Brinker Award basic science recipient discusses neoadjuvant therapy response predictors in breast cancer

//

Estimated Read Time:

3 minutes
Carlos Caldas, MD, FMedSci
Carlos Caldas, MD, FMedSci

This year’s recipient of the Susan G. Komen® Brinker Award for Scientific Distinction in Basic Science, Carlos Caldas, MD, FMedSci, is being honored for his groundbreaking work in breast cancer genomics.

In his address to SABCS attendees on Tuesday, Dr. Caldas shared new data from TransNEO, a translational study of the New Primary Endocrine-therapy Origination Study (NEOS). His award lecture, Multi-omic Machine Learning Predictor of Breast Cancer Therapy Response, will be available to registered symposium participants for on-demand viewing until March 10, 2022.

“We increasingly consider tumors as ecosystems where not only malignant cells but also the tumor microenvironment play a crucial role in determining tumor dormancy, tumor metastasis, therapy response, and therapy resistance,” Dr. Caldas said.

His lab designed TransNEO as a platform to identify predictors of response and resistance to therapy in breast cancer patients.

“We have taken advantage of the neoadjuvant therapy setting to acquire clinical data in multi-platform profiling of tumors, evaluated the response at the time of surgery, and then in two steps developed the framework to predict response to neoadjuvant therapies,” said Dr. Caldas, Senior Group Leader at the Cancer Research UK Cambridge Institute and Chair of Cancer Medicine, University of Cambridge, United Kingdom; Honorary Consultant Medical Oncologist at Addenbrooke’s Hospital; and Director of the Cambridge Breast Cancer Research Unit.

The first step identified 34 features associated with response to therapy — among them tumor size, age at diagnosis, homologous recombination deficiency score, and lymphocyte density. These features were divided into four classifications — clinical, DNA, RNA, and digital pathology — and reflect cell autonomous compartment features, immune activation, and immune evasion, Dr. Caldas explained. The second step integrated the individual features into a machine learning model that was then validated in an independent dataset.

“You can predict therapy with higher areas under the ROC (receiver operating characteristic) curve as you include more features, but you can still predict, although with less than precision, if you have fewer features,” Dr. Caldas said. This could help determine whether standard-of-care or experimental neoadjuvant therapy should be used for a given patient.

“You deploy the model in patients that are candidates for neoadjuvant therapy and ask the question using the model: Are they likely to have pathCR (pathologic complete response)?” Dr. Caldas said. If so, the patient should receive standard therapy. If not, enrollment in neoadjuvant clinical trials should be considered, he said.

“The response is determined by the totality of the baseline tumor ecosystem captured through data integration in machine learning. Machine learning models can integrate either fewer or more features, and they are a framework by which you can bring in new features,” Dr. Caldas said, citing levels of circulating tumor DNA, diversity of T cell and B cell receptors, and more sophisticated spatial architecture information as examples.

In HER2-positive tumors, response to treatment appears to be independent of proliferation, but the same is not true of HER2-negative tumors, Dr. Caldas said. The reason for this discrepancy is not known.

Overall, response to breast cancer treatment is modulated by the pre-treated tumor ecosystem, he noted.

“Residual disease is monotonically associated with point mutations in copy number aberration landscapes, tumor proliferation, immune infiltration, and T cell dysfunction and exclusion,” Dr. Caldas said.