Cerebrospinal fluid cytokines and chemokines exhibit distinct profiles in bacterial meningitis and viral meningitis.

Differential diagnosis of bacterial meningitis (BM) and viral meningitis (VM) is a critical clinical challenge, as the early and accurate identification of the causative agent determines the appropriate treatment regimen and markedly improves patient outcomes. Clinical and experimental studies have demonstrated that the pathogen and the host immune response contribute to mortality and neurological sequelae. As BM is associated with the activation of an inflammatory cascade, the patterns of pro- and anti-inflammatory cytokines/chemokines (CTs/CKs) present in the cerebrospinal fluid (CSF) in response to the immune assault may be useful as sensitive markers for differentiating BM from VM. In the present study, the ability of CTs/CKs in the CSF to differentiate between BM and VM was investigated. For this, biochemical markers and CT/CK profiles were analysed in 145 CSF samples, divided into three groups: BM (n=61), VM (n=58) and the control group (C; n=26) comprising patients with meningism. The CSF concentrations of monocyte chemoattractant protein-1, interleukin (IL)-8, IL-1β, IL-6, macrophage inflammatory protein-1α (MIP-1α), epithelial-neutrophil activating peptide, IL-10, tumour necrosis factor-α (TNF-α), proteins and white blood cells were significantly higher and the CSF glucose level was significantly lower in the BM group compared with the VM and C groups (P<0.01). Correlation analysis identified 28 significant correlations between various CTs/CKs in the BM group (P<0.01), with the strongest positive correlations being for TNF-α/IL-6 (r=0.75), TNF-α/MIP-1α (r=0.69), TNF-α/IL-1β (r=0.64) and IL-1β/MIP-1α (r=0.64). To identify the optimum CT/CK patterns for predicting and classifying BM and VM, a dataset of 119 BM and VM samples was divided into training (n=90) and testing (n=29) subsets for use as input for a Random Forest (RF) machine learning algorithm. For the 29 test samples (15 BM and 14 VM), the RF algorithm correctly classified 28 samples, with 92% sensitivity and 93% specificity. The results show that the patterns of CT/CK levels in the CSF can be used to aid discrimination of BM and VM.