Main

Breast cancer is the leading cause of cancer-related mortality in women globally and the most common of any cause of death in UK women aged 35–64 years[1](https://www.nature.com/articles/s41586-025-09684-7#ref-CR1 “Ferlay, J. et al. Global cancer observatory: cancer today. International Agency for Research on Cancer https://gco.iarc.who.int/today

(2024).“) (https://www.ons.gov.uk/). In both mouse and human mammary glands, progesterone-induced proliferation of stem and progenitor cells results in increased branching and ductal complexity3,4. This proliferation is mediated through paracrine signals secreted from progesterone receptor (PR)-positive ‘luminal mature’ cells that act on PR-negative ‘luminal progenitor’ cells, the postulated cell of origin for basal (triple-negative) breast cancer2,4,5,6,7. In premenopausal women, breast epithelial cell proliferation is highest during the progesterone-dominant luteal phase of the menstrual cycle and can be reduced by anti-progestins such as mifepristone8,9. Supplementation of progestin, as a contraceptive or hormone replacement therapy, increases breast cancer incidence10,11,12 and stimulates epithelial proliferation and hyperplasia in preclinical models13. Conversely, inhibiting PR or its downstream pathways in mouse models results in a substantial reduction in mammary carcinogenesis through suppression of mammary luminal progenitor and stem cell activity7,14,15,16,17, with clinical window studies also showing reduced proliferation in normal and cancerous breast tissue9,18,19,20.

One of the challenges of primary prevention studies is identifying clinically relevant surrogate indicators of risk reduction. Mammographic density is one of the strongest risk factors for breast cancer21 and is a reliable clinical measure across a range of methods, including automated volumetric analysis22. Magnetic resonance imaging (MRI) measurements of fibroglandular volume (FGV) correlate well with automated volumetric mammographic density, and FGV is greater in the luteal than in the follicular phase of the menstrual cycle23,24. Mammographic density also declines through menopause and increases in post-menopausal women using progestin-containing hormone replacement therapy25,26. Mammographically dense areas contain increased epithelial and fibroblast cell numbers as well as collagen27,28. Breast stroma has a role in cancer initiation and progression by regulating epithelial cell proliferation29; in rodent models, stromal crosslinked fibrillar collagen increases the incidence of invasive tumour formation30,31. Periductal tissue stiffness positively correlates with increased collagen fibril alignment in human breast tissue with high mammographic density32. The question that we set out to address was how anti-progestin therapy might prevent luminal progenitor cells undergoing oncogenic transformation through both direct effects on the epithelium and indirect effects on the microenvironment structure, composition and stiffness that could potentially be appreciated radiologically.

Here we report findings from the BC-APPS1 study (NCT02408770) that demonstrate the profound effects of 12 weeks of ulipristal acetate (UA) therapy on normal breast composition in 24 premenopausal women at increased risk of breast cancer. We conducted multi-OMICs analyses on paired vacuum-assisted breast biopsy (VAB) tissues, before and after treatment, alongside critical clinical correlates such as mammographic density or FGV. Our comprehensive analyses of the primary tissues at cellular, molecular and functional levels have exposed powerful dependencies of the extracellular matrix (ECM) and breast epithelial progenitor fractions on hormone-dependent stromal triggers. This work demonstrates that critical components of the mammary progenitor cell niche and mammographic density determinants can be altered with anti-progestins. Together, targeting PR signalling may be a valuable strategy in preventing aggressive breast cancers in premenopausal women at increased risk.

Anti-progestin prevention study participants

Between 29 March 2016 and 11 March 2019, 32 women with an increased risk of breast cancer due to their family histories consented to the BC-APPS1 study. Six failed screening owing to inability to time the luteal phase of the menstrual cycle (P4 of less than 15 nmol l−1). Of the 26 eligible participants who received UA therapy, two underwent baseline investigations but subsequently withdrew from the study before the second VAB: one participant owing to anxiety related to a small biopsy-associated haematoma and one participant owing to drug-induced anxiety. Therefore, 26 participants were included in toxicity analyses, and 24 with paired VAB samples were included in molecular analyses of response to UA therapy. Downstream OMICs analyses of VAB samples were applied to selected samples depending primarily on the tissue availability for the technology utilized. Baseline VAB was timed to the luteal phase of the menstrual cycle owing to the profound effect of cycling ovarian hormones on breast biology and epithelial dynamics. The trial schema in Fig. 1a outlines our systematic multi-tiered workflow of OMICs analyses. Participant demographics are presented in detail in Supplementary Table 1. In summary, the 24 participants with paired samples had a median age of 39 years (range of 34–44 years), median BMI of 26 kg m−2 (range of 21–42) and a median remaining lifetime breast cancer risk of 25.5% (range of 17–38.3%; Tyrer Cuzick v7.02). Treatment was generally well tolerated with no grade 3 or 4 adverse events (Supplementary Table 2).

Fig. 1: Anti-progestin treatment reduces luminal progenitor activity.

a, Trial schema of the BC-APPS1 study. A VAB was collected in the luteal phase (baseline), and repeated in the opposite breast after 12 weeks of UA (5 mg daily). AFM, atomic force microscopy; IF, immunofluorescence; IHC, immunohistochemistry; IMC, imaging mass cytometry. The trial schema was created using BioRender (https://biorender.com). b, Percentage of Ki67-positive cells in 24 paired breast tissue samples before (baseline) and after (post-treatment) 3 months of UA therapy. Representative staining is shown. c, Proportion of epithelial area per lobule area before (baseline) and after (post-treatment) 3 months of UA therapy (n = 19 tissue pairs). Examples of lobule epithelial areas (green outlines) are shown. d, Flow cytometry analysis of luminal mature (LM; CD49f−EpCAM+), luminal progenitor (LP; CD49f+EpCAM+), basal (BA; CD49f+EpCAM−/low) and stromal (S; CD49f−EpCAM−) cells. The graph shows the percentage of epithelial populations (LP, LM and BA) in 17 tissue pairs. NS, not significant. e, Percentage of luminal, mixed or basal colonies in 18 breast tissue sample pairs before and after UA therapy. Representative examples of clonogenic assay colonies are shown above. f, MFE data expressed as a percentage for 19 tissue pairs. Horizontal dotted line, 0. A representative example of a mammosphere is shown above. g, Percentage of SOX9 and Ki67 double positive cells in eight tissue pairs quantified by immunofluorescence. The arrow in the representative images above indicates a cell expressing both SOX9 and Ki67. In all plots, boxplot centre lines represent median values and box bounds indicate the 25th and 75th percentiles, with connecting lines between paired data points. P values were calculated with two-sided Wilcoxon matched-pairs signed-rank test (b–g). Scale bars, 50 μm (b,c,e,f) and 10 μm (g).

Anti-progestin treatment reduces luminal progenitor activity

The primary end point of the BC-APPS1 study was epithelial proliferation assessed by Ki67 immunohistochemistry, chosen primarily to power the study statistically, as Ki67 is not a recognized surrogate for breast cancer risk. The study met its primary end point with a significant reduction in proliferation between baseline (8.2%; 95% confidence interval (CI) 5.2–11.2%) and 12-week samples (2.9%; 95% CI 2.1–3.7%; P < 0.0001; Fig. 1b). Mean serum progesterone levels reduced with treatment from 36 nmol l−1 (95% CI 29.4–41.6 nmol l−1) at baseline to less than 3 nmol l−1 (95% CI 0.3–4.6 nmol l−1; P < 0.0001; Extended Data Fig. 1a), effectively abrogating the luteal phase. Both the epithelial area within each lobule (Fig. 1c) and the average area of acinar structures (Extended Data Fig. 1b) were significantly reduced with UA treatment; however, the mean number of acini per lobule did not change (Extended Data Fig. 1c). Next, flow cytometry analysis showed a significant reduction in the luminal progenitor (CD49f+EpCAM+) fraction with treatment from 43% (95% CI 35–52%) to 30% (95% CI 21–39%; P < 0.001), with no significant changes detected in luminal mature (CD49f−EpCAM+) or basal (CD49f+EpCAM−/low) populations (Fig. 1d and Extended Data Fig. 1d). Epithelial colony-forming assays used to enumerate progenitor activity yield three distinct colony phenotypes: myoepithelial/basal, luminal and mixed (where mixed colonies represent bi-lineage differentiation potential)33. Anti-progestin treatment reduced the proportion of mixed colonies from 70% (95% CI 60–80%) to 55% (95% CI 44–67%; P < 0.05; Fig. 1e and Extended Data Fig. 1e). Mammosphere-forming efficiency (MFE), another measure of luminal progenitor activity, was also reduced by UA (baseline 0.29%; 95% CI 0.19–0.39% versus 12 weeks 0.16%; 95% CI 0.04–0.28%; P < 0.01; Fig. 1f). In vitro treatment of baseline cell suspensions with UA and an alternative anti-progestin (onapristone) similarly reduced MFE (Extended Data Fig. 1f,g). SOX9 is a marker of luminal progenitor cells34, and both the overall percentage of SOX9+ (Extended Data Fig. 1h) and proliferating SOX9+ cells (dual staining for SOX9 and Ki67) were reduced with UA treatment (SOX9+Ki67+ at baseline 4.4%; 95% CI 1.6–7.2% versus 12 weeks 1.3%; 95% CI 0.7–1.9%; P < 0.05; Fig. 1g). Overall, these data demonstrate that anti-progestin treatment reduces the proportion, proliferation and activity of luminal progenitor cells in the normal breast tissue of women at increased breast cancer risk. Given that luminal progenitors are the putative cell of origin in basal (triple-negative) breast cancers, abrogation of this breast cancer precursor pool is pertinent for targeted breast cancer prevention.

Luminal mature cells regulate the basal cell/fibroblast matrisome

To evaluate transcriptional changes with treatment, bulk tissue RNA sequencing (RNA-seq) analysis was performed. RNA quality was sub-optimal in at least one sample from each of 14 participants, and data are presented for the paired samples from 10 participants that met quality standards. UA treatment resulted in differential expression of 50 genes (log2[fold change] (log2FC) > 1.5, P < 0.05; Extended Data Fig. 2a), including two established PR target genes (TNFSF11 and CXCL13) that were significantly downregulated with treatment (Extended Data Fig. 2b). Gene Ontology term analysis of the top 50 differentially expressed genes showed that almost half of these genes (23) were associated with the extracellular space (Extended Data Fig. 2a).

To define the molecular changes in diverse breast cell types after anti-progestin treatment, single-cell RNA-seq (scRNA-seq) profiling of six paired samples was performed (Fig. 2a). Single-cell transcriptomes of 115,875 cells were obtained after quality filtering for gene coverage, read counts and mitochondrial reads (see Methods). Uniform manifold approximation and projection (UMAP) analysis of the combined 12 samples revealed seven major cell populations (Fig. 2b). Using previously published gene signatures35, we identified three epithelial (luminal adaptive secretory precursor (LASP), luminal hormone sensing (LHS) and basal-myoepithelial (BMYO)) and four stromal (fibroblasts, endothelial, perivascular and immune) cell types (Fig. 2c). A similar number of cells from baseline (56,014) and post-treatment (59,861) were analysed, and all seven cell populations were present in each of the 12 samples (Supplementary Table 3). Using differential abundance testing, we did not observe any significant changes in cellular abundance following UA treatment across all seven broad cell populations (Fig. 2d). However, when considering only the total epithelial population, we observed a significant reduction in the proportion of LASPs post-treatment (Fig. 2d). This reduction was seen in five of six paired samples analysed, with no consistent trend for LHS and BMYO populations (Extended Data Fig. 3a). Participant samples with both flow cytometry (Fig. 1d) and scRNA-seq data showed a strong correlation (r = 0.762; P = 0.0055) between the percentage of luminal progenitor cells detected by flow cytometry and LASP cells identified by scRNA-seq (Extended Data Fig. 3b), indicating that luminal progenitor and LASP cells are largely the same population. Here we use ‘luminal progenitor/LASP cells’ to refer to luminal progenitor or LASP cells, defined by the specific assays. To explore further granularity within each cell type, we performed Leiden subclustering to match the ‘level 2’ annotations used in the integrated Human Breast Cell Atlas (iHBCA), the largest integrated breast scRNA-seq dataset35. This identified several subclusters within each major cell type in our dataset, consistent across all six paired samples (Extended Data Figs. 4 and 5). iHBCA clusters BMYO2 and LASP4 (ref. 35) were not identified in the BC-APPS1 dataset, and differential abundance testing did not reveal any significant changes in the abundance of individual iHBCA-annotated subclusters following UA treatment (Extended Data Fig. 3c). This included the three LASP subclusters 1, 2 and 3, even when analysis was restricted to epithelial populations, suggesting that LASP subgroup response to UA is variable between individual participants (Extended Data Fig. 3d). Pairwise differential expression analysis for each of the major cell populations in response to UA revealed that the majority of differentially expressed genes were observed in LHS cells (Fig. 2e and Extended Data Fig. 6a), in which pathway analysis showed mainly downregulation of predominantly cell-intrinsic RNA processing pathways (Extended Data Fig. 6b). The known PR target genes TNFSF11 and CXCL13 did not meet stringent cell number and expression thresholds in this analysis (see the section ‘Memento differential expression analysis’ in Methods), but both showed significant per-participant downregulation in LHS cells following UA treatment (Extended Data Fig. 6c). Although other cell types also exhibited significant gene expression changes in response to UA treatment, these were considerably less pronounced than those observed in LHS cells. The list of differentially expressed genes for each of the seven cell populations is provided in Supplementary Table 4, and analyses of the major pathways that are upregulated and downregulated in each population after UA treatment are also included (Extended Data Figs. 6b and 7a). Given that paracrine signalling is known to have a critical role in normal mammary gland development, we next investigated differentially expressed ligands following UA treatment. LHS cells, but also fibroblast and BMYO cells, showed a high number of downregulated ligands (Fig. 2f). The number of upregulated ligands was lower overall, but higher in LHS cells than in the other cell types (Extended Data Fig. 6d). The list of differentially expressed ligands for each of the seven cell populations is provided in Supplementary Table 5.

Fig. 2: Transcriptome network analyses reveal that luminal mature cells orchestrate the matrisome landscape of basal and fibroblast cells.

a, Workflow for paired biopsy single-cell transcriptomics from six participants at baseline and 12 weeks after UA treatment. The workflow was created using BioRender (https://biorender.com). b, UMAP of breast tissue cells annotated by broad cell type. n = 115,875 cells. c, Dot plot of broad cell-type marker genes. The columns correspond to key markers (normalized per gene) with brackets detailing the cell type, and the rows correspond to the cell population identified in the dataset. d, Proportionality fold change (post-treatment to baseline) across broad cell types (left) and restricted to epithelial cells (right). Positive or negative changes denote enrichment or depletion post-treatment, respectively. Boxplot centre lines represent median values, box bounds indicate the 25th and 75th percentiles, and whiskers extend to the extreme datapoint within 1.5 times the interquartile range (IQR) of the boxplot hinges. Significance was calculated with a two-sided Student’s t-test adjusted P value for multiple comparisons using Benjamini–Hochberg correction. n = 6 tissue pairs. e, UpSet plot depicting downregulated genes (less than −0.25 logFC, P < 0.05, Memento analysis) post-treatment across broad cell types. The intersection size indicates the number of genes uniquely regulated within a single cell type or shared across multiple cell types. f, UpSet plot depicting downregulated genes that encode proteins that act as ligands (less than –0.25 logFC, P < 0.05, Memento analysis) post-treatment across broad cell types. g, CellChat analysis of incoming–outgoing interaction strength between broad cell types at baseline. The node size represents the number of interactions in each cell type. h,i, Differential L–R pathway signalling changes (post-treatment to baseline) across broad cell types (h) or within basal (BMYO1) and fibroblast (FB1–3) cell states (i). Negative values represent a decrease in L–R signalling post-treatment. j, Chord diagram of pairwise downregulated collagen gene signalling post-treatment (less than –0.25 logFC, P < 0.05) from basal and fibroblast cell states (sender cells) to all breast cell states (receiver cells with 5% or more receptor expression), highlighting the epithelial populations in light blue.

To investigate how UA treatment affects cell communication networks across broad and granular cell states, Cell Chat36 was used to model potential ligand–receptor (L–R) interactions between cell populations. After normalizing cell numbers to infer ‘per cell’ L–R interaction signalling strengths (ISSs) at baseline, BMYO and fibroblast populations had far greater incoming and outgoing ISSs than LASP or LHS populations (Fig. 2g). Annotating L–R pairs by established pathways revealed outgoing collagen ISS to be most markedly affected in BMYO and FB cells post-treatment, suggesting that UA therapy diminishes their role as sources of collagen signals, with a reduction in collagen incoming ISS seen in all seven cell states (Fig. 2h). To further corroborate these findings, gene set enrichment analysis of pairwise differentially expressed genes in BMYO and fibroblast cells demonstrated robust overrepresentation of ECM-related terms including ‘ECM organization’, ‘degradation of the ECM’, ‘collagen degradation’ and ‘assembly of collagen fibrils’ (Extended Data Fig. 7a). When we restricted the analysis of pairwise differentially expressed genes to the ‘Reactome ECM organization’ gene set, fibroblast and BMYO cells exhibited a higher number of downregulated ECM genes than other cell populations, including many genes encoding collagen proteins (Extended Data Fig. 7b and Supplementary Table 6). By contrast, LHS cells displayed a greater number of upregulated ECM regulatory genes, with 4 out of the 13 genes encoding matrix metalloproteinases (MMP1, MMP3, MMP10 and MMP12), which are known to have key roles in ECM degradation (Extended Data Fig. 8a,b). The complete list of ECM-related differentially expressed genes, both downregulated and upregulated, across the seven cell populations is provided in Supplementary Table 6. These results point to the ECM as a prime target downstream of UA treatment.

To determine whether specific subpopulations of fibroblast cells are driving reduced collagen signalling post-UA treatment, we assessed L–R networks within each cell subcluster. BMYO1 and fibroblast 1 (FB1) cells were confirmed as the primary sender subclusters exhibiting the most pronounced reduction in collagen signalling compared with FB2 and FB3 cell states (Fig. 2i). Analysis of collagen gene expression across all cell subclusters revealed that genes encoding collagen I, collagen IV and collagen VI are the most abundantly expressed in the human breast, with FB1–3 cells being primary producers of collagen I (COL1A1 and* COL1A2*) and collagen VI (COL6A1, COL6A2 and COL6A3), whereas BMYO1 cells primarily express collagen IV (COL4A1 and COL4A2) and collagen VI (COL6A1 and* COL6A2*; Extended Data Fig. 8c). As the most profound changes were in BMYO1 and FB1 cells, we examined the inferred differential collagen L–R interactions between these cells and all other subclusters. Collagen gene expression was downregulated after UA treatment in FB1 (specifically COL1A1, COL1A2, COL4A1, COL4A2, COL6A1,* COL6A3* and COL6A6), FB2 (specifically COL1A1, COL1A2, COL4A1, COL4A2 and COL6A1), FB3 (COL6A3) and BMYO1 cells (specifically COL4A1, COL4A2 and COL6A1; Extended Data Figs. 9 and 10d). This reduction in collagen expression potentially affects autocrine and paracrine interactions of numerous cell types given the collagen receptor expression across subclusters, most notably in the epithelial subclusters (Fig. 2j). A list of the collagen L–R interactions shown in Fig. 2j, along with the percentage of cells within each target population expressing collagen receptors, is provided in Supplementary Table 7. We then interrogated whether the observed collagen gene expression changes in FB1 and BMYO1 cells could be mediated by ligands secreted from LHS cells, the PR-expressing targets of UA treatment. We used NicheNet analysis37 to investigate ligands predicted to be secreted by LHS cells (sender cell) that influence the expression of collagen target genes across FB1–3 and BMYO1 (receiver cell). Among the downregulated collagen genes, only COL1A2 and COL6A3 could be linked to LHS ligands. Downregulation of WNT5A and RARRES1 in LHS cells were ligands predicted to regulate COL6A3 expression in FB1 and FB3 cells, whereas APOD was predicted to regulate COL1A2, specifically in FB1 cells. No LHS ligands could be associated with regulation of collagen IV or other differentially expressed collagens in BMYO1 or FB2 cells, suggesting that FB1 and FB3 expression of COL1A2 and COL6A3 are key targets of UA-driven alterations in LHS paracrine signalling (Extended Data Fig. 10a,b). WNT5A is expressed at higher levels in the LHS1 subcluster but was significantly downregulated in both LHS1 and LHS2 after UA treatment, whereas COL6A3 was similarly expressed across FB1 and FB3 subclusters and was significantly downregulated in both post-treatment (Extended Data Fig. 10c,d). Fibroblasts (FB1–3) express nine receptors for WNT5A, which could mediate WNT5A-dependent regulation of COL6A3 (Extended Data Fig. 10e).

Thus, beyond the known paracrine PR signals from luminal mature or LHS cells to luminal progenitor/LASP cells4,7, we identified LHS-secreted progesterone ligands that are prime candidates for downregulation of key collagen genes in human fibroblasts and basal cells, potentially shaping the matrisome landscape. Consistent with this, steroid hormones have recently been shown to stimulate ECM-remodelling fibroblasts, probably increasing mammary gland stiffness in mice[38](https://www.nature.com/articles/s41586-025-09684-7#ref-CR38 “Pascual, R. et al. Fibroblast hierarchy dynamics during mammary gland morphogenesis and tumorigenesis. EMBO J. https://doi.org/10.1038/s44318-025-00422-3

(2025).“). Altogether, we identified striking cell–cell communication network alterations with major changes in fibroblast and basal cell matrisome components, probably mediated by a reduction in LHS-secreted ligands in response to 12 weeks of anti-progestin treatment in women at increased risk of breast cancer.

Anti-progestin treatment remodels the breast matrix

To investigate the effects of UA therapy on breast tissue in proximity to luminal mature cells, we undertook laser capture microdissection (LCM) of breast lobules and peri-lobular stroma of four paired BC-APPS1 samples (Fig. 3a). Proteomic analysis of the tissue revealed the detection of 8,197 unique peptides corresponding to 1,519 proteins. Among these 1,519 proteins, 1,454 (96%) were consistently detected before and after treatment (data not shown) with 1,373 (90%) identified in all four participants (Fig. 3b). We identified 65 proteins regulated by UA treatment with q < 0.05 (Fig. 3c). Collagen α2 (VI) chain (COL6A2) and collagen α3 (VI) chain (COL6A3) were the most significantly downregulated proteins after treatment, whereas several histones (for example, histone H4 (HIST1H4A)) were the most significantly upregulated proteins. Gene set enrichment analysis using Reactome Pathway annotations revealed many pathways related to ECM and collagen (for example, ‘ECM organization’, ‘ECM proteoglycans’, ‘collagen formation’ and ‘assembly of collagen fibrils’) that were downregulated with UA treatment (Fig. 3d), in line with scRNA-seq data. Of the 65 proteins that were differentially abundant after treatment (Extended Data Fig. 11a), 27 (41.5%) were ‘matrisome’ proteins, comprising thirteen collagens, seven glycoproteins, three proteoglycans, three ECM regulators and one secreted factor (Fig. 3e), consistent with extensive remodelling of the ECM.

Fig. 3: Anti-progestin treatment remodels the breast matrix.

a, Lobular epithelium and peri-lobular stroma (within 25 μm of the observable edge of the epithelium) were laser capture microdissected from haematoxylin and eosin-stained paired tissue sections before (baseline) and after (post-treatment) UA treatment. A representative example of undissected tissue (left) and tissue after laser ablation (right) is shown. n = 4 tissue pairs. Scale bar, 100 μm. b, Venn diagram representing the distribution of the total proteins detected (1,519) in the four participants (P06, P12, P17 and P31) used for LCM proteomics. c, Volcano plot shows differential protein abundance analysis following UA treatment. Matrisome (structural ECM or ECM-modifying) proteins among the significantly altered proteins are colour coded according to their respective subcategories. d, Gene set enrichment analysis of LCM proteomics data using the Reactome Pathways reference set, showing pathways significantly altered by UA treatment. e, Heatmap of the 27 matrisome proteins identified as significantly differentially abundant after UA treatment. ECM proteins are grouped by their structural and functional properties. f, Imaging mass cytometry was performed on paired tissue sections before (baseline) and after (post-treatment) UA treatment. Representative images show staining with metal-conjugated antibodies to E-cadherin, SOX9 and collagen VI. Nuclei were visualized using a metal-tagged DNA intercalator. The yellow boxes indicate regions corresponding to the zoomed-in inserts. n = 8 tissue pairs. Scale bars, 100 µm and 10 µm (insets). g, Single-cell neighbourhood analysis of pericellular collagen VI abundance in SOX9high and SOX9low cell populations across paired BC-APPS1 samples at baseline (B) and post-treatment (PT) timepoints. Tissue images were segmented into single-cell objects, and cells were classified based on expression of specific markers. Analysis was performed on E-cadherin+ cells classified as either SOX9high or SOX9low. For each selected cell, collagen VI staining intensity was quantified within a 10-µm radius. Scale bar, 100 µm. Boxplot centre lines represent median values, box bounds indicate the 25th and 75th percentiles, and whiskers denote minimum and maximum values. Statistical analysis was performed using a repeated measure one-way analysis of variance (ANOVA) followed by Sidak’s multiple comparisons test. n = 8 tissue pairs.

To examine the spatial location of luminal progenitor cells in relation to these specific stromal components and their perturbation in response to UA, Hyperion imaging mass cytometry was performed. Metal-conjugated antibodies for collagen I, collagen VI and fibronectin (FN1) were used in combination with markers of epithelial (E-cadherin) and luminal progenitor/LASP cells (SOX9), as well as Ki67. Eight paired BC-APPS1 samples with plentiful lobules were selected. Initial analysis confirmed decreased expression of collagen I, collagen VI and FN1 (Extended Data Fig. 11b) with UA treatment as previously observed by LCM proteomics (Fig. 3e). Single-cell neighbourhood analysis of SOX9high and SOX9low cells at baseline identified the SOX9high cells to be in close proximity to regions of high collagen VI and FN1 but not collagen I expression compared with SOX9low cells, a finding that persisted following UA treatment (Fig. 3f,g and Extended Data Fig. 11c). In both baseline and post-treatment conditions, Ki67+ cells were significantly more prevalent in the SOX9high than SOX9low populations, confirming their higher proliferative activity, although UA treatment reduced proliferation in both populations (Extended Data Fig. 11d). The widespread staining pattern of collagen VI (a non-fibrillar collagen) is consistent with its expression in both stromal and epithelial cells (Extended Data Fig. 8c). These data identify stromal remodelling as an early event in breast tissue perturbed by anti-progestin treatment, although the persistent spatial association of SOX9high cells with collagen VI and FN1 after treatment suggests some continued colocalization despite short-term UA therapy (Extended Data Fig. 11c).

Anti-progestins reduce stiffness-driven luminal progenitor activity

Increased elastic force (stiffness) between cells expressing oncogenes and their surrounding ECM have been shown to induce signals that promote epithelial transformation39,40. Increased matrix stiffness also enhances the enrichment of cancer stem cells and the induction of chemoresistance in patients with breast cancer41. Given the robust downregulation of multiple collagens post-treatment (Figs. 2 and 3), we next investigated the effects of a stiff microenvironment on breast tissue. Organoids (3D microstructures) from six women at higher risk of breast cancer were grown for 1 week in collagen-mimetic hydrogels with ‘soft’ (600–900 Pa) or ‘stiff’ (1,800–3,000 Pa) conditions (Supplementary Table 8). Expression of the PR target gene TNFSF11 and luminal progenitor markers SOX9 and KIT were increased in stiff hydrogels, which was confirmed at the protein level for SOX9 and KIT expression, and accompanied by increased MFE after extraction and dissociation of the cells (Fig. 4a–c). Anti-progestin treatment of breast microstructures using UA or onapristone blocked stiffness-induced increases in SOX9 and KIT, as well as MFE; however, onapristone did not reduce MFE under soft conditions (Fig. 4b,c and Extended Data Fig. 12a,b; for gel source data, see Supplementary Fig. 1). Overall, these results establish that anti-progestin treatment attenuates stiffness-induced upregulation of progesterone signalling and progenitor cell activity, and also reduces the basal level of PR activity seen in softer gels in this in vitro system.

Fig. 4: Tissue stiffness-amplified progesterone response and luminal progenitor activity are inhibited by anti-progestins.

a, Real-time PCR gene expression of TNFSF11, KIT and SOX9 in breast tissue microstructures from women at increased cancer risk, cultured in ‘soft’ and ‘stiff’ hydrogels. Data are shown as mean fold change ± s.d., with individual points. n = 6 breast samples. b, KIT and SOX9 protein in breast microstructures (sample 1989N) cultured in soft (S) and stiff (ST) hydrogels, treated with UA (2 nM) or onapristone (ON; 100 nM). Densitometry normalized to β-actin is shown above the bands. n = 3 breast samples. c, MFE after culture in soft and stiff hydrogels with UA (2 nM) or ON (100 nM). Data are shown as mean fold change ± s.d., with individual points. n = 6 breast samples. d, Collagen coherency was assessed in peri-lobular regions (three lobules per sample) with representative PSR-stained sections shown at baseline and post-treatment. The ellipse indicates fibre alignment: examples of aligned (baseline) and non-aligned (post-treatment) collagen are shown in the insets. The graph shows mean collagen coherency for n = 22 paired samples. Scale bars, 100 μm. e, Reduced modulus of peri-lobular regions at baseline (B) and post-treatment (PT) measured by AFM indentation. At least three 100 μm2 regions per sample were measured as shown in the representative images. n = 4 tissue pairs. Scale bars, 100 μm. f, MRI annotation in ITK-snap: black denotes the background, opaque red indicates fatty tissue, and bright red shows the fibroglandular tissue. The FGV percentage was calculated by dividing the number of fibroglandular pixels by the total number of fibroglandular and fat pixels across slices. n = 12 paired MRI scans. Scale bars, 1 cm. g, Percentage of Ki67+ cells before treatment and post-treatment stratified by mammographic density. Participants were grouped using Volpara density grades to approximate BI-RADS categories (A/B denotes low MD, n = 6 tissue pairs; C/D indicates high MD, n = 17 tissue pairs). h, Heatmap of whole-tissue RNA-seq showing the differentially expressed genes between high MD (BI-RADS C/D; dark grey) and low MD (BI-RADS A/B; light grey) breast tissue at baseline (n = 9; FC > 3, P < 0.05). VST, variance-stabilizing transformation. i, Illustration shows that progesterone paracrine signalling regulates luminal progenitor/LASP (SOX9+) cells and fibroblasts, driving ECM remodelling and stiffness. Stiffness amplifies PR signalling, establishing a feedback loop. Anti-progestins disrupt this by inhibiting luminal cell-derived ligands (for example, WNT5A), lowering fibroblast collagen (for example, COL6A3), decreasing stif