Characterization and validation of a methylation signature as a prognostic marker in osteosarcoma.

Abstract

10016 Background: Despite extensive research efforts, outcomes for patients with osteosarcoma have not substantially improved since the 1980s. Moreover, no prognostic features are used clinically to risk adapt therapy. Here, we performed methylation and RNA sequencing of pre-treatment osteosarcoma tumor samples to identify distinct features of tumor biology and to generate and validate a prognostic signature. Methods: All patients diagnosed with osteosarcoma at the Children’s Hospital of Philadelphia (CHOP) before March 15, 2024 with tissue available for analysis were eligible. Of the 100 eligible patients, three were excluded for lack of clinical data and one for initial diagnostic uncertainty. Methylation and RNA sequencing were performed on DNA and RNA extracted from FFPE or frozen tumor. Unsupervised analyses characterized global methylation structure and differential methylation analyses compared differences between progression groups. Tumor and immune cell fractions were estimated using a reference-based deconvolution algorithm. CpGs were filtered using an unsupervised variability criterion (top 20% MAD) and a ridge-penalized Cox model was used to derive a methylation risk score as a linear predictor. The publicly available TARGET-OS cohort was used for validation with median-based risk stratification defined in the CHOP cohort. Results: Of 96 patients, 44 patients had pre-treatment samples that passed methylation QC and 13 passed RNA sequencing QC. Within the CHOP cohort, 34/44 (77%) had localized disease with a median age of 13 years. Variability in methylation profiles was driven by leukocyte infiltration, supported by strong correlations between global methylation levels and mitotic index estimates. Tumors from patients with progressive disease displayed 12,338 significantly different CpGs (FDR < 0.01) and enriched for hypomethylated CpGs. A methylation-derived risk score was established for overall survival (OS) in the CHOP cohort. In multivariable Cox models, the methylation risk score remained significantly associated with OS (HR 1.54 [1.05–2.26]; p = 0.026) together with metastatic status (HR 5.13 [2.49–10.59]; p < 0.001), yielding improved model discrimination (C-index 0.77 [0.70–0.83]). This was replicated in event-free survival (EFS, HR 2.49 [1.49,4.19], p<0.01). In the TARGET-OS cohort, after adjustment for metastatic status, methylation risk score remained independently associated with EFS (HR 1.6 [1.15, 2.23], p = 0.005). and OS (HR 1.54 [1.05, 2.26], p = 0.003). Conclusions: In this single-institution cohort analysis, a methylation-based risk score predicts survival and adds prognostic value when combined with metastasis status. Integration of CpG methylation with downstream pathway activation via RNA sequencing is ongoing and will be reported. Further validation of this prognostic signature should be considered in future osteosarcoma trials.