henhao
Sign in
BiochemistrybioRxiv
Preprint

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

Omar N. MohdiD, Lin WangiD, Barbara Sardella, David JouiD, Gerburg M. WulfiD, Frank John SlackiD, Yujing Jan HengiD, Patrick S. DoyleiD

Abstract

BRCA1/2-mutated breast cancers exhibit homologous recombination deficiency (HRD) and are initially sensitive to poly(ADP-ribose) polymerase (PARP) inhibitors, but 40-70% of patients develop resistance, creating a need for predictive biomarkers that capture treatment-associated spatial heterogeneity. Using the K14-Cre Brca1f/fTrp53f/fmodel with tumors that acquired PARP inhibitor resistance, we evaluated PARP inhibitor combinations with either PI3K inhibition or Poly(I:C) in vivo. To determine how treatment altered tumor spatial microRNA (miRNA) profiles, we applied our hydrogel-based, nanoliter well array in situ miRNA assay to quantify and spatially profile miRNAs on FFPE sections from tumors treated for 10 days and developed spatial miRNA analysis frameworks integrating latent Dirichlet allocation (LDA) and principal component analysis (PCA). We also incorporated immune architecture using Structural Similarity Index Measure (SSIM) maps to assess co-localization of immune infiltration and miRNA topics. Both combinations improved antitumor activity compared to PARP inhibition alone. The resulting spatial miRNA topics stratified early tumors according to subsequent PARP inhibitor sensitivity or resistance and distinguished their treatment regimens, while SSIM analysis revealed co-localization of immune infiltration and miRNA topics. This integrative spatial miRNA assay and analysis identified a let-7a-dominant topic associated with PARP inhibitor resistance, indicating that spatial miRNA profiling may inform therapeutic stratification in BRCA1/2-related breast cancers.

PDFDOICodeData
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
0
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (20 Jul 2026) · CC BY 4.0

Latent Dirichlet allocation topics from hydrogel-array spatial miRNA maps of day-10 treated K14-Cre Brca1/Trp53 mouse breast tumors separated olaparib-sensitive (miR-21-dominated) from resistant (let-7a-dominated) tumors.

Roughly 40% to 70% of patients with BRCA1/2-mutant breast cancer fail or become resistant to PARP inhibitors, and bulk miRNA assays discard intratumoral spatial heterogeneity that might inform response. Extending the group's earlier endpoint-tumor study of seven miRNAs, this work profiles tumors early in therapy and adds LDA/PCA topic stratification, a priming refinement, and immune co-localization via SSIM.

Key results

  1. ShownAdding alpelisib or Poly(I:C) to olaparib prolonged time to the 20 mm tumor endpoint in olaparib-sensitive orthotopic tumors, but in acquired-resistant tumors only the alpelisib combination gave a significant gain.

    Sensitive: olaparib 75-78 vs control 19-23 days (p<0.001); olaparib+alpelisib 115 vs 75 days and olaparib+Poly(I:C) 120 vs 78 days (both p<0.001). Resistant: olaparib+alpelisib 39 vs 23 days (p=0.018); olaparib+Poly(I:C) 30 vs 24 days (p=0.567). Values are median days to 20 mm; survival curves were compared by log-rank test. · Combination therapies produced promising…; Animal experiments and analysis

  2. ShownWithin this cohort, a 4-topic LDA model trained on all day-10 tumors, with per-tumor topic percentages passed to PCA, separated olaparib-sensitive from resistant tumors on PC1; a miR-21-dominated topic loaded positively and a let-7a-dominated topic negatively.

    21 tumors, over 115,248 spatial data points; PC1 explained 96% of variance; median PC1 0.42 (sensitive) vs −0.083 (resistant), two-sided Mann-Whitney U p<0.001, n=9 per group. Bulk tests: only let-7a differed (lower with olaparib+Poly(I:C) vs olaparib alone, adj.p=0.01). · Spatial miRNA analysis; Case-study-wide spatial LDA implementation; PCA statistical methods

  3. SuggestedAmong resistant tumors, the global model separated olaparib-only from combination-treated tumors, with combinations increasing the miR-21-dominated topic share toward a sensitive-like profile; sensitive tumors showed little separation by regimen.

    Qualitative, based on per-tumor topic percentages and PCA placement; no statistic is reported for treatment-group separation. In sensitive tumors this lack of separation contrasted with roughly two-fold longer survival under combination therapy. · Spatial miRNA analysis; Discussion

  4. SuggestedPriming sensitive-tumor inputs with per-miRNA maximum topic probabilities from a sensitive-only LDA model, then reapplying the global model, separated sensitive tumors by regimen: let-7a topic higher with olaparib alone, miR-21 topic higher with combinations.

    Separation was almost entirely along PC1. Priming weights: let-7a 0.92, miR-21 0.88 (let-7a about 4% higher), other miRNAs lower. In olaparib-only tumors, regions with near-equal miR-21 and let-7a switched from topic 2 to topic 1, driving the split. No test statistic; relies on known sensitivity status. · Refined spatial miRNA analysis of PARP…; Priming-based specialized LDA analysis for…

  5. SuggestedStructural similarity (SSIM) maps between the two main miRNA topics and a normalized immune-to-tumor cell ratio from CD45 IHC, summarized by a 3-topic LDA, showed immune infiltration aligned with the let-7a topic in most tumors but with the miR-21 topic in sensitive olaparib+Poly(I:C) tumors.

    SSIM used an approximately 7×7 pixel window on 28×28 tumor maps; results are qualitative per-tumor SSIM-topic percentages without statistics. In sensitive olaparib+Poly(I:C) tumors (e.g., SOP3), miR-21-immune similarity was strongest near let-7a-immune boundaries; in resistant tumors, olaparib+Poly(I:C) differed little from olaparib alone. · Spatial associations between miRNAs and tumor…; Discussion; SSIM map formulation; LDA-SSIM implementation

Methods, data and reproducibility

Olaparib-sensitive or acquired-resistant K14-Cre Brca1/Trp53 tumor fragments grown orthotopically received olaparib alone or plus alpelisib or Poly(I:C); day-10 FFPE sections were profiled with a 7-plex hydrogel nanoliter-well in situ miRNA array and analyzed with latent Dirichlet allocation (LDA) topics, PCA, data priming, and SSIM versus CD45+ immune maps.

  • Orthotopic transplants of K14-Cre Brca1^(f/f)Trp53^(f/f) mammary tumor fragments (olaparib-sensitive and derived olaparib-resistant) in female FVB/NJ mice for survival studies · mouse
  • Day-10 treated FFPE tumor sections (olaparib, olaparib+alpelisib, olaparib+Poly(I:C); vehicle for sensitive tumors) profiled with the 7-plex in situ miRNA assay · mouse · n = 21 tumors; n=9 per group in sensitive vs resistant PC1 test · 10.5281/zenodo.18499891
  • H&E images for tumor/stroma cell classification, plus CD45 IHC on the serial section cut right after each miRNA-assayed section for immune cell counts · mouse
Methods & tools
Hydrogel nanoliter-well in situ miRNA assay with rolling circle amplification, Latent Dirichlet allocation (collapsed Gibbs sampling), Principal component analysis (PCA), Structural Similarity Index Measure (SSIM), MATLAB built-in function, Priming-based specialized LDA (weights from sensitive-only model), MATLAB (MathWorks), QuPath, GraphPad Prism, CD45 IHC (rat anti-mouse clone 30-F11), Two-sided Mann-Whitney U test, Log-rank test with Kaplan-Meier curves, Two-way ANOVA with Šídák's multiple comparisons
Code
https://doi.org/10.5281/zenodo.18499891
Data & accessions
https://doi.org/10.5281/zenodo.18499891
Reagents & environment
MATLAB (MathWorks) for preprocessing, LDA, and SSIM; QuPath for H&E/IHC cell classification; GraphPad Prism for statistics. No software versions or containers stated.
Experimental conditions
Case-study-wide LDA training with collapsed Gibbs sampling took approximately 21 minutes; hardware not stated.
Reproducibility gaps
Code and data share one Zenodo record; the analysis depends on proprietary MATLAB, and assay device fabrication and protocol are described only by reference to prior publications. Software versions are not given.
Limitations (6)
  • Only one BRCA1-related TNBC mouse model and a single 10-day treatment timepoint were profiled; the authors call for other treatment stages and BRCA1/2 tumor models to test whether the spatial topics persist. · stated by the authors
  • The priming analysis used prior knowledge that tumors were olaparib-sensitive, so the authors state it is not a fully prognostic standalone classifier but a way to reveal which spatial patterns separate regimens. · stated by the authors
  • Linking let-7a- and miR-21-dominated topics to resistance and sensitivity does not explain the underlying biology; the authors note that added spatial biomarkers and histology of miRNA-defined regions are needed. · stated by the authors
  • Spatial topic maps are collapsed into per-tumor percentages before PCA, discarding spatial detail; the authors note that larger preclinical and clinical BRCA1/2 cohorts are needed before any formal predictive classification. · stated by the authors
  • LDA, PCA, and priming were fit and evaluated on the same 21 tumors (n=9 per group in the PC1 test) with no held-out set or cross-validation, and only seven preselected miRNAs were measured. · noted in this summary
  • Survival reporting omits per-arm mouse numbers and gives inconsistent medians for olaparib-sensitive tumors (75-78 vs 75-58 days); there is no matched 10-day vehicle cohort for resistant tumors. · noted in this summary

Who should read:Read this if you develop spatial biomarkers of PARP inhibitor response or want an LDA/PCA and SSIM workflow for low-plex spatial miRNA maps from FFPE tumor sections.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISBhin '22Lynce '26Reeves '19Tatarova '21Frederick '25Parvin '20
61% semantic matchCancer Biology
J. Bhin, M. Paes Dias, E. Gogola+18 more·bioRxiv·Sep 7, 2022