Transcriptomic Profiling of Mouse Mammary Tumors Enables Prognostic and Predictive Biomarker Discovery for Human...

Area of Science:

• Oncology
• Genomics
• Immunotherapy
• Biomarker Discovery

Background:

• Breast cancer biomarker development is hindered by a lack of comprehensive datasets linking tumor molecular profiles to clinical outcomes.
• Existing datasets often lack sufficient diversity in genetic backgrounds, tumor subtypes, and immune microenvironments, limiting generalizability.

Purpose of the Study:

• To create and validate a large, well-annotated preclinical dataset of mouse mammary tumor models for prognostic and predictive biomarker discovery.
• To assess the utility of mouse models in predicting human breast cancer patient survival and response to immune checkpoint inhibition (ICI) and chemotherapy.

Main Methods:

• Generated 26 immunocompetent mouse mammary tumor models with diverse characteristics.
• Collected survival data under no treatment, ICI, and chemotherapy (carboplatin/paclitaxel).
• Performed RNA sequencing on baseline and 7-day on-treatment tumor samples.
• Trained machine learning models (Elastic Net, XGBoost, random forests, support vector regression) using murine gene expression data to predict outcomes in human datasets.

Main Results:

• A machine learning model trained on baseline murine tumor gene expression predicted survival in human breast cancer datasets comparably to existing assays.
• Models predicting response to immune checkpoint inhibition (ICI) using murine data showed predictive power on human ICI-treated datasets, with 7-day on-treatment models performing better.
• A chemotherapy response predictor performed well in mice but did not generalize to human cohorts, suggesting differences in chemosensitivity mechanisms.
• Elastic Net demonstrated the best performance and interpretability among tested machine learning algorithms for survival prediction.

Conclusions:

• Mouse models can effectively recapitulate conserved cancer biology relevant to prognosis and immune checkpoint inhibition response in human breast cancer.
• The generated preclinical dataset serves as a valuable resource for discovering and validating novel prognostic and predictive biomarkers.
• Further research is needed to develop robust predictors for chemotherapy response based on conserved mechanisms.