NCI Division of Cancer Biology Junior Investigators Meeting (Abstract): Submission #10

Submission information
Submission Number: 10
Submission ID: 194041
Submission UUID: 59ef355f-95c9-44e7-af49-e3d87943e526
Submission URI: /dcb/ji-meeting/abstract

Created: Tue, 09/08/2026 - 09:06
Completed: Tue, 09/08/2026 - 09:06
Changed: Tue, 09/08/2026 - 09:06

Remote IP address: 10.208.28.41
Submitted by: Anonymous
Language: English

Is draft: No
First Name Anna
Middle Initial
Last Name Michmerhuizen
Degree(s) Ph.D.
Position/Title/Career Status Postdoctoral Research Associate
Organization University of North Carolina at Chapel Hill
Organization Address Chapel Hill, NC
Email anna_michmerhuizen@med.unc.edu
Abstract Category Use my abstract for team formation only (do not consider it for a presentation)
Abstract Keywords breast cancer, metastasis, metastatic tropism, computational modeling, precision oncology
Abstract Title Identifying drivers of organ-specific breast cancer metastasis
Abstract Metastatic breast cancer typically progresses to grow in critical organs and is the primary cause of breast cancer-related deaths. We and others have shown that organotrophic metastatic behavior is associated with the molecular subtype of the primary tumor. Specifically, the basal-like “intrinsic” subtype metastasizes preferentially to the lung and brain, HER2-enriched subtype to the liver, and luminal subtypes to the bone. Using data from 2486 human breast tumors where many have progressed to metastases (bone, brain, liver, lung, or multi-metastases [2+ sites]), we performed supervised machine learning to develop logistic regression models predictive of site-specific metastasis employing elastic net regularization for selection of interpretable features and model building. First, models predictive of distant metastasis were successfully built to classify patients based on whether their tumor will metastasize (AUC = 0.67). We next trained models predictive of site-specific metastasis, including a luminal bone metastasis model (AUC = 0.75), where selected features identify patients with bone metastases only (i.e. no visceral metastatic sites [brain, liver, lung]). Tumor cell intrinsic and microenvironment extrinsic features were selected in the final model. A model for multi-metastases was also developed (AUC = 0.68), and features associated with multi-site metastasis include hypoxia, clinical HER2 status, correlation to the HER2-enriched molecular subtype, and lymphovascular invasion. These overall modeling results suggest two distinct metastatic phenotypes (bone and viscera), which are supported by the performance of published site-specific metastasis signatures. Our models of site-specific metastasis may provide clinically useful biomarkers and nominate unique features for mechanistic studies.