Researchers at City of Hope and the University of California, Berkeley, have developed a microfluidic and machine learning platform that assesses breast cancer susceptibility by analyzing the mechanical behavior of individual mammary epithelial cells. Their findings, published in eBioMedicine, could enable earlier, personalized risk stratification that complements existing models.

“Most women who develop breast cancer have no known genetic mutation or strong family history, leaving them without a reliable way to assess their personal risk,” lead author Stefan Hinz, PhD, of the Department of Population Sciences, Beckman Research Institute, City of Hope, Duarte, California, told ASCO AI in Oncology. “MechanoAge addresses this gap by measuring how individual breast cells physically respond to stress, a functional readout of biologic aging that appears linked to cancer susceptibility.”

The platform generates a risk score based on the mechanical properties of a woman’s own cells, he explained, even when standard genetic testing results are unremarkable. Using the approach, the investigators identified women at elevated risk of breast cancer, despite a normal histology, and linked the associated mechanical aging signatures to a cytoskeletal protein.

Study and Model Methods

Primary human mammary epithelial cells from women spanning different ages and breast cancer risk backgrounds were analyzed using mechano-node-pore sensing, a high-throughput microfluidic platform that evaluates how individual cells deform, recover, and respond to mechanical stress induced by passage through a liquid-filled constriction channel.

Using these data, the investigators developed two outputs: the machine learning classifier MechanoAge, which estimates chronological age (< 35 vs > 50) from cellular mechanical phenotypes, and the biologic age–based risk index Mechano-RISQ. The model was developed using data from 18 women, comprising 1,381 cells in the training set and 661 cells in the validation set, and incorporated 10 cell-level mechanophenotypic features.

For MechanoAge, the investigators used a stacked ensemble framework combining machine learning algorithms of bagged decision trees, random forests, and extremely randomized trees to model the relationship between predictor variables and age categories, with generalized boosted regression as the meta-learner. Model training was performed using repeated k-fold cross-validation to ensure performance estimation optimized for receiver operating characteristic area under the curve (ROC AUC). MechanoAge achieved an AUC of 0.95 in validation and 0.91 in an independent external sample cohort comprising 281 cells from 5 samples. Mechano-RISQ was calculated by comparing the proportion of cells classified as “older” by the model to a baseline misclassification rate derived from average-risk human mammary epithelial cells, thereby quantifying age-discordant mechanical states at single-cell resolution.

To explore potential molecular drivers of these mechanical states, the investigators further evaluated the cytoskeletal protein keratin 14 (KRT14)—differentially expressed in luminal epithelial cells of older women—through overexpression and knockdown experiments.

Key Findings

Epithelial cells derived from histologically normal tissue of young BRCA1/2 mutation carriers (n = 4), individuals with a family history of breast cancer (n = 3), and contralateral breast tissue from patients with unilateral tumors (n = 9) exhibited higher Mechano-RISQ scores vs age-matched controls (n = 18). These elevated scores are consistent with accelerated biologic aging, the investigators wrote.

Overexpression of KRT14 was found to drive a biologically aged phenotype in cells derived from younger women, whereas its knockdown partially restored a more youthful state in cells from older women. Mass cytometry by time-of-flight profiling and modeling revealed that modulation of KRT14 alters protein expression signatures linked to aging and disease risk.

“Together with prior molecular and epigenetic studies, these findings support a model in which accelerated biologic aging of mammary epithelia may underpin breast cancer susceptibility across genetic and nongenetic risk groups,” the investigators concluded. “Mechanical phenotyping captures an integrative cellular state that reflects underlying molecular networks rather than single biomarkers.”

Dr. Hinz added, “If validated in broader, more diverse populations, this approach could provide a more individualized way to identify women at elevated risk before cancer develops.”

DISCLOSURES: The study was funded by grants from the National Institutes of Health and the American Cancer Society—Fred Ross Desert Spirit Postdoctoral Fellowship. For full disclosures of the study authors, funding information, and data and code availability, visit thelancet.com.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.