Using prognostic signatures and machine learning to identify core features associated with response to CDK4/6 inhibitor-based therapy in metastatic breast cancer

CDK4/6 inhibitors in combination with endocrine therapy are widely used to treat HR+/HER2- metastatic breast cancer leading to improved progression-free survival (PFS) compared to single agent endocrine therapy. Over 300 patients receiving standard-of-care CDK4/6 inhibitor combination therapy for metastatic disease were enrolled at a single institution. Clinical, pathological, and gene expression data were employed to define determinants for PFS duration. Visceral disease (HR 1.55, p=0.0013), prior endocrine therapy (HR 2.34, p<0.001), and the type of endocrine therapy (HR 2.16, p<0.001) were highly associated with PFS duration. Multiple pre-defined gene expression signatures were employed to determine association with response to CDK4/6 inhibitor-based therapy. Random survival forest was applied to define key gene expression and clinical features associated with PFS and develop a predictive model. The time to progression predicted by this model was related to the median PFS observed in PALOMA-2/3 and PEARL studies. Interrogating genes identified as highly significant across all studies indicated common enrichment of gene networks associated with cell cycle and estrogen receptor signaling. These findings indicate that there are common features from real-world use of CDK4/6 inhibitors that could be used to infer time to progression and better inform treatment. Patient tissue samples were obtained during standard of care treatment from various clinical timepoints and were sent to HTG Molecular Diagnostics, Inc. for targeted sequencing using their HTG EdgeSeq Oncology Biomarker Panel (OBP) consisting of 2549 cancer-associated genes. Raw data was assembled from five batches of separate HTG runs. Batch effect removal was performed using the ComBat-seq function from the sva R package (v3.44.0). Data was then normalized on this batch effect-corrected raw read count matrix using the edgeR Bioconductor package (V3.38.4). Finally, a pseudo-count of 1 was added to each value of the normalized data matrix and then log2 transformed. *************************************************************** the raw data files are unavailable from an out-of-business vendor. ***************************************************************