A Combined Survival Model Integrating Gene Expression and Alternative Splicing Events Provides Higher Predicative Power for Risk Stratification

Abstract 1929

Considerable efforts have been spent evaluating impact of global gene expression profile on clinical outcome. Although significant correlation has been described between outcome and expressed gene signature, the overall predictability of such models have reached a plateau. Biologically this is expected as gene function is modulated at multiple levels. Besides the change in level of expression, post-transcriptional changes such as alternate splicing alter specificity of gene function and may affect the eventual outcome. Although various genes have normal alternate spliced form, dysregulation of alternative splicing that alters protein function has been implicated in number of disease processes including cancer. We have observed significant level of dysregulated splicing events in multiple myeloma (MM). We hypothesize that a combined model that includes dysregulated splicing events, besides level of expressed genes, may provide superior survival model in MM. To develop a combined model we have hybridized RNA isolated from CD138+ purified MM cells collected at the time of diagnosis from 170 newly-diagnosed patients treated homogeneously in tandem transplantation IFM trials, 23 MM cell lines and 6 Healthy donors on Affymetrix Exon 1.0 ST GeneChip arrays. Exon array not only provides an accurate measure of expression levels for genes, but also allows simultaneous identification of alternative splicing events. Pre-processing and normalization methods in aroma, affymetrix and robust multichip analysis model in FIRMA, followed by t-tests with Benjamini-Hochberg multiple hypothesis corrections were used respectively to identify differential expression and alternative splicing. We identified 1454 differentially expressed genes and 759 differential splicing events between healthy donors and MM patients, and 5476 differentially expressed genes and 4012 differential splicing events between healthy donors and MM cell lines. There are 1071 differentially expressed genes and 286 alternative spliced exons shared between MM samples and MM cell lines.

Univariate survival analysis using FIRMA scores of exons identified a total of 89 genes with more than 10 alternative splicing events between healthy donors and MM patients associated with survival with Cox proportional hazard model and log-rank tests. We have now built 3 different survival models considering: 1) gene expression only, 2) alternative splicing only, and 3) a combined model integrating gene expression and alternative splicing events. We utilized refined regularized variable selection methods to handle these high-dimensional feature space. Our analysis suggests that composite model using gene expression and alternative splicing information performs significantly better than the gene expression only model in identifying high-risk patients, when the data were divided in median or quartiles. Specifically, the difference in overall survival is 32.6 months to 38.5 months using the median survival, and 18 months vs 23 months for median event free survival. We are currently in the process of validating the combined model. Our data suggests the need for inclusion of modifiers of transcriptome to develop a comprehensive model that will have higher predicative power for risk stratification as well as for selection of therapeutic intervention.