Predicting epirubicin response in Danish patients with breast cancer.

Abstract

539 Background: Anthracyclines are highly effective chemotherapy drugs. However, they are toxic and feared for their cardiotoxicity. There is a need for improved personalized prediction of treatment benefit. For example, state-of-the-art guidelines (NCCN Guidelines Version 1.2024 and meta-analysis PMID: 37061269) mark anthracyclines as optional adjuvant treatment of estrogen receptor (ER) positive, HER2 negative early high-risk breast cancer. A biomarker to predict whether a patient will benefit from anthracycline treatment could potentially lead to increased cure rate and overall reduced toxicity. However, no such test exists. Here we introduce a clinically validated, predictive test based on clinical tumor transcriptome data and the cell line panel NCI60 that can forecast epirubicin response in breast cancer patients with high accuracy. Methods: Microarray gene expression levels and growth inhibition levels (GI50) from the NCI60 cell line panel were acquired from NCBI to build a prediction algorithm. A tumor-specific variance-based filter based on >6.000 patient tumors was applied to select the most relevant genes. AUC (Area Under Curve) was used to evaluate the performance of numerous configurations and combinations of parameters that led to different gene-signature sets. Response scores based on the gene expression level of all genes in each signature were generated for each sample in a published cohort (PMID: 23340299) and a final signature with 141 genes was the best-performing signature (Progressive Disease versus Response AUC 0.81 and ANOVA-test p-value < 0.01). The model was validated on a cohort of Danish breast cancer patients treated with epirubicin in the metastatic setting (n=153), with time-to-progression (TTP) as endpoint (PMID: 31654283, PMID: 30099635). Finally, we used an established method (PMID: 30289602) to test if the gene signature was correlated between microarray- and RNA sequencing-based datasets. Results: The hazard ratio (HR) for the best-performing signature was 0.31 when comparing patients with a difference of at least 50 in response score (two-sided p-value = 0.012, 95% confidence interval 0.12-0.77, endpoint TTP, continuous scoring). This was consistent in a multivariate model including ER and HER2. Using the data and method provided by Pedersen et al (PMID: 30289602), we also saw a high correlation between matching microarray and RNA sequencing datasets for the genes in our model (mean Pearson’s r = 0.87). Conclusions: Here is presented a development and clinical validation of an accurate epirubicin response signature for breast cancer patients. The expression of the genes in the signature is highly correlated between microarray- and RNA sequencing data. Moving forward, we will seek to validate the model independently on RNA sequencing data. Our aim is that these models will be accepted for prospective clinical tests at local hospitals in Denmark in early 2025.