Transcriptome signatures for the identification of bevacizumab responders in ovarian cancer

Background: Bevacizumab is widely used as an anti-angiogenic maintenance therapy in ovarian cancer; however, there are currently no validated clinical criteria to guide patient selection for its use. Methods: To satisfy the urgent need for bevacizumab response biomarkers, we created a novel RNA-seq dataset (n = 244) and applied unsupervised and supervised machine learning to identify expression signatures associated with benefit from adding bevacizumab to standard treatment and validated our findings using a previously published microarray dataset (n = 377). Additionally, we validated the existence of the discovered signatures using RNA-seq data from the TCGA-OV cohort (n = 426) and performed public expression data mining to provide a biological interpretation of the prioritized signature. Results: Among expression signatures reproducibly detected in independent datasets, one was prioritized as a potential predictive biomarker for bevacizumab benefit. Further stratified analysis revealed that over-expression of this signature was associated with improved overall survival in patients who received bevacizumab in addition to standard chemotherapy in both novel (HR = 0.41, 95% CI: (0.23–0.74), adj.p-value = 0.008) and previously published cohorts (HR = 0.51, 95% CI: (0.34–0.75), adj.p-value = 0.003), while no significant survival benefit from bevacizumab was observed in patients negative for this signature. We hypothesize that this signature may be associated with stemness-like features, possibly driven by CTCFL. In addition, we identified several other signatures reproducible in independent datasets and not related to known molecular subtypes of ovarian cancer, which may also represent biomarker candidates and require further validation in additional RNA-seq data. Conclusions: We identified a previously undescribed expression signature with potential predictive value for bevacizumab benefit, and revealed transcriptional heterogeneity of ovarian cancer that extends beyond current molecular classifications. Given the high heterogeneity of ovarian cancer and that the novel signature only partially explains variation in survival outcomes under bevacizumab treatment, larger RNA-seq datasets are required to further improve predictive models.