Transcriptomic Characterisation and Diagnostic Performance of S100A8, S100A9, S100B, and ENO2 (Neuron-Specific Enolase) in Pediatric Bacterial Meningitis: A Cross-Platform Re-analysis of Two Public GEO Datasets

Abstract

Background: Bacterial meningitis (BM) remains a leading cause of mortality and neurological disability in children worldwide. Despite advances in antimicrobial therapy, neurological sequelae persist in up to 50% of survivors. Host-blood transcriptomic biomarkers have shown promise for early differentiation of bacterial from non-bacterial causes of meningitis; however, no cross-platform transcriptomic analysis of the three biologically distinct S100 protein-family members (S100B, S100A8, S100A9) together with ENO2 (encoding NSE) in pediatric BM has been reported. Methods: Gene expression profiles were obtained from the public Gene Expression Omnibus (GEO) database. The discovery cohort (GSE80496) included 24 children with bacterial meningitis/sepsis and 21 age-matched healthy controls (platform GPL6883, Illumina HumanRef-8 v2.0). The validation cohort (GSE248261) comprised 5 BM children and 5 controls sequenced on the DNBSEQ-T7 platform. Expression levels of four candidate diagnostic transcripts—S100A8 and S100A9 (which together form the calcium-dependent calprotectin heterodimer complex), S100B (an independent astroglial and Schwann-cell injury marker), and ENO2 (encoding NSE, a neuronal glycolytic enzyme)—were extracted, log2-transformed, and compared between groups using the limma package. Diagnostic performance was assessed using ROC curve analysis, decision curve analysis, and leave-one-out cross-validation (LOOCV). Correlation analysis, gene set enrichment analysis (GSEA), protein-protein interaction (PPI) network construction, and subgroup analyses were performed as complementary analyses. Results: All four transcripts were significantly upregulated in the bacterial meningitis group compared with healthy controls: S100A8 (linear fold change = 3.21, p < 0.001), S100A9 (linear fold change = 2.87, p < 0.001), ENO2 (linear fold change = 2.14, p < 0.001), and S100B (linear fold change = 2.19 [log₂FC = 1.10], p = 0.012). The combined four-transcript panel achieved a cross-validated AUC of 0.864 (95% CI 0.756–0.936) for discriminating BM from controls in the discovery cohort, with sensitivity 83.3% and specificity 81.0% at the Youden-optimal cut-off, and replicated at 0.840 in the independent RNA-sequencing validation cohort. Decision curve analysis confirmed greater net clinical benefit of the combined panel across the clinically relevant threshold range. In exploratory analyses, S100A8 and ENO2 transcript levels were positively associated with neurological-impairment severity (ρ = 0.816 and 0.664, both p < 0.001); however, these severity associations are hypothesis-generating and were not designed as primary endpoints. Conclusion: S100A8, S100A9, S100B, and ENO2 transcript levels were significantly upregulated in children with bacterial meningitis, and a combined four-transcript panel showed reproducible diagnostic discrimination in two independent cohorts. The transcript level findings warrant paired serum/CSF protein-level validation and prospective evaluation before any clinical inference about disease severity or outcome can be drawn.