66P - Prognostic classification of lung adenocarcinoma by integrated miRNA-mRNA expression profiles

Date 17 April 2015
Event ELCC 2015
Session Poster lunch
Topics Non-Small-Cell Lung Cancer, Early Stage
Translational Research
Presenter Shirley Tam
Citation Annals of Oncology (2015) 26 (suppl_1): 18-23. 10.1093/annonc/mdv048
Authors S. Tam1, M. Pintilie2, N. Liu1, J.D. McPherson3, M.S. Tsao4
  • 1Ontario Cancer Institute, Princess Margaret Hospital, M5G 1L7 - Toronto/CA
  • 2Biostatistics, Princess Margaret Hospital, Toronto/CA
  • 3Genome Technologies, Ontario Institute for Cancer Research, Toronto/CA
  • 4Pathology, Princess Margaret Hospital, M5G 2M9 - Toronto/CA



The current staging system for prognostication does not fully encompass the morphological and molecular heterogeneity observed amongst non-small cell lung cancer (NSCLC) patients. Molecular prognostic markers have largely been developed based on single molecular platforms, such as gene mutation, copy number, mRNA, miRNA or protein. We hypothesize that integrating miRNA and mRNA data may better capture the molecular heterogeneity and multivariate nature of the molecular changes involved in NSCLC, resulting in more robust classification.


Gene expression profiling by Affymetrix U133 Plus 2.0 and miRNA-sequencing was performed on RNA isolated from tumors of 174 resected early-stage NSCLC patients. The cohort includes 71% adenocarcinomas (ADC), 24% squamous cell carcinomas (SQCC), 1% adenosquamous and 4% large cell carcinoma. Unsupervised machine learning was applied to identify molecular subtypes of ADC using integrated miRNA and mRNA expression profiles, and compared to the results obtained using single-dimensional data alone.


Initial clustering of the dataset (n = 174) identified three distinct NSCLC patient subgroups. Despite showing a statistically significant difference in survival (p = 0.0015), these groups were driven by histological differences between SQCC and ADC (Fisher's exact test of independence: p = 3.564 × 10−10). A subset analysis was performed on ADCs (n = 124), which separated the patients into one low-risk (nlow = 48) and two overlapping high-risk (nhigh1 = 29, nhigh2 = 47) groups. The high-risk groups were collapsed into a single group, which had significantly poorer survival than the low-risk group (HR 4.5, CI 2.0-10.1, p < 0.001). Preliminary validation of our integrated approach for tumor classification in a subset of lung ADCs (n = 108) from the TCGA project revealed similar patient subgroups (HRhigh_vs_low 2.45, CI 0.89-6.17, p = 0.086).


Reproducible ADC subtypes of significantly different prognosis can be identified using integrated miRNA-mRNA profiles, which differ from those identified using either data types alone.


All authors have declared no conflicts of interest.