Prediction of early Toxoplasma gondii infection during pregnancy combining host biomarkers and machine learning

Authors:
Preprint
Publisher website
Google Scholar
Abstract
Toxoplasma gondii is a major cause of reproductive failure and severe gestational complications in both humans and small ruminants and early diagnosis is essential to prevent adverse outcomes. In sheep, congenital toxoplasmosis can trigger early fetal death (EFD) during the acute phase of the infection (days 7-14 post-infection) when conventional diagnosis methods are ineffective. Therefore, this study aimed to identify early infection biomarkers using a machine learning approach to support EFD diagnosis and providing a translational foundation for early diagnostic strategies in human toxoplasmosis. For that, 27 serum biomarkers (cytokines, acute-phase proteins, oxidative stress and hemostatic parameters) were evaluated during the early stage of infection in 22 pregnant sheep experimentally challenged with T. gondii and 6 uninfected controls. To assess diagnostic specificity, models were tested using sera from animals experimentally infected with other abortifacient agents ( Neospora caninum and Chlamydia abortus ) and validated with samples from 19 natural abortion outbreaks. A four-biomarkers panel, encompassing CXCL9, CXCL10, IFN-γ, and serum amyloid A, achieved over 74% accuracy in the specific detection of early T. gondii infection and 100% of sensitivity when studying natural infection cases. This approach might support the development of rapid diagnostic strategies, improving early intervention and disease management in livestock production systems. Its translation to human medicine could also aid in addressing current diagnostic gaps and facilitate timely interventions to prevent congenital toxoplasmosis and pregnancy loss in women. Importance: This study addresses a critical veterinary and public health challenge: the diagnostic gap in early infection during congenital toxoplasmosis. In sheep, early fetal death (EFD) is a clinical outcome than remain underdiagnosed under field conditions due to the absence of maternal IgG antibodies and placental or fetal parasite DNA at the time of abortion. By pioneering the integration of host-response biomarkers with machine learning, this work successfully predicted acute infection with 100% diagnostic sensitivity and 85% accuracy under real-world field conditions (natural infections). Furthermore, this study provides clues for unraveling the pathogenic mechanism involved by demonstrating that EFD is driven mainly by oxidative processes rather than by the initially proposed increase in pro-inflammatory cytokines. In addition, the proposed approach may have a translational potential, both in terms of the selected biomarkers and the machine learning-based methodology, with the purpose of enabling early therapeutic interventions in human congenital toxoplasmosis.

This publication has 80 references indexed in Scilit: