Introduction: Intermediate-risk pediatric rhabdomyosarcomas (IR-RMSs) have heterogeneous outcomes, suggesting an inherent inability of clinical staging to accurately stratify a large proportion of patients. This study profiled IR-RMS for expressions of coding and non-coding transcripts to construct prognostic signatures that reflect underlying tumor biology and provide better risk stratification than routine clinicopathologic parameters. Methods: RNAs extracted from 80 prospectively-obtained primary IR-RMS and 40 low/high-risk RMS (non-IR-RMS) patients were profiled on Affymetrix Human Exon microarrays. Expressions of 1,393,765 probe selection regions (PSRs) representing annotated and unannotated transcripts were analyzed. Cox regression and leave-n-out cross validation were used to derive and finalize the weighted signatures. Potentials of the coding and non-coding signatures to predict overall survival were compared using areas under receiver operating characteristic curves that provided a measure of predictive accuracy. Results: Standard pathologic prognosticators such as histologic subtype and PAX-FKHR fusion status were unable to predict survival in this subset of IR-RMS (p=0.94 and 0.66, respectively). Tumor site was the only clinical predictor of outcome in this cohort of IR-RMS (p=0.041). Iterative cox regression on 17,045 coding transcripts identified a 34-gene meta-feature (34gMF) that this was able to predict survival in IR-RMS (p=0.001). Analysis of PSRs corresponding to unannotated transcripts identified a 39-PSR meta-feature (39ncMF) that also predicted survival (p<0.001). To eliminate feature redundancy, multiple PSRs interrogating the same unannotated genomic locus were replaced by a representative PSR that reduced the meta-feature size to 34 PSRs (34ncMF), which was still able to predict outcome (p<0.001). In multivariable models in IR-RMS that included tumor site, the genomic meta-features were the only significant predictors of outcome (34gMF, p=0.008; 39ncMF, p=0.019; 34ncMF; p=0.007). When applied on non-IR-RMS cases, the meta-features retained their prognostic abilities (34gMF, p=0.026; 39ncMF, p=0.034; 34ncMF, p=0.027). In IR-RMS, predictive accuracy of 39ncMF was significantly higher than 34gMF (96.4% vs. 71.1%, p<0.001). However, predictive accuracy of the former was comparable to the non-redundant 34ncMF (96.7%, p=0.54). Conclusions: A concise non-coding RNA meta-feature was able to better predict outcome in IR-RMS than a coding gene meta-feature, where most standard clinical prognosticators failed. The prognostic value of these meta-features is also observed in non-IR-RMS. These observations point to the possible 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, Sheetal A. Mitra, Jonathan D. Buckley, Philipp Kapranov, James R. Anderson, Stephen X. Skapek, Douglas S. Hawkins, Timothy J. Triche. Prognostic value of coding and non-coding genomic meta-features in rhabdomyosarcoma. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 17. doi:10.1158/1538-7445.AM2013-17