Discovery and validation of plasma-based protein biomarkers for the aetiological distinction of bacterial and non-bacterial febrile infections in African children

BackgroundIn many low resource settings, the clinical management of children with febrile infections is hindered by poor access to diagnostic tools to determine whether the cause of an infection is bacterial, viral or parasitic. As a result, many clinicians resort to the default prescription of antibiotics as a safety precaution, contributing to the alarming spread of antimicrobial resistance. Commonly used biomarkers for identification of bacterial sepsis such as CRP lack aetiological specificity and are frequently elevated by non-bacterial infections including malaria. We set out to discover and validate new biomarkers for the characterization of the microbial aetiology of febrile acute infections in Kenyan children. MethodsWe recruited a discovery cohort comprising of children who had been admitted to hospital with a variety of severe acute infections. Diagnostic identification of viral infections was done using a 15-target virus PCR panel, bacterial infections were identified using blood culture while malaria infections were identified by microscopy. Using mass spectrometry analysis, we identified a set of 76 plasma proteins whose abundance varied significantly by the microbial aetiology of infection and used machine learning to generate a shortlist of candidate biomarkers that had the highest diagnostic performance in distinguishing aetiologies. To validate these candidate biomarkers, we recruited a separate validation cohort where the plasma levels of the shortlisted biomarkers were assayed among children with different infectious aetiologies using a custom protein microarray. ResultsIn the discovery study, six candidate biomarkers whose plasma abundance was significantly different in children with bacterial and viral infections were shortlisted by random forest for cross-cohort validation (AGT, HRG, LBP, PON1, SERPINA1, SERPINA3). In the validation study, we found that of the six biomarkers, only AGT compared favourably to CRP and identified febrile bacterial infections with a sensitivity of 72.4% (95% CI 48.4% - 83.6%) compared to CRP which distinguished febrile bacterial infections with a sensitivity of 69.5% (30.8% - 88.2%). Plasma AGT was superior to CRP in distinguishing children with febrile bacterial infections from those with febrile malaria episodes, with a sensitivity of 72.5% (40% - 84.6%) for AGT and 26% (15% - 32.8%) for CRP. ConclusionsWe report the discovery of AGT, as a sensitive plasma biomarker for the identification of febrile bacterial infections among African children living in a malaria endemic setting.