Longitudinal Transcriptomics Reveals Heterogeneous Dynamics through the Course of Disease and Therapy in Breast Cancer

Background

Tumors are continuously evolving through their course of progression and treatment, a major process contributing to resistance to therapy. Increasingly, it becomes clear that unraveling the dynamics of gene expression over time, during stages of cancer progression and therapy, is fundamental to interpretation of tumor evolution and resistance mechanisms. To understand the emergence of resistance it is important to comprehensively profile matched biopsies from individual patients accompanied with their long-term clinical follow-up.

Methods

We collected triplets of archived samples from 33 individual patients that underwent neo-adjuvant (preoperative) treatment. Matched biopsies of pre-treatment and post-treatment tumor, as well as adjacent normal epithelium were included, together with normal breast tissues from 6 healthy individuals. Full transcriptome analysis was performed by mRNA sequencing, optimized for archived samples. Comprehensive clinical and pathological information was collected. A dedicated longitudinal pattern analysis method was developed to follow dynamic expression fluctuations of individual patients. Pathifier was used to calculate pathway deregulation scores for > 1300 canonical pathways.

Results

Principal component analysis showed clustering of the samples according to their type. Dynamic fluctuations across the 3 time-points were classified into 8 patterns, each representing a different scenario through the tumor progression and treatment stages. Genes were divided into two main types: 1. Genes sharing a common temporal expression pattern across most patients. These genes were associated with tumor progression pathways. 2. Genes that were divided into two or three dominant patterns and this division showed correlation with pathological response score. We identified 150 genes that their dynamics through the course of disease was significantly associated with response to treatment (Wilcoxon rank sum test). Furthermore, calculating pathway deregulation scores at each time point enabled to follow individual fluctuations in specific pathways. The heterogeneous dynamics in the pathway deregulation was correlated with response to therapy, in an attempt to pinpoint resistance-related pathways.

Conclusions

The longitudinal approach reveals heterogeneous dynamic behavior across patients through the course of disease. This individual dynamics has higher sensitivity than single-time point measurements in detecting clinically relevant genes that are associated with resistance to therapy and with tumor progression.