A team from Western University, London, Ontario used the Olink® Explore 3072 platform for high-throughput protein biomarker discovery to compare plasma protein profiles in patients suffering from Long-COVID with those from acute COVID-19 patients and healthy controls. Up to 30% of survivors of acute COVID-19 suffer prolonged, diffuse symptoms post-infection. These symptoms include fatigue, dyspnea, neurological symptoms, chest pain and gastrointestinal upset and are collectively referred to as “Long-COVID”. A lack of Long-COVID biomarkers and pathophysiological mechanisms currently limits effective diagnosis, treatment and disease surveillance.
In total, 119 proteins were found to be useful in classifying Long-COVID outpatients compared to acute COVID inpatients & healthy controls (109 upregulated, 10 downregulated). Even individually, these 119 proteins had excellent individual classification abilities, with AUCs of 0.91 or better, and when combined, could separate Long-COVID from the other groups with a classification accuracy of 100% (AUC=1.00). In order to provide more specific research targets for future studies, the researchers then applied machine learning methods (recursive feature elimination) to look for high-performance protein models composed of fewer biomarkers. Using different threshold levels for inclusion in the models, they identified models containing 9 proteins (CXCL5, AP3S2, MAX, PDLIM7, EDAR, LTA4H, CRACR2A, CXCL3, FRZB) and 5 proteins (CXCL5, AP3S2, MAX, PDLIM7, FRZB) that retained an AUC=1.00 for Long-COVID diagnosis.
AI-based Natural Language Processing (NLP) of the UniProt database was also used to examine the organ and cell-type protein expression characteristics of the 119 proteins associated with Long-COVID. This suggested the involvement of diffuse organ systems in Long-COVID (including digestive lymphatic and nervous systems), as well as highlighting specific cell types, such as leukocytes and platelets. The authors concluded that both the optimal protein models and individual proteins identified in this study hold the potential for accurate diagnosis of Long-COVID, as well as for future targeted therapeutics.