Cancer-associated fibroblast heterogeneity in axillary lymph nodes drives metastases in breast cancer through complementary mechanisms

Although fibroblast heterogeneity is recognized in primary tumors, both its characterization in and its impact on metastases remain unknown. Here, combining flow cytometry, immunohistochemistry and RNA-sequencing on breast cancer samples, we identify four Cancer-Associated Fibroblast (CAF) subpopulations in metastatic lymph nodes (LN). Two myofibroblastic subsets, CAF-S1 and CAF-S4, accumulate in LN and correlate with cancer cell invasion. By developing functional assays on primary cultures, we demonstrate that these subsets promote metastasis through distinct functions. While CAF-S1 stimulate cancer cell migration and initiate an epithelial-to-mesenchymal transition through CXCL12 and TGFβ pathways, highly contractile CAF-S4 induce cancer cell invasion in 3-dimensions via NOTCH signaling. Patients with high levels of CAFs, particularly CAF-S4, in LN at diagnosis are prone to develop late distant metastases. Our findings suggest that CAF subset accumulation in LN is a prognostic marker, suggesting that CAF subsets could be examined in axillary LN at diagnosis.


Metastatic LN
(a) PCA analysis of CAF-S1 and CAF-S4 primary cultures (n = 4 pairs of CAF-S1 and CAF-S4, each pair being isolated from the same patient). Left, PCA based on the 500 most variant genes from RNAseq data of primary cultures. Right, PCA based on top 500 gene-signatures from RNAseq data of fresh samples (as shown in Fig. 3). (b) Same PCA as in (a, left) showing the patient and tissue of origins of each CAF-S1 and CAF-S4 primary cultures. (c) Hierarchical clustering using Ward's method with Pearson distances on the same primary CAF-S1 and CAF-S4 as in (a). Columns represent in vitro cultured CAF-S1 and CAF-S4 cells and lines stand for genes of CAF-S1 and CAF-S4 top 500 gene-signatures from samples.
Color saturation shows gene expression deviation from the mean (above in red, below in blue). (d) Representative FACS plots of FAP, CD29, PDGFRβ, FSP1 and αSMA protein levels from primary CAF-S1 and CAF-S4 cultures (paired samples). (e) Cell exclusion zone assay (independent experiment using other CAF-S1 and CAF-S4 cells as those tested in Fig.   4c). Left, Representative images of a free-zone (t0, top) closing after 40 h (bottom) of CAF subset migration. Scale bar, 200 μm. Graphs show velocity (left), persistence (middle) and direction (|sin(α)|, right), which were assessed by tracking 40 CAF-S1 and 40 CAF-S4 cells. boxplots are medians ± 25%-75% quantiles, whisker values range 1.5 × IQR above 75th or below 25th percentiles. Source data are provided as a Source Data file, as well as R scripts used to generate the figure panels.
Supplementary Figure 4 CAF-S1 and CAF-S4 modulate BC cell phenotype and invasion capabilities (related to Fig. 5 and 6) (a) Impact of CAF-S1 or CAF-S4 co-culture with MCF7 (left, n = 7 independent experiments) and T47D (right, n = 6) co-culture with CAF-S1 or CAF-S4 on the total number of viable (Dapiassessed by FACS) CAF subset relative to control (-, CAF-S1 without cancer cells (red bar) considered as the reference). p values from Wilcoxon signed rank test. (b) CAF-S1 and CAF-S4 chemo-attraction capacities evaluated by Transwell assays on MCF7 (left, n ≥ 10 independent experiments) and T47D (right, n ≥ 8), corrected by their pro-proliferation properties. p values from Wilcoxon signed rank test. (c) Images of F-actin (green, top) or Vinculin (red, bottom) with DAPI (blue) staining in MCF7 (left) and T47D (right) cultured alone, or in presence of CAF-S1 or CAF-S4. Merge images are shown in Fig. 5e. Scale bars, 20 μm. Quantifications show F-Actin and Vinculin staining in CAF subset per CAF cell area (F-Actin and Vinculin signal areas defined on ImageJ, divided by CAF surface and expressed in arbitrary units (a.u.). Evaluation on at least 5 images per CAF subset (CAF subset cocultured with MCF7 (left) or T47D (right)). p values from Mann-Whitney test (F-Actin far left graph) and Student's t-test (all other 3 graphs). (d) Representative images (different from those shown in Fig. 5f) showing staining of E-Cadherin (red) and DAPI (blue) (top), or costaining of E-Cadherin (red), F-actin (green) and DAPI (blue) (bottom) in MCF7 or T47D cultured alone, or in presence of CAF-S1 or CAF-S4. Scale bars, 20 μm. (e) Proportion of BC cells that invaded (in Transwell inverted assay) above 30 μm (left) and 40 μm (right) in CAF-S1-or CAF-S4-embedded collagen, relative to CAF-free condition (n = 3 independent experiments, ~500 BC cells per analyzed z-stack). p values from paired t-test. (f) CAF-S1 and CAF-S4 velocity assessed in tumor-on-chip devices. The graphs display 2 different pairs, each dot represents a single cell (n ≥ 27 per CAF subset). p values from Mann-Whitney test. Boxplots are median ± 25%-75% quantiles, whisker values range 1.5 × IQR above 75th or below 25th percentiles. Barplots are mean ± SEM. At least 2 CAF-S1 and CAF-S4 pairs have been tested. Source data are provided as a Source Data file, as well as R scripts used to generate the figure panels. Figure 5 TGFb and CXCL12 pathways are involved in CAF-S1-mediated BC cell invasion (related to Fig. 7) (a) CXCL12 mRNA levels in CAF-S1 and CAF-S4 (n = 10) assessed by RNAseq. p values from DESeq2. (b) CXCL12 mRNA levels after silencing of CXCL12 in CAF-S1 and CAF-S4.

Supplementary
Data are shown as fold change to control (n = 3). p value from Student t-test. (c) Representative images showing immunofluorescence co-staining of E-Cadherin (red), Factin (green) and DAPI (blue) in MCF7 (Up) or T47D (Down) cultured alone (-), or in presence of CAF-S1 or CAF-S4 transfected with non-targeting siRNA (siCTL) or with siRNA targeting CXCL12 (siCXCL12). Scale bars, 20 μm. (d) Impact of CXCL12 silencing in CAF-S1 on CAF-S1 capacities to chemo-attract MCF7 (left, n = 3) and T47D (right, n = 3) assessed by Transwell assays. Data are shown as fold change to control and p values are from one sample t-tests. (e) Impact of 24 h treatment with conditioned medium (CM) from CAF-S1 transfected with non-targeting siRNA (siCTL) or from CAF-S1 silenced for CXCL12 (siCXCL12) on the total number of viable MCF7 (left, n = 4) and T47D (right, n = 4) (assessed by Resazurin staining) relative to control (without CAF-S1). p values from paired ttests. (f) CXCR4 mRNA levels after silencing of CXCR4 in MCF7 (left) and T47D (right). Data are shown as fold change to control (n = 4 independent experiments). p value from one sample t-tests. (g) Gene set enrichment analysis of TGFβ pathway in CAF-S1 versus CAF-S4 transcriptomic signatures. p refers to false discovery rate q-value. (h) TGFBR1, -2 and -3 mRNA levels in CAF-S1 and CAF-S4 (n = 10) assessed by RNAseq. p values from DESeq2.
(i) Same as in (c) with/without CAF-S1 or CAF-S4 and with/without TGFβR inhibitor (LY2109761). In all panels, at least 3 CAF-S1 and CAF-S4 pairs have been analyzed. Each pair of CAF-S1 and CAF-S4 was isolated from the same patient. Source data are provided as a Source Data file, as well as R scripts used to generate the figure panels.
Barplots mean ± SEM. In all panels, at least 3 CAF-S1 or CAF-S4 cell lines have been tested, except S5C (MCF7) where 2 CAF-S1 cell lines and S5K where 1 CAF-S4 cell line were used. (i) Levels of cytokine release in supernatants collected from CAF-S1 and CAF-S4 fibroblasts cultured either alone (-) or in presence of MCF7 or T47D, as indicated. Cytokine release is measured using cytokine antibody-paired array and assessed by the following ratio: Cytokine intensity mean (intensity mean of 2 spots per cytokine per experiment) / Control mean intensity (2 spots of 3 internal controls). P values from Mann-Whitney test (n = 3). In all panels, at least 2 CAF-S1 and CAF-S4 pairs have been analyzed. Each pair of CAF-S1 and CAF-S4 was isolated from the same patient. (j) Boxplots showing RNA levels of the genes tested in the cytokine antibody array shown in (i) in CAF-S1 and CAF-S4 cells (n = 20). Data are from RNA-seq. P values from Mann-Whitney test. Expression of G-CSF, GM-CSF and MIF was also analyzed but was not detected at either protein or RNA levels.
Source data are provided as a Source Data file, as well as R scripts used to generate the figure panels.
Source data are provided as a Source Data file, as well as R scripts used to generate the figure panels.
Source data are provided as a Source Data file, as well as R scripts used to generate the figure panels. Supplementary