Africa-wide evaluation of host biomarkers in QuantiFERON supernatants for the diagnosis of pulmonary tuberculosis

Abstract

We investigated host-derived biomarkers that were previously identified in QuantiFERON supernatants, in a large pan-African study. We recruited individuals presenting with symptoms of pulmonary TB at seven peripheral healthcare facilities in six African countries, prior to assessment for TB disease. We then evaluated the concentrations of 12 biomarkers in stored QuantiFERON supernatants using the Luminex platform. Based on laboratory, clinical and radiological findings and a pre-established algorithm, participants were classified as TB disease or other respiratory diseases(ORD). Of the 514 individuals included in the study, 179(34.8%) had TB disease, 274(51.5%) had ORD and 61(11.5%) had an uncertain diagnosis. A biosignature comprising unstimulated IFN-γ, MIP-1β, TGF-α and antigen-specific levels of TGF-α and VEGF, identified on a training sample set (n = 311), validated by diagnosing TB disease in the test set (n = 134) with an AUC of 0.81(95% CI, 0.76-0.86), corresponding to a sensitivity of 64.2%(95% CI, 49.7-76.5%) and specificity of 82.7%(95% CI, 72.4-89.9%). Host biomarkers detected in QuantiFERON supernatants can contribute to the diagnosis of active TB disease amongst people presenting with symptoms requiring investigation for TB disease, regardless of HIV status or ethnicity in Africa.

Figure 1

STARD flow diagram showing the study design and classification of study participants. CRF, case report form; TB, Pulmonary tuberculosis; ORD, Individuals presenting with symptoms and investigated for pulmonary TB but in whom TB disease was ruled out; ROC, Receiver operator characteristics; GDA, General discriminant analysis.

Figure 2

Accuracy of multi-marker models in the diagnosis of TB disease. Receiver operator characteristics (ROC) curve showing the accuracy of the most accurate four-marker biosignature (IFN-γ_nil + TGF-α_nil + IL-1ra_Ag-nil + MIP-1β_Ag-nil) in the diagnosis of TB disease regardless of HIV infection status when all host markers evaluated were considered (251 study participants) (A), frequency of analytes in the top 20 general discriminant analysis (GDA) models that most accurately classified study participants as TB disease or ORD irrespective of HIV status when all host markers evaluated were considered (B), ROC curve showing the accuracy of the most accurate five-marker biosignature (IFN-γ_nil + MIP-1β_nil + TGF-α_nil + TGF-α_Ag-nil + VEGF_Ag-nil) in the diagnosis of TB disease regardless of HIV status when analysis was done only on the host markers that were evaluated on all study participants (i.e., excluding IL-1ra, IFN-α2 and TNF-α) (C), and frequency of analytes in the top 20 GDA models that most accurately diagnosed TB disease regardless of HIV status when analysis was done only on the host markers that were evaluated on all study participants (D). The bar graphs (B,D) indicate the frequency of analytes in the most accurate GDA models.

Figure 3

Accuracy of multi-marker models in the diagnosis of TB disease in the absence of HIV infection. Frequency of analytes in the top 20 general discriminant analysis (GDA) models that most accurately classified study participants regardless of HIV infection status, when all host markers evaluated were considered (limited numbers of study participants) (A), frequency of analytes in the top 20 GDA models that most accurately classified study participants as TB disease or ORD irrespective of HIV status when all study participants (limited numbers of host markers) were considered (B), ROC curve showing the accuracy of the most accurate seven-marker biosignature (IFN-γ_N, TGF-α_N, IL-1α_N, MMP-2_N, EGF_Ag-N, VEGF_Ag-N, TGF-α _Ag-N) in the diagnosis of TB disease regardless of HIV status when analysis was done only on the host markers that were evaluated on all study participants (i.e., excluding IL-1ra, IFN-α2 and TNF-α) (C). The bar graphs (A,B) indicate the frequency of analytes in the top 20 most accurate GDA models.