Integrating multi-omics, imaging, and longitudinal data to predict radiation response in cervical cancer

PROJECT SUMMARY/ABSTRACT Cervical cancer is among the most common cancer diagnoses among women, and treatment failure of standard of care chemoradiation therapy (CRT) for locally advanced cervical cancer (LACC) is as high as 30-50%. Since recurrent and metastatic diseases are not curable, there is a pressing need to identify patients at risk of treatment failure as early as possible to allow for personalized treatment, rather than after a failure and progression. While TCGA’s molecular stratification of cervical cancer using genomic data failed to associate to patient outcomes, we recently published on integrating genomic and imaging data to improve LACC risk stratification after CRT. Therefore, in this study we intend to use multi-omics data to define and validate LACC risk groups and identify group-specific treatment targets. Based on our preliminary data that indicate distinct biological mechanisms drive CRT resistance in patients with different levels of lymph node (LN) involvement at presentation, we will stratify patients by LN status to develop and validate novel radiogenomic biomarkers. Prognostic models will be developed using gene expression data from pre-treatment tumor biopsy and radiomic features from pre- treatment PET imaging data. Upstream driver and/or feature genes will be validated at the RNA and protein levels by qRT-PCR, Western blotting, and tissue microarray (TMA). One such gene identified from our preliminary data using a radiogenomic approach is nuclear factor erythroid 2–related factor 2 (NRF2), which has not been previously characterized in LACC, since it is not frequently mutated in cervical cancer. We will perform functional analysis to study NRF2 biology in LACC via clonogenic survival assay and other standard assays. In addition to pre-treatment biomarkers, we will leverage radiomic features from our time course MR images and on-treatment gene expression data to develop novel radiogenomic biomarkers to assess a patient's evolving risk of treatment failure over the course of CRT, informing adjustment of therapy at mid-treatment. The pre-treatment model will be further refined by applying deep learning to identify predictive features for CRT outcome directly from clinical PET images to inform intensified treatment from the beginning. Finally, we will apply multi-omics approaches (scRNA-seq, proteomics, metabolomics) to characterize the biology related to LACC CRT radiogenomic biomarkers. Taken together, we expect fulfillment of these aims will create a series of optimized, validated recurrence biomarkers at presentation and over the course of 6 weeks of CRT treatment, and will indicate targets for personalized alternative treatment regimens. Beyond the specific application to LACC, our proposal will generate novel methods to integrate multi-omics data to improve hypothesis-driven cancer research.