Abstract
Background: Follicular lymphoma (FL) is the second most common type of non-Hodgkin lymphoma in the Western world. It is generally an indolent disease, however some patients experience rapid clinical progression. Identification of this subset of patients at the time of initial diagnosis would allow for more informed decisions to be made regarding clinical management.
Methods: We investigated whether a multi-dimensional profiling approach using gene expression microarrays, a tissue microarray (TMA) and baseline clinical parameters, would allow for survival prediction for FL patients. Sixty-seven cases of FL were identified, of which high quality gene expression data were obtained with a minimum of 5 years of follow-up for 41 patients. Expression data were subjected to Predictive Interaction Analysis (PIA) to identify pairs of interacting genes that predict poor outcome, defined as death within five years of diagnosis. A TMA of all 67 patients was subjected to immunohistochemistry for markers routinely used in lymphoma diagnosis, and for numerous proteins whose relevance to oncogenesis is well-established, including p53, bcl-2, bcl-6, MUM1, p16 and p65.
Results: Gene expression analysis revealed numerous genes that are highly predictive of clinical outcome. Many of these genes are known to be involved in pathways that regulate apoptosis, cell survival, proliferation and hematological function. The highly predictive single genes included BMX, NOTCH2, TFF3, BIRC4 and RIPK5, which have established roles in promoting or antagonizing apoptosis. The PIA approach further identified numerous pairs of genes that together possess greater predictive power than their individual constituent genes. Subsequent Kaplan-Meier analysis indicated that segregation of the cases according to the top performing gene pair, LOXL3 and NTS, produced two groups of cases with significantly different survival. This gene pair was able to further differentiate patient outcomes following stratification of the cases according to the Follicular Lymphoma International Prognostic Index (FLIPI), indicating its utility in providing supplementary information to the FLIPI. Upon analysis of the TMA results, detectable expression of p53 in lymphoma cells, along with clinical involvement of multiple nodal sites and B symptoms emerged as significant predictors of overall survival. Further IHC results examining expression of proteins identified as highly predictive at the transcript level will be presented.
Conclusions: Our results support the utility of our profiling approach for the identification of candidate biomarkers in follicular lymphoma. Queen’s University and Biosystemix Ltd are co-owners of the intellectual property and are respectively the assignees of a provisional patent application filed at the US PTO in Sept. 2007. Roland Somogyi PhD and Larry D. Greller PhD as founding directors retain ownership positions in Biosystemix Ltd, a privately held company incorporated in Canada. Queen’s University, Biosystemix, and all the co-authors believe and agree to the best of our respective knowledge that there are no conflicts of interest in how the study was initiated, conducted, analyzed, and reported.
Personally Tailored Survival Prediction of Patients With Follicular Lymphoma Using Machine Learning Transcriptome-Based Models
