Identification and validation of NETs-related biomarkers in active tuberculosis through bioinformatics analysis and machine learning algorithms

Siyu Xia, Qi An, Rui Lin, Yi Tu, Chen Zhu, Dongmei WangFrontiers in Immunology 2025 · DOI

TL;DR

Researchers have identified a three-gene signature (CD274, IRF1, and HPSE) related to Neutrophil Extracellular Traps (NETs) that can accurately distinguish active tuberculosis (ATB) from latent infection (LTBI). By integrating transcriptomic data with machine learning, the study provides a robust diagnostic framework and identifies 46 potential drug candidates for host-directed therapy.

What Problem Does This Paper Address?

Tuberculosis (TB) remains a global health crisis, but a major hurdle in controlling it is the "diagnostic gap." Current standard tests, like the tuberculin skin test (TST) and interferon-gamma release assays (IGRAs), are excellent at detecting infection but struggle to differentiate between Active TB (ATB)—the contagious, symptomatic stage—and Latent TB Infection (LTBI), where the bacteria are dormant.

Recent research suggests that neutrophils, the body's first responders, play a dual role in TB: they help fight the bacteria but can also cause massive tissue damage through the release of Neutrophil Extracellular Traps (NETs). NETs are web-like structures of DNA and proteins intended to trap pathogens, but their overproduction is a hallmark of ATB immunopathology. This paper seeks to identify specific NETs-related gene signatures that can serve as reliable biomarkers for early and accurate diagnosis.

Methods

The authors employed a multi-stage bioinformatics pipeline:

  1. Differential Expression Analysis: They compared transcriptomic data from ATB and LTBI patients to find genes that were significantly up- or down-regulated.
  2. Intersection Mapping: These differentially expressed genes (DEGs) were intersected with a curated list of 123 known NETs-related genes (NRGs).
  3. Machine Learning (ML) Triangulation: To ensure robustness, they didn't rely on a single model. Instead, they used three distinct algorithms—LASSO, Random Forest (RF), and SVM-RFE—to identify the most influential "hub genes."
  4. Immune Profiling: They used the CIBERSORT algorithm to estimate the proportions of 22 immune cell types in each sample, correlating gene expression with actual cellular infiltration.
  5. Network Analysis: They constructed regulatory networks involving Transcription Factors (TFs) and miRNAs, and used the DGIdb database to predict potential drug-gene interactions.

Datasets

The study utilized three publicly available transcriptomic datasets from the Gene Expression Omnibus (GEO):

  • GSE19491 (Training): 69 LTBI and 54 ATB samples.
  • GSE62525 (Validation): 14 LTBI and 14 ATB samples.
  • GSE28623 (Validation): 25 LTBI and 49 ATB samples.
  • Reference List: A manually curated list of 123 NETs-related genes derived from previous literature.

Key Insights and Ideas

  • The "Hub" Trio: Out of 123 candidates, only three genes—CD274 (PD-L1), IRF1, and HPSE—consistently emerged across all three machine learning models.
  • Biological Plausibility: These aren't just statistical noise; they have clear roles in TB. CD274 is involved in immune exhaustion/escape; IRF1 is a master regulator of the Type I Interferon response (critical in TB); and HPSE (Heparanase) facilitates the breakdown of the extracellular matrix, likely aiding in NET extrusion and tissue remodeling.
  • Neutrophil Correlation: The expression of these hub genes strongly correlated with neutrophil infiltration levels ($r > 0.6$). This reinforces the theory that the transition to active disease is driven by a "hyper-inflammatory" neutrophil state.
  • Master Regulators: The study identified FOXC1 and GATA2 as key transcription factors, and hsa-miR-106a-5p as a core microRNA, that likely orchestrate the expression of these biomarkers.

Key Results

  • Diagnostic Accuracy: The three-gene signature showed exceptional performance. In the training set, the Area Under the Curve (AUC) for the individual genes ranged from 0.865 to 0.930.
  • The Nomoscore: When combined into a single diagnostic "Nomoscore," the accuracy jumped to an AUC of 0.949, with sensitivity and specificity both exceeding 80%.
  • Validation Consistency: The model held up across independent datasets, maintaining AUCs above 0.75 even in smaller, heterogeneous cohorts.
  • Therapeutic Leads: The drug-gene interaction analysis prioritized 46 candidate drugs. Notably, PD-1 inhibitors (like Pembrolizumab) and Heparin (which can inhibit hepcidin and interfere with bacterial replication) were highlighted as potential candidates for repurposing.

Limitations and Future Work

While the results are promising, the authors acknowledge several hurdles:

  • Observational Data: This study is based on transcriptomic snapshots. It shows correlation, but not necessarily causation. It remains unproven if these genes drive the transition to ATB or are simply markers of it.
  • Sample Heterogeneity: The validation datasets were relatively small and came from different geographic populations, which can introduce noise.
  • Clinical Translation: The predicted drugs require rigorous preclinical validation in animal models before they can be considered for TB therapy.

Future Directions: The authors suggest using spatial omics to see exactly where these genes are expressed within lung granulomas and employing CRISPR screening to knock out these genes in primary human neutrophils to definitively prove their role in NETosis during TB infection.

Summary updated

Abstract

CD274, IRF1, and HPSE represent promising NETs-derived diagnostic biomarkers for ATB. Their dual roles in neutrophil-mediated immunity highlight therapeutic potential, though drug predictions require preclinical validation. Future studies should leverage spatial omics and CRISPR screening to elucidate mechanistic pathways.
BibTeX
@inproceedings{xia2025identification,
  author = {Siyu Xia and Qi An and Rui Lin and Yi Tu and Chen Zhu and Dongmei Wang},
  title = {{Identification and validation of NETs-related biomarkers in active tuberculosis through bioinformatics analysis and machine learning algorithms}},
  booktitle = {Frontiers in Immunology},
  year = {2025},
  doi = {10.3389/fimmu.2025.1599667},
}
Figure