Abstract

Predicting Taxane Response in Danish Breast Cancer Patients

Background:

Taxanes (docetaxel and paclitaxel) are standard-of-care in early and metastatic breast cancer (MBC); yet, clinical benefit is limited to ~30% of patients. Predictive biomarkers are needed to optimize therapeutic selection and avoid unnecessary toxicity. Here, we present a machine learning (ML)-based predictor of taxane sensitivity, integrating in vitro pharmacogenomics with transcriptomic profiles from patient tumors.

Methods:

Multiple gene signatures were generated by (1) correlating gene expression data from Cancer Cell Line Encyclopedia (CCLE) with half-maximal inhibitory concentration (IC50) values for docetaxel from the Genomics of Drug Sensitivity in Cancer (GDSC) and (2) using a variance filter based on >6,000 patient biopsies. L2-regularized Cox regression was used to train and cross-validate models using data from a Danish MBC cohort (PMID: 31654283) treated with docetaxel (n = 164). Across 100 iterations, the best-performing model achieved a hazard ratio (HR) of 0.33 when stratifying by time to progression (TTP) at a 6-month cutoff. Independent validation was performed in paclitaxel-treated patients from the same cohort (n = 41) with no overlapping patients.

Results:

In the independent validation, a continuous model showed HR = 0.26 for a 50-point difference in score (two-sided p-value = 0.014, 95% CI = 0.09-0.76, TTP as endpoint) and HR = 0.35 in a model comparing patients with response scores >50 versus ≤50 (two-sided p-value = 0.0505, 95% CI = 0.12-1.00, TTP as endpoint). Pathway analysis showed enrichment of genes involved in mitotic cell cycle, chromosome segregation, and microtubule organization. In explorative analyses, the model did not predict outcomes of patients treated with endocrine therapy (fulvestrant, letrozole, exemestane), but was associated with response in those receiving epirubicin.

Conclusions:

We have developed and clinically validated a predictor of taxane response in MBC. This predictor integrates cell line data with patient tumor expression profiles and time-based disease progression data using ML. The model aligns with taxane mechanism of action and appears to capture drug-specific rather than prognostic effects. Future work will focus on validating this model in independent datasets of women receiving taxane therapy for MBC.

Authors and affiliations:

Troels Dreier Christensen (presenting author): Christensen, T.D. (1)

Jan Nart: Nart, J. (2)

Anna Sofie Buhl Rasmussen: Rasmussen, A.S.B. (3)

Jacob Hansen Niklassen: Niklassen, J.H. (2)

Beatrice Hahn: Hahn, B. (2)

Bent Ejlertsen: Ejlertsen, B. (4, 5)

Ann Søegaard Knoop: Knoop, A.S. (4, 5)

Frederik Otzen Bagger: Bagger, F.O. (6)

Simon Grund Sørensen: Sørensen, S.G. (7)

Ulla Hald Buhl: Buhl, U.H. (2)

Ida Kappel Buhl: Buhl, I.K. (2)

Emma van Boven: van Boven, E. (2)

Peter Buhl Jensen: Jensen, P.B. (2)

1: Department of Oncology, Herlev and Gentofte Hospital, Denmark

2: Aida Oncology ApS, Denmark

3: Paediatric Oncology Research Laboratory, Rigshospitalet, Denmark

4: Department of Oncology, Rigshospitalet, Denmark

5: Danish Breast Cancer Group, Rigshospitalet, Denmark

6: Department of Genomic Medicine, Rigshospitalet, Denmark

7: Department of Molecular Medicine, Aarhus University Hospital, Denmark

Contact

Jan Nart: jan@aidaoncology.com