Phenotype-driven clustering of ocular manifestations reveals prognostic and genetic heterogeneity in Marfan syndrome: a real-world longitudinal cohort study.

backgroundMarfan syndrome (MFS) patients exhibit diverse ocular phenotypes with distinct genetic and prognostic characteristics, posing significant challenges for clinical management, especially in pediatric patients. This study aimed to identify clinically meaningful subtypes of MFS based on ocular phenotypes using unsupervised clustering and to explore their associated genetic characteristics and prognostic outcomes.

methodsThis real-world, prospective longitudinal cohort study included MFS patients with confirmed pathogenic or likely pathogenic FBN1 mutations and ectopia lentis (EL). Comprehensive ocular biometric and demographic data were collected and analyzed. Unsupervised hierarchical cluster analysis was performed on phenotypic variables using agglomerative hierarchical clustering. FBN1 variants were classified based on molecular characteristics. Statistical analyses were conducted to evaluate inter-cluster differences in genotype, visual prognosis, axial length (AL) growth, refractive growth, and postoperative complications. A web-based clustering tool was developed for clinical application.

resultsAmong 476 patients in the cohort, Z‑AL, AL, log₁₀(Age + 0.6), corneal center thickness, and corneal curvature radius were included in the cluster analysis, which revealed four distinct phenotypic subgroups among patients with MFS and EL. Cluster A primarily consisted of pediatric patients with ocular parameters close to the normative range. Cluster B included pediatric patients characterized by long AL or thinner/flatter corneas. Cluster C was composed mainly of adolescents and adults with moderate axial elongation and demonstrated the best postoperative visual acuity. Cluster D encompassed patients with extremely long AL and a higher incidence of posterior capsular opacification. Significant inter-cluster differences were observed in AL, age distribution, and corneal biometry. Genotypically, Cluster A was enriched with dominant-negative (DN) variants affecting non-critical residues, while Cluster D showed a greater proportion of haploinsufficiency variants and DN mutations located within the neonatal and TGF-β–regulating regions of FBN1. Visual prognosis and the occurrence of posterior capsular opacification also varied significantly among clusters, with Cluster C exhibiting the most favorable outcomes.

conclusionPhenotype-driven unsupervised clustering revealed four clinically and genetically distinct subgroups of MFS with EL. These subtypes exhibit differential prognosis and genetic architecture, suggesting that data-driven clustering can augment traditional genotype-based approaches in guiding personalized clinical management.