三阴性乳腺癌的细胞生态型与新辅助化疗应答

Fig. 1: Study design and main cell types in TNBC. a , Experimental workflow for scRNA-seq and spatial transcriptomic profiling of patients with TNBC. b , c , Uniform manifold approximation and projection (UMAP) of the unintegrated scRNA-seq data from 427,857 cells coloured by the patients ( b ) and the main cell types ( c ). d , Dot plot showing the seven top DEGs in each cell type. e , UMAP of the unintegrated scRNA-seq data coloured by NAC response. NA, not applicable f , Bar plot showing the average cell-type percentages in patients with pCR and RD. g , h , Patient P66 Xenium data showing a dot plot of the five top DEGs in each cell type ( g ), and the haematoxylin and eosin (H&E) image with spatial locations of main cell types ( h ). Scale bar, 1,000 μm. Panel a created in BioRender; Yan, Y. https://biorender.com/zjf9vol (2026).

Fig. 2: Patient-level archetypes of patients with TNBC identified from cancer cells. a , Unintegrated UMAP showing 49,275 single cancer cells that are coloured by patients ( N = 97). b , Relative expressions of top 15 marker genes across patients grouped by archetypes. BL, basal-like; IFR, interferon responsive; LSL, luminal secretory-like. c , Relative expressions of the known normal breast epithelial lineage genes from HBCA in the archetypes. d , Jaccard similarity of DEGs between archetypes and previous TNBC subtypes from bulk expression studies 7 , 9 . BLIA, basal-like immune-activated; BLIS, basal-like immune-suppressed; IM, immunomodulatory; LAR, luminal AR; M, mesenchymal; MES, mesenchymal; MSL, mesenchymal stem-like. e , Gene enrichment of the archetype signatures against the MSigDB cancer hallmark database. f , Relative expression of the known cancer genes that were differently expressed across archetypes. g , Xenium data of four representative patients for archetypes showing the H&E images, the locations of cancer cells (red dots) and the overall expression scores of archetype gene signatures in cancer cells. h , Contingency table showing the number of patients grouped by archetype and therapy response in this study (left) and the BrighTNess cohort (right). i , Multivariate Cox proportional-hazards model testing the associations between the archetype gene signature expressions and the overall survival in the METABRIC cohort. NPI, Nottingham prognostic index. j , Normalized RNA expression of clinically targetable genes across archetypes. Group-wise comparisons per gene were conducted by Wilcoxon rank-sum test and P values were adjusted for false discovery rate. Only comparisons showing adjusted P values <0.05 are labelled. CPM, counts per million. Box plots show the median (centre line), interquartile range (box limits) and 1.5× the interquartile range (whiskers). Scale bars 1,000 µm.

Fig. 3: Intra-tumoural metaprograms of cancer cells. a , Jaccard index between all ITH gene-expression programs that were grouped by metaprograms. b , Bar plot showing the number of patients in whom the metaprograms were detected. c , Relative expressions of the metaprogram marker genes across single cells in a patient example (P54). d , Xenium data in a patient example (P31) showing the H&E images, the locations of the main cell types and the in situ expressions of EPCAM with the marker genes of the M6-HLA and M5-interferon metaprograms. e , f , Box plots showing the average cellular abundance ( e ) and the average module score ( f ) of metaprograms in the comparison of pCR and RD. g , Box plot showing the normalized pseudo-bulk RNA expression of three representative genes of M7-S. h , Bar plot showing the average cell abundance of the computationally estimated cell cycling phases in patients with pCR and RD. i , Box plot comparing the percentage of cells with positive immunohistochemistry staining for Ki-67 (Wilcoxon test exact P value). j , Heatmap showing the relative gene expression of the major histocompatibility complex class I and class II genes across patients. k , Box plot showing the normalized pseudo-bulk RNA expression of three representative interferon-related genes of M5-interferon. l , Box plot showing the normalized pseudo-bulk RNA expression of three representative genes of M13-ER stress. In e , f , Wilcoxon rank-sum tests were conducted with P values adjusted by the Benjamini–Hochberg method. In g , j – l , the Wald test was conducted by DESeq2 with P values adjusted by the Benjamini–Hochberg method. Asterisks indicate adjusted P < 0.05. Box plots show the median (centre line), interquartile range (box limits) and 1.5× the interquartile range (whiskers). Scale bars, 100 μm.

Fig. 4: Immune and stromal cell states and their associations with NAC response. a , UMAP plot of the unintegrated scRNA-seq data showing 355,334 immune cells from 3 main cell types: myeloid cells, T/NK cells and B cells, with cell-state names and marker genes labelled. B mem -switched, switched-memory B cell; B N , naive B cell; GC-B, germinal centre B cell; mDC, mature dendritic cell; pDC, plasmacytoid dendritic cell. b – d , Cellular percentage difference in cell states of myeloid cells ( b ), T/NK cells ( c ) and B cells ( d ) in a comparison of patients with pCR and RD. e , UMAP plot of the unintegrated scRNA-seq data showing 23,248 stromal cells from 3 main cell types: fibroblast, endothelial cells and perivascular cells, with cell-state names and marker genes labelled. f – h , Cellular percentage difference in cell states of fibroblast cells ( f ), endothelial cells ( g ) and perivascular cells ( h ) in a comparison of patients with pCR and RD. In b – d and f – h , Wilcoxon rank-sum test was conducted with P values adjusted by the Benjamini–Hochberg method. Statistically significant values are coloured in red.

Fig. 5: Ecotypes and spatial niches of cancer metaprograms and TME cell states. a , Network graph showing the co-occurrences of cell identities (that is, cancer-cell metaprograms and immune and stromal cell states) denoted by nodes. Edge width denotes Spearman correlation coefficient (SCC) of cellular percentages of two connected nodes among patients. b , c , Relative co-occurrence of cell identities across archetypes ( b ) and response groups ( c ). Edge width denotes the geometric mean of the averaged z -score cellular percentages of the two connected nodes. Only edges with non-zero widths are shown. d , Heatmap showing the relative abundance of cell identities across the ten spatial niches N1 to N10. The hierarchical clustering tree of niches (columns) was built on the basis of the CellTrek’s spatial colocalization result. e , Bar plot showing the number of patients in which the spatial niches were detected. f , Bar plot showing the Pearson residuals of the chi-squared test between the number of cells assigned to ecotypes and niches. g , Xenium data in two patient examples (P56, P90) showing H&E images, spatial niches and the locations of four expanded windows (W1–W4). h – k , Xenium data of the expanded windows W1–W4 showing H&E images, spatial niches and cell identities together with the bar plot of cell counts within the spatial niche examples for the niche N2 ( h ), N9 ( i ), N4 ( j ) and N8 ( k ). Scale bars, 100 μm.