Multidimensional data integration to capture intrinsic and extrinsic mechanisms driving treatment benefit in breast cancer

Breast cancer is one of the most common tumours and the leading cause of cancer-related death among women worldwide. It can be classified into molecular subtypes based on the expression of estrogen receptor (ER), progesterone receptor (PR), and the amplification of human epidermal growth factor receptor 2 (HER2). However, even within the same subtype, tumours show differences in molecular profiles, clinical behaviour, and treatment response. This variability highlights the urgent need for the identification of new biomarkers capable of capturing the biological heterogeneity and developing personalised therapeutic strategies. Beyond malignant cells, tumours can be depicted as dynamic and evolving ecosystems, characterised by the interaction of different cell populations and biological components, such as stromal and immune cells, blood vessels, and the extracellular matrix. Within the tumour microenvironment (TME), cancer cells' interactions form a network of biological signals that collectively drive tumour progression, metastasis, and drug resistance. Understanding how cancer cells and their surrounding microenvironment interact represents a major challenge to develop novel therapies. High-throughput transcriptomic technologies have revolutionised the study of tumour complexity, enabling genome-wide gene expression profiling and integration with clinical and genomic data. This doctoral thesis leveraged transcriptomic analyses from multiple breast cancer clinical trials to characterise tumour heterogeneity across subtypes and therapeutic contexts, with a focus on translational applicability. In ER+/HER2– breast cancer, integrating two metagene-based scores—MKS (proliferation) and ERS (estrogen signalling)—captured the prognostic and therapeutic variability within this subtype. MKShi/ERSlo tumours showed poor outcome, endocrine resistance, high immune activation, and frequent PIK3CA mutations, suggesting potential benefit from PI3K inhibitors or immunotherapy. In HER2+/ER+ breast cancer, analysis of samples from the NA-PHER2 Phase II trial—testing a chemo-free combination of dual HER2 blockade and CDK4/6 inhibition—identified low ESR1 expression and high immune infiltration as the strongest predictors of pathological complete response, while TP53 mutations correlated with persistent proliferation. These results highlight biological bases for differential outcomes and support tailored treatment strategies. To address confounding factors in bulk transcriptomics, we developed the Breast Cancer Purity Score (BCPS), a robust and platform-independent metric to estimate tumour cellularity. Complementarily, a computational deconvolution pipeline was established to infer the relative abundance of TME populations from bulk RNA-seq data, leveraging breast cancer single-cell references to enhance interpretability. In TNBC, the transcriptional effects of different chemotherapy regimens revealed that anthracycline- and nab-paclitaxel–based treatments induce strong immunogenic responses, while carboplatin-based therapies may exert immunosuppressive effects, providing a rationale to optimise chemo-immunotherapy combinations. Finally, a trajectory-based longitudinal clustering method was developed to capture temporal evolution of gene set activities across treatment timepoints. Validated on synthetic and clinical datasets, this approach uncovered heterogeneous molecular trajectories and provided insights into therapy resistance mechanisms. Overall, this thesis demonstrates how transcriptomics, integrated with genomic and clinical data, can elucidate breast cancer complexity and its microenvironment. The developed methods and biomarkers—including tumour purity estimation, deconvolution, and longitudinal clustering—offer practical tools for patient stratification and therapeutic decision-making, contributing to the advancement of precision oncology and personalised breast cancer care.