An imprinted non-coding genomic cluster at 14q32 defines clinically relevant molecular subtypes in osteosarcoma across multiple independent datasets

Article English OPEN
Katherine E. Hill ; Andrew D. Kelly ; Marieke L. Kuijjer ; William Barry ; Ahmed Rattani ; Cassandra C. Garbutt ; Haydn Kissick ; Katherine Janeway ; Antonio Perez-Atayde ; Jeffrey Goldsmith ; Mark C. Gebhardt ; Mohamed S. Arredouani ; Greg Cote ; Francis Hornicek ; Edwin Choy ; Zhenfeng Duan ; John Quackenbush ; Benjamin Haibe-Kains ; Dimitrios Spentzos (2017)
  • Publisher: BMC
  • Journal: Journal of Hematology & Oncology, volume 10 (issn: 1756-8722, eissn: 1756-8722)
  • Related identifiers: doi: 10.1186/s13045-017-0465-4., doi: 10.1186/s13045-017-0465-4, pmc: PMC5433149
  • Subject: Molecular subtypes | Methylation | Diseases of the blood and blood-forming organs | RC254-282 | Loss of imprinting | Osteosarcoma prognosis | MicroRNA expression | RC633-647.5 | Research | Neoplasms. Tumors. Oncology. Including cancer and carcinogens

Background: A microRNA (miRNA) collection on the imprinted 14q32 MEG3 region has been associated with outcome in osteosarcoma. We assessed the clinical utility of this miRNA set and their association with methylation status. Methods: We integrated coding and non-coding RNA data from three independent annotated clinical osteosarcoma cohorts (n = 65, n = 27, and n = 25) and miRNA and methylation data from one in vitro (19 cell lines) and one clinical (NCI Therapeutically Applicable Research to Generate Effective Treatments (TARGET) osteosarcoma dataset, n = 80) dataset. We used time-dependent receiver operating characteristic (tdROC) analysis to evaluate the clinical value of candidate miRNA profiles and machine learning approaches to compare the coding and non-coding transcriptional programs of high- and low-risk osteosarcoma tumors and high- versus low-aggressiveness cell lines. In the cell line and TARGET datasets, we also studied the methylation patterns of the MEG3 imprinting control region on 14q32 and their association with miRNA expression and tumor aggressiveness. Results: In the tdROC analysis, miRNA sets on 14q32 showed strong discriminatory power for recurrence and survival in the three clinical datasets. High- or low-risk tumor classification was robust to using different microRNA sets or classification methods. Machine learning approaches showed that genome-wide miRNA profiles and miRNA regulatory networks were quite different between the two outcome groups and mRNA profiles categorized the samples in a manner concordant with the miRNAs, suggesting potential molecular subtypes. Further, miRNA expression patterns were reproducible in comparing high-aggressiveness versus low-aggressiveness cell lines. Methylation patterns in the MEG3 differentially methylated region (DMR) also distinguished high-aggressiveness from low-aggressiveness cell lines and were associated with expression of several 14q32 miRNAs in both the cell lines and the large TARGET clinical dataset. Within the limits of available CpG array coverage, we observed a potential methylation-sensitive regulation of the non-coding RNA cluster by CTCF, a known enhancer-blocking factor. Conclusions: Loss of imprinting/methylation changes in the 14q32 non-coding region defines reproducible previously unrecognized osteosarcoma subtypes with distinct transcriptional programs and biologic and clinical behavior. Future studies will define the precise relationship between 14q32 imprinting, non-coding RNA expression, genomic enhancer binding, and tumor aggressiveness, with possible therapeutic implications for both early- and advanced-stage patients. Electronic supplementary material The online version of this article (doi:10.1186/s13045-017-0465-4) contains supplementary material, which is available to authorized users.