MicroRNA spatial profiling for assessing drug efficacy in BRCA1-related triple-negative breast tumors

Reviewer #1 (Public review):

Summary:

This manuscript presents an innovative approach combining spatial miRNA profiling with computational analysis to characterize treatment-associated tumor states in a BRCA1-deficient breast cancer model.

Strengths:

(1) The integration of latent Dirichlet allocation-based topic modeling and Structural Similarity Index Measure maps analysis provides a potentially valuable framework for studying tumor heterogeneity.

(2) The innovation is high, both conceptually and on technical aspects. The work is a potentially important technical development as spatial transcriptomics for miRNAs is needed, and an interesting manuscript for a large spectrum of readers.

Weaknesses:

The method analysed a limited set of microRNAs, although all are functionally important and well published.

Reviewer #2 (Public review):

Summary:

The authors apply a hydrogel nanoliter-well in situ microRNA assay to tissue sections from a mouse model of BRCA1-related triple-negative breast cancer, in which tumors had acquired resistance to a PARP inhibitor, and test two drug combinations. They develop a spatial analysis that groups wells into microRNA "topics" and relates these topics to treatment sensitivity and to immune infiltration, aiming to show that the spatial arrangement of microRNAs can report on drug efficacy and stratify tumors by their eventual sensitivity or resistance.

Strengths:

(1) The in vivo combination-therapy experiments are technically careful, and the finding that adding Poly(I:C) to olaparib improves antitumor activity is new.

(2) The data and the analysis code are openly deposited on Zenodo.

(3) Whether the spatial organization of microRNAs carries treatment-relevant information remains a worthwhile question.

Weaknesses:

(1) It is not clear what the spatial measurement adds. The discrimination of sensitive vs resistant tumors was already reported in the authors' prior work, and in this dataset the separation reduces to the relative amount of two microRNAs, let-7a and miR-21, that the bulk analysis had already nominated.

(2) The platform is on the well-level rather than single-cell, and the effective spatial resolution (well size and the number of cells per well) is not stated, so the meaning of "spatial" is unclear, and the sensitivity and specificity of the assay are not established in this work.

(3) The sensitive vs resistant separation is an in-sample description of labels that were fixed in advance, on roughly 21 tumors with about three per treatment arm and nothing held out; the model is refit for each analysis rather than frozen, so it cannot be evaluated as a classifier.

(4) The headline association of a let-7a topic with resistance is correlative, unvalidated, and runs opposite to the canonical roles of let-7 as a tumor suppressor and miR-21 as an oncomiR, and it may reflect the abundance ratio of the two dominant probes.

(5) The topic model is simplistic, collapses to two informative topics, and does not use the existing morphology or H&E information already available on the same sections; the co-localization analysis relies on an image-quality metric that is not appropriate for this purpose and lacks a null.

(6) The conclusions are dependent on a single mouse model at a single timepoint with no human data, but the framing attempts to extend to patients.

Reviewer #3 (Public review):

Summary:

This manuscript investigates strategies to overcome resistance to PARP inhibitors (olaparib) in a mouse model of BRCA1-related triple-negative breast cancer. They test whether combining olaparib with a PI3K inhibitor (alpelisib) or with an immune-stimulating agent (Poly(I:C)) can improve outcomes in both drug-sensitive and drug-resistant tumors, and find that both combinations extend survival in sensitive tumors, while only the PI3K combination partially overcomes established resistance. To understand the biological basis for these differences, the authors apply a previously developed hydrogel-based assay that measures seven microRNAs directly within tumor tissue sections, preserving their spatial location. They then build a set of computational tools-based on a topic-modeling method called latent Dirichlet allocation (LDA), principal component analysis (PCA), and an image-similarity measure called SSIM-to interpret these spatial microRNA patterns and relate them to tumor drug sensitivity, treatment type, and the surrounding immune cell environment.

Strengths:

(1) The in vivo survival experiments are well powered and rigorously analyzed, with appropriate statistical testing (log-rank tests) clearly supporting the central efficacy claims: both combination therapies benefit sensitive tumors, and PI3K inhibition (but not Poly(I:C)) partially rescues resistant tumors.

(2) The spatial microRNA measurement technology, while previously published, is thoughtfully applied here to an earlier treatment timepoint (10 days) than prior work, which is a reasonable and useful extension aimed at capturing biology before advanced tumor changes complicate interpretation.

(3) The core finding that a microRNA topic dominated by miR-21 associates with drug sensitivity, while a topic dominated by let-7a associates with resistance, is supported by a formal statistical test (a Mann-Whitney U test comparing principal component scores between sensitive and resistant tumors), giving reasonable confidence in this specific result.

(4) The idea of layering an immune-infiltration similarity map (SSIM) onto the microRNA topic maps is a creative and potentially broadly useful way to connect molecular spatial patterns with tissue architecture, and could be adapted to other spatial biomarker studies beyond this one.

Weaknesses:

(1) Several of the paper's central spatial and immune-colocalization findings are supported only by visual inspection of small numbers of samples (commonly three tumors per treatment group) rather than by formal statistical comparison. This applies to the treatment-type separation shown in Figure 3B.iii and to the immune-microRNA co-localization patterns shown in Figure 4B.ii, both of which are described narratively without accompanying statistical tests.

(2) The claim that Poly(I:C) fails to alter the immune composition of resistant tumors, unlike PI3K inhibition, is stated qualitatively ("little difference") without a quantitative comparison, which weakens confidence in this specific interpretation.

(3) All spatial analyses derive from a single mouse model at a single treatment timepoint, with no independent tumor cohort used to test whether the identified microRNA topics and their treatment/sensitivity associations generalize. This is an appropriate scope for an initial proof-of-concept study but means the biomarker potential of the approach remains unproven outside this specific dataset.

Overall, the authors achieve their stated aims, and the evidence broadly supports the paper's central conclusions distinguishing drug-sensitive from drug-resistant tumors. The additional analyses linking treatment regimen and immune infiltration to spatial microRNA patterns are plausible and consistent with the survival data, though they rely on small sample sizes and, in places, qualitative rather than statistical comparisons, and would benefit from further validation. The analytical framework itself - particularly the SSIM-based integration of molecular and cellular spatial data - is a useful and creative contribution with potential utility for other groups working on spatial biomarker discovery.