This prospective cohort study used convenience sampling. Eligible participants were adults with MDR/RR-TB who were initiating kanamycin-based treatment at the participating hospitals. Patients with middle-ear pathology were excluded from the study.

The sample-size estimate was based on detecting a clinically relevant 10 dB difference in pure-tone audiometric thresholds between baseline and follow-up assessments, using American Speech-Language-Hearing Association (ASHA) Significant Threshold Shift (STS) criteria and assuming a common standard deviation of 10 dB. Based on this calculation, a sample size of 60 participants would provide 99% power using a two-sided test at the 5% significance level. However, this calculation was based on changes in audiological threshold rather than on detecting associations between rare mitochondrial variants and cochleotoxicity. Because no prior frequency estimates were available for m.15312T>C or m.10114T>C in this South African MDR/RR-TB population, the study was not formally powered for genetic association testing. This limitation is acknowledged in the interpretation of the genetic findings.

At the time of the study, the standard MDR/RR-TB regimen consisted of pyrazinamide, moxifloxacin, kanamycin, terizidone, and either ethionamide or isoniazid. Ethambutol was included according to local guidelines when resistance was considered low. Participants received kanamycin daily, six times per week, at 15 mg/kg per dose according to South African Department of Health guidelines. Kanamycin dosage was adjusted for renal dysfunction at the discretion of the treating clinician. Renal function was assessed at 4 weeks, 8 weeks and 12 weeks after treatment initiation using the Cockcroft-Gault method to calculate creatinine clearance.

Audiological monitoring was performed at baseline and at approximately 4 weeks, 8 weeks and 12 weeks after treatment initiation. The assessment included case history, otoscopic examination, tympanometry and pure-tone audiometry, including ultra-high-frequency audiometry up to 16 kHz. Pure-tone thresholds were measured across conventional frequencies and ultra-high frequencies. For reporting the distribution of threshold shifts, frequencies were grouped into low conventional frequencies, 0.25 kHz – 2 kHz; high conventional frequencies, 3 kHz – 8 kHz; and ultra-high frequencies, 9 kHz – 16 kHz.

Cochleotoxicity was determined by comparing each follow-up audiogram with the participant’s baseline audiogram using ASHA STS criteria (ASHA, 1994). Cochleotoxicity was defined as a 20 dB decrease at any one test frequency, a 10 dB decrease at any two adjacent frequencies or loss of response at three consecutive frequencies where responses were previously obtained. Participants required a minimum of two valid audiograms to be included in the hearing analysis. Results were shared with the patient, resident audiologist and medical doctor for appropriate management or treatment modification.

Blood samples were collected for genetic analysis, immediately centrifuged, and the buffy coat was stored at −70 °C. Deoxyribonucleic acid (DNA) was extracted using the chemagic 360 instrument according to the manufacturer’s instructions. Deoxyribonucleic acid quality and quantity were assessed using the Nanodrop ND100 spectrophotometer (Nanodrop Technologies, Wilmington, DE, United States [US]) and DNA integrity was confirmed using a 1% agarose gel.

Direct cycle sequencing was performed using custom primers for MT-CYB gene (mitochondrially encoded cytochrome B) and MT-ND3 gene (mitochondrially encoded NADH: ubiquinone oxidoreductase core subunit 3) (Table 1). Primer specificity was evaluated using Primer-Basic Local Alignment Search Tool (BLAST) (Ye et al., 2012) and in silico polymerase chain reaction (PCR). Primer annealing was performed at 58.5 °C for MT-CYB and 54 °C for MT-ND3 for 30 s. Sequencing of PCR products was performed using the BigDye Terminator v3.1 Cycle Sequencing Kit (Thermo Fisher Scientific, Applied Biosystems, Waltham, MA, US) and an ABI 3130XL Genetic Analyzer (Applied Biosystems and Hitachi, Tokyo, Japan).

Data were captured in Microsoft Excel and double-checked for accuracy before analysis. Descriptive statistics were used to summarise demographic and clinical characteristics, hearing outcomes, and variant frequencies. The minor allele frequency was calculated for both target variants.

Fisher’s exact tests were used to explore associations between each target variant and cochleotoxicity because of low expected cell counts. Results were interpreted cautiously because the variants were rare and the number of participants without cochleotoxicity was small. No multivariable model was fitted because the number of variant carriers was insufficient to support adjusted analysis. Therefore, potential contributors to cochleotoxicity, including human immunodeficiency virus (HIV) status, renal function, previous MDR/RR-TB treatment, baseline hearing status and treatment exposure, could not be controlled for in an adjusted model.

Ethical clearance to conduct this study was obtained from the University of Cape Town Human Research Ethics Committee (No. HREC 065/2015 and HREC 595/2018). Written informed consent was obtained from each participant in their preferred language (English, Afrikaans, or isiXhosa). Consent for publication of anonymised data was also obtained.

Of the 102 participants with analysable hearing data, 84 participants (82.4%) demonstrated a STS on their final available audiogram according to the ASHA cochleotoxicity criteria (ASHA, 1994). Among these 84 participants, 61 participants (73%) demonstrated bilateral STS, and 17 participants (20%) demonstrated unilateral STS. A further six participants (7%) demonstrated unilateral STS, while the opposite ear had middle-ear pathology and was excluded from analysis.

The frequency range affected by STS changed over the 12-week monitoring period (Table 3). At 4 weeks, 58 participants (57%) demonstrated STS, with 43 participants (42%) showing STS in the ultra-high-frequency range. By 8 weeks, 72 participants (71%) demonstrated STS; by 12 weeks, 84 participants (82%) demonstrated STS. The number of participants classified as having STS limited to the ultra-high-frequency range decreased over time, while conventional frequency involvement increased from 15 participants (15%) at 4 weeks to 33 participants (33%) at 8 weeks and 48 participants (47%) at 12 weeks.

The degree of hearing impairment in the conventional frequency range, 0.25 kHz – 8 kHz, was classified according to the WHO grading system (WHO, 2008) using the most severely affected ear (Table 4). Among the 84 participants with STS on their last audiogram, 47 participants (56%) had no conventional-frequency hearing impairment, while 37 participants (44%) had slight to severe hearing impairment. All 37 participants with conventional-frequency impairment had sensorineural hearing loss.

Cochleotoxicity severity was further described using Common Terminology Criteria for Adverse Events version 5 (CTCAEv5), TUNE ototoxicity grading system, and University of Cape Town (UCT) criteria (National Cancer Institute, 2017; Ramma, 2016; Theunissen et al., 2014) (Table 5). Common Terminology Criteria for Adverse Events version 5 and TUNE were used as established ototoxicity grading systems, while the UCT criteria were used as a locally used clinical/research grading approach in this setting. The proportion of participants classified with grade greater than 0 varied across grading systems: 45 participants (53.5%) using CTCAEv5, 65 participants (77%) using TUNE and 84 participants (100%) using UCT criteria. These results are presented descriptively to show how cochleotoxicity classification varied according to the grading system applied.

The m.15312T>C variant in MT-CYB was detected in three of 78 successfully sequenced participants (Table 6; Figure 1). All three participants developed cochleotoxicity. The variant was not detected among participants without cochleotoxicity. Fisher’s exact test did not demonstrate a statistically significant association between m.15312T>C and cochleotoxicity, 3 of 66 versus 0 of 12, p = 1.000.

The m.10114T>C variant in MT-ND3 was detected in 4 of 80 successfully sequenced participants (Table 6; Figure 2). All four participants developed cochleotoxicity. The variant was not detected among participants without cochleotoxicity. Fisher’s exact test did not demonstrate a statistically significant association between m.10114T>C and cochleotoxicity, 4 of 65 versus 0 of 15, p = 1.000.

Non-target variants are summarised in Table 7. Two participants carried the m.15301G>A variant near the m.15312T>C site, and seven participants carried the m.10115T>C variant near the m.10114T>C site.

These results show that both target variants were observed only in participants who developed cochleotoxicity, but statistical evidence of association was not demonstrated in this cohort.

The high incidence of cochleotoxicity in this cohort is consistent with the established ototoxic risk of injectable aminoglycosides and may also reflect the sensitivity of ultra-high frequency audiometry (Jiang et al., 2017; Petersen & Rogers, 2015). The 82.4% incidence observed in this study is higher than the 57% high frequency hearing loss reported by Harris et al. (2012) among South African MDR-TB patients treated with aminoglycosides, and higher than the 48% sensorineural or mixed hearing loss reported by Ghafari et al. (2015) among children with tuberculosis receiving ototoxic medication. This difference may partly reflect this study’s use of ASHA STS criteria and ultra-high frequency audiometry up to 16 kHz, which are sensitive to early cochlear changes. The expanded hearing results showed that STS was detected earliest in the ultra-high frequency range and later extended into conventional frequencies in a substantial proportion of participants. This pattern supports the value of ultra-high frequency audiometry for early ototoxicity monitoring, as conventional frequency impairment may occur only after earlier ultra-high frequency changes (Konrad-Martin et al., 2005).

The hearing results also show the importance of distinguishing between STS and clinically graded conventional-frequency hearing impairment. Although 84 participants met ASHA criteria for STS, 47 did not yet meet criteria for conventional frequency hearing impairment (ASHA, 1994; WHO, 2008). This suggests that ASHA-based monitoring and ultra-high-frequency audiometry may identify early cochlear changes before they are reflected in conventional hearing-impairment categories.

Cochleotoxicity severity varied according to the grading system applied. Common Terminology Criteria for Adverse Events version 5 classified fewer participants as having grade greater than 0 compared with TUNE and UCT criteria (National Cancer Institute, 2017; Ramma, 2016; Theunissen et al., 2014). This variation indicates that cochleotoxicity prevalence and severity estimates are influenced by the sensitivity and classification structure of the grading scale used. In this cohort, the UCT criteria classified the highest proportion of participants as having cochleotoxicity; however, because these criteria are locally applied and are not as widely standardised internationally as CTCAEv5 or TUNE, UCT-based classifications should be interpreted cautiously and as contextual descriptive information.

The target variants m.15312T>C in MT-CYB and m.10114T>C in MT-ND3 were detected only among participants who developed cochleotoxicity. This observation is consistent with the possibility that these variants may represent candidate susceptibility markers, as previously suggested in South African work on aminoglycoside-associated hearing loss (Human, 2009). However, the findings should not be interpreted as confirmation of pathogenicity. The number of variant carriers was small, and Fisher’s exact tests did not demonstrate statistically significant associations for either variant. The results are therefore exploratory and hypothesis-generating and require validation in larger cohorts with broader genetic panels.

The absence of statistically significant associations is important for interpretation. Although the variants were observed only in participants with cochleotoxicity, the high overall incidence of cochleotoxicity and the small number of participants without cochleotoxicity reduced the ability to distinguish genetic susceptibility from the broader ototoxic effect of kanamycin exposure. Therefore, the findings should be viewed as preliminary evidence supporting further investigation rather than evidence for clinical implementation.

This study was conducted in the context of kanamycin-based MDR/RR-TB treatment in use at the time of recruitment, in accordance with South African guidance then in force (South African Department of Health, 2013). Current treatment recommendations have shifted towards shorter all-oral and injectable-sparing regimens, reducing the routine use of injectable aminoglycosides (South African National Department of Health, 2023; WHO, 2022). However, this does not eliminate the relevance of aminoglycoside-related ototoxicity research. Amikacin, a structurally related aminoglycoside with recognised cochleotoxic potential, may still be considered in selected MDR/RR-TB patients when susceptibility is demonstrated and adequate monitoring is available (Ramirez & Tolmasky, 2017; Ristuccia & Cunha, 1985; WHO, 2022). Therefore, evidence from historical kanamycin-exposed cohorts remains relevant for understanding aminoglycoside cochleotoxicity and potential susceptibility mechanisms.

The study also contributes to the limited evidence base on mitochondrial variation and aminoglycoside-induced hearing loss in African populations. Given the genetic diversity of South African populations and the limited availability of local genetic susceptibility data, exploratory studies such as this can help identify candidate variants for future validation (Choudhury et al., 2018; Krause, 2015).

This study has several strengths. Firstly, the prospective cohort design enabled monitoring of hearing thresholds from baseline through follow-up, rather than relying on retrospective auditory status data. Secondly, cochleotoxicity was determined by comparing each participant’s follow-up audiogram with their own baseline audiogram using ASHA STS criteria (ASHA, 1994). Thirdly, the inclusion of ultra-high frequency audiometry up to 16 kHz increased sensitivity to early cochlear changes before progression into the conventional frequency range (Konrad-Martin et al., 2005). Fourthly, the study contributes local evidence from a South African MDR/RR-TB cohort, addressing limited data on mitochondrial variation and aminoglycoside-induced cochleotoxicity in African populations.

This study also has limitations. Convenience sampling may have introduced selection bias and may limit generalisability. Attrition was high, with 45 of 147 recruited participants excluded because they did not complete the minimum number of valid hearing assessments. Participants who died, withdrew, were discharged early or were too ill to complete follow-up may have differed clinically from those retained in the analysis, introducing possible attrition bias.

The sample size calculation was based on detecting a clinically relevant 10 dB difference in pure-tone audiometric thresholds between baseline and follow-up assessments. Therefore, the study was designed to detect changes in audiological thresholds, but it was not powered for rare-variant genetic association analysis. Deoxyribonucleic acid was unavailable, or sequencing was unsuccessful for a proportion of participants, further reducing the sample size for genetic analysis. The low number of variant carriers limited statistical power and prevented adjusted modelling.

Known aminoglycoside-associated mitochondrial variants, including m.1555A>G and m.1494C>T, were not screened. This limits the interpretation of the independent contribution of m.15312T>C and m.10114T>C, because established mitochondrial variants have previously been associated with aminoglycoside-induced hearing loss (Gao et al., 2017; Lu et al., 2010). Potential confounders, including HIV status, renal function, previous MDR/RR-TB treatment, baseline hearing status, cumulative aminoglycoside exposure, and other ototoxic exposures, could not be fully controlled. If the UCT grading criteria are retained, UCT-based classifications should also be interpreted cautiously because these criteria are locally applied and are not as widely standardised internationally as CTCAEv5 or TUNE.

These limitations mean that the genetic findings should be interpreted cautiously as exploratory and hypothesis-generating.

Future studies should use larger, adequately powered cohorts and broader genetic panels that include established aminoglycoside associated variants, such as m.1555A>G and m.1494C>T, as well as candidate variants identified in African populations. Where possible, future analyses should adjust for relevant clinical risk factors, including HIV status, renal function, prior MDR/RR-TB treatment, baseline hearing status, cumulative aminoglycoside exposure, and concurrent ototoxic medications.

The present findings do not justify routine clinical screening for m.15312T>C and m.10114T>C alone. However, they support further investigation of genetic susceptibility to aminoglycoside-induced cochleotoxicity. Clinically, the high incidence of cochleotoxicity observed in this cohort reinforces the importance of structured ototoxicity monitoring, including ultra-high frequency audiometry where available, in patients exposed to aminoglycosides.