Integrating Machine Learning-Predicted Circulating Tumor Cells (CTCs) and circulating tumor DNA (ctDNA) in Metastatic Breast Cancer: a proof of principle study on endocrine resistance profiling.

Abstract

The study explored endocrine resistance by leveraging machine learning to establish the prognostic stratification of predicted Circulating tumor cells (CTCs), assessing its integration with circulating tumor DNA (ctDNA) features and contextually evaluate the potential of CTCs-based transcriptomics. 1,118 patients with a diagnosis of luminal-like Metastatic Breast Cancer (MBC) were characterized for ctDNA through NGS before treatment start, predicted CTCs were computed through a K nearest neighbor algorithm. Differences across subgroups were analyzed through chi square or Fisher's exact test according to sample size and corrected for False Discovery Rate. Differences in survival were tested by log-rank test and uni- and multivariable Cox regression. CTCs transcriptomics was performed through RNAseq after sorting with DEPArray NxT. Univariable and multivariable analysis adjusted for ctDNA alterations revealed a significant impact of CTCs predictive stratification on both progression-free survival (PFS) and overall survival (OS). Alterations in RTK and ER pathways were significantly correlated with predicted-Stage IV_aggressive. The combined impact of CTCs stratification and RTK/ER pathway alterations influenced patient outcomes, with predicted-Stage IV_aggressive having a negative impact on PFS regardless of the mutational status. The pilot exploratory CTCs transcriptomics analysis showed transcriptional changes linked to cell proliferation such as under expression of MALAT1 and overexpression of GREM1, GPR85 and OCM. Our data underline the potential of an integration between ctDNA and CTCs, both through quantification and transcriptomic analysis, for a deeper understanding of tumor biology and treatment response in HR-positive, HER2-negative MBC.