Background: The angiogenesis (ANG1/2) inhibitor trebananib (TR) was one of the experimental agents evaluated in I-SPY 2. In I-SPY 2, all patients received at least standard chemotherapy (paclitaxel followed by doxorubicin/cyclophosphamide: T->AC). HER2- patients were randomized to receive TR+T->AC vs. T->AC. For HER2+ patients, TR was administered with trastuzumab (TR+H+T->AC vs. H+T->AC). We hypothesized that genes/signatures in the ANG/TIE signaling axis may specifically predict response to TR, and tested expression levels of 11 genes: TIE1/2, ANGPT1/2/4, AGPTNL1/3, VEGFA, ICAM1, PECAM1 and MMP2. We also evaluated angiogenesis and hypoxia expression signatures, based on the hypothesis that hypoxic tumors with a fragile blood supply may be vulnerable to drugs in this class.

Methods: Data from 266 patients (TR: 134 and concurrent controls: 132) were available. Pre-treatment biopsies were assayed using Agilent 44K (32627) or 32K (15746) expression arrays; and these data were combined using ComBat. All I-SPY 2 qualifying biomarker analyses follow a pre-specified analysis plan. We use logistic modeling to assess biomarker performance. A biomarker is considered a specific predictor of TR response if it associates with response in the TR arm, and if the biomarker x treatment interaction is significant (likelihood ratio test, p<0.05). This analysis is also performed adjusting for HR and HER2 status as covariates, and within receptor subsets. Additional exploratory global transcriptomic analysis was performed using DAVID. Our statistics are descriptive rather than inferential and do not adjust for multiplicities of other biomarkers outside this study.

Results: ANGPT1, a direct target of trebananib, associates with pCR in the TR arm but not the control arm, and shows a significant interaction with treatment that retains significance in a model adjusting for HR and HER2. In addition ICAM1, expressed on endothelial and immune cells, strongly associates with response in the TR arm, but also in the control arm in the population as a whole. In the HR+HER2- subset, both ICAM1 and PECAM1 associate with pCR in the TR arm and not the control arm, with a trend toward treatment interaction. Interestingly, in the TN subset, where pCR rates were highest in the TR arm relative to control, these mechanism-of-action biomarkers did not appear to predict response. Rather, in exploratory whole genome analysis, response of TN's strongly associates with immune related genes (e.g. HLA's, IL21R, CCL13).

Conclusion: Following our pre-specified analysis, ANGPT1 succeeds as a specific predictor of response to trebananib in I-SPY 2. In addition, ICAM1 and PECAM1 associate with response in the HR+HER2- subset; and in exploratory analysis immune signaling predicts response in the TN subset. These biomarkers may merit further evaluation in future trials.