Machine Learning Identifies HNRNPA2B1, RPE, and TAF15 as Potential Stress Granule-Related Biomarkers for Childhood Asthma
核心洞察
An integrated bioinformatics and machine learning study identified HNRNPA2B1 (搜索), RPE (搜索), and TAF15 (搜索) as potential stress granule-related biomarkers for childhood asthma (搜索).
The three genes showed significantly elevated expression in childhood asthma (搜索) blood samples across independent training and validation datasets, with AUC values above 0.7.
RT-qPCR confirmed significantly higher expression of all three genes in childhood asthma (搜索) patients compared to healthy controls (p < 0.001).
A research team has identified three stress granule (SG)-related genes—HNRNPA2B1 (搜索), RPE (搜索), and TAF15 (搜索)—as potential biomarkers for childhood asthma (搜索) (CA), using an integrated bioinformatics and machine learning approach combined with experimental validation. The study, which mined publicly available transcriptomic datasets and validated findings through RT-qPCR, offers a systematic investigation into the previously unexplored link between stress granules and childhood asthma.
Childhood asthma (搜索) is one of the most prevalent chronic respiratory diseases in children, affecting approximately 14% of children worldwide. The disease is characterized by heterogeneous and reversible airway inflammation, hyperresponsiveness, and remodeling, manifesting as recurrent wheeze, shortness of breath, chest tightness, and cough. While inhaled corticosteroids serve as the core treatment, commonly used clinical biomarkers—including exhaled nitric oxide (FeNO), blood eosinophil count, and serum total IgE—have limited sensitivity and specificity in childhood asthma, often failing to accurately predict acute exacerbations or guide individualized treatment.
Stress granules are dynamic, membraneless cytoplasmic condensates formed in eukaryotic cells under stressors such as oxidative stress, viral infection, and heat shock. Their formation is driven primarily by liquid-liquid phase separation (LLPS) and involves the accumulation of untranslated mRNAs, RNA-binding proteins, and translation initiation factors. Although SG dysfunction has been linked to neurodegenerative disorders, cancer, and inflammatory conditions, the specific link between SGs and childhood asthma (搜索) has remained largely unexplored.
Study Design and Biomarker Screening
The researchers retrieved childhood asthma (搜索) datasets from the Gene Expression Omnibus (GEO) database. The training set GSE27011 consisted of 17 CA samples and 18 healthy control samples, while the validation set GSE40732 encompassed 194 blood samples (97 CA and 97 healthy controls). A separate SG-related gene dataset, GSE99304, included three SGcoreRNA and three TotalRNA samples from U-2 OS cells following arsenite-induced stress granule formation.
Differential expression analysis identified 422 significantly differentially expressed genes in the training set (215 up-regulated and 207 down-regulated), and a final SG gene set of 10,589 genes. Overlapping these two gene sets yielded 196 candidate genes. A protein-protein interaction (PPI) network was constructed using the STRING database, comprising 136 nodes and 196 edges. The Maximal Clique Centrality (MCC) algorithm selected the top 50 genes as key candidates, which were then screened using three machine learning methods—Random Forest (RF), Lasso, and Boruta.
The RF analysis yielded 31 characteristic genes, the Lasso algorithm identified 16 characteristic genes, and Boruta analysis confirmed 22 characteristic genes. Integrating the results from all three algorithms produced 11 candidate biomarkers: CCNH, CD247, HNRNPA2B1 (搜索), KRT5, NOTCH2, RPE (搜索), TAF15 (搜索), TAS2R50, DOX20, MKI67, and PAICS.
Cross-Dataset Validation and Diagnostic Performance
In the training set, eight candidate biomarkers (CCNH, CD247, HNRNPA2B1 (搜索), KRT5, NOTCH2, RPE (搜索), TAF15 (搜索), and TAS2R50) showed significantly elevated expression in the CA group, while DOX20, MKI67, and PAICS showed significantly reduced expression (p < 0.05). However, in the validation set, only HNRNPA2B1, RPE, and TAF15 maintained significantly higher expression in the CA group with expression trends completely consistent with the training set.
Receiver operating characteristic (ROC) curve analysis based on the training dataset demonstrated AUC values above 0.7 for all three genes: HNRNPA2B1 (搜索) (0.745), RPE (搜索) (0.781), and TAF15 (搜索) (0.886). A nomogram constructed from the three biomarkers achieved an AUC of 0.941, with a calibration curve p-value of 0.906 (> 0.05), indicating excellent calibration and high predictive reliability.
Experimental Validation and Biological Context
RT-qPCR verification using blood samples collected from Shanghai Children's Hospital confirmed that all three genes showed significantly higher expression in CA blood samples compared to controls (p < 0.001). The samples were drawn from six CA exacerbation patients and six healthy controls aged 5–14 years.
The study provides the first report of RPE (搜索) upregulation in the blood of children with asthma. RPE, or ribulose-phosphate 3-epimerase, is a metabolic enzyme in the pentose phosphate pathway that supplies NADPH, protecting cells from oxidative stress damage. HNRNPA2B1 (搜索) encodes a core component of stress granules involved in RNA processing and metabolism, while TAF15 (搜索) belongs to both the TBP-related factor family and the hnRNP family and is located in the chromosomal region 17q21.31, an important susceptibility locus for childhood asthma (搜索).
Drug Prediction and Immune Correlations
Drug-compound targeting prediction revealed that quercetin and acetaminophen can target all three potential biomarkers simultaneously, while cobaltous chloride and valproic acid can jointly target HNRNPA2B1 (搜索) and RPE (搜索). The top drug-gene pairs included naftopidil with RPE (P value = 0.003296, Combined Score = 2718.11) and lasalocid with TAF15 (搜索) (P value = 0.003296, Combined Score = 2718.11).
Immunoinfiltration analysis identified four immune cell types with significant differences between CA and control groups: CD56dim natural killer cells, central memory CD8+ T cells, immature B cells, and monocytes. Correlation analysis revealed negative associations between central memory CD8+ T cells and both HNRNPA2B1 (搜索) and TAF15 (搜索).
Study Limitations
The authors emphasize that these findings are exploratory and based on a limited sample size derived from public datasets. "At this stage, these genes can only be considered potential association markers rather than clinically validated diagnostic indicators," the researchers note. The AUC values obtained from ROC analysis "merely reflect the discriminatory performance within the current dataset."
Several limitations are acknowledged, including the small sample sizes of both the training set and the RT-qPCR validation cohort, technical heterogeneity between microarray platforms, and the need for prospective cohort evaluation of the nomogram model. The authors conclude that "the above findings should be considered hypothesis-generating, and the identified genes should be regarded as potential SG-related markers rather than validated clinical biomarkers." Future research should prioritize multicenter, large-scale clinical cohort studies encompassing diverse asthma phenotypes and severity strata, combined with cellular and animal models to investigate the specific molecular mechanisms by which these markers regulate SG dynamics and the immune microenvironment.