Fig. 1: Workflow exploring composition and communication of phenotypically diverse cancer and non-cancer cells within the tumor microenvironment of early-stage ER+ breast cancer patient tumors resistant or sensitive to CDK4/6i and endocrine therapy.

Serial single-cell RNA-seq data was generated for 62 patients by applying 10x Genomics to 173 tumor biopsy samples, collected over 3 treatment time points (Pre-treatment baseline (Day 0), Early follow-up (Day 14) and Post-treatment (Day 180)). A total of 424,581 high quality cells were transcriptionally profiled, with cancer and non-cancer cell types classified using established machine learning classifiers (see methods). Figure S1/2 show UMAP dimension reduction plots of single cell gene expression profiles, supporting cell type classification. The TME compositions of CDK4/6i-resistant/sensitive tumors were contrasted based on cell type frequencies, using pairwise distance-based dimension reduction. This identified archetypical tumor ecosystem compositions and their association with CDK4/6i response. Phenotypically diverse cell type subpopulations were resolved using cell type specific UMAP dimension reduction of ssGSEA profiles. Communication pathways through which cell subpopulations may signal were defined by 1444 ligand-receptor communication pairs with known protein-protein interaction. Networks of communication between the phenotypically diverse populations of cancer and non-cancer cell types constituting the tumor were measured, accounting for both composition and phenotype (see methods and schematic overview in Figure S3). Networks of diverse ligand-receptor communications between cancer and non-cancer cell types were compared between treatment-resistant (growing) and sensitive (shrinking) tumors. Divergent aspects of cell type communication were identified using a bootstrap comparison and verified in the independently profiled validation cohort. Specific ligand-receptor communications associated with resistance were used to predict and subsequently verify consequences on the phenotype and abundance of signal receiving cell types. The CDK4/6i treatment effects on immune cell abundance in the tumor microenvironment were then compared to temporal changes in peripheral blood mononuclear cells (PBMC) during treatment in the same patient cohort. In vitro experiments were then conducted to validate predicted effects/side-effects of CDK4/6i on cancer/non-cancer cell proliferation. Finally, we examined whether modulation of communications associated with CDK4/6i-resistance in patient tumors can overcome CDK4/6i side-effects on non-cancer cells and improve control of cancer cell growth.