According to a 2015 WHO report, tuberculosis remains one of the top 10 causes of death worldwide. Despite considerable efforts by the United Nations to eradicate the disease by 2030, a global TB epidemic still persists. Its causative agent, the bacterium Mycobacterium tuberculosis, an obligate pathogen, has been plaguing humanity since it originated, and has coevolved with its main host, Homo sapiens, over thousands of years. Contemporary tuberculosis strains exhibit a structured phylogeographic pattern, carrying the genetic print of their geographic origin. The Koch bacillus infects and kills in large numbers, in poor and developing countries, where fragile health care systems, combined with high HIV prevalence, facilitate epidemic spread. In western countries, the major current threats are the multiplication and propagation of antibiotic resistant strains (MDR/XDR) coming predominantly from former Soviet republics. In this thesis, I unravel the evolutionary history, propagation, and acquisition of drug resistance-conferring mutations in different settings, by implementing multiple genetic and genomic data sets. First, focusing on Central Asia, using whole genome sequencing and Bayesian statistics, I assess the effects of a treatment campaign on the development of MDR strains and highlight key mutations in successful strains. More importantly, the success of DOTs campaigns was compromised by the genetic make-up of these outbreak clades (pre-treatment low frequency resistance SNPs). Special attention was also given to a particular outbreak of MDR strains, i.e. the Russian W148 clone. I present its westward spatial and temporal propagation at a continental scale during the last century, and underline the key contribution of compensatory mutations in its epidemic success. However, tuberculosis does not only infect humans, but also has experienced successive mammalian host jumps. To decipher the adaptive constraints accompanying such secondary events, a systemic gene screen with selection signature-detecting algorithms was implemented to identify putative targets during diversifying selection. Finally, novel mathematical tools and indices that reflect the epidemicity of a strain were developed, jumping from a population-driven approach to a strain specific one, with broader epidemiological applications. This allows us to correlate strain fitness with patient, lineage, and socio-economic information.