Unsupervised clustering of RNA-Seq data to identify molecular subtypes of sarcomas.

Abstract

11547 Background: Sarcomas are a rare and heterogeneous group of malignancies, which include over 100 histological subtypes. Their histologic diversity present significant diagnostic and therapeutic challenges. This study aimed to define clinically relevant transcriptional subtypes using RNA-Sequencing (RNA-seq) data, and to further characterize the immune and metabolic microenvironment. Methods: This study uses data from the “Cohort to Augment the Understanding of Sarcoma Survivorship Across the Lifespan” (CAUSAL) study, which has enrolled sarcoma patients, treated at VUMC, from 2002-2026 for investigation of health outcomes. We analyzed RNA-seq data from tumor samples of 430 participants representing 42 unique histological sarcoma subtypes. We utilized sparse K-means clustering to identify stable subtypes and developed a classifier using a 310-gene panel. The immune microenvironment was characterized by applying CIBERSORTx deconvolution. Sarcoma subtype reproducibility and prognostic validity were evaluated using the Cancer Genomic Atlas data. Results: Among 430 samples, consensus matrices and diminishing gains in Cumulative Distribution Function (CDF) behavior yielded six distinct sarcoma molecular subtypes (SMS), each demonstrating strong correlation with known sarcoma histologies. SMS1 (labeled as “Cytotoxic-Adaptive”, n=30) and SMS6 (“Interferon-Activated”, n=97) were characterized by robust immune infiltration and interferon signaling, suggesting a potential sensitivity to immune checkpoint blockade. SMS2 (“Myeloid-Dominant”, n=93) primarily grouped undifferentiated pleomorphic sarcoma, giant cell sarcoma, and myxofibrosarcoma, and exhibited an immune-suppressive, macrophage-rich phenotype. SMS3 (“Early-Onset”, n=58) mostly grouped Ewing and synovial sarcoma and comprised mostly fusion-driven tumors with an immune-desert landscape. SMS4 (“EMT-Stromal”, n=107) mostly comprised desmoid and leiomyosarcoma and was defined by epithelial-mesenchymal transition and fibrosis, correlating with the poorest 10-year overall survival ( p=0.022 ). SMS5 (“Metabolic”, n=45) consisted exclusively of Gastrointestinal Stromal Tumors, with distinct metabolic programming and favorable prognosis. Conclusions: Each SMS transcends histological boundaries, grouping tumors based on shared immune, stromal, and metabolic features. This classification highlights distinct vulnerabilities in the tumor microenvironment, offering a functional framework to guide future precision treatment interventions, such as immunotherapy for SMS1/SMS6 and stromal-targeting agents for SMS4.