| dc.description.abstract | Glycans introduce complexity to the proteins to which they are attached. These modifications vary during the progression of many diseases; thus, they serve as potential biomarkers for disease diagnosis and prognosis. The immense structural diversity of glycans makes glycosylation analysis and quantitation difficult. Therefore, a better understanding of various glycosylation profiling strategies; their strengths and weaknesses, is important towards selecting the best approach for a given clinical glycomics study. Not only that, successful application of glycomics analysis methods in the clinical glycomics field depends on using effective sample preparation strategies and better classification systems to accurately classify glycomics samples. Among many analytical methods in the glycoproteomics analysis field, LC-MS analysis of glycopeptides is a frequent choice, as it provides information of both the glycans and their attachment sites. Numerous software tools have been developed to assist the glycopeptide identification workflow; however, those tools typically do a sub-optimal job when the glycopeptides of interest are in low abundance or when they are poorly ionized. Therefore, in such incidences, expert targeted analysis approaches, where LC-MS data is manually interpreted to confidently identify the recalcitrant glycopeptides would be beneficial. Thus, chapter 2 of this dissertation introduces a simple, expert analysis method, the peak alignment approach, which relies on high-resolution MS data and chromatographic retention times to assign the glycosylation sites. The method identifies a set of co-eluting glycopeptides in an LC-MS experiment using a reverse phase column; these glycopeptides are extracted based on a limited N-linked glycan library, and once the co-eluting glycopeptides are identified, they are verified by using high-resolution MS data and confirmed by using MS/MS data. The developed method successfully quantified many of the glycosylation sites of a heavily glycosylated human plasma glycoprotein within a single LC-MS run while requiring less sample amount and less analysis time, compared to the state-of-the-art competing analysis method. Sample preparation is a vital step in all glycomics analysis studies, as it affects both the sensitivity and the selectivity of the analysis. Altered glycosylation of specific proteins can serve as a biomarker for diverse diseases. Uromodulin is one such glycoprotein; it is a biomarker for kidney health. Current strategies of uromodulin glycosylation analysis are time-consuming and tedious; they involve complex steps to enrich uromodulin, label glycans, followed by post sample clean-up, which limit the utility of these methods in clinical glycomics studies. Therefore, chapter 3 of this dissertation introduces a simple and straightforward direct ESI-MS analysis performed in the negative ion mode to quantify N-linked glycans of uromodulin, enriched from urine samples of two different biological states. The developed method enriches uromodulin directly from urine via ultrafiltration performed with a 50 kD molecular weight cut-off filter; it omits any labeling steps that require post-sample clean-up and includes steps to reduce the salt contents of the samples to minimize the ion suppression during the direct ESI-MS detection. The method proved to be highly reproducible over multiple samples’ preparations and over multiple analyses; it was useful for accurately quantifying uromodulin glycans and classifying the samples of different biological states into clearly distinguishable groups by PCA. Sample classification based on the whole glycomic profile, instead of selecting a single glycan feature or a few glycan features, could benefit the sample classification through identifying underlying trends of the glycomics data. The Aristotle Classifier is one such supervised classification algorithm that uses not only all the individual glycans abundances, but also their relative proportions to each other, to classify samples. Once this classifier was built, its’ classification ability needed to be challenged and compared with standard classification methods, like PCA. However, acquiring large sets of real glycomics samples with known glycosylation differences is difficult; thus, we chemically generated large sets of IgG glycomics data in-house, to mimic two different biological states as healthy and disease. Therefore, chapter 4 of this dissertation describes the optimization of both the sample preparation and LC-MS conditions to generate large sets of IgG glycopeptides’ data to mimic samples of a healthy state and a disease state. Of these samples, the healthy state was represented by samples with a native IgG glycosylation profile while the disease state was represented by samples with a slightly altered IgG glycosylation profile. The generated data were quantified, but the samples could not be classified into healthy versus disease based on any individual glycopeptide of the samples. Therefore, the data proved to be challenging; thus, they were submitted to both the Aristotle Classifier and to a principal components analysis (PCA), to challenge each approach’s classification ability. The generated results showed that the Aristotle Classifier outperformed the PCA classification in multiple data sets. | |
