Deciphering the spatial landscape and plasticity of immunosuppressive fibroblasts in breast cancer

Abstract

Although heterogeneity of FAP+ Cancer-Associated Fibroblasts (CAF) has been described in breast cancer, their plasticity and spatial distribution remain poorly understood. Here, we analyze trajectory inference, deconvolute spatial transcriptomics at single-cell level and perform functional assays to generate a high-resolution integrated map of breast cancer (BC), with a focus on inflammatory and myofibroblastic (iCAF/myCAF) FAP+ CAF clusters. We identify 10 spatially-organized FAP+ CAF-related cellular niches, called EcoCellTypes, which are differentially localized within tumors. Consistent with their spatial organization, cancer cells drive the transition of detoxification-associated iCAF (Detox-iCAF) towards immunosuppressive extracellular matrix (ECM)-producing myCAF (ECM-myCAF) via a DPP4- and YAP-dependent mechanism. In turn, ECM-myCAF polarize TREM2+ macrophages, regulatory NK and T cells to induce immunosuppressive EcoCellTypes, while Detox-iCAF are associated with FOLR2+ macrophages in an immuno-protective EcoCellType. FAP+ CAF subpopulations accumulate differently according to the invasive BC status and predict invasive recurrence of ductal carcinoma in situ (DCIS), which could help in identifying low-risk DCIS patients eligible for therapeutic de-escalation.

Fig. 1. In silico analysis of FAP+ CAF plasticity in human breast cancer.

A UMAP combining FAP+ CAF from BC (Left) and fibroblasts from healthy breast tissues (Right, n = 15,667 cells), colored by cluster identity. B PI16 gene expression. C Differential gene expression between Detox-iCAF and normal fibroblasts. p values from two-sided Wilcoxon rank sum test. In red, genes with adjusted p value < 0.05. Trajectory inferences on BC FAP+ CAF scRNA-seq dataset from ref. inferred by PAGA tree (10 random downsampling) (D), STREAM trajectory showing cluster proportion along pseudotime (E) and Monocle 3 (F). G Expression of FAP+ CAF cluster markers according to Monocle 3 pseudotime and colored according to cluster identity. H Left, velocities from scVelo on the FAP+ CAF UMAP. Right, spliced/unspliced phase portrait and expression on UMAP for COL5A2 (Top) and COL1A2 (Bottom) genes. I Percentages of iCAF (ANTXR1−) and myCAF (ANTXR1+) clusters among FAP+ CAF cultured on collagen-coated or plastic dishes (n = 6 independent experiments). p value from two-sided Fisher’s Exact test. J Same as (I) for FAP+ CAF cluster identity using specific markers by flow cytometry (n = 6). K Flow cytometry plots showing FAP+ CAF cluster-specific surface markers in sorted primary FAP+ CAF. L Percentage of each FAP+ CAF cluster among FAP+ CAF (flow cytometry data) after co-culture of Detox-iCAF with MCF7. Timepoints indicate the duration of co-culture (n = 3 independent experiments). M–O Same as (L) for co-culture of Detox-iCAF with T47D (M), MDA-MB-231 (N) and MCF10A (O). P, Q Same as (L) for co-culture of IL-iCAF (P) or IFNγ-iCAF (Q) with MCF7. All data are mean ± SEM. R Left, UMAP of fibroblasts from scRNA-seq data following TN BC cell injection from ref. colored by cell identity (n = 3363 fibroblasts). Right, quantification at 0, 14 and 28 days after tumor implantation. S Same as (R) for fibroblasts from scRNA-seq data following subcutaneous injection of PDAC cell line in WT mice (n = 23,675 fibroblasts) (Left) and in Lrrc15-diphteria toxin receptor knock-in mice (n = 21,306 cells) (Right) from ref. . Quantification after tumor implantation at 0, 10, 17, 24 and 31 days. Source data are provided as a Source Data file.

Fig. 2. Identification of DPP4- and YAP-1-dependent transitions of Detox-iCAF into ECM-myCAF.

A Volcano plot showing differential gene expression in FAP+ CAF from the direct transition (red cells in Supplementary Fig. 2J) compared to other CAF from the BC scRNA-seq dataset. p values from two-sided Wilcoxon rank sum test. In red, genes with adjusted p value < 0.05. B DPP4 expression in a subset of Detox-iCAF (transitional Detox-iCAF). C DDP4 expression in FAP+ CAF clusters from scRNA-seq data of BC (Left) and PDAC (Right) mouse models^,. D Topmost variable transcription factors in CAF in direct/indirect transition and in other CAF. E Expression of TEAD/YAP1-target genes in FAP+ CAF clusters. F Same as (D) for scRNA-seq data from ref. (Left) and ref. (Right). G Representative western blot showing DPP4 silencing in Detox-iCAF at the beginning of the co-culture (t0) from three independent experiments. Actin is internal control for protein loading. H Percentages of Detox-iCAF, Wound-myCAF and ECM-myCAF clusters among FAP+ CAF after co-culture of MCF7 with Detox-iCAF silenced (siDPP4) or not (siCtrl) for DPP4 (n = 3 independent experiments). p values from two-sided Student’s t test. I Same as (H) showing the fraction of each FAP+ CAF cluster with/without DPP4 silencing (n = 3). p values from two-sided Fisher’s Exact test. J Same as (G) for YAP1 silencing. K Same as (H) after co-culture of MCF7 with Detox-iCAF silenced (siYAP1) or not (siCtrl) for YAP1 (n = 3). L Same as (I) with/without YAP1 silencing (n = 3). M Same as (G) showing DPP4 and YAP1 silencing. N Same as (H) after co-culture of MCF7 with Detox-iCAF silenced (siDPP4/siYAP1) or not (siCtrl) for both DPP4 and YAP1 (n = 3). O Same as (I) with/without DPP4 and YAP1 silencing (n = 3). P Western blots showing DPP4 and YAP1 protein levels in Detox-iCAF silenced either for DPP4 or YAP1 at 3 timepoints of the co-culture with MCF7. Q Same as (H) with Detox-iCAF silenced (siTGFBRII) or not (siCtrl) for TGFBRII (n = 3). R Same as (I) with/without TGFBRII silencing (n = 3). All data are mean ± SEM. Source data and exact p values are provided as a Source Data file.

Fig. 3. Spatial organization of breast cancer microenvironment.

A H&E images of representative Lum BC sections processed by Visium and annotated by pathologists. Tumor areas are colored in red, with cancer epithelial cells in dark red and intra-tumor stroma in light red. Invasive margins are highlighted in gray. Normal peritumor tissues include normal ducts and lobules in brown and peritumor stroma in yellow. T lymphocyte aggregates are in blue. (N = 17 sections in total). Scale bars = 500 μm. B UMAP of 73,426 cells from 43 patients (34 BC patients and 9 healthy donors) encompassing 39 different cell types and states and composing a comprehensive BC cellular atlas. C Deconvolution at single cell-like resolution based on the cellular atlas showed in (B) on a representative BC section (see also Supplementary Fig. 5 for deconvolution of additional BC sections). Each dot shows one single cell and the different colors represent distinct cell types and states. Black arrowheads indicate normal lobules and ducts (Panel epithelial cells) and their co-localization with TGFβ-myCAF (Panel FAP+ CAF/myCAF clusters). Dashed lines delineate the invasive margin. Scale bars = 500 μm. D Heatmap of the median proportion of each cell state among the corresponding cell type within each pathological compartment (Tumor cell-enriched areas; Intratumor stroma; Peri-tumor stroma; Lymphocyte aggregates and Normal ducts and lobules), as shown in (A). Exact p values are shown for significant (p < 0.05) enrichment (red and orange) or depletion (yellow) (% indicated on scale bars) of a cell state compared to the others within a particular pathological annotation. p values from two-sided Wilcoxon (one versus all) rank sum test. Source data are provided as a Source Data file.

Fig. 4. Identification of shared spatial cellular compositions across patients, called EcoCellTypes.

A Distribution of FAP+ CAF, CAP and TAM populations according to the distance to cancer cells (in µm). Distances are computed between the closest cancer cell identified by deconvolution and FAP+ CAF clusters, CAP and TAM in the 17 sections. Cell types are ranked based on their median distance to cancer cells. B Spatial distribution of 11 cellular niches in 3 representative BC patients. C Heatmap showing the mean cell type composition per niche identified on the 17 sections. Values are centered and scaled per cell type and state. Hierarchical clustering in rows defines 10 different groups of co-occurring cell types in the cellular niches referred to as EcoCelltypes (ECT). D–F Data from the METABRIC cohort (N = 1234 BC patients). D Proportions of each ECT in Lum A (N = 487), Lum B (N = 368), HER2 (N = 193) and Basal-like TN (N = 186) BC subtypes. p values from Mann–Whitney test. E Heatmap and clustering of all BC samples (columns) from the METABRIC cohort showing 4 subgroups of patients (C1 = 263, C2 = 483, C3 = 227, C4 = 261) with different ECT enrichments. F Left, Kaplan–Meier curves showing overall survival of the 4 BC patient subgroups (C1–C4) stratified in the heatmap. p value from Log-rank test. Right, Distribution of the BC molecular subtypes within the four subgroups (C1–C4) of BC patients. In all boxplot the center line, box limits and whiskers indicate the median, upper and lower quartiles and 1.5 × interquartile range. Source data are provided as a Source Data file.

Fig. 5. Interactions between FAP+ CAF clusters and immune cells in BC.

A CellChat dominant sender and receiver plot showing incoming and outgoing interaction strength for each cell type identified in the BC atlas (73,426 cells from 43 patients). The size of each circle corresponds to the total number of significant interactions, colored per cell type. B Scatter plots with linear regression lines showing correlations in the content of FAP+ CAF clusters quantified by flow cytometry in BC (N = 87 patients). p values from two-sided Pearson’s correlation test. C Same as (B) analyzing correlations between the content in ECM-myCAF, TGFβ-myCAF, Wound-myCAF and Detox-iCAF with TREM2+ or FOLR2+ macrophages in BC (N = 25). D Bar plot showing the percentages (%) of migration of CD14+ monocytes after 6 h of transwell co-culture with FAP+ CAF clusters. Data are mean ± SEM (n = 4 independent experiments). p values from two-sided Student’s t test. E % of CD14+ CD16+ myeloid cells among total CD14+ monocytes after 24 h of co-culture with FAP+ CAF clusters (Detailed gating strategy in Supplementary Fig. 9A). Data are mean ± SEM (n = 9). p values from two-sided Student’s t test. F Same as (E) for TREM2+ macrophages. p values from two-sided Student’s t test. G Same as (E) for FOLR2+ macrophages. p values from two-sided Mann–Whitney test. H % of FOXP3+ regulatory T cells among CD4+ CD25+ T lymphocytes after 16 h of co-culture with FAP+ CAF clusters (Detailed gating strategy in Supplementary Fig. 9B). Data are mean ± SEM (n = 8). p values from two-sided Mann–Whitney test. I Same as (H) for the % of PD-1+ FOXP3+ T lymphocytes. p values from two-sided Mann–Whitney test. J Same as (H) for the % of CTLA-4+ FOXP3+ T lymphocytes. p values from two-sided Mann–Whitney test. K % of Perforin+ among total CD16+ NK cells after 24 h of co-culture with FAP+ CAF clusters (Detailed gating strategy in Supplementary Fig. 9C). Data are mean ± SEM (n = 7). p values from two-sided Student’s t test. L Same as (K) for Granzyme B+ NK cells. p values from two-sided Student’s t test. M Same as (K) for CD16+ CD56^high NK cells. p values from two-sided Student’s t test. N Same as (K) for NKG2A+ NK cells. p values from two-sided Student’s t test. Source data are provided as a Source Data file.

Fig. 6. BC cellular composition is associated with tumor invasiveness.

A–D, H Data from the INVADE cohort (N = 55 BC patients), including Ductal Carcinoma in Situ (DCIS) lesions (N = 18), micro-invasive DCIS (MI-DCIS) (i.e. DCIS lesions with invasive foci not exceeding 1 mm (N = 17) and Invasive Breast Cancer (IBC) (N = 20)). A Boxplot showing Shannon index from cell type fractions in DCIS, MI-DCIS and IBC samples. p value from Mann–Whitney test. B Bar plots showing the relative composition of epithelial, stromal, endothelial and immune cells per patient (Left) and according to BC invasive status (Right). p value from chi-squared test. C Boxplots of the relative proportions of FAP+ CAF, CAP, endothelial, myeloid and lymphoid cells in DCIS, MI-DCIS and IBC samples. p values from Mann–Whitney test. D Bar plots showing the relative proportions of clusters among FAP+ CAF, CAP, endothelial, myeloid and lymphoid cells in DCIS, MI-DCIS and IBC samples. p values from Fisher’s exact test. E Communications from cancer cells to Detox-iCAF mediated by TGFβ2-TGFBR2 interaction in space on one section. Colored dots represent cells from a given cell type expressing either the receptor for Detox-iCAF (in yellow) or the ligand for cancer cells (in red). Arrows highlight cells close enough to communicate through the selected ligand-receptor, as inferred by SpaTalk. F Network of downstream pathways upregulated in Detox-iCAF following TGFβ2-TGFBR2 interaction in the section shown in (E) inferred using the atlas. G Left, Cancer cell abundance in each spot of a DCIS section inferred by deconvolution. Right, Manual selection by a pathologist of spots in direct contact with cancer cell-enriched spots (in red); other spots farther away from cancer cells are in gray. H Bar plot showing the relative proportions of FAP+ CAF clusters (assessed by deconvolution) in spots directly surrounding cancer cell-enriched spots (red dot) compared to other spots (gray dot) in the DCIS section. p value from Fisher’s exact test. In all boxplot the center line, box limits and whiskers indicate the median, upper and lower quartiles and 1.5 × interquartile range. Source data are provided as a Source Data file.

Fig. 7. FAP+ CAF composition predicts DCIS progression.

A, C–E Data from the TBCRC cohort of DCIS patients (N = 216 patients). A Boxplots of the relative proportions of FAP+ CAF clusters among FAP+ CAF in DCIS patients with DCIS recurrence (N = 66), with IBC recurrence (N = 55) or without recurrence (N = 95). p values from two-sided Mann–Whitney test. B Relative proportions of Detox-iCAF and TGFβ-myCAF between DCIS with (N = 5) or without recurrences (N = 13) in the INVADE cohort (N = 18). p values from two-sided Mann–Whitney test. C Left, Kaplan–Meier curve of time to recurrence (DCIS and IBC) stratified by the median of Detox-iCAF content in DCIS at diagnosis. p value from Log-rank test. Right, Forest plot for multivariate Cox proportional hazards model considering Detox-iCAF content (median stratification) and PAM50 classification. D Kaplan–Meier curves of time to recurrence (IBC only) stratified by the median of Detox-iCAF (Left) or TGFβ-myCAF (Right) content among FAP+ CAF in DCIS at diagnosis. p values from Log-rank test. E Left, Kaplan–Meier curve of time to recurrence (IBC only) stratified in low risk patients (defined as high Detox-iCAF and low TGFβ-myCAF content) and high risk patients (other patients). p value from Log-rank test. Right, Forest plot for multivariate Cox proportional hazards model including low/high risk patients and PAM50 classification. In all boxplot the center line, box limits and whiskers indicate the median, upper and lower quartiles and 1.5 × interquartile range. In the forest plots, the center points shows the hazard ratio (HR) and lines represent 95% confidence interval (CI). Source data are provided as a Source Data file.

Fig. 8. Schematic model.

CAF heterogeneity and plasticity shape a structured organization of the tumor micro-environment in BC. In this paper, we describe spatial organization, plasticity and interactions of FAP+ CAF clusters with neighboring cells by combining analysis of single-cell data, spatial transcriptomics and functional assays. We identify spatially organized cellular EcoCellTypes, which are precisely localized within tumors and composed of specific FAP+ myCAF or iCAF clusters. Distances to cancer cells induce a gradient of FAP+ CAF cluster identities. Detox-iCAF are found around blood vessels composed of ap-EC and in close proximity to FOLR2+ TAM. Detox-iCAF serve as a reservoir and can give rise to ECM-myCAF in presence of cancer cells, either directly or indirectly through the Wound-myCAF cluster, by DPP4- and YAP1/TEAD-dependent mechanisms. ECM-myCAF localize close to tumor cells, where they can reach a TGFβ-myCAF phenotype in presence of T lymphocytes. In addition, specific TAM are found in different FAP+ CAF cluster-enriched territories. While FOLR2+ TAM are close to Detox-iCAF, TREM2+ and SPP1+ TAM are enriched in ECM-myCAF, IFNαβ-myCAF and TGFβ-myCAF-enriched niches. Our data show that spatial organization in BC tumors is related to reciprocal interactions of FAP+ CAF clusters with cancer and immune cells in specific spatial domains. Importantly, we identify that the content in Detox-iCAF and TGFβ-myCAF at DCIS diagnosis is a predictive factor for the recurrence of DCIS into invasive breast cancer. The figure was created with Biorender.com.