A four-marker-protein signature to predict ramucirumab efficacy in combination with TAS-102 in advanced colorectal cancer: Translational analysis of the RAMTAS trial of the German AIO (AIO-KRK-0316/IKF643).

Abstract

3588 Background: The combination of TAS-102 and bevacizumab is standard in the treatment of advanced colorectal cancer (aCRC). Ramucirumab (Ram) is a mAb that blocks VEGFR2. Ram in combination with Tas-102 did not improve OS compared to Tas-102 alone in the ITT population of the RAMTAS trial. Subgroups such as female patients (pts) and pts with left-sided tumors appeared to benefit from the combination (DOI:10.1016/j.annonc.2024.08.2264). Here we aimed at identifying angiogenesis-related, soluble, predictive biomarkers defining a RAMTAS subpopulation benefitting from the addition of Ram to Tas-102. Methods: Serum samples were available from 194 of 428 pts from the RAMTAS study (96 pts randomized to TAS-102, 98 pts to TAS-102 plus Ram). Serum levels of 85 angiogenesis-related proteins obtained prior to treatment with Tas-102 +/- Ram (baseline, BL) and at cycle 2 day 1 prior to the start of the next cycle (treatment, TR) were quantified via multiplex-ELISA by Myriad-RBM. Multivariate Cox regression models were developed within the combination arm using stepwise selection and ridge regularization. Model performance was assessed via C-index and validated using 10 x 5-fold cross-validation (CV) as well as full quality control for multicollinearity and proportional hazard assumptions. To distinguish predictive from general prognostic effects, the identified signatures were applied to the monotherapy control arm. Results: A TR model incorporating just 4 serum proteins demonstrated robust discrimination in the combination arm (Training C-index: 0.712; CV C-index: 0.708 ± 0.063). This model successfully stratified patients into a “Ram benefit” and a “no Ram benefit” group with a corresponding median PFS of 7.0 mo and 1.9 mo, respectively (HR 6.97, p < 0.001). A Baseline (BL) model yielded a C-index of 0.699, also utilizing just 4 serum proteins. Application of both models to the TAS-102 mono arm resulted in poor discrimination (C-index ~0.60) indicating that the signatures are specific predictors for the combination. In a comparative survival analysis, "Ram-benefit" TR model patients achieved a significantly longer PFS with the combination compared to monotherapy (7.00 vs. 2.14 mo; HR 0.558, p=0.0078). Conversely, "no-Ram-benefit" patients derived no significant benefit from the addition of Ram (PFS 1.91 vs. 1.91 mo; HR 0.946, p=0.4204). Conclusions: Oligo-marker serum signatures, when incorporating early on-treatment dynamics after one cycle of Tas-102/Ram, may effectively predict a survival benefit for aCRC-pts receiving TAS-102 in combination with Ram. We propose that a two-stage “Screen & Monitor” (BL+TR) strategy will help to identify primary refractory patients who gain no benefit from the addition of Ram. Such an approach could maximize therapeutic utility while reducing potential toxicity. Clinical trial information: NCT03520946 .