Tuberculosis (TB) remains a leading cause of death, but antibiotic treatments for tuberculous meningitis, the deadliest form of TB, are based on those developed for pulmonary TB and not optimized for brain penetration. Here, we performed first-in-human dynamic 18F-pretomanid positron emission tomography (PET) studies in eight human subjects for three-dimensional, multi-compartmental in situ visualization of antibiotic concentration-time exposures (area under the curve - AUC), demonstrating preferential brain (AUCtissue/plasma 2.25) versus lung (AUCtissue/plasma 0.97) tissue partitioning. Preferential, antibiotic-specific partitioning into brain or lung tissues of antibiotics active against MDR strains were confirmed in experimentally-infected mice and rabbits, using dynamic PET with chemically identical antibiotic radioanalogs, and postmortem mass spectrometry measurements. PET-facilitated pharmacokinetic modeling predicted human dosing necessary to attain therapeutic brain exposures in human subjects. These data were used to design optimized, pretomanid-based regimens which were evaluated at human equipotent dosing in a mouse model of TB meningitis, demonstrating excellent bactericidal activity without an increase in intracerebral inflammation or brain injury. Importantly, several antibiotic regimens demonstrated discordant activities in brain and lung tissues in the same animal, correlating with the compartmentalized tissue exposures of the component antibiotics. These data provide a mechanistic basis for the compartmentalized activities of antibiotic regimens, with important implications for the development of antimicrobial regimens for meningitis and other infections in compartments with unique antibiotic penetration.