---
title: "Formononetin targets GSTM2 to reshape the prostate cancer immune microenvironment via ROS/STAT3 ch"
id: "frontiers-in-immunology-9-integrated-pharmacology-and-validation-reveal-formononetin-upregulates-gstm2-to"
canonical_url: "https://medichelpline.com/clinical-feed/frontiers-in-immunology-9-integrated-pharmacology-and-validation-reveal-formononetin-upregulates-gstm2-to"
content_type: "clinical_feed_article"
specialty: "Oncology"
source_name: "Frontiers in Immunology"
source_url: "https://www.frontiersin.org/articles/10.3389/fimmu.2026.1964912"
published_at: "2026-10-23T00:00:00.000Z"
evidence_level: "Journal Feed"
license: "CC-BY-NC-4.0 / Informational Use"
---
# Formononetin targets GSTM2 to reshape the prostate cancer immune microenvironment via ROS/STAT3 ch
## Provenance & Clinical Metadata
- **Canonical URL:** https://medichelpline.com/clinical-feed/frontiers-in-immunology-9-integrated-pharmacology-and-validation-reveal-formononetin-upregulates-gstm2-to
- **Specialty:** [Oncology](https://medichelpline.com/clinical-feed/oncology.md)
- **Primary Source:** Frontiers in Immunology
- **Source URL:** [Original Journal Publication](https://www.frontiersin.org/articles/10.3389/fimmu.2026.1964912)
- **Published At:** 2026-10-23T00:00:00.000Z
- **Evidence Rating:** Journal Feed
## Executive GIST (TL;DR)
- Investigators integrated five public databases (TCMSP, CTD, HERB, SwissTarget, PharmMapper) to generate 413 unique **Formononetin** target candidates and found 335 overlapped with prostate cancer–related genes. - Transcriptomic filtering and WGCNA in the DKFZ_2018 cohort reduced the list to module-derived genes; differential expression in TCGA_PRAD provided additional refinement, producing 22 candidate Formononetin-related genes in PCa. - An ensemble of machine-learning models (LASSO Cox, random survival forest, SVM-RFE) and survival analyses pointed to **GSTM2** as the highest-priority target; investigators reported correlations between GSTM2 expression, immune infiltration, and predicted sensitivity to Bicalutamide. - Single-cell RNAseq analysis (GSE157703) showed GSTM2-positive malignant cells had upregulated glutathione metabolism pathways and downregulated respiratory-burst signatures. - Structural work included AutoDock Vina docking and 100-ns molecular dynamics using Amber99SB-ILDN/GAFF2 force fields; the complex was characterized by RMSD/RMSF/Rg/SASA and FEL analyses, and investigators described stable binding of Formononetin to GSTM2. - In vitro functional assays in LNCaP and VCaP lines demonstrated that **Formononetin** upregulated **GSTM2**, reduced intracellular **ROS**, suppressed **STAT3** phosphorylation, and required GSTM2 to inhibit proliferation, colony formation, migration, and invasion. - Chemokine assays revealed upregulation of **CXCL10** and downregulation of **CCL22** following Formononetin treatment in a GSTM2-dependent manner, consistent with a shift toward a more cytotoxic immune milieu. - The report did not include in vivo efficacy data; investigators concluded that in vivo validation is warranted before translational claims can be made.
## Clinical Analysis & Structured Key Points
Frontiers | Integrated pharmacology and validation reveal Formononetin upregulates GSTM2 to modulate prostate cancer immune microenvironment via ROS/STAT3-driven chemokine reprogramming ORIGINAL RESEARCH article Front. Immunol. , 23 October 2026 Sec. Molecular Innate Immunity Volume 17 - 2026 | https://doi.org/10.3389/fimmu.2026.1964912 Published in Frontiers in Immunology Molecular Innate Immunity 7 impact factor 11.3 citescore Editor & Reviewers Edited by J W Jia-feng Wang Reviewed by Y W Yaoan Wen A S Aakansha Singh Outline Figures and Tables Figure 1 View in article Figure 2 View in article Figure 3 View in article Figure 4 View in article Figure 5 View in article Figure 6 View in article Figure 7 View in article Figure 8 View in article Figure 9 View in article Figure 10 View in article ORIGINAL RESEARCH article Front. Immunol. , 23 October 2026 Sec. Molecular Innate Immunity Volume 17 - 2026 | https://doi.org/10.3389/fimmu.2026.1964912 Integrated pharmacology and validation reveal Formononetin upregulates GSTM2 to modulate prostate cancer immune microenvironment via ROS/STAT3-driven chemokine reprogramming T Y Tianlun Yu 1 † Y Z Yu Zhou 2 † * F H Fangkai He 2 † Y W Yufan Wu 2,3 S T Shiyun Tang 2 Z Z Zhouhan Zhu 2 W Q Wei Qian 2,4 * 1. Department of Clinical Laboratory, Kunshan Rehabilitation Hospital, Kunshan, China 2. Central Laboratory, Kunshan Hospital of Traditional Chinese Medicine, Kunshan Affiliated Hospital of Nanjing University of Chinese Medicine, Kunshan, China 3. Department of Urology, Kunshan Hospital of Traditional Chinese Medicine, Kunshan Affiliated Hospital of Nanjing University of Chinese Medicine, Kunshan, China 4. Kunshan Key Laboratory of Immune Regulation in Health and Disease, Kunshan, China See more Article metrics View details Abstract Background: Prostate cancer (PCa) is responsible for a substantial proportion of male cancer cases worldwide. Biochemical recurrence (BCR) representing a critical challenge in its disease management. Formononetin, a natural isoflavone derived from traditional Chinese medicine, has demonstrated potential anticancer properties. We sought to determine the translational potential of Formononetin-targeted molecules in PCa. Methods: A list of putative targets of Formononetin were identified through five public databases. Overlapping candidates underwent further refinement using an ensemble of bioinformatics tools, encompassing differential expression profiling, co-expression network construction, and multiple machine-learning classifiers. Docked complexes and complementary analyses were subjected to evaluate the binding stability. To functionally validate GSTM2, we assessed its functional role on PCa cell growth, motility, and invasiveness in vitro , as well as the expression of CXCL10 and CCL22 chemokines. Results: Integrative analysis identified 413 Formononetin targets, with overlapped 335 PCa-associated genes. Through weighted gene co-expression network analysis and differential expression analysis, 22 candidate genes were identified as Formononetin-related genes in PCa. Machine learning algorithms and survival analysis further suggested GSTM2 as the most important target. GSTM2 was positively correlated with enhanced immune infiltration and sensitivity to Bicalutamide. Single-cell analysis revealed that GSTM2-positive malignant cells exhibited upregulated glutathione metabolism and suppressed respiratory burst pathways. Stable complex formation between Formononetin and GSTM2 was demonstrated by computational docking. Cell-based assays demonstrated that Formononetin upregulated GSTM2 and inhibited PCa malignancy in a GSTM2-required manner. In terms of mechanism, Formononetin attenuated ROS levels and suppressed STAT3 phosphorylation via GSTM2, leading to reprogramming of chemokine expression characterized by upregulated CXCL10 and downregulated CCL22. Conclusion: Formononetin reprograms chemokine expression through the GSTM2-ROS-STAT3 axis to regulate the immune microenvironment, suggesting that it might have immune-modulating effects and is worth further in vivo studies. 1 Introduction Biochemical recurrence (BCR) can happen in up to 46% of prostate cancer (PCa) patients within ten years after radical treatment, but right now, there are no approved drugs specifically for this post-surgery stage ( 1 , 2 ). Androgen deprivation therapy is currently the standard approach for high-risk cases. It can buy some time, but the cost is metabolic toxicity and inevitable development of castration-resistant prostate cancer (CRPC) ( 3 ). Is there a well-tolerated natural compound that could fill this gap? Traditional Chinese Medicine (TCM) active ingredients have exhibited unique properties in prevention and treatment, due to their multi-targeting, low toxicity, and synergistic properties ( 4 – 6 ). Here, we focus on Formononetin, which is an isoflavone from Astragalus . It is known for its antioxidant activity ( 7 – 9 ), its specific molecular targets and regulatory networks in PCa were lack of systematic exploration. Network pharmacology could help decipher the systematic elucidation of the multi-target synergistic mechanisms of TCM active ingredients. In cancer research, bioinformatics algorithms, such as weighted gene co-expression network analysis (WGCNA), machine learning models are broadly adopted. By integrating these technical approaches, we can screen for potential biomarkers and actionable drug targets. While computational docking and dynamics were employed to characterize the binding stability of small molecules to target proteins at the atomic level. Under this approach, we identified a key target gene for Formononetin in PCa. As a mu class glutathione S-transferase, GSTM2 encoded enzyme catalyzes glutathione conjugation to electrophilic substrates, thereby participating in the detoxification of carcinogens, environmental toxins, and oxidative stress products, while maintaining intracellular redox balance ( 10 – 12 ). In a whole-genome methylation analysis of 58 prostate tumor tissues, Angulo et al. found GSTM2 promoter hypermethylation in 79.3% of cases, which independently predicted BCR after radical prostatectomy ( 13 ). Moreover, GSTM2 is significantly overexpressed in enzalutamide-resistant CRPC and drives resistance to second-generation androgen receptor inhibitors by counteracting oxidative stress and activating the p38 MAPK pathway, suggesting its roles may vary across disease stages ( 14 ). As a key antioxidant enzyme, GSTM2 regulates intracellular ROS scavenging ( 15 ). ROS, acting as second messengers, activate JAK/STAT signaling, influencing tumor cell proliferation, survival, and immune cell recruitment ( 16 , 17 ). The STAT1 and STAT3 are the most frequently activated by ROS among STAT family members. They may alter the immune microenvironment in a way of regulating downstream chemokines. They led in distinct directions: STAT1 mainly upregulates CXCL10 to recruit CD8 + T, NK and Th1 cells ( 18 – 20 ), whereas STAT3 upregulates CCL22 to recruit Treg and Th2 cells ( 21 , 22 ). To some extent, the CXCL10/CCL22 ratio reflects the differences of immune cell infiltration ( 23 ). However, whether GSTM2 influences early-stage PCa progression through the ROS/STAT/chemokine axis remains unexplored, and no study has investigated GSTM2 as a direct target of Formononetin. 2 Methods 2.1 Targets of formononetin A heterogeneous data acquisition strategy was employed to identify putative targets of Formononetin. Initially, TCMSP ( https://www.tcmsp-e.com/ ) was queried for the keyword “Formononetin” ( 24 ), yielding 39 related targets, which were subsequently manually curated to convert them into their corresponding human gene symbols. Concurrently, targets were acquired from CTD ( https://ctdbase.org/ ) and HERB ( http://herb.ac.cn/ ) databases ( 25 , 26 ), resulting in 91 and 34 targets, respectively. To further expand the target landscape, structure-based computational predictions were conducted via SwissTarget ( https://swisstargetprediction.ch/ ) and PharmMapper ( https://www.lilab-ecust.cn/pharmmapper/index.html ) web servers ( 27 , 28 ). Briefly, the canonical SMILES string and the SDF V2000 structural file (both obtained from PubChem, CID: 5280378) were uploaded to these platforms, which predicted 101 and 239 potential targets, respectively. 2.2 Access and analysis of online cancer database resources Using “prostate cancer” as the search term, PCa-related genes were obtained from GeneCards ( https://www.genecards.org/ ) and OMIM ( https://www.omim.org/ ) databases. A relevance threshold of 1.0 was applied in the GeneCards. This approach yielded 7,508 and 180 genes from each database, respectively, which after deduplication resulted in a final set of 7,515 unique genes. Transcriptome data were sourced from multiple sources, including DKFZ_2018, TCGA_PRAD and GSE54460 (Information in Supplementary Table S1 ). For different expression genes (DEGs) analysis, the result of TCGA_PRAD cohort was using “DESeq2” package based on raw count data. For WGCNA, gene expression in TPM format from the DKFZ_2018 cohort was subjected to filtering, retaining only genes with an inter-sample standard deviation > 0.5 to exclude those with low variation. An approach involving β values from 1 to 30 was employed, with the final parameter selected by balancing scale-free R² and mean connectivity to meet the network assumption. The optimal power ( 12 ) was selected as the value that maximized R² while maintaining sufficient connectivity, ensuring both network fitness and biological interpretability. Subsequently, modules with a correlation coefficient |r| > 0.3 with BCR were extracted. 2.3 Enrichment characterization and PPI network analysis of candidate targets The R package “clusterProfiler” was applied for functional annotation of the candidate genes, with GO annotations from “GO.db” and KEGG pathways retrieved via “createKEGGdb” ( 29 ). Pathway visualizations were generated using “pathview” ( 30 ). For PPI analysis, candidates were submitted to STRING ( https://string-db.org ) database, and interactions passing the confidence threshold of 0.4 were retained ( 31 ). The node connectivity (degree) was recalculated to construct the final PPI network. 2.4 Machine learning algorithms LASSO Cox regression, random survival forest, and SVM-RFE were employed as three complementary machine learning approaches. (i) LASSO Cox regression was performed to select the most robust prognostic genes from the candidate set by “glmnet” R package ( 32 – 34 ). LASSO regression applies a penalty to shrink coefficients, discarding variables with zero coefficients while retaining the rest. The optimal penalization parameter (λ) was selected via 10-fold cross-validation using the λ.min criterion. (ii) “randomForestSRC” R package was used for random survival forest model. Bootstrap resampling was applied to generate training subsets, with model performance internally validated against out-of-bag (OOB) samples. The Nelson-Aalen estimator was used to estimate cumulative hazard, and the final prediction was aggregated from all trees via weighted voting. The predictive performance of the model was reflected by the C-index ( 35 ). (iii) SVM-RFE, an embedded machine learning algorithm for binary classification, was applied to screen key features using the mSVM-RFE function. The algorithm recursively eliminates features with the smallest contribution to the classification hyperplane, and the optimal subset was determined via cross-validation ( 36 ). All candidate genes were included as initial features with default settings. 2.5 Prognostic validation of hub genes The hazard contribution of each hub gene’s expression to BCR-free survival was estimated by univariate Cox regression, forest plots were generated to present the results. Patients were dichotomized by median expression, and log-rank comparison of survival curves was performed in Kaplan–Meier analysis. 2.6 Immune infiltration and drug sensitivity analysis The tumor microenvironment (TME) characteristics were generated via the “estimate” R package ( 37 ). The ssGSEA method was also served to quantify the composition of diverse immune cell types. We adopted “oncoPredict” R package to assess chemo-response to Bicalutamide ( 38 ). The GDSC2 database version 8.5 was employed as a training dataset. Genes with low variation across samples, defined as those with a coefficient of variation below 0.2, were filtered out to improve the robustness of the analysis. 2.7 Single-cell transcriptomic data Single-cell RNAseq data from PCa samples were obtained from GSE157703 ( 39 ). Raw count matrices were processed with Seurat V5, including quality filtering (cells with 25% mitochondrial reads), variable feature selection (2,000 genes via vst) following log-normalization (scale factor = 10,000), scaling, and Harmony batch correction. Dimensionality reduction was performed using PCA and UMAP on was constructed to cluster cells at a resolution of 0.5. The Human Primary Cell Atlas served as a guide for SingleR-based cell-annotation, followed by manual refinement based on canonical marker genes ( 40 , 41 ). For functional annotation, we ranked all expressed genes using FindMarkers function with Wilcoxon rank-sum test ( p 1 and adjusted p -value 0.3 as the cutoff, 1,324 genes were retrieved from modules ( Figur
## Related Clinical Research

- [Editorial Expression of Concern: Image concerns in EphA2–arachidonic acid prostate cancer paper](https://medichelpline.com/clinical-feed/british-journal-of-cancer-0-editorial-expression-of-concern-ligand-independent-activation-of-epha2-by.md)
- [Cribriform on Follow-up Biopsy Predicts Active Surveillance Failure in MRI-Era Prostate Cancer](https://medichelpline.com/clinical-feed/medrxiv-21-cribriform-status-at-follow-up-biopsy-is-critical-for-predicting-prostate.md)
- [Galleri multicancer blood test: expert views on performance, limits, and clinical use](https://medichelpline.com/clinical-feed/healthcare-dive-2-5-doctors-weigh-in-on-the-promise-limitations-of-a-closely-watched-multicancer.md)
- [SRSF1 controls transcription and alternative splicing to alter p53 signaling and genome maintenance](https://medichelpline.com/clinical-feed/biorxiv-4-srsf1-coordinates-transcription-and-alternative-splicing-to-modulate-p53-and.md)
- [Prostate cancer whole-genome profiling in African American men identifies novel drivers](https://medichelpline.com/clinical-feed/plos-one-12-personalized-genomic-landscape-of-prostate-cancer-in-african-american-men.md)

## Navigation
- [← Back to Oncology Feed](https://medichelpline.com/clinical-feed/oncology.md)
- [← All Clinical Specialties](https://medichelpline.com/clinical-feed.md)
## Medical & Regulatory Disclaimer

> [!CAUTION]
> MedicHelpline content is structured for research, educational, and professional discovery purposes. It does not constitute individual medical advice, clinical diagnosis, or treatment recommendations.
> Always verify dosing, contraindications, and regulatory alerts against official product labeling and primary regulatory sources before clinical decision-making.