BACKGROUND: Bedaquiline and clofazimine are important drugs in the treatment of drug-resistant tuberculosis and are commonly used across southern Africa, although drug susceptibility testing is not routinely performed. In this study, we did a genotypic and phenotypic analysis of drug-resistant Mycobacterium tuberculosis isolates from cohort studies in hospitals in KwaZulu-Natal, South Africa, to identify resistance-associated variants (RAVs) and assess the extent of clofazimine and bedaquiline cross-resistance. We also used a comprehensive dataset of whole-genome sequences to investigate the phylogenetic and geographical distribution of bedaquiline and clofazimine RAVs in southern Africa. METHODS: In this study, we included M tuberculosis isolates reported from the PRAXIS study of patients with drug-resistant tuberculosis treated with bedaquiline (King Dinuzulu Hospital, Durban) and three other cohort studies of drug-resistant tuberculosis in other KwaZulu-Natal hospitals, and sequential isolates from six persistently culture-positive patients with extensively drug-resistant tuberculosis at the KwaZulu-Natal provincial referral laboratory. Samples were collected between 2013 and 2019. Microbiological cultures were done as part of all parent studies. We sequenced whole genomes of included isolates and measured bedaquiline and clofazimine minimum inhibitory concentrations (MICs) for isolates identified as carrying any Rv0678 variant or previously published atpE, pepQ, and Rv1979c RAVs, which were the subject of the phenotypic study. We combined all whole-genome sequences of M tuberculosis obtained in this study with publicly available sequence data from other tuberculosis studies in southern Africa (defined as the countries of the Southern African Development Community), including isolates with Rv0678 variants identified by screening public genomic databases. We used this extended dataset to reconstruct phylogenetic relationships across lineage 2 and 4 M tuberculosis isolates. FINDINGS: We sequenced the whole genome of 648 isolates from 385 patients with drug-resistant tuberculosis recruited into cohort studies in KwaZulu-Natal, and 28 isolates from six patients from the KwaZulu-Natal referral laboratory. We identified 30 isolates with Rv0678 RAVs from 16 (4%) of 391 patients. We did not identify any atpE, pepQ, or Rv1979c RAVs. MICs were measured for 21 isolates with Rv0678 RAVs. MICs were above the critical concentration for bedaquiline resistance in nine (43%) of 21 isolates, in the intermediate category in nine (43%) isolates, and within the wild-type range in three (14%) isolates. Clofazimine MICs in genetically wild-type isolates ranged from 0·12-0·5 μg/mL, and in isolates with RAVs from 0·25-4·0 μg/mL. Phylogenetic analysis of the extended dataset including M tuberculosis isolates from southern Africa resolved multiple emergences of Rv0678 variants in lineages 2 and 4, documented two likely nosocomial transmission events, and identified the spread of a possibly bedaquiline and clofazimine cross-resistant clone in eSwatini. We also identified four patients with pepQ frameshift mutations that may confer resistance. INTERPRETATION: Bedaquiline and clofazimine cross-resistance in southern Africa is emerging repeatedly, with evidence of onward transmission largely due to Rv0678 mutations in M tuberculosis. Roll-out of bedaquiline and clofazimine treatment in the setting of limited drug susceptibility testing could allow further spread of resistance. Designing strong regimens would help reduce the emergence of resistance. Drug susceptibility testing is required to identify where resistance does emerge. FUNDING: Wellcome Trust, National Institute of Allergy and Infectious Diseases and National Center for Advancing Translational Sciences of the National Institutes of Health.