Experimental versus clinical studies capture distinct types of relapse in tuberculosis.

Tuberculosis (TB) remains a global health concern, as Mycobacterium tuberculosis (Mtb) currently infects a quarter of the world's population. Though many TB patients cure infection with short-course treatment, regimens shorter than 4 months risk post-treatment relapse. This is facilitated by the presence of granulomas, spatially heterogeneous immune structures that form during infection that harbor a caseated core that sequesters Mtb in a non-replicating state and pharmacokinetically impedes antibiotics. Typically, clinical relapse is described as TB recurrence <2 years after a misdiagnosis of cure upon treatment completion (MDxC). Comparatively, many experimental models cannot (or do not) screen for cure and examine relapse ~2 months after treatment completion. Understanding how relapse occurs, and whether it is impacted by experimental versus clinical study designs, may improve treatment. Here, we examine two hypothesized mechanisms of relapse: (i) reservoir, where treatment kills all Mtb except those in a non-replicating, diagnostically inaccessible host reservoir; and (ii) threshold, where replicating Mtb remain alive below detectable levels within a test-accessible reservoir. We use our computational model capturing whole-host Mtb infection dynamics, HostSim. Our simulations uncover rates of reported relapse correlating with how studies are performed. Experimentally, relapse is likely driven by incomplete sterilization of replicating bacteria, whereas clinical relapse is most likely driven by gradual expansion of non-replicating Mtb from caseum into granuloma cellular areas. This suggests that TB patients who relapse may best be re-treated with regimens including caseum-penetrating antibiotics, such as rifamycin-class antibiotics, and that experimental animal relapse models better translate to clinical when "cure" is established at treatment completion.IMPORTANCEIncomplete treatment of TB leads to risks relapse, which may occur years later. The threat of relapse is the main reason TB treatment takes 4-9 months. Predictors of relapse are not well-defined given variability in technical definitions of relapse and nuances of study designs. Here, we use biologically based computation to simulate both clinical and experimental TB relapse studies, employing multiple diagnostic tests and relapse definitions. We find that two hypothesized types of relapse, each possibly requiring different re-treatment strategies, are simultaneously at play. Simulations suggest that relapse after a "cure" diagnosis (most often clinical studies) is caused by reactivation of non-replicating bacteria hidden from treatment within necrotic granuloma tissue or other sites, whereas experimental systems that cannot or do not test for cure post-treatment are likely to report relapse caused by incomplete sterilization of replicating Mtb. Predictions also depend on the currently unclear relationship between bacterial burden and clinical symptoms.