Evaluation of New Biomarker-based Approaches for Improving the Diagnosis of Childhood Tuberculous Meningitis

Introduction:

te ongoing research, early and cost-effective diagnostic tools for TBM are lacking. The detection of Mycobacterium tuberculosis (Mtb) in cerebrospinal fluid (CSF) is the gold standard for diagnosing TBM. Unfortunately the sensitivity of both smear microscopy and culture for TBM is low. Depending on the reference standard employed, the sensitivity of the GeneXpert test (Cepheid Inc, USA) for TBM is approximately 50-60%, and improved to 72% when centrifuged CSF was used in one study. In a more recently published study conducted on HIV positive adults, however, the GeneXpert performed with a sensitivity of 43% or 45%, compared to 43% or 45% for culture and 70% or 95% for the GeneXpert Ultra, depending on which of the two reference standards were used. Despite the relatively high roll-out of the GeneXpert test across South Africa, the test is currently mostly offered at centralised laboratories. The availability of the test in other African countries is limited. The diagnosis of TB relies on the poorly sensitive symptom screening and smear microscopy, especially at rural health centres. Mtb culture facilities are often only available at referral level laboratories and results might take up to 42 days. The need for multiple health care visits leads to loss of follow-up and delayed diagnosis, fuelling the spread of TB and advanced lung damage. In the case of TBM in particular, proper diagnosis is only made upon admission in a tertiary level referral center. In routine clinical practice, diagnosis is mostly based on a combination of clinical findings, multiple laboratory tests on the CSF, imaging findings and the exclusion of common differential diagnoses. Most of these techniques are unavailable in many high-burden, but resource-constrained settings in most of sub-Saharan Africa. Children seen at primary and secondary healthcare facilities often have multiple missed opportunities, up to six visits, before eventual diagnosis of TBM is made in a relatively well-resourced setting in South Africa. Findings from the CSF can be highly variable. Recently, international experts have proposed new uniform case definitions that should be employed in future research to replace the many different definitions in the literature. New tests are therefore urgently needed for the diagnosis of TBM.

The investigators investigated the potential of host markers detected in CSF samples from children suspected of having TBM as diagnostic candidates for TBM . The investigators evaluated the levels of the host biomarkers present in a standard BioPlex 27plex multiplex cytokine kit (Bio Rad Laboratories) and other protein biomarkers in CSF and serum samples. An unsupervised hierarchical clustering and principal component analysis, using the Glucore Omics explorer, revealed significant clustering of patients with TBM by the biomarkers detected in the CSF. A 3-marker host protein biosignature comprising vascular endothelial growth factor (VEGF), interleukin (IL)-13 and the antibacterial peptide cathelicidin, LL-37, showed potential as a diagnostic biosignature for TBM (international patent application: PCT/IB2015/052751), diagnosing TBM with an area under the receiver operator characteristics curve (AUC) of 0.91, with sensitivity of 52%, but with good specificity of 95%. Since the publication of this biosignature, the investigators have evaluated the diagnostic potential of >70 host biomarkers in serum and plasma samples from adults suspected of having active pulmonary TB in 5 different African countries (South Africa, Namibia, Malawi, Uganda and Ethiopia) in an EDCTP-funded trial (AE-TBC). The investigators identified, patented (PCT/IB2015/051435 and PCT/IB2017/052142), and published 6- and 7-marker protein biosignatures with strong diagnostic potential for TB.

In a more recent study (South African Provisional Patent application; Manyelo et al 2019), the investigators hypothesized that at least some of the host biomarkers comprising our adult protein biosignatures may be useful for TBM diagnostics. Funded by the South African Technology Innovation Agency (PI: Chegou), the investigators prospectively enrolled a new cohort of children suspected of having TBM at the Tygerberg Academic Hospital, Western Cape, and determined the concentrations of 66 host biomarkers, in CSF samples from these children. The investigators also included the 3 biomarkers that comprised our previous CSF biosignature for TBM (VEGF, IL-13 and cathelicidin LL-37) for validation purposes in this new study; a total of 69 host protein biomarkers.

With the exception of VEGF (AUC of 0.81), the accuracy of the individual markers in the previous 3-marker signature was poor (AUCs of 0.58 and 0.55, respectively, for IL-13 and LL-37) but when used in combination the discrimination between TBM and no-TBM by the 3-marker model was confirmed [AUC of 0.67 (95% CI: 0.52-0.83); sensitivity of 75% and specificity of 65%]. Forty-seven of the additional markers showed significant differences between the TBM and no TBM groups (Mann Whitney U test), with 28 showing strong diagnostic potential, even as individual markers (AUC ≥ 0.80). These markers include interferon (IFN)-γ, CCL18(MIP-4), CXCL9, CCL1, CCL5(RANTES), IL-6, tumour necrosis factor (TNF)-α, myeloperoxidase (MPO), matrix metalloproteinase 9 (MMP), MMP-8, complement C2 (CC2), IL-10, total plasminogen activator inhibitor 1 (PAI-1), CXCL8, IL-1β, alpha-2-antitrypsin(A1AT), CXCL10, granulocyte colony stimulating factor (G-CSF), CC4, CC4b, granulocyte-macrophage colony stimulating factor (GM-CSF), platelet-derived growth factor (PDGF)-AB/BB, apolipoprotein A1 (apoA1), mannose-binding lectin (MBL), ferritin, CC5a, serum amyloid P (SAP), and CC5.

Combinations of these biomarkers were investigated and using Linear Discriminant Analysis (LDA) models. A 4-marker CSF biosignature comprising soluble intracellular adhesion molecule (sICAM)-1, MPO, CXCL8 and IFN-γ diagnosed TBM with an AUC of 0.97 (95% CI: 0.92-1.00), with a sensitivity of 87% (20/23) and specificity of 95.8% (23/24). After leave-one-out cross validation, there was no change in the sensitivity and specificity of the 4-marker biosignature. Further optimization of the 4-marker biosignature by the selection of better cut-off values resulted in a sensitivity and specificity of 96% and 96%, respectively.

As VEGF performed well in single-marker analyses (19), the investigators evaluated the potential accuracy of other biosignatures that included VEGF. A 3-marker model comprising VEGF, IFN-γ and MPO discriminated with high accuracy between the children with and without TBM. In leave-one-out cross validation and optimizations of best cut-off values, the sensitivity and specificity of the 3-marker VEGF-based signature were 92% and 100%, respectively.

All 69 host markers investigated in CSF samples were also investigated on serum samples using the Luminex multiplex platform. The median serum levels of 17 analytes [sVCAM1, CCL2, IL-4, TNF-α, CCL4, adipsin, SAP, CC5, CFH, G-CSF, IL-10, Apo-CIII, IL-17A, PAI-1(total), PDGF AB/BB, MBL and NCAM1] were significantly different (p<0.05; Mann Whitney U test) between children with and without TBM. When the diagnostic potential of individual serum biomarkers was assessed by ROC curve analysis, 13 of the markers had promising AUC ≥ 0.70. LDA demonstrated that optimal diagnosis of TBM was achieved using 3 markers. The most accurate 3-marker serum biosignature for the diagnosis of TBM [adipsin (complement factor D), Ab42 and IL-10] diagnosed TBM with an AUC of 0.84 (95% CI: 0.73-0.96), a sensitivity of 82.6% (19/23) and specificity of 75% (18/24). In leave-one-out cross validation, the sensitivity remained 82.6% (19/23) with the specificity decreasing to 70.3% (17/24). Further optimisation of the biosignature by selection of better cut-off values resulted in an improved sensitivity and specificity of 83% and 83%, respectively.

OVERALL OBJECTIVE The main objective is to validate previously identified host serum and CSF biomarkers and to develop a biosensor-based POC test for the diagnosis of TBM, based on these biomarkers.

The investigators propose to identify a panel of correlated biomarkers that showed potential in previous studies. This will be done to identify biomarkers which can be substituted with each other as the transition from a laboratory-based technological platform such as Luminex to a POC test using a biosensor-based technology is likely to be faced by the loss of some of the markers due to technical reasons or due to unavailability of some of the markers due to antibody ownership or cost issues. Highly correla-ted markers can then substitute such markers. The investigators will test which set of biomarkers works best in the POC diagnostic test platform. Finally, the investigators will evaluate the prototype test prospectively in a new cohort of 300 study participants with suspected TBM as described below.

The prototype test will be based on the best biosignature of CSF or serum biomarkers, depending on which performs best. However, developing the test based on serum biomarkers may be advantageous as CSF samples are difficult to collect. Furthermore, a test based on serum biomarkers may be easily converted to a fingerprick based test, which will be much easier to implement in resource-constrained settings. The investigators are currently evaluating a fingerprick screening test for adult TB based on host biomarkers discovered and validated in serum samples as part of an EDCTP2-funded consortium (www.screen-tb.eu).