Mixed-strain infection and heteroresistance in drug-resistant tuberculosis: insights from a whole-genome sequencing pilot.

BACKGROUND: Mixed-strain Mycobacterium tuberculosis (M. tuberculosis; Mtb) infections and heteroresistance are increasingly recognized as under-detected contributors to drug resistance dynamics. In high-burden rural settings such as the Eastern Cape, these genomic features may reflect both within-host diversity and ongoing transmission. However, their epidemiological and clinical implications remain incompletely understood, particularly in contexts with high HIV co-infection. METHODS: We conducted an exploratory whole-genome sequencing (WGS) analysis of 28 drug-resistant M. tuberculosis isolates obtained through routine diagnostic workflows. Mixed-strain candidacy was defined using a composite genomic criterion, including multi-lineage assignments, ≥2 heteroallelic variants [allele frequency (AF): 0.10-0.90], or heteroresistance across multiple drug classes. A Random Forest (RF) model with Leave-One-Out Cross-Validation (LOOCV) was used as a feature-prioritization tool to identify genomic characteristics associated with mixed-strain candidacy. Given the limited sample size, analyses were designed to generate hypotheses rather than support causal inference. Clinical and epidemiological metadata, including HIV status and treatment outcomes, were not available. RESULTS: Lineage 4 (46%) and Lineage 2 (43%) predominated, with 10.7% of isolates classified as probable mixed-strain candidates. Heteroallelic variants were most frequently observed in fabG1/inhA and rpoB, with a median AF of 0.22, consistent with subclonal diversity and possible within-host microevolution. Additional heteroallelic variants were identified in non-resistance-associated genes, suggesting broader genomic heterogeneity. The RF model demonstrated high discriminatory performance (AUC = 1.00), although this likely reflects the small sample size and should be interpreted cautiously. Feature importance analysis identified heteroresistance burden as the most prominent predictor of mixed-strain candidacy. CONCLUSION: This pilot study provides preliminary genomic evidence of mixed-strain infection and heteroresistance in a high-burden rural setting. While these findings highlight the potential role of subclonal diversity in shaping resistance patterns, their clinical and epidemiological relevance remains uncertain due to the absence of patient-level data and limited sample size. Future studies integrating genomic, clinical, and epidemiological data in larger cohorts and incorporating benchmarking against established tools are required to validate these findings and clarify their implications for tuberculosis control.