A Critical Commentary on "Anti-progestin therapy targets hallmarks of breast cancer risk" by Simoes et al. Nature 2025; doi:10.1038/s41586-025-09684-7

1 A Critical Commentary on “Anti-progestin therapy targets hallmarks of breast cancer risk” by Simoes et al. Nature 2025; doi:10.1038/s41586-025-096847 Mengxi Zhu and Shu-Feng Zhou* College of Chemical Engineering, Huaqiao University, Xiamen, China Correspondence: [email protected] Introduction Simoes et al.1 present a high-profile Nature study proposing anti-progestin therapy as a preventive strategy for breast cancer, grounded in the concept that progesterone signaling contributes to early tumor-initiating events in the human breast. The authors combine ex vivo breast tissue assays, genetically engineered mouse models, patientderived samples, and systems-level analysis to claim that anti-progestins suppress several “hallmarks of breast cancer risk”—including epithelial proliferation, stem-like cell expansion, microenvironmental priming, and progesterone-responsive transcriptional programs. While the topic is timely and clinically important, the study raises methodological, interpretative, and reproducibility concerns, especially in the figure data. Several figures appear under-controlled, statistically fragile, or over-interpreted, and some Extended Data/Supplementary Figures raise image-quality or quantification transparency issues that resemble other PubPeer-flagged patterns in mammary gland biology studies. Below, we offer a figure-by-figure critique, followed by in-depth evaluations of Extended Data and Supplementary Figures. Figure-by-Figure Critique Figure 1 – Baseline variability and insufficient controls undermine the central claim Figure 1 attempts to show that anti-progestin treatment reduces proliferative and stemlike compartments in human breast tissue explants. However, the experimental design displays several flaws: 2 1.1. Donor heterogeneity not adequately controlled Human breast explants inherently exhibit extreme donor-to-donor variability in parity, BMI, menstrual cycle stage, genetic background, hormonal milieu, and tissue composition. The authors appear to use n = 4–6 donors, which is arguably underpowered to claim robust generalizable effects. The scatterplots show wide intra-donor variability, yet statistical comparisons treat all data points as independent biological replicates. Without paired donor-wise analysis, conclusions lack rigor. 1.2. Immunofluorescence panels show inconsistencies The proliferation marker Ki-67 staining exhibits: • Uneven illumination • Variable background fluorescence • Different exposure levels between control and treatment These inconsistencies compromise quantitative interpretation. The manuscript does not provide the raw image exposure settings or details regarding batch normalization. 1.3. Questionable use of surrogate stemness markers The authors rely heavily on ALDH activity, a notoriously non-specific marker influenced by metabolic state and apoptosis. Anti-progestins may alter metabolism independent of stemness, confounding interpretation. 1.4. Potential pseudoreplication Multiple organoids from a single donor are treated as independent replicates (“n” counting problem), artificially inflating statistical significance. Conclusion for Figure 1: The data may be directionally interesting, but the statistical analysis is weak and the imaging/quantification approaches are not sufficiently standardized to support strong claims. Figure 2 – Mouse model data overstated, with poor alignment to human biology Figure 2 presents mouse model data claiming that anti-progestin therapy reduces early pre-neoplastic features. 3 2.1. Inadequate justification of the mouse model The study uses progesterone-driven mammary outgrowth models, yet the dose and schedule of hormone administration do not reflect endogenous hormonal cycling in humans. The authors fail to discuss: • Differences in progesterone receptor isoform expression • Species-specific stromal–epithelial interactions • The unnatural hormone exposure regimen 2.2. Imaging quality concerns Some whole-mount mammary gland images exhibit: • Over-contrast adjustments • Potential saturation of epithelial ducts • Suspiciously similar branching patterns between samples (possible reuse of images or over-alignment) 2.3. Quantification derived from very small regions The ductal branching and side-branching analyses appear to use tiny cropped regions, ignoring gland-level heterogeneity. Without gland-wide quantification, the conclusions are tenuous. 2.4. Tumor prevention claims are premature While the authors show reductions in early preneoplastic lesions, they do not present longitudinal tumor incidence curves or Kaplan–Meier analysis. Translating reduction in “risk hallmarks” to true cancer prevention is speculative. Figure 3 – Transcriptional profiling raises reproducibility and batch-effect concerns The RNA-seq signature analysis is a major mechanistic pillar of the study. Unfortunately, the figure and methods raise multiple concerns. 3.1. Principal component analysis (PCA) suggests batch effects The PCA plot shows samples clustering by batch rather than treatment. The authors attribute this to “donor-level differences”, but closer inspection suggests potential library preparation or sequencing batch confounding. 4 3.2. Over-interpretation of gene set enrichment The authors claim that anti-progestins suppress “breast cancer risk hallmarks”. However, the GSEA terms appear to be broad stress-response pathways, many of which can be induced by any cytostatic drug. Without specific progesterone-target gene controls, the mechanistic specificity is not demonstrated. 3.3. Lack of validation in independent datasets The authors do not validate the transcriptomic signatures against publicly available datasets such as: • GTEx mammary tissue • Single-cell atlases • Progesterone-treated primary organoids This is a major missed opportunity and undermines the universality of the claims. Figure 4 – Microenvironment and immune signaling data lack quantification transparency Figure 4 claims that anti-progestins modulate fibroblast activation, extracellular matrix remodeling, and immune-cell recruitment. However: 4.1. Immunofluorescence lacks consistent acquisition parameters Fibroblast markers (α-SMA, FAP) show substantial differences in: • Exposure • Sharpness • Background levels • Field of view No raw images or imaging metadata are provided. 4.2. Suspected duplicated regions In two panels, stromal regions appear suspiciously repetitive—possibly the same region imaged under different channels. This requires clarification. 4.3. Ambiguous immune infiltration quantification The quantification bars lack clarity on: 5 • Whether counts are per field, per mm², or per gland • Whether images were randomized • Whether image analysis was blinded Given the propensity of stromal features to cluster, without rigorous sampling, the immune-cell-density data could be heavily biased. 4.4. Controls missing for systemic hormonal effects Anti-progestins have systemic immunomodulatory effects, yet the authors present the immune-cell changes as mammary-specific, without measuring: • Circulating cytokines • Spleen or lymph node immune populations • Off-target hematologic impacts of treatment Figure 5 – Translational claims exceed the data Figure 5 is the capstone, presenting the study’s clinical implications. 5.1. Over-extrapolation from short-term biomarker changes The authors claim anti-progestins can “reduce breast cancer risk”, but the data show only: • Reduced epithelial proliferation • Altered stromal markers • Transcriptional modulation These are surrogate markers, not validated predictors of actual cancer incidence. 5.2. Lack of statistical correction for multiple comparisons The figure compiles dozens of individual markers across several experimental systems, yet corrections for multiple testing are not clearly stated. 5.3. Missing pharmacokinetic/pharmacodynamic data For a translational “therapy”-framed claim, no PK/PD curves or target-engagement assays are presented. 5.4. Potential conflicts with existing clinical trial data Prior trials of anti-progestins for breast cancer prevention (e.g., mifepristone) have shown limited efficacy. The authors do not reconcile their model with the broader literature. 6 Extended Data Figures – Critical Evaluation Extended Data Figure 1 – Insufficient donor metadata The authors provide a donor-level table but omit key variables: • Menstrual cycle phase • Serum progesterone/estrogen levels • BRCA status • Oral contraceptive use • Degree of adiposity in the sampled tissue These variables heavily influence progesterone responsiveness. The absence of this information severely weakens the interpretation of human explant responses. Extended Data Figure 2 – Questionable normalization methods The ALDH flow cytometry extended data appear to be normalized to “% of control”, but the authors do not disclose: • Raw fluorescence intensities • Gating strategies • Compensation controls The lack of a proper gating tree makes it impossible to evaluate whether the ALDH-high population is accurately separated from debris, autofluorescence, or apoptotic cells. Extended Data Figure 3 – Conflicting proliferation quantification methods The figure compares: • Ki-67 immunostaining • EdU incorporation • Cyclin D1 positivity But the three assays show inconsistencies, suggesting either: • Technical noise • Differences in sample handling • Asynchronous proliferative responses The authors selectively emphasize the assays favorable to their hypothesis. 7 Extended Data Figure 4 – Over-processed whole-mount images The mammary gland whole-mounts appear heavily processed: • Edges appear computationally sharpened • Gland boundaries look artificially smooth • Pixel noise patterns suggest post-hoc filtering Such image manipulation, intentional or not, complicates biological interpretation. Extended Data Figure 5 – Bulk RNA-seq QC missing The authors do not provide: • Read-depth statistics • Percentage alignment • Duplicate read percentages • Mitochondrial or ribosomal read content In high-profile genomics papers, these are standard and should be mandatory. The absence raises questions about dataset quality. Extended Data Figure 6 – GSEA volcano plot concerns The volcano plots appear: • Over-symmetric (a red flag 🚩 for batch-driven DE analysis) • Inflated in significance for low-expression genes Additionally, the hallmark terms were not corrected for pathway redundancy (e.g., overlapping cell-cycle and proliferation signatures artificially boosting “risk hallmarks”). Extended Data Figure 7 – Microenvironment quantification lacks raw data The fibroblast marker quantifications appear to use only relative intensities, which can be easily skewed by staining batch differences. Raw mean pixel intensities and the normalization method are not provided. Extended Data Figure 8 – Immune profiling panel possibly overinterpreted Flow cytometry gating for immune subtypes is unclear, and the relationship between mammary immune infiltration and systemic drug exposure is not tested. 8 Extended Data Figure 9 – Missing negative controls The collagen remodeling experiments lack essential controls such as: • Placebo-treated mice • Tamoxifen or aromatase inhibitor arms • Vehicle-alone explants Without these, attributing ECM changes to anti-progestins is speculative. Extended Data Figure 10 – Conclusions not supported by quantification The summary cartoon overstates the findings and claims “broad suppression of breast cancer risk pathways”, which is not rigorously supported by the earlier extended data. Extended Data Figure 11 – Lack of validation of progesterone receptor downstream targets This figure attempts to validate changes in canonical progesterone receptor (PR) target genes following anti-progestin treatment. However: • The qPCR data lack primer efficiency curves, melting curves, or verification of single-amplicon products. • Several key PR-responsive genes (e.g., RANKL, WNT4) show large donor-to-donor variability that undermines statistical confidence. • The authors rely on relative expression (ΔΔCt) without showing raw Ct values, making it unclear whether observed differences reflect true biological changes or noise near the detection limit. Moreover, the range of genes tested seems selective, focusing only on those more likely to change, which may introduce confirmation bias. Extended Data Figure 12 – Potential overfitting of machinelearning classifiers The study introduces a classifier to distinguish treated vs. untreated explants based on transcriptional signatures. However: • The classifier appears to be trained and evaluated on the same dataset, raising concerns about overfitting. • The number of samples (n ≈ 20–30) is too small for robust machine-learning modeling. • No cross-validation, independent test set, or external dataset is used to evaluate generalizability. 9 Given these methodological weaknesses, the reported accuracy metrics likely overstate the model’s predictive power. Extended Data Figure 13 – Incomplete stromal cell-type characterization This figure attempts to characterize fibroblast and myoepithelial subsets using bulk or single-cell RNA-seq. However: • The clustering resolution appears too high, generating many small clusters lacking biological justification. • Marker gene expression is inconsistent across replicates, hinting at batch effects or sample degradation. • The authors do not benchmark their stromal clusters against published mammary stromal atlases, which is essential for validation. In addition, the absence of independent validation using immunostaining (IHC or IF) weakens confidence in these transcriptomic-defined stromal states. Extended Data Figure 14 – Morphological quantification is underpowered The figure presents quantification of epithelial thickness, ductal lumen size, and sidebranch density across treatment groups. Yet: • Sample sizes are unclear. • Variance is high and distribution non-normal, yet authors apply parametric statistics without justification. • No correction for multiple measurements within the same animal is performed. The statistical fragility indicates that morphological differences may not be as robust as claimed. Extended Data Figure 15 – Apoptosis assays lack controls and raw data TUNEL assays are sensitive to: • Tissue fixation • Permeabilization • Enzymatic reaction time • Autofluorescence Yet the authors do not show: 16 Until such improvements are made, the study’s central claims should be interpreted with caution. Reference 1 Simoes, B. M. et al. Anti-progestin therapy targets hallmarks of breast cancer risk. Nature (2025). https://doi.org/10.1038/s41586-025-09684-7