Abstract LB-A05: Utilizing methylation and expression signatures to construct predictive models and personalized cancer treatments

Abstract

Abstract Background: Data from high-throughput genomic arrays available through The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO) have been helpful in elucidating the role of methylome and transcriptome level variations in the oncogenesis and progression of various cancers. Methylome remodeling has been observed across most cancers, and several studies have revealed a novel role for methylation in regulating alternative splicing. Alternative splicing contributes to the malignancy and aggressiveness of cancers, and intragenic methylation of alternatively spliced exons (ASEs) can serve as an effective biomarker for prognosis and personalized treatment. This study focuses on identifying and exploiting clinically relevant intragenic methylation signatures in cancer. Methods: In this study, methylation beta values from the Illumina HumanMethylation450 (HM450) platform, exon expression values from the Illumina HiSeq 2000 platform and patient survival data were downloaded from TCGA. Values were then sorted into methylated and unmethylated groups at each intragenic HM450 probe using an algorithm for dichotomizing microarray data. Probes were selected if their methylation status significantly affected both exon expression and patient mortality and sorted by the difference in patient mortality between the two groups. Biomarkers with beta values falling within the lower mortality group were considered “positive biomarkers.” A series of univariate analyses were used to determine groupings of patients with similar outcomes based on the total number of positive biomarkers in a sample. This methodology was applied to multiple TCGA cohorts including breast cancer, non-small cell lung cancer (NSCLC) and metastatic melanoma. Prognostic models were constructed using multivariate regressions for each TCGA cohort and independently validated with data from GEO. Using clinical data from TCGA and Broad GDAC Firehose, each group of patients was also analyzed based on treatment including surgery, radiation and chemotherapy. Results: In each type of cancer, an unweighted index was created from the top biomarkers. In breast cancer nine probes were included in the index. Patients with seven to nine positive biomarkers had similar overall survivals (HR=1.000), as did patients with four to six positive biomarkers (HR=2.336, 95% CI=1.507-3.621) and zero to three positive biomarkers (HR=8.223, 95% CI=4.103-16.482). Benefits of radiation therapy were seen in the group with four to six positive biomarkers (p<.01). On average, normal breast tissue had more positive biomarkers than tumor tissue (p<.0001). In NSCLC, shorter survival rates were associated with patients in the top biomarker group treated with radiation therapy (p<.05). Patients in the second NSCLC group responded best to cisplatin (p<.05) while patients in the third group responded best to carboplatin (p<.05). The full results of the study will be made available online through an interactive web tool for physicians and researchers. Conclusions: This study has identified biomarkers of genes and pathways involved in alternative splicing in various cancers. These biomarkers are powerful predictors of prognosis and therapeutic response in the tested cohorts and may ultimately be useful for predicting personalized patient prognosis and guiding effective therapy. Citation Format: Caleb Reagor, Rebecca Conway. Utilizing methylation and expression signatures to construct predictive models and personalized cancer treatments [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2017 Oct 26-30; Philadelphia, PA. Philadelphia (PA): AACR; Mol Cancer Ther 2018;17(1 Suppl):Abstract nr LB-A05.