Salivary proteome investigation for Autoimmune Liver Diseases classification and biomarker discovery

Autoimmune Liver Diseases (AILDs) are pathologies afflicting the liver caused by an inappropriate immune response against self-antigens and bringing, if not properly managed, to hepatic damage, chronic inflammation, and systemic repercussions. Despite progress in understanding the etiopathogenesis, diagnostic and therapeutic approach of AILDs, there are still critical issues concerning early diagnosis, risk stratification of disease progression and identification of response to therapy predictors. Indeed, there are several confounding factors involved in the initiation of hepatic autoimmune and inflammatory phenomena and also similarities and overlap syndrome among the three main types of autoimmune liver diseases: Autoimmune Hepatitis (AIH), Primary Biliary Cholangitis (PBC), and Primary Sclerosing Cholangitis (PSC). Saliva, as a mirror of oral and systemic health, represents a promising biofluid for biomarker discovery. Several studies evidenced that various systemic disorders affected qualitatively and quantitatively the salivary proteome. The present thesis reports the analysis of the salivary proteome from patients affected by AIH and PBC in comparison with healthy controls (HCs) using an integrated top-down/bottom-up proteomic pipeline with the purpose of identifying qualitative and quantitative variations in the proteomic profile useful for diagnostic, prognostic purposes, and for potential biomarker discovery. The first part of the thesis reports the innovative text-mining approach used to retrieve the existing literature on proteomic data of AIH and PBC and create a co-occurrence network of terms associated with these two pathologies. The aim was to provide an overall understanding of the past and current scenario of publications related to AIH and PBC proteomic studies and to highlight the molecular functions and biological pathways of the proteins identified. The experimental workflow exploited in Part II and III, combining acidification and centrifugation of whole salivary samples provided two fractions: an acidic-soluble fraction submitted to a top-down analysis and an acidic-insoluble fraction analyzed by a gel-based bottom-up approach. In both cases MS data were subjected to statistical analysis by exact Mann-Whitney and Kruskal-Wallis tests which evidenced quantitative variations among groups. Random Forest (RF), one of the most widely used supervised machine learning algorithms for MS data, multidimensional scaling (MDS) and linear discriminant analysis (LDA) identified a panel of salivary proteins, useful to accurately classify the subjects based on AIH or PBC occurrence.