Abstract A23: A non-coding RNA panel predicts intermediate-risk childhood rhabdomyosarcoma prognosis better than standard pathologic criteria and coding genes

Abstract

Abstract Introduction: Rhabdomyosarcoma (RMS) is the most common pediatric soft-tissue sarcoma, stratified by the Children's Oncology Group (COG) into low/intermediate/high risk based on clinical outcomes. However, most patients are categorized as intermediate-risk where survival is highly heterogeneous, thus suggesting an inability to accurately stratify a majority of patients. We profiled intermediate-risk RMS's for levels of coding and non-coding transcripts to construct prognostic signatures. The goal was to identify panels of RNAs that reflect underlying tumor biology and provide better risk stratification than routine clinicopathologic parameters. Methods: Transcriptomes from 79 prospectively-obtained primary tumors from intermediate-risk RMS patients under COG clinical trial protocols were profiled on Affymetrix Human Exon 1.0 ST microarrays. Expressions of 1,400,033 probe sets representing annotated and unannotated transcripts were analyzed using Genetrix suite of microarray analysis tools. Cox regression and leave-n-out cross validation were used to derive and finalize the expression signatures. An effort was made compare individual prognostic potentials of the coding and non-coding signatures, and that of a signature that combined both features. Results: Standard pathologic prognosticators such as histologic subtype classification (alveolar versus embryonal) and PAX-FKHR fusion gene status were unable to predict outcome in this cohort (p=0.40 and 0.45, respectively). Cox regression analysis on 17,049 coding transcripts created a 42-gene meta-feature that was able to predict survival (p=0.00024). Leave-n-out cross validation of this meta-feature upheld its prognostic ability (p=0.00030). Analysis of probe set regions (PSRs) corresponding to unannotated “dark matter” transcripts identified a 32-PSR meta-feature that also predicted survival with greater significance than PSRs corresponding to coding transcripts (p<0.00001). To reduce feature redundancy, multiple PSRs interrogating the same genomic locus were replaced by a representative PSR that shrunk the meta-feature size to 24 PSRs, which was still able to predict survival better than the coding gene meta-feature (p<0.00001). A meta-feature that combined coding and non-coding RNA features retained its ability to predict outcome (p=0.00002), with non-coding RNA features contributing towards the bulk of its prognostic potential. Conclusions: A more concise non-coding RNA meta-feature was able to better predict outcome than a larger coding gene meta-feature in intermediate-risk RMS, where standard pathologic prognosticators failed. This suggests the role of non-coding transcripts in regulating and determining RMS biology and aggressiveness, and their potential to serve as novel prognostic indicators. Citation Format: Anirban P. Mitra, Jonathan D. Buckley, Sheetal A. Mitra, Elai Davicioni, James R. Anderson, Philipp Kapranov, Stephen X. Skapek, Douglas S. Hawkins, Timothy J. Triche. A non-coding RNA panel predicts intermediate-risk childhood rhabdomyosarcoma prognosis better than standard pathologic criteria and coding genes [abstract]. In: Proceedings of the AACR Special Conference on Noncoding RNAs and Cancer; 2012 Jan 8-11; Miami Beach, FL. Philadelphia (PA): AACR; Cancer Res 2012;72(2 Suppl):Abstract nr A23.