Rapid host species identification from dried bloodspots of Culex mosquito blood meals using mid-infrared spectroscopy

Abstract Background Understanding mosquito blood-feeding preferences can help infer vector-borne pathogen transmission dynamics and improve intervention strategies for important human and animal pathogens. The gold-standard tests for identifying mosquito blood meals are expensive, time-consuming methods and decrease in accuracy as blood meals are digested. Mid-infrared spectroscopy with machine-learning (MIRS-ML) can offer a rapid, cost-effective alternative approach to identifying host blood meals but has never been used to analyse blood meals from Culex spp., an increasingly important genus of vectors of several high-priority multi-host pathogens such as orthoflaviviruses and filariasis. Methods In this study, laboratory-reared Culex quinquefasciatus were fed blood from six hosts (human, pig, chicken, cattle, dog and horse). Dried bloodspots were created 6, 12, 24 and 30 h after digestion in the mosquito by rolling abdominal contents onto filter paper and then scanned using a Bruker ALPHA Fourier-transform infrared spectrometer. The spectra generated were used to train a multilayer perceptron algorithm to predict blood meals from these six host species and then tested on field-collected mosquito blood meals. Results For laboratory-reared mosquitoes, the model was 93.6% accurate in predicting host species in comparison to 16.7% random classification, with a small decline in accuracy after 30 h of digestion (6%). While the accuracy for field-collected bloodspots was lower (67.9%), this likely reflects differences in blood source from wild mosquitoes, such as no anticoagulants and direct feeding from hosts rather than artificial feeders. Model performance was further enhanced by applying a probability threshold, which increased accuracy to 78% in field samples while removing 21% samples. Conclusions This study demonstrates the potential application of MIRS-ML with dried bloodspots as a practical, scalable tool for blood meal analysis with scope for improving accuracy through further model development. This method may be of particular use for large-scale surveillance programmes in low-resource settings due to low running and labour costs. Further development of this method using samples from field-collected mosquitoes could enhance classification accuracy and expand the range of identifiable host species.