Portal de Periódicos da CAPES

New soft tissue sarcoma (STS) transcriptomic clusters to unveil STS subsets with unique biological characteristics and refine the accuracy of overall survival (OS) prediction.

2024; Lippincott Williams & Wilkins; Volume: 42; Issue: 16_suppl Linguagem: Inglês

10.1200/jco.2024.42.16_suppl.11545

1527-7755

Miguel Esperança‐Martins, Hugo Vasques, Manuel Sokolov Ravasqueira, Maria Manuel Lemos, Filipa Fonseca, Diogo Coutinho, Jorge Antonio López, Richard S.P. Huang, Sérgio Dias, Lina M. Gallego-Paez, Luís Costa, Nuno Abecasis, Emanuel Gonçalves, Isabel Fernandes,

Sarcoma Diagnosis and Treatment

11545 Background: The currently used histopathological classification (HPC) is eminently morphological and error prone, fragmenting STS in 80 subtypes. Adding fragmentation to STS’s rarity limits the deployment of pre-clinical and clinical research, hampering drug discovery and development. Finding new classification tools that better reflect STS’s biology and that show predictive and prognostic value is an unmet need. Methods: FFPE samples of 25 dedifferentiated liposarcomas (DDLPS), 25 leiomyosarcomas (LMS), and 52 undifferentiated pleomorphic sarcomas (UPS) were analyzed along with clinical data from the patients. DNA and RNA were isolated. RNA sequencing was performed using the FoundationOne RNA assay. Normalized gene expression (GE) data (RUO) was analyzed using unsupervised machine learning and consensus clustering. 4 transcriptomic clusters (TC) were identified. A subsequent differential GE analysis between TC was done using limma. The Cox Proportional Hazards Model (CPHM) was used to evaluate the predictive ability of clinical variables, including HPC, and TC on OS, followed by an ANOVA test over CPHM results (AoC). The TCGA-SARC dataset was used for validation, classifying samples with our TC specific gene signatures via ssGSEA. Samples were classified to the TC with the lowest significant FDR adjusted p-value. This classification was integrated into a CPHM analysis of TCGA-SARC OS. Results: TC 1 (C1) (52.4% DDLPS, 19.0% LMS, 28.6% UPS) is characterized by overexpression (OE) of genes as CDK4 and under-expression (UE) of homologous recombination repair genes, such as BRCA. Over representation analysis identified UE of cell cycle and proliferation pathways. TC 2 (C2) (12.5% DDLPS, 29.2% LMS, 58.3% UPS) is portrayed by OE of MAGE genes, and OE of pathways linked with transcriptional regulation. TC 3 (C3) (5% DDLPS, 10% LMS, 85% UPS) is marked by OE of HLA genes (and immune related pathways) and UE of the β-catenin pathway. TC 4 (C4) (11.1% DDLPS, 22.2% LMS, 66.7% UPS) is distinguished by OE of cell components pathways. The CPHM revealed that C2, C3 and C4 are negative prognostic factors (C2 (HR 5.10; 95% CI 1.810-14.34; P= 0.002), C3 (HR 4.47; 95% CI 1.386-14.45; P= 0.01), C4 (HR 7.66; 95% CI 2.056-28.53; P= 0.002)).The CPHM and the AoC revealed that TC display the best ability to predict OS, being the only variable with a statistically significant correlation with OS (AoC (P<0.01)). Validation with TCGA-SARC showed an enrichment to C1 and C3. C3 enriched samples show a worse prognosis (HR 2.28; 95% CI 1.228-4.2; P= 0.009) and TC emerged, again, as the most significant variable for OS prediction (AoC (P<0.01)). Conclusions: RNA sequencing and a machine learning-based analysis identified 4 TC with distinct molecular characteristics (with potential specific predictive value) and superior accuracy for OS estimation when compared with HPC.