Tamoxifen Response at Single Cell Resolution in Estrogen Receptor-Positive Primary Human Breast Tumors

Abstract

In ER+/HER2- breast cancer, multiple measures of intra-tumor heterogeneity are associated with worse response to endocrine therapy. To investigate heterogeneity in response to treatment, we developed an operating room-to-laboratory pipeline for the collection of live human tumors and normal breast specimens immediately after surgical resection for processing into single-cell workflows for experimentation and genomic analyses. We demonstrate differences in tamoxifen response by cell type and identify distinctly responsive and resistant subpopulations within the malignant cell compartment of human tumors. Tamoxifen resistance signatures from 3 distinct resistant subpopulations are prognostic in large cohorts of ER+ breast cancer patients and enriched in endocrine therapy resistant tumors. This novel ex vivo model system now provides a foundation to define responsive and resistant sub-populations within heterogeneous tumors, to develop precise single cell-based predictors of response to therapy, and to identify genes and pathways driving resistance to therapy.

Extended Data Fig. 1.. Effect of time in suspension on cell population abundance and gene expression.

(A) UMAP plot of all cells from Normal_01 with immediate library creation and control suspension color coded by cell type. (B) UMAP plot of cells color coded by sample and demonstrating clustering by cell type and not by time in suspension. (C) Feature plots of canonical breast epithelial markers. (D) Bar chart comparing the abundance of discrete cell populations in the immediately sequenced sample and after 12 hours in suspension demonstrating fewer cells after time in suspension without systematic depletion of any cell population. (E) Volcano plot comparing gene set enrichment and depletion in up and down regulated genes with time in suspension. Significant gene sets were defined as a gene ratio of ≥ 0.2 and a log10qvalue of < −2 or > 2.

Extended Data Fig. 2.. Cell type-specific tamoxifen regulated genes in normal human breast tissue.

(A) Venn diagram showing overlapping number of up-regulated genes between basal, luminal, and mature luminal normal breast cells treated with tamoxifen. Significance was defined as q-value < 0.01. (B) Venn diagram showing overlapping number of down-regulated genes between basal, luminal, and mature luminal normal breast cells treated with tamoxifen. (C) Cell type-specific heatmaps of top up- and down-regulated genes by cell type.

Extended Data Fig. 3.. Tumor-specific cell annotations and tamoxifen-regulated genes.

(A) Bar chart comparing the total number of cells sequenced per treatment condition, color-coded by cell type. Malignant cells were further classified into PAM50 intrinsic subtypes by the nearest neighbor method using centroid vectors obtained from single cells defined by a single-cell method of intrinsic subtype classification (SCSubtype)40. (B) Bar chart showing number of up- and down-regulated (significance defined as log2fold change > 0.25 and q-value < 0.01) with tamoxifen treatment in the computationally identified malignant tumor cells by tumor.

Extended Data Fig. 4.. Cell type-specific response to tamoxifen in primary human breast tumors.

Venn diagrams of (A) up- and (B) down-regulated genes with tamoxifen treatment by cell type. Most tamoxifen-regulated genes are cell type-specific with minimal overlap across compartments.

Extended Data Fig. 5.. Characteristic gene expression profile of tamoxifen resistant clusters.

(A) Heatmaps showing differentially expressed genes between resistant clusters (3,12, 19) and the tamoxifen-sensitive cluster 2. (B) Volcano plot comparing gene set enrichment and depletion of resistant clusters. Significant gene sets were defined as a gene ratio of ≥ 0.2 and a log10qvalue of < −2 or > 2. To simplify visualization, only gene sets with names containing relevant terms were plotted. Supplementary Table 7–9 provides a comprehensive list of the enriched and depleted gene sets for the tamoxifen-resistant clusters (Clusters 3, 12, and 19) compared to the tamoxifen response cluster (Cluster 2).

Extended Data Fig. 6.. Prognostic signatures of resistant cell populations.

(A) Clinically annotated transcriptional data from ER+ patients in the METABRIC and SCAN-B studies was analyzed to evaluate the prognostic significance of resistant cell-specific signatures. Kaplan-Meier (KM) curves demonstrate that signature scores from cluster 3 and cluster 12 were associated with worse survival in SCAN-B but not in METABRIC. The cluster 19 signature was prognostic in METABRIC (HR 2.2, p< 0.001) and SCAN-B (HR1.9, p=0.002). Significance was determined using the log-rank test. (B) We obtained data of clinically annotated endocrine therapy sensitive and resistant tumors from Xia et al51. All three tamoxifen resistance signatures were enriched in resistant tumors compared to sensitive. Significance was determined using the Wilcoxon Sum-Rank test, ***p<0.001.

Figure 1.. Characterization of cell type specific response to tamoxifen in normal human breast tissue.

(A) Graphical representation of operating room to single-cell sequencing workflow. The breast schematic was created using BioRender.com. (B) UMAP plot of all scRNA-seq cells from two normal breast tissue samples. Cells were color-coded by cell type. (C) UMAP plot of scRNA-seq cells color-coded by sample and treatment condition. (D) Feature plots showing the expression of select epithelial markers in normal breast cells. (E) Bar chart comparing the total number of normal breast cells sequenced per treatment condition, color-coded by cell type. (F) Heatmap of gene sets enriched and depleted in differentially regulated genes in tamoxifen-treated cells relative to control cells by cell type demonstrating distinct biologic activity of tamoxifen in different cell compartments. Gene set enrichment was determined using enricher and p-values reported after correction for false discovery.

Figure 2.. T47D response to tamoxifen at a single cell level.

(A) Illustration showing T47D scRNA-seq workflow. (B) UMAP plot of T47D scRNA-seq cells color-coded by groupA (luminal A-like subpopulation) or groupB (luminal B-like subpopulation). (C) UMAP plot of T47D scRNA-seq cells color-coded by treatment condition. (D) Volcano plot showing enriched and depleted gene sets after tamoxifen treatment in T47D cells. Significant gene sets were defined as a gene ratio of > 0.125 and a log10qvalue of < −4.5 or > 4.5, and a thresholding limit of 10 was applied when log10qvalue > 10 for visualization. (E) Feature plots showing the expression of luminal epithelial markers and proliferation score in T47D cells.

Figure 3.. Tamoxifen response in 10 primary ER+/HER2- breast tumors.

(A) UMAP plot of all scRNA-seq cells from 10 breast tumor samples. Cells were color-coded by cell type. Malignant epithelial cells (Epi. Tumor) were distinguished from normal epithelial cells (Epi. Non-tumor) by inferred copy number changes using InferCNV. (B) UMAP plot of all cells color-coded by tumor and treatment condition. (C) Feature plots showing the expression of luminal epithelial markers and proliferation score. (D) Box-plots of early estrogen response genes (CCND1, PGR, and GREB1) in the InferCNV+ malignant cells comparing expression in control and tamoxifen treated cells. Significance was determined using the Wilcoxon Sum-Rank test *p-value<0.05, **p-value<0.01. Sample PAM50 subtypes determined on bulk mRNA sequencing are denoted by color-code. (E) Gene set enrichment heatmap showing distinct tamoxifen response within InferCNV+ malignant cells across tumors.

Figure 4.. Targeted analysis of 4 tamoxifen responsive tumor pairs.

(A) UMAP plot of scRNA-seq cells from 4 tumors that demonstrated depletion of scTAM-response-T47D signature. Cells are color-coded by cell type. (B) UMAP plot of scRNA-seq cells from 4 tumor pairs, color-coded by tumor and treatment condition. (C) Heatmap of upregulated and downregulated gene sets in tamoxifen-treated tumor cells relative to control cells. (D) Volcano plot showing enriched and depleted gene sets after tamoxifen treatment in malignant cells from 4 ER+HER2- tumor pairs. Significant gene sets were defined as a gene ratio of > 0.085 and a log10qvalue of < −10 or > 10 for visualization. (E) Kaplan-Meier (KM) curve for overall survival using two independent clinically annotated datasets with transcriptional data. ER+ patients were assigned a centroid score of our malignant cell-specific tamoxifen resistance signature (scTAM-resistance-M) and stratified into high and low signature score. High signature score is associated with significantly worse overall survival in patients in METABRIC (HR 1.63, p=0.023) and SCAN-B (HR 2.94, p=0.002). Statistical significance was assessed by the log-rank test and the estimates of survival probabilities and cumulative hazard with a univariate Cox proportional hazards model.

Figure 5.. Identification and characterization of resistant tumor cell subpopulations.

(A) Individual cells were assigned a score based from the T47D tamoxifen response signature compared to cluster matched untreated signature score. (A) Application of response score to UMAP plot demonstrated 3 distinct clusters with enriched signature score on treatment (Cluster 3, 12, 19) and one tamoxifen sensitive cluster with depletion of response score (Cluster 2). (B) Stacked bar chart showing distribution of cells within 22 distinct clusters, color-coded by tumor and treatment condition. (C) UMAP plot of scRNA-seq cells from 4 tumors with depleted T47D tamoxifen response score. Cells were color-coded by cluster. (D) Kaplan-Meier (KM) curve demonstrates the prognostic significance of the cluster 19 signature (scTAM-resistance-C19), which was evaluated using the METABRIC dataset. Higher signature score predicted worse overall survival (HR 2.17, p<0.001). Survival curve differences were calculated by the log-rank test and the estimates of survival probabilities and cumulative hazard with a univariate Cox proportional hazards model.