Biomarker Analysis Demonstrates Potential for Predicting Breast Cancer Response

A biomarker analysis of participants in a phase II breast cancer trial demonstrated potential for identifying tumor markers to predict susceptibility to specific therapies.

Denise Wolf, PhD

A biomarker analysis of participants in a phase II breast cancer trial demonstrated potential for identifying tumor markers to predict susceptibility to specific therapies.

Though based on a limited number of patients, the analysis found that the PARPi-7 gene signature, the 77-gene BRCAness signature, and the MammaPrint 1 and 2 risk categories predicted response to combination therapy with the PARP inhibitors veliparib and carboplatin. PARP1 protein and the CIN70 chromosomal instability assay were not predictive.

Investigators further honed the results to determine the patients who were MammaPrint 2 and PARPi7-high had tumors with even greater sensitivity to the veliparib-carboplatin combination.

The analysis also suggested that a “voting” method can be used to combine multiple biomarkers for the same treatment, Denise Wolf, PhD, reported at the San Antonio Breast Cancer Symposium.

“Our sample size is small. We will need to validate the findings in larger trials and in patients who received carboplatin as single treatment,” said Wolf, a computational biologist at the Helen Diller Family Comprehensive Cancer Center and University of California, San Francisco. “Ongoing and future work focus on developing predictive biomarkers for other agents used in the I-SPY-2 trial.”

The I-SPY-2 trial program has a standing platform for conducting phase II, adaptively randomized studies of neoadjuvant therapy for high-risk breast cancer. As many as 4 investigational treatment arms are compared simultaneously against a standard neoadjuvant chemotherapy regimen.

In the trial design, investigators try to match therapies with the most responsive breast cancer subtypes, as defined by hormone receptor (HR) status, HER2 expression, and the MammaPrint 70-gene signature. The phase II investigations have the common primary endpoint of pathologic complete response. I-SPY-2 agents “graduate” if trial results in the most responsive patient subset suggest a >85% predictive probability of success in a phase III trial.

The I-SPY-2 trial program includes a biomarker component that employs available assays to examine biomarkers associated with the mechanism of action of each investigational treatment, as well as the predefined patient subsets.

Wolfe reported findings from the biomarker analysis of an I-SPY-2 trial that evaluated the veliparib-carboplatin combination in patients with HER2-negative breast cancer. The combination subsequently graduated in the triple-negative breast cancer subset.

The biomarker analysis focused on the DNA-damaging mechanism of carboplatin and veliparib’s inhibition of DNA repair. The analysis included five types of tests: BRCA1/2 germline mutation, PARP1 protein (and cleaved protein levels, RPPA), PARPi-7, BRCAness (which distinguishes BRCA1 from wild type), and CIN70.

The analysis involved 116 patients treated with veliparib-carboplatin (N=72) and the concurrent control arms (N=44). The analysis comprised a 3-step process, the first of which addressed 3 questions: Is the biomarker associated with response to the investigational treatment? Is the biomarker associated with response in the control arm? Does a treatment-biomarker interaction exist (p<0.05)?

Positive answers to the questions in the first step led to the second step, which addressed the question of whether there was a treatment-biomarker interaction. A positive answer led to the third step, which answered two questions: What are the estimated pCR rates in the veliparib-carboplatin and control arms? What is the predictive probability of success in a 300-patient phase III trial?

Summarizing the results, Wolf said that 15 patients had BRCA1/2 germline mutations, including 11 patients with triple-negative breast cancer. Of 12 BRCA-positive patients assigned to veliparib-carboplatin, 9 achieved pCR representing a significant association (p=0.006). All three BRCA-positive patients were assigned to control therapy did not achieve pCR, but the small number precluded definitive analysis.

Application of the PARPi-7 signature showed that 33 patients in the veliparib-carboplatin group had high scores, and 13 of the 20 achieved pCR (p=0.00025). In contrast, 4 of 20 patients with PARPi-7-high results in the control group had pCR. Investigators found that a significant biomarker-treatment interaction existed (p=0.03) and persisted after adjustment for HR status (p=0.025).

Statistical modeling predicted the veliparib-carboplatin regimen would achieve a pCR rate of 59% in a phase III trial compared with 23% for control, associated with >90% probability of success.

“The results showed that PARPi-7 is a specific predictor of veliparib-carboplatin response,” Wolf said.

After determining that BRCAness and MP1/2 class predicted response to veliparib-carboplatin, the investigators asked whether combining the results of the predictive biomarkers would identify a subset of patients with an even greater likelihood of response to veliparib-carboplatin.

Analysis of concordance between paired tests yielded moderate rates of 50-67%. The results suggested that the tests did not identify the same types of patients, Wolf said.

Investigators then performed a simple “voting” analysis, comparing biomarkers with respect to tumors’ resistance or sensitivity to veliparib-carboplatin. As an example, 40% of triple-negative tumors were associated with MammaPrint 2 and PARPi-7-high, both suggesting sensitivity. Combining the 2 biomarkers for that subgroup of patients yielded a predicted pCR probability of 79% with veliparib-carboplatin versus 23% for the control arm.

Voting analyses that included negative results for one biomarker (such as MammaPrint 1 or PARPi-7-low) predicted resistance (35% probability of pCR with veliparib-carboplatin versus 29% for control).