Application of Cerebrospinal Fluid Host Protein Biosignatures in the Diagnosis of Tuberculous Meningitis in Children from a High Burden Setting

Abstract

Background: The diagnosis of tuberculous meningitis (TBM) especially in children is challenging. New tests are urgently needed for the diagnosis of the disease, especially in resource-limited settings.

Methods: We collected cerebrospinal fluid (CSF) samples from children presenting with symptoms requiring investigation for meningitis at a tertiary hospital in Cape Town, South Africa. Children were later classified as TBM or no TBM using published case definitions. Using a multiplex platform, we investigated the concentrations of biomarkers comprising a previously established 3-marker biosignature (VEGF, IL-13, and LL-37) and other potentially useful host biomarkers as diagnostic candidates for TBM.

Findings: Out of 47 children, age, 3 months to 13 years, 23 were diagnosed with TBM and six (16%) were HIV-infected. We validated the previously identified CSF biosignature (sensitivity of 95.7% (95% CI, 79.0-99.2%) and specificity of 37.5% (95% CI, 21.2-57.3%)). However, substitution of IL-13 and LL-37 with IFN-γ and MPO, respectively, resulted in improved accuracy (area under the ROC curve (AUC) = 0.97, 95% CI, 0.92-1.00, up to 91.3% (21/23) sensitivity and up to 100% (24/24) specificity). An alternative four-marker biosignature (sICAM-1, MPO, CXCL8, and IFN-γ) also showed potential, with an AUC of 0.97.

Conclusion: We validated a previously identified CSF biosignature and showed that refinement of this biosignature by incorporation of other biomarkers diagnosed TBM with high accuracy. Incorporation of these biomarkers into a point-of-care or bedside diagnostic test platform may result in the improved management of TBM in children.

Figure 1

STARD diagram showing the study design and classification of study participants. CRF: case report form; TBM: tuberculous meningitis; no TBM: individuals presenting with symptoms and investigated for TB but TBM ruled out. The no TBM group included bacterial meningitis (n = 2), viral meningitis (n = 2), and children with other diagnoses (Table 1).

Figure 2

Representative plots showing the concentrations of biomarkers detected in CSF samples from children with and without TBM and ROC curves showing the accuracies of these biomarkers in the diagnosis of TBM. Error bars in the scatter-dot plots indicate the median and interquartile ranges. Representative plots for six analytes with AUC ≥ 0.80 are shown. The accuracies of all host biomarkers evaluated in the study are shown in Supplementary Table 2.

Figure 3

Accuracy of the 4-marker CSF biosignature (sICAM-1, MPO, CXCL8, and IFN-γ) in the diagnosis of TBM. Scatter plot showing the ability of the 4-marker signature to classify children as TBM or no TBM (a). ROC curve showing the accuracy of the 4-marker biosignature (b). Red squares: children with TBM. Blue circles: children with no TBM.

Figure 4

Accuracy of a new VEGF-based 3-marker CSF biosignature in the diagnosis of TBM. Scatter plot showing the ability of the 3-marker signature to classify children as TBM or no TBM (a). ROC curve showing the accuracy of the 3-marker biosignature (VEGF, IFN-γ, and MPO) (b). Red squares: children with TBM. Blue circles: children with no TBM.