0.172
Age, y	43.49 ± 16.78	48.71 ± 16.77	0.067
Body mass index, kg/m2	25.9 (21.6–28.6)	24.2 (21.6–27.9)	0.355
Systolic BP, mm Hg	107.0 (96.0–121.5)	116.0 (101.0–137.2)	0.007*
Diastolic BP, mm Hg	70.0 (61.2–77.0)	73.0 (63.0–84.0)	0.140
Heart rate, bpm	80.0 (73.0–98.5)	75.0 (66.0–86.2)	0.008*
Smoking, %	27 (42.2)	31 (37.3)	0.619
Drinking, %	21 (32.8)	25 (30.1)	0.791
New York Heart Association classification			0.024*
I	2 (3.1)	6 (7.2)
II	32 (50.0)	57 (68.7)
III	20 (31.3)	16 (19.3)
IV	10 (15.6)	4 (4.8)
Diabetes, %	15 (23.4)	19 (22.9)	0.993
Hypertension, %	19 (29.7)	35 (42.2)	0.096
Atrial fibrillation/flutter, %	11 (17.2)	19 (22.9)	0.359
Ventricular arrhythmias, %	21 (32.8)	25 (30.1)	0.791
Left bundle‐branch block, %	5 (7.8)	9 (10.8)	0.508
Chronic kidney disease, %	9 (14.1)	14 (16.9)	0.603
Estimated glomerular filtration rate, mL/min/1.73m2	98.3 (86.9–109.2)	92.4 (71.1–103.9)	0.077
N‐terminal pro‐brain natriuretic peptide, pg/mL	1665.0 (635.5–3058.0)	634.0 (334.5–1569.5)	0.001*
Echocardiography
LVEF, %	30.4 ± 7.2	36.2 ± 9.1	<0.001*
LV end‐diastolic diameter, mm	65.8 ± 8.9	60.9 ± 8.2	0.001*
Left atrial diameter, mm	46.4 ± 7.2	42.7 ± 6.6	0.002*
LV mass index, g/m2	109.6 (97.9–134.5)	107.3 (89.6–133.1)	0.388
Interventricular septal thickness, mm	8.0 (7.0–9.0)	8.0 (7.5–10.0)	0.097
LV posterior wall thickness, mm	8.0 (7.0–9.0)	8.0 (7.5–9.0)	0.352
Early mitral inflow to mitral annular early diastolic velocity ratio	14.5 (11.0–20.0)	12.0 (9.0–16.5)	0.007*
Pulmonary artery systolic pressure, mm Hg	32.0 (23.0–45.0)	28.0 (22.0–33.5)	0.030*
Tricuspid annular plane systolic excursion, mm	15.5 (13.5–18.5)	18.0 (16.5–21.0)	<0.001*
Mitral regurgitation [moderate–severe (%)]	30 (46.9)	31 (37.3)	0.293
Tricuspid regurgitation [moderate–severe (%)]	18 (28.1)	14 (16.9)	<0.001*
Cardiac magnetic resonance
LVEF, %	23.2 (18.9–28.4)	34.3 (29.9–42.2)	<0.001*
LVEDVI, mL/m2	148.3 (117.2–172.6)	121.3 (101.2–149.5)	0.002*
RVEF, %	29.7 (24.2–36.6)	49.5 (44.6–58.8)	<0.001*
RVEDVI, mL/m2	93.2 (75.3–113.5)	73.0 (58.4–89.2)	<0.001*
Midwall late gadolinium enhancement, %	41 (64.1)	45 (54.2)	0.284
Therapy
Angiotensin‐converting enzyme inhibitor/angiotensin receptor blocker/angiotensin receptor neprilysin inhibitor, %	58 (90.6)	73 (88.0)	0.862
β‐blocker, %	60 (93.8)	73 (88.0)	0.345
Spironolactone, %	60 (93.8)	77 (92.8)	0.764
Sodium‐glucose cotransporter 2 inhibitor, %	55 (85.9)	64 (77.1)	0.246
Diuretics, %	49 (76.6)	41 (49.4)	0.001*
Digoxin, %	20 (31.3)	5 (6.0)	<0.001*

BP indicates blood pressure; EDVI, end‐diastolic volume index; EF, ejection fraction; LV, left ventricular; LVEDVI, left ventricular end‐diastolic volume index; LVEF, left ventricular ejection fraction; and RV, right ventricular; RVEDVI, right ventricular end‐diastolic volume index; RVEF, right ventricular ejection fraction; SGLT2i, sodium‐glucose cotransporter‐2 inhibitors; TAPSE, tricuspid annular plane systolic excursion; and TR, tricuspid regurgitation.

^* P<0.05.

Urinary Proteomic Profiling and Differential Protein Expression Analysis

Using data‐independent acquisition‐based mass spectrometry, 3579 urinary proteins were initially identified from 147 quality‐controlled urine samples. After removing proteins missing in >50% of samples in both the RVD and non‐RVD groups, 2094 proteins remained. Further filtering based on a coefficient of variation >0.5 reduced the data set to 1643 proteins. Differential expression analysis then identified 156 proteins with a false discovery rate <0.05. Of these, 46 proteins met the criteria of fold change >1.5 or <0.67, including 23 upregulated and 23 downregulated proteins in the group with RVD (Table S1 and Figure 1).

Figure 1. Volcano plot depicting the differentially expressed proteins between RVD and non‐RVD.

Proteins upregulated in RVD vs non‐RVD are shown in red and downregulated in blue. Dotted lines indicate the thresholds for a 1.5‐fold change in expression (vertical) and statistical significance (P _adjusted=0.05; horizontal). Gene names are shown instead of the full‐length protein names. RVD indicates right ventricular dysfunction.

Correlations Between Differential Proteins and Clinical or Imaging Parameters

In patients with iDCM, urinary proteomic analysis identified multiple proteins significantly associated with right ventricular structure, function, and clinical status (Figure 2). GSK3A (glycogen synthase kinase 3 alpha) demonstrated the most consistent and extensive associations, correlating with better RV performance—reflected by higher RVEF and TAPSE—and with lower markers of volume overload and clinical severity, including RV end‐diastolic volume index, pulmonary artery systolic pressure, tricuspid regurgitation, New York Heart Association class, and NT‐proBNP. Similarly, ABCA2 (ATP‐binding cassette subfamily A member 2) was positively associated with RVEF and inversely related to RV end‐diastolic volume index, pulmonary artery systolic pressure, tricuspid regurgitation, and NT‐proBNP. IGHV5‐10‐1 (immunoglobulin heavy variable 5 to 10) also showed favorable correlations with both RVEF and TAPSE, indicating a link with preserved RV systolic function. In contrast, proteins such as RARRES1 (retinoic acid receptor responder protein 1) and WDR91 (WD repeat domain 91) were closely linked to adverse RV remodeling and elevated filling pressures, evidenced by negative correlations with RVEF and positive associations with RV end‐diastolic volume index, pulmonary artery systolic pressure, and tricuspid regurgitation. MVB12B (multivesicular body subunit 12B) was likewise inversely related to RVEF. Detailed correlation results are provided in Table S2.

Figure 2. Correlations of differential urinary proteins with clinical and imaging parameters in patients with iDCM.

Correlogram showing pairwise Pearson correlations between the abundance levels of differentially expressed urinary proteins (columns) and key clinical/imaging traits (rows) in patients with iDCM. Only significant correlations are marked with asterisks (*P<0.05; **P<0.01; ***P<0.001). Cell colors represent the strength and direction of Pearson’s correlation coefficient (blue=negative, red=positive), and only significance levels are overlaid. Protein names are ordered based on hierarchical clustering. E/e indicates early mitral inflow to mitral annular early diastolic velocity ratio; eGFR, estimated glomerular filtration rate; iDCM, idiopathic dilated cardiomyopathy; IVS, interventricular septal thickness; LAD, left atrial diameter; LGE, late gadolinium enhancement; LVEDVI, left ventricular end‐diastolic volume index; LVEF, left ventricular ejection fraction; LVMI, left ventricular mass index; LVPW, left ventricular posterior wall thickness; MR, mitral regurgitation; NP, N‐terminal pro‐B‐type natriuretic peptide; NYHA, New York Heart Association class; PASP, pulmonary artery systolic pressure; RVEDVI, right ventricular end‐diastolic volume index; RVEF, right ventricular ejection fraction; TAPSE, tricuspid annular plane systolic excursion; and TR, tricuspid regurgitation.

Functional and Pathway Enrichment Analysis of Differentially Expressed Urinary Proteins

Gene Ontology enrichment analysis revealed that differentially expressed proteins associated with RVD are predominantly involved in antigen processing and presentation, metabolic processes (including glycolysis, redox regulation, and lipid metabolism), cytoskeletal dynamics, exosome‐mediated transport, and organelle function (Figure 3A). These proteins are mainly localized in the cytoplasm, organelles, and exosomes, indicating key roles in cell signaling, metabolic regulation, membrane trafficking, and the maintenance of organelle homeostasis. Kyoto Encyclopedia of Genes and Genomes pathway analysis further demonstrated significant enrichment of these differentially expressed proteins in metabolic pathways (eg, glycolysis/gluconeogenesis), immune‐related pathways (such as antigen processing and presentation, and type 1 diabetes), endocytosis, proteasome function, and RNA degradation (Figure 3B). Notably, metabolic dysregulation, immune imbalance, impaired proteostasis, and altered cell adhesion likely represent core mechanisms contributing to the pathogenesis of RVD. In summary, integrated Gene Ontology and Kyoto Encyclopedia of Genes and Genomes analyses suggest that RVD is characterized by profound disturbances in metabolism, immune regulation, protein homeostasis, and membrane dynamics, which may collectively drive disease onset and progression.

Figure 3. Functional enrichment analysis of differentially expressed proteins associated with right ventricular dysfunction.

A, Gene Ontology enrichment analysis reveals significant enrichment in biological processes related to antigen processing, metabolism, and cell signaling, cellular components such as organelles and exosomes, and molecular functions including transport and metabolic regulation. B, Kyoto Encyclopedia of Genes and Genomes pathway analysis shows significant enrichment in metabolic pathways, immune‐related pathways, endocytosis, and proteasome function. BP indicates biological processes; CC, cellular components; GO, Gene Ontology; MF, molecular functions; and NAD, nicotinamide adenine dinucleotide.

Identification of Candidate Urinary Biomarkers for RVD Using Multivariable Modeling

We performed binary multivariable logistic regression analysis on 46 differentially expressed proteins, adjusting for clinical confounders including age, sex, body mass index, LVEF, NT‐proBNP, and systolic blood pressure for each protein individually. As shown in Table 2, after adjustment, 15 proteins were independently associated with RVD. Specifically, MVB12B, WDR91, PAXX (PAXX non‐homologous end joining factor), PGP (phosphoglycolate phosphatase), RARRES1, B4GALT4 (beta‐1,4‐galactosyltransferase 4), CUTA (copper transport accessory protein), KLK11 (kallikrein‐11), and JAG2 (Notch ligand jagged 2) were significantly upregulated in the RVD group. In contrast, IGHV5‐10‐1, ARPC2 (actin‐related protein 2/2B), GSK3A, KRT80 (keratin 80), and ABCA2 were significantly downregulated.

Table 2.

Multivariate Logistic Regression Identifies Differentially Expressed Urinary Proteins Associated With RVD in iDCM

Protein	Gene	Beta	P value	FDR
A0A0J9YXX1	IGHV5‐10‐1	−1.81	2.68E‐06	3.26E‐05
O15144	ARPC2	−2.53	3.57E‐06	3.26E‐05
P49840	GSK3A	−2.20	2.07E‐06	3.26E‐05
Q6KB66	KRT80	−2.28	2.41E‐06	3.26E‐05
Q9BZC7	ABCA2	−1.71	3.81E‐06	3.26E‐05
Q9H7P6	MVB12B	1.63	4.26E‐06	3.26E‐05
A4D1P6	WDR91	2.36	5.92E‐06	3.89E‐05
Q9BUH6	PAXX	1.93	8.05E‐06	4.63E‐05
A6NDG6	PGP	1.47	2.63E‐05	1.30E‐04
P49788	RARRES1	1.98	2.82E‐05	1.30E‐04
O60513	B4GALT4	1.33	4.17E‐05	1.74E‐04
O60888	CUTA	0.74	7.15E‐03	2.74E‐02
Q9UBX7	KLK11	0.61	9.47E‐03	3.11E‐02
Q9Y219	JAG2	0.78	9.20E‐03	3.11E‐02
Q06323	PSME1	−0.68	1.05E‐02	3.21E‐02

Adjusted for age, sex, body mass index, left ventricular ejection fraction, N‐terminal pro–brain natriuretic peptide, and systolic blood pressure. FDR indicates false discovery rate; iDCM, idiopathic dilated cardiomyopathy; and RVD, right ventricular dysfunction.

Diagnostic Utility of a 3‐Protein Panel for RVD Stratification in iDCM

To develop a robust diagnostic model, all 147 patients with iDCM were randomly divided into a training set (n=109) and a validation set (n=38), with stratification for age and sex. Following stratification, the training set included 48 patients with RVD and 61 without RVD, and the validation set comprised 16 patients with RVD and 22 without RVD. Least absolute shrinkage and selection operator regression identified 11 out of 15 candidate proteins at the λ.1se threshold, excluding PSME1 (proteasome activator complex subunit 1), JAG2, KLK11, and CUTA (Figures 4A and 4B). Spearman’s correlation analysis of these 11 differentially expressed proteins showed no strong collinearity (ρ<0.6), indicating their relative independence in the predictive model (Figure 4C). Among the retained proteins, GSK3A, MVB12B, and RARRES1 were ultimately selected for the final model based on their biological and diagnostic significance.

Figure 4. Feature selection of diagnostic biomarkers for RVD in patients with DCM.

LASSO regression was performed on the training set to identify informative protein features. A total of 11 proteins were retained using the λ.1se criterion. Among these, GSK3A, MVB12B, and RARRES1 were selected for the final diagnostic model based on both statistical performance and biological relevance. A, LASSO path plot showing the coefficient profiles of 15 candidate proteins across a range of log(λ) values. The 2 vertical dashed lines indicate λ.min (yielding the minimum cross‐validated error) and λ.1se (yielding the most parsimonious model within 1 SE of λ.min); the latter was selected for final model construction. B, Tenfold cross‐validation plot displaying binomial deviance vs log(λ). The same 2 vertical lines denote λ.min and λ.1se, with λ.1se chosen to balance model simplicity and performance. C, Correlation matrix of the 11 LASSO‐selected DEPs, with color intensity representing the strength of the Spearman’s correlation (red for positive, blue for negative). DCM indicates dilated cardiomyopathy; DEP, differentially expressed proteins; GSK3A, glycogen synthase kinase 3 alpha; LASSO, least absolute shrinkage and selection operator; MVB12B, multivesicular body subunit 12B; RARRES1, retinoic acid receptor responder protein 1; and RVD, right ventricular dysfunction.

As shown in Figure 5 and summarized in Table S3, the diagnostic performance of RARRES1, GSK3A, and MVB12B was first evaluated in the training data set. Their AUCs were 0.797 (95% CI, 0.712–0.882), 0.801 (95% CI, 0.705–0.896), and 0.747 (95% CI, 0.652–0.841), respectively. The combined model integrating all 3 proteins significantly outperformed the individual markers, NT‐proBNP, and TAPSE, achieving an AUC of 0.946 (95% CI, 0.908–0.983), with a sensitivity of 0.854, specificity of 0.918, positive predictive value of 0.891, and negative predictive value of 0.889. In contrast, NT‐proBNP alone showed poor performance (AUC, 0.676 [95% CI, 0.574–0.777]), TAPSE performed slightly better (AUC, 0.702 [95% CI, 0.602–0.802]), and the combination of NT‐proBNP and TAPSE modestly improved diagnostic accuracy (AUC, 0.716 [95% CI, 0.617–0.815]), with a sensitivity of 0.708, specificity of 0.639, positive predictive value of 0.607, and negative predictive value of 0.736; however, all 3 models were clearly inferior to the combined protein model. In the validation data set, the combined model again demonstrated superior diagnostic performance, yielding an AUC of 0.946 (95% CI, 0.878–1.000), notably higher than NT‐proBNP (0.662), TAPSE (0.636), and their combination (0.668). To further confirm these findings, ELISA assays were conducted for RARRES1 and GSK3A, both of which have commercially available kits, in the same patient cohort. The results were consistent with the proteomic data, further supporting their diagnostic utility (Table S4 and Figure S2).

Figure 5. Diagnostic performance of the three‐protein model and comparison with NT‐proBNP, TAPSE, and their combination in training and validation data sets.

The diagnostic model integrating GSK3A, MVB12B, and RARRES1 demonstrated superior performance compared with individual proteins, NT‐proBNP, TAPSE, and the NT‐proBNP + TAPSE combination, achieving high AUC, sensitivity, and specificity in both training and validation sets. A, ROC curves comparing individual proteins, the 3‐protein model, NT‐proBNP, TAPSE, and the NT‐proBNP + TAPSE combination in the training set. B, ROC curves comparing the same models in the validation set. AUC indicates area under the curve; GSK3A, glycogen synthase kinase 3 alpha; MVB12B, multivesicular body subunit 12B; NT‐proBNP, N‐terminal pro‐brain natriuretic peptide; RARRES1, retinoic acid receptor responder protein 1; ROC, receiver operating characteristics; and TAPSE, tricuspid annular plane systolic excursion.

Prognostic Implications of the 3‐Protein Panel in Patients With iDCM

Among 147 patients with iDCM, 37 experienced composite end point events during a median follow‐up of 19.3 months (interquartile range, 10.9–28.1 months). A multivariable Cox regression model incorporating a 3‐biomarker urinary proteomic panel (GSK3A, MVB12B, and RARRES1), adjusted for age, sex, and NT‐proBNP, was used to calculate individual risk scores. Based on the median value of these risk scores (cutoff=0.86), patients were stratified into high‐ and low‐risk groups. Kaplan–Meier survival analysis demonstrated that patients in the high‐risk group had a significantly increased risk of experiencing the composite end point compared with those in the low‐risk group (HR, 3.24; 95% CI, 1.56–6.71, P=0.002; E‐value=5.92; Figure 6A). The model achieved a C‐index of 0.63 (95% CI, 0.57–0.80). When the proteins were examined individually, only urinary GSK3A remained significantly associated with the end point (P=0.001), emphasizing the added value of the integrated panel (Figure S3).

Figure 6. Kaplan–Meier survival curves based on a Cox regression‐derived risk score incorporating 3 urinary biomarkers (GSK3A, MVB12B, and RARRES1) for predicting adverse outcomes in iDCM.

Patients were stratified into high‐ and low‐risk groups using the median risk score. A, For the composite end point of cardiovascular death, LVAD implantation, heart transplantation and heart failure rehospitalization (cutoff=0.86), the high‐risk group exhibited a significantly higher risk compared with the low‐risk group (HR, 3.24 [95% CI, 1.56–6.71], P=0.002). B, For cardiovascular death, LVAD implantation, or heart transplantation (cutoff=0.872), the high‐risk group showed significantly worse outcomes (HR, 4.17 [95% CI, 1.17–14.79], P=0.027). C, For heart failure hospitalization (cutoff=0.973), the high‐risk group also exhibited a significantly increased risk (HR, 2.56 [95% CI, 1.03–6.34], P=0.043). GSK3A indicates glycogen synthase kinase 3 alpha; HR, hazard ratio; iDCM, idiopathic dilated cardiomyopathy; LVAD, left ventricular assist device; MVB12B, multivesicular body subunit 12B; and RARRES1, retinoic acid receptor responder protein 1.

We next dissected the composite into its 2 clinical components. For the hard‐events cluster (cardiovascular death, LV assist device implantation or cardiac transplantation) the high‐risk group had an HR of 4.17 (95% CI, 1.17–14.79, P=0.027; Figure 6B) and a C‐index of 0.72 (95% CI, 0.65–0.90). For heart‐failure hospitalization, the corresponding values were HR 2.56 (95% CI, 1.03–6.34, P=0.043) and C‐index 0.66 (95% CI, 0.59–0.86; Figure 6C), confirming consistent prognostic performance across distinct outcomes.

Further mediation analysis of the composite end point indicated that baseline RVD explained part of the panel’s prognostic effect, yet a proportion of risk prediction was independent of RVD (Figure S4).

DISCUSSION

RVD is a prevalent and prognostically significant phenotype in iDCM, associated with advanced biventricular remodeling and adverse outcomes. ¹⁸ , ¹⁹ In this study, using CMR‐defined assessment of RVD, urinary proteomic analysis identified 46 differentially expressed proteins, implicating pathways related to metabolic dysregulation, protein homeostasis, and immune signaling. Based on these findings, we developed a 3‐protein urinary panel comprising RARRES1, MVB12B, and GSK3A, which demonstrated robust discriminatory performance and significantly outperformed NT‐proBNP and TAPSE in both training and validation cohorts. Furthermore, this panel enabled risk stratification through a proteomics‐derived score, with patients in the high‐risk group exhibiting a substantially increased risk of adverse outcomes compared with those in the low‐risk group. Collectively, these findings highlight the potential of urinary proteomics as a noninvasive tool for enhanced phenotyping and risk stratification in iDCM.

Proposed Pathophysiological Mechanisms of Urinary Proteomic Biomarkers in Right Ventricular Dysfunction

Previous studies have primarily highlighted extracellular and systemic mechanisms in the pathogenesis of RVD, particularly in the context of heart failure. For instance, tissue proteomic analyses identified matrix proteins such as fibulin‐5 and fibromodulin as structural hallmarks of adverse RV remodeling, ²⁰ whereas plasma‐based profiling revealed elevated levels of FGF‐23 (fibroblast growth factor 23), implicating systemic phosphate imbalance and oxidative stress. ²¹ In contrast, our urinary proteomic analysis uncovered a distinct pattern centered on intracellular processes. The differentially expressed proteins associated with CMR‐defined RVD were enriched in pathways related to proteasomal degradation, immune signaling, mitochondrial homeostasis, and metabolic regulation. These findings suggest that cell‐autonomous stress responses—rather than solely extracellular matrix remodeling or systemic hormonal changes—may represent key contributors to RV maladaptation in iDCM. A recent transcriptomic study identified the Wipi1–Hspb6–Map4 axis as a molecular hallmark of right ventricular failure, implicating dysregulated autophagy and mitochondrial oxidative stress, which is consistent with the core biological themes of proteostasis, immune dysregulation, and mitochondrial dysfunction revealed by our proteomic data. ²² Together, these findings underscore a multifaceted and RV‐specific pathophysiology of RVD and provide a rationale for therapeutic strategies targeting these intracellular mechanisms.

Clinical Implications of the 3‐Protein Urinary Biomarker Panel for DCM‐RVD

We identified a urinary protein biomarker panel comprising GSK3A, RARRES1, and MVB12B, which not only distinguishes patients with iDCM with concomitant RVD with high specificity but also serves as a valuable tool for prognostic stratification. Each protein reflects distinct yet interconnected pathophysiological processes that contribute to disease progression.

GSK‐3α is a serine/threonine protein kinase with distinct cell‐specific and context‐dependent functions in the heart. Early global knockout or constitutively active knock‐in models showed that both sustained loss and activation of GSK‐3α worsen LV dysfunction and remodeling under pressure overload or ischemia, ²³ , ²⁴ suggesting a dual regulatory role in cardiac homeostasis. In contrast, cardiomyocyte‐specific knockout models revealed no effect on baseline cardiac function but demonstrated reduced cell death, attenuated remodeling, and improved function following myocardial infarction or pressure overload. ²⁵ , ²⁶ These benefits are linked to increased cyclin E1 expression and delayed mitochondrial permeability transition pore opening, enhancing mitochondrial function. ²⁶ In fibroblasts, GSK‐3α deletion suppresses fibrosis by downregulating RAF–MEK–ERK signaling, thereby inhibiting myofibroblast activation and extracellular matrix production. ²⁷ Together, these findings suggest that under stress, GSK‐3α promotes apoptosis, fibrosis, and mitochondrial dysfunction in both cardiomyocytes and fibroblasts. However, these mechanistic insights are primarily derived from studies in the LV and whether similar roles exist in the RV remains unknown. Given our findings that GSK‐3α was significantly downregulated in urinary proteomics of iDCM with RVD and was strongly associated with adverse outcomes, it is plausible that this protein may play an important role in RV remodeling and the associated pathophysiological changes. The specific contribution of RV versus LV remodeling to urinary GSK‐3α excretion, as well as its temporal dynamics during disease progression, warrants further investigation.

In patients with iDCM and concomitant RVD, we found for the first time that urinary RARRES1 is markedly elevated. This rise likely reflects integrated cardio‐renal crosstalk rather than a single‐organ signal. Three non–mutually exclusive pathways appear contributory: (1) mechanical overload in the left ventricle upregulates RARRES1 in cardiac endothelial cells, triggering NF‐κB (nuclear factor kappa B)/Axl signaling, inflammatory cytokine release and fibroblast activation that collectively drive ventricular fibrosis and remodeling ²⁸ ; (2) soluble RARRES1 released by renal podocytes can enter the systemic circulation, disturb tubular lipid metabolism and generate oxidative stress, leading to lipotoxic intermediates that aggravate myocardial lipid accumulation and mitochondrial dysfunction ²⁹ , ³⁰ ; and (3) intrinsic overexpression of RARRES1 within cardiomyocytes itself impairs fatty‐acid oxidation, compromising energy supply under high workload conditions. ³⁰ , ³¹ Collectively, urinary RARRES1 indexes a convergent inflammatory–metabolic axis operational across the heart–kidney interface in RVD. Dissecting the relative contribution of each cellular reservoir and its chamber specificity is now required to refine RARRES1 from a composite risk signal to a precision tool capable of discriminating cardiac from renal injury.

MVB12B is predominantly expressed in vascular endothelial cells and cardiac fibroblasts within the cardiovascular system. ³² Although it was not found to be independently prognostic in our survival analyses, MVB12B likely plays a mechanistic role in cellular adaptation to myocardial stress. As a key component of the endosomal sorting complex required for transport, MVB12B appears to be upregulated as a compensatory mechanism to promote membrane repair and maintain protein homeostasis under stress conditions. ³³ However, sustained overexpression may contribute to disease progression by activating NF‐κB–mediated fibrosis, ³⁴ , ³⁵ impairing autophagy‐lysosome function, ³³ and amplifying TGF‐β (transforming growth factor beta) signaling, contributing to maladaptive remodeling. ³⁶ Notably, most functional studies on MVB12B have been conducted in non‐RV‐specific contexts, such as vascular models or LV‐based disease models, and whether similar mechanisms are operative in RV pathophysiology remains to be established. Nonetheless, elevated urinary MVB12B within a multiprotein panel may reflect broader pathophysiological processes such as cardio–renal dysfunction, systemic inflammation, and hemodynamic deterioration. ²⁰ , ³⁷ Its prognostic value in the panel likely relates to involvement in signaling pathways associated with increased RV afterload, ³⁸ biventricular decompensation, ³⁹ and maladaptive interorgan crosstalk. ⁴⁰ , ⁴¹

Together, these 3 biomarkers provide a multifaceted perspective on the pathophysiological processes underlying iDCM‐RVD—GSK‐3α as a stress‐responsive kinase linked to apoptosis and fibrosis, with reduced urinary levels potentially reflecting disease progression or systemic alterations, RARRES1 as a modulator of inflammation and metabolic dysregulation, and MVB12B as a marker of stress response and cardiac remodeling. Their combined assessment as a biomarker panel presents a promising, noninvasive approach for disease monitoring and risk stratification, warranting further validation in clinical studies.

Mediation analysis suggested that the prognostic effect of the 3‐protein urinary panel was only partly explained by baseline RVD, indicating that the panel may provide incremental prognostic information beyond RVD. This finding raises the possibility that urinary proteomics reflects additional biological processes, such as systemic inflammation, myocardial remodeling, or extracardiac organ involvement, which are not fully captured by imaging‐derived RVD assessment. Although these results highlight the potential complementary role of urinary biomarkers in risk stratification for iDCM, further studies are needed to confirm the underlying mechanisms and temporal dynamics.

Limitations

This study has certain limitations. First, the high cost and technical complexity of CMR limited our sample size in this single‐center study. Second, although we employed an internal validation strategy by dividing the data set into training and validation sets, we did not include an independent external validation cohort. Third, although urinary proteomics offers a noninvasive approach to biomarker discovery, its applicability is limited in patients with severely impaired renal function (estimated glomerular filtration rate <30 mL/min/1.73 m²), hematuria, or proteinuria. Fourth, the mediation analysis was exploratory, and the nonsignificant indirect effect with wide CIs highlights the need for larger, well‐powered studies to clarify the potential mediating role of RVD. Future research should also incorporate multicenter validation, histopathological correlation, and functional assays to further elucidate the pathophysiological significance of urinary biomarkers in iDCM with RVD.

Conclusions

This study identified a 3‐biomarker urinary proteomic panel (RARRES1, MVB12B, and GSK3A) that reliably differentiates RVD from non‐RVD in patients with iDCM. The biomarker panel exhibited superior diagnostic sensitivity and specificity for RVD relative to conventional measures (NT‐proBNP, TAPSE) and demonstrated significant prognostic utility for adverse clinical outcomes. These findings highlight the potential of urinary proteomics as a noninvasive, effective tool for both diagnosing and predicting outcomes in patients with iDCM with RVD, offering significant promise for improved clinical management and patient care.

Sources of Funding

This work was supported by the National Natural Science Foundation of China (82170397), National High Level Hospital Clinical Research Funding (2025‐PUMCH‐C‐035) and Peking Union Medical College Hospital Talent Cultivation Program (Category C) No. UBJ 06514.

Disclosures

None.

Supporting information

Data S1

References [42–45]