DNA repair deficiency biomarkers and the 70-gene ultra-high risk signature as predictors of veliparib/carboplatin response in the I-SPY 2 breast cancer trial

Abstract

Veliparib combined with carboplatin (VC) was an experimental regimen evaluated in the biomarker-rich neoadjuvant I-SPY 2 trial for breast cancer. VC showed improved efficacy in the triple negative signature. However, not all triple negative patients achieved pathologic complete response and some HR+HER2- patients responded. Pre-specified analysis of five DNA repair deficiency biomarkers (BRCA1/2 germline mutation; PARPi-7, BRCA1ness, and CIN70 expression signatures; and PARP1 protein) was performed on 116 HER2- patients (VC: 72 and concurrent controls: 44). We also evaluated the 70-gene ultra-high risk signature (MP1/2), one of the biomarkers used to define subtype in the trial. We used logistic modeling to assess biomarker performance. Successful biomarkers were combined using a simple voting scheme to refine the 'predicted sensitive' group and Bayesian modeling used to estimate the pathologic complete response rates. BRCA1/2 germline mutation status associated with VC response, but its low prevalence precluded further evaluation. PARPi-7, BRCA1ness, and MP1/2 specifically associated with response in the VC arm but not the control arm. Neither CIN70 nor PARP1 protein specifically predicted VC response. When we combined the PARPi-7 and MP1/2 classifications, the 42% of triple negative patients who were PARPi7-high and MP2 had an estimated pCR rate of 75% in the VC arm. Only 11% of HR+/HER2- patients were PARPi7-high and MP2; but these patients were also more responsive to VC with estimated pathologic complete response rates of 41%. PARPi-7, BRCA1ness and MP1/2 signatures may help refine predictions of VC response, thereby improving patient care.

Fig. 1

a I-SPY 2 TRIAL design schematic. Only patients with HER2− disease were eligible for randomization to the VC arm. b Consort diagram for the VC arm and HER2− concurrent controls, showing data availability for biomarker analysis

Fig. 2

Biomarker analysis approach. Qualifying biomarker candidates are evaluated as specific predictors of response to VC using a predefined 3-step Qualifying Biomarker Evaluation (QBE) methodology, as shown in this flow diagram

Fig. 3

Qualifying biomarker performance. a Ordered heatmap showing the prevalence of all dichotomized biomarkers evaluated in this study, stratified by HR status. b–d Mosaic plots showing patient response stratified by treatment arm and b PARPi-7, c BRCA1ness (this figure panel is also jointly published in ref. [41]), and d MP1/2 signatures, respectively. e–g Bayesian estimated pCR probability distributions by treatment arm, for e PARPi7-high, f BRCA1ness, and g MP2 patients

Fig. 4

Combining VC-sensitivity markers in TN patients. a Simple voting scheme for combining biomarkers to refine sensitivity prediction. b Venn diagram showing overlap between VC-sensitivity biomarkers, including the graduating TN signature. c–f Bayesian estimated pCR probability distributions by treatment arm, for c unselected triple negative [TN], d TN/MP2, e TN/PARPi7-high, and f TN/BRCA1ness patients, respectively. g, h Bayesian estimated pCR probability distributions by treatment arm, for g predicted sensitive (TN/MP2/PARPi7-high) and h predicted resistant (TN/(MP1 or PARPi7-low)) triple negative patients. i Pie chart showing relative proportion of predicted sensitive vs. resistant TN patients

Fig. 5

Combining VC-sensitivity markers in HR+HER2− patients. a Venn diagram showing overlap between VC-sensitivity biomarkers in the HR+HER2− subset. b–g Bayesian estimated pCR probability distributions by treatment arm, for b unselected HR+HER2−, c HR+HER2−/MP2, d HR+HER2−/PARPi7-high, and e HR+HER2−/BRCA1ness patients, respectively, and for HR+HER2− patients who are f predicted sensitive (HR+HER2−/MP2/PARPi7-high) and g predicted resistant (HR+HER2−/(MP1 or PARPi7-low)) by our simple voting scheme. h Pie chart showing relative proportion of predicted sensitive vs. resistant HR+HER2− patients