Abstract
Hypertrophic cardiomyopathy (HCM) is characterized by significant phenotypic heterogeneity, complicating risk assessment and clinical management. This study introduces a data-driven taxonomy of HCM by analyzing the three-dimensional (3D) structure of the left ventricle using machine learning. By mapping the relationship between cardiac morphology, genetic variants, and clinical outcomes, the researchers identified four distinct phenotypic branches. The findings reveal that pathogenic variants are associated with specific structural patterns and a significantly higher mortality risk, providing a framework for more personalized approaches to HCM.
Introduction
Hypertrophic cardiomyopathy (HCM) is a primary myocardial disorder and a significant cause of sudden cardiac death and heart failure. Despite its clinical importance, HCM is notoriously heterogeneous; patients exhibit a wide range of heart wall thickening patterns, genetic backgrounds, and clinical trajectories. Traditional classification systems rely on subjective visual assessments of two-dimensional images, often categorizing morphology into broad types like "sigmoid" or "reverse curve." However, these labels frequently fail to reflect the underlying genetic drivers or predict long-term outcomes accurately.
To address this, a recent study utilized machine learning to develop a systematic, 3D taxonomy of HCM. By integrating high-resolution imaging with genetic data, the researchers aimed to bridge the gap between a patient's physical heart structure (phenotype) and their genetic risk (genotype).
Methodology
The study analyzed a discovery cohort of 436 HCM patients from the NIHR Royal Brompton Hospital Cardiovascular Biobank, a validation cohort of 60 patients from Singapore, and a reference population of 16,691 individuals from the UK Biobank.
The researchers employed a multi-step computational pipeline:
- 3D Cardiac Modeling: Deep learning was used to segment Cardiac Magnetic Resonance (CMR) images, creating high-resolution 3D models of the left ventricle.
- Unsupervised Learning: A technique called Discriminative Dimensionality Reduction via Learning a Tree (DDRTree) was applied to project these complex 3D shapes into a 2D "phenotypic tree." This approach allows for the visualization of how different heart shapes transition along a continuum.
- Genomic Integration: Patients were stratified based on the presence of rare pathogenic or likely pathogenic (P/LP) variants. Additionally, Polygenic Risk Scores (PGS) were calculated to assess the impact of common genetic variations.
- Survival Analysis: Mortality data were projected onto the phenotypic tree to identify high-risk structural patterns.
A New Taxonomy of HCM
The unsupervised model identified four primary branches of disease expression, representing distinct structural trajectories:
- Non-sarcomeric Hypertrophy: Characterized by thickening in the mid-to-apical regions, often associated with co-existing hypertension.
- Diffuse and Basal Asymmetric Hypertrophy: Associated with left ventricular outflow tract obstruction (LVOTO).
- Isolated Basal Hypertrophy: A milder form of thickening, potentially related to aging.
- Milder Non-obstructive Hypertrophy: This branch was highly enriched for familial sarcomeric HCM, showing a strong association with high-risk genetic mutations.
Key Findings
The Genotype-Phenotype Paradox
One of the study's most striking findings was that carriers of P/LP variants (SARC-P/LP) actually exhibited lower overall left ventricular mass compared to genotype-negative individuals. However, their hypertrophy was more localized, specifically targeting the basal septum. This suggests that the pattern of remodeling, rather than the total amount of thickened muscle, is a more critical indicator of genetic risk.
Mortality and Genetic Risk
The clinical implications of this taxonomy are significant. Patients carrying P/LP variants had a 2.66-fold higher risk of death compared to genotype-negative patients. The median lifespan for these genetic carriers was 67 years, compared to 76 years for those without such variants. Branch 4 of the phenotypic tree was identified as the most "genetic" cluster, with an odds ratio of 2.18 for carrying a pathogenic variant.
Generalizability and Polygenic Influence
The model's robustness was confirmed using the Singaporean cohort, where it achieved a trustworthiness score of 0.86–0.88. This indicates that the 3D structural patterns of HCM are consistent across different ancestries. Furthermore, the study highlighted that even in the absence of a single rare mutation, common genetic variations (polygenic risk) significantly influence the severity and shape of cardiac remodeling.
Conclusion
This research moves HCM classification from subjective categories toward a continuous, data-driven framework. By mapping the 3D structure of the heart onto a phenotypic tree, clinicians can better understand the relationship between morphology, genetics, and prognosis. This taxonomy provides a foundation for future "multi-omic" integration and the development of clinical tools that can provide personalized risk profiles based on standard cardiac imaging.
Summary updated