Degree: Master of Science Abstract: Tuberculosis (TB) remains a global health problem. Incorrect or missed serodiagnosis for TB, especially in resource-limited settings, is one of the main reasons for TB epidemics. Clinically, serological tests that use circulating antibodies are attractive because antibodies in serum are present at higher concentration than disease biomarkers, making detection of disease-specific antibodies easier. However, the accuracy of a serological test is highly dependent on the antigens used in the test; existing serological tests for TB provide inconsistent specificities and sensitivities. This issue is further aggravated by the heterogeneity and complexity of the glycolipid antigen across mycobacteria species, which makes rational design of discriminatory antigens difficult. This thesis approaches these challenges, i.e. the need for simpler and more efficient antigens with high specificity for TB diagnoses, by combining fragment-based discovery with phage display. Through anchoring each genetically-encoded displayed peptide on phage to a small molecule fragment, I constructed glycopeptide libraries that demonstrated their application for the selection of promising antigens for antibodies associated with mycobacterial infections. By using an anti-LAM antibody CS-35 as a model target, I discovered novel hexasaccharide-peptide conjugates with KD value 10-fold better than that of the hexasaccahride itself (Chapter 2). The selectivity and affinity of these conjugates enhanced modestly when constructed in multivalent presentation and tested in multivalent binding assays. Interestingly, when the hexasaccharide on these conjugates were truncated to monosaccharides, their binding affinity drastically diminished, suggesting that some, if not all, monosaccahrides in the hexasaccharide must be present to elicit constructive antigen-antibody interactions (Chapter 3). Consequently, I demonstrated a site-selective dual modification strategy to create the first phage displayed macrocyclic glycopeptide library that present two carbohydrate moieties per peptide. The peptide library can be glycosylated first using sodium periodate cleavage on N-terminal serine followed by oxime ligation; the second glycosylation was achieved through alkylation on disulfides using a dichloro-oxime derivative. These libraries show potential for use in fragment-based ligand discovery, and selection for TB-specific antibodies. Finally, I developed a deep-panning system that combines single-round selection with deep sequencing to afford rapid and unbiased selection. The system has been validated on model target ConA and antiFLAG antibody, and will strongly complement the fragment-based ligand discovery to identify disease-specific antigens using serum from diseased patients.