The combination of circulating Ang1 and Tie2 levels predicts progression-free survival advantage in bevacizumab-treated patients with ovarian cancer

Abstract

Purpose: Randomized ovarian cancer trials, including ICON7, have reported improved progression-free survival (PFS) when bevacizumab was added to conventional cytotoxic therapy. The improvement was modest prompting the search for predictive biomarkers for bevacizumab.

Experimental design: Pretreatment training (n=91) and validation (n=114) blood samples were provided by ICON7 patients. Plasma concentrations of 15 angio-associated factors were determined using validated multiplex ELISAs. Our statistical approach adopted PFS as the primary outcome measure and involved (i) searching for biomarkers with prognostic relevance or which related to between-individual variation in bevacizumab effect; (ii) unbiased determination of cutoffs for putative biomarker values; (iii) investigation of biologically meaningfully predictive combinations of putative biomarkers; and (iv) replicating the analysis on candidate biomarkers in the validation dataset.

Results: The combined values of circulating Ang1 (angiopoietin 1) and Tie2 (Tunica internal endothelial cell kinase 2) concentrations predicted improved PFS in bevacizumab-treated patients in the training set. Using median concentrations as cutoffs, high Ang1/low Tie2 values were associated with significantly improved PFS for bevacizumab-treated patients in both datasets (median, 23.0 months vs. 16.2; P=0.003) for the interaction of Ang1-Tie2 treatment in Cox regression analysis. The prognostic indices derived from the training set also distinguished high and low probability for progression in the validation set (P=0.008), generating similar values for HR (0.21 vs. 0.27) between treatment and control arms for patients with high Ang1 and low Tie2 values.

Conclusions: The combined values of Ang1 and Tie2 are predictive biomarkers for improved PFS in bevacizumab-treated patients with ovarian cancer. These findings need to be validated in larger trials due to the limitation of sample size in this study.

©2014 American Association for Cancer Research.

Figures

Figure 1

Multivariate Fractional Polynomial Analysis for Interactions (MFPI) of treatment × Ang1 (upper panel) and treatment × Tie2 (lower panel) interactions keeping the biomarkers continuous, fitted by FP1 functions with power 1.0 (i.e. linear). Functions were estimated within multivariable models adjusting for age. In the left-hand plot: solid line, estimated effect of biomarker in patients not treated with Bevacizumab; dashed line, estimated effect of biomarker in patients treated with Bevacizumab; In the right-hand plot: effect of Bevacizumab by biomarker status, with 95% pointwise CI. Horizontal dashed lines denote zero and the main effect of Bevacizumab in the absence of an interaction. The vertical red lines denote the medians of the respective biomarker distributions (p50: 50th percentile).

Figure 2. Flow Diagram showing predictive effects of different Ang1 and Tie2 concentrations

Patients are classified according to whether they had pre-treatment plasma concentrations of Ang1. There was no predictive value in patients who had low Ang1 (right hand graph). Those with Ang1 concentrations above the median are then categorized according to their plasma Tie2 concentrations before treatment. Benefit from bevacizumab was observed in the arm with high Ang1 and low Tie2 (left graph). Those with high Ang1 and high Tie2 do not benefit from bevacizumab (middle graph)

Figure 3. Kaplan-Meier analysis of the risk groups defined for the validation patients using the Cox model including all prognostic and predictive biomarkers

The Cox survival model of the training set, including all prognostic and predictive biomarkers, was validated by assigning prognostic indices to the additional validation patients and subsequently divided them into two risk groups: “high probability” and “low probability” for disease progression. Kaplan-Meier analysis was carried out for the risk groups showing that they are significantly different. Not possible to calculate probability in 8 patients.