structure but included the maximum likelihood-inferred ancestral nucleotide positions from a virtual ancestor 11 , by using the Burrows-Wheeler Aligner. SNP calls were made with SAMtools and VarScan (coverage of at least 20×). We kept only the homozygous calls (present in at least 90% of the reads in a specific position) and filtered out potential calling errors by omitting variants detected in repetitive regions, phages, PE/PPE regions, and SNPs close to indels (10 bp window) or in areas with an anomalous accumulation of variants (three or more SNPs in 10 bp). Alignments and SNP variants were visualized and checked with the IGV program. Multiple comparisons between the SNPs from the clustered isolates were made using an in-house script written in R software (R Foundation for Statistical Computing, Vienna, Austria 2011, www.R-project.org). The median-joining networks were constructed from the SNP matrix generated using NETWORK 5.0.0.1. Median vectors (mv) were defined when the distribution of SNPs indicated the existence of a non-sequenced node corresponding to non-sampled genotype in the cluster. The chronology of acquisition of the SNPs is represented from left to right in the networks.