Summary

We present the first Mycobacterium tuberculosis (MTB) complete genome dataset designed to resolve genetic diversity across three evolutionary scales: population, epidemiological, and within-host. We generated 216 high-quality complete genomes from long-read sequencing to recover the 5–10% of the genome excluded by short-read mapping. At the population level, we uncover diversity hotspots, yielding a 1.44-fold higher evolutionary rate than estimated from short reads. They are particularly concentrated in pe/ppe genes and largely driven by gene conversion, a process we show can also affect T-cell epitopes of vaccine candidates. At the epidemiological scale, complete genomes refine transmission network resolution by incorporating previously masked SNPs, indels, and structural variants. Finally, using patient- and cluster-specific reference genomes for mapping, we show that most non-fixed variants detected by conventional pipelines using a common reference are false positives, prompting a reassessment of within-host MTB diversity. This integrated multi-scale approach establishes a more accurate framework for understanding MTB evolution, transmission, and host–pathogen interactions.