Effect of molecular mechanisms mediating bevacizumab (BEV) and vanucizumab (VAN) on gastrointestinal perforation: Use of artificial neural networks for integrated data analysis.

Abstract

Listen

Translate article

e15108 Background: VAN is a novel bispecific mAb targeting VEGF & ANG-2. A randomized phase 2 study (McCAVE) investigating VAN & BEV, each combined with chemotherapy (mFOLFOX-6), in 1L metastatic colorectal cancer patients was completed. Reported gastrointestinal perforation (GIP) events (including GI fistula & abdominal abscess) in both arms (approx. 10%) exceeded those expected with BEV+CTx. The present work focused on identifying the explicit molecular pathways mediating VAN/BEV associated GIP. Methods: Clinical data from McCave (safety outcomes & gene expression profiling) was integrated with data from other clinical/preclinical VAN/BEV studies. Machine learning was applied to the dataset to computationally infer GIP mechanisms. The methodology was constrained by relevant biological & clinical knowledge. Literature reviews & database queries mapped 194 unique proteins to 8 biological processes previously hypothesized to play a role in GIP. Identified proteins guided construction of a GIP protein interaction network which was transformed into an artificial neural network and was fitted to the dataset to predict molecular pathways/biological processes most likely linking VEGF-A and/or ANG-2 inhibition to GIP. Results: VAN/BEV were predicted to enhance GIP risk via 4 out of 8 previously suggested biological effects: (i) reduced GI mucosal integrity; (ii) impaired wound healing; (iii) inhibition of angiogenesis (iv) increased thromboembolic risk. The specific proteins & associated interactions predicted to mediate these 4 effects were identified as solely VEGF-A inhibition driven. These effects may drive increased susceptibility to GIP triggered by secondary insults, e.g. preclinical ulcer & intramural lesion. Candidate GIP biomarkers for monitoring these effects were identified, e.g., aPTT, citrulline & I-FABP. Conclusions: The molecular pathways mediating GIP in VAN/BEV treated patients are likely driven by VEGF inhibition with no contribution from ANG-2 inhibition. It is proposed, that in addition to VAN/BEV, one or more secondary insults to the GI tract wall are required to drive perforation.