Background Haematopoietic stem cells expressing the CD34 surface marker have been posited as a niche for Mycobacterium tuberculosis complex bacilli during latent tuberculosis infection. Our aim was to determine whether M tuberculosis complex DNA is detectable in CD34-positive peripheral blood mononuclear cells (PBMCs) isolated from asymptomatic adults living in a setting with a high tuberculosis burden. Methods We did a cross-sectional study in Ethiopia between Nov 22, 2017, and Jan 10, 2019. Digital PCR (dPCR) was used to determine whether M tuberculosis complex DNA was detectable in PBMCs isolated from 100 mL blood taken from asymptomatic adults with HIV infection or a history of recent household or occupational exposure to an index case of human or bovine tuberculosis. Participants were recruited from HIV clinics, tuberculosis clinics, and cattle farms in and around Addis Ababa. A nested prospective study was done in a subset of HIV-infected individuals to evaluate whether administration of isoniazid preventive therapy was effective in clearing M tuberculosis complex DNA from PBMCs. Follow-up was done between July 20, 2018, and Feb 13, 2019. QuantiFERON-TB Gold assays were also done on all baseline and follow-up samples. Findings Valid dPCR data (ie, droplet counts >10 000 per well) were available for paired CD34-positive and CD34-negative PBMC fractions from 197 (70%) of 284 participants who contributed data to cross-sectional analyses. M tuberculosis complex DNA was detected in PBMCs of 156 of 197 participants with valid dPCR data (79%, 95% CI 74–85). It was more commonly present in CD34-positive than in CD34-negative fractions (154 [73%] of 197 vs 46 [23%] of 197; p<0·0001). Prevalence of dPCR-detected M tuberculosis complex DNA did not differ between QuantiFERON-negative and QuantiFERON-positive participants (77 [78%] of 99 vs 79 [81%] of 98; p=0·73), but it was higher in HIV-infected than in HIV-uninfected participants (67 [89%] of 75 vs 89 [73%] of 122, p=0·0065). By contrast, the proportion of QuantiFERON-positive participants was lower in HIV-infected than in HIV-uninfected participants (25 [33%] of 75 vs 73 [60%] of 122; p<0·0001). Administration of isoniazid preventive therapy reduced the prevalence of dPCR-detected M tuberculosis complex DNA from 41 (95%) of 43 HIV-infected individuals at baseline to 23 (53%) of 43 after treatment (p<0·0001), but it did not affect the prevalence of QuantiFERON positivity (17 [40%] of 43 at baseline vs 13 [30%] of 43 after treatment; p=0·13). Interpretation We report a novel molecular microbiological biomarker of latent tuberculosis infection with properties that are distinct from those of a commercial interferon-γ release assay. Our findings implicate the bone marrow as a niche for M tuberculosis in latently infected individuals. Detection of M tuberculosis complex DNA in PBMCs has potential applications in the diagnosis of latent tuberculosis infection, in monitoring response to preventive therapy, and as an outcome measure in clinical trials of interventions to prevent or treat latent tuberculosis infection.