Optimisation and validation of capture mNGS for predicting antimicrobial resistance

BACKGROUND: Antibiotic resistance critically compromises bacterial infection treatment. While antimicrobial susceptibility testing (AST) remains the standard for resistance assessment, its culture dependence is time-consuming. Clinical metagenomic next-generation sequencing (mNGS) offers rapid pathogen detection and antibiotic resistance gene (ARG) profiling. However, low ARG detection sensitivity and unclear genotype-phenotype correlations limit its clinical utility. METHODS: We developed capture mNGS approach with probe-based ARG enrichment and a host-attribution algorithm for precise ARG-bacteria linkage. Its ARG detection sensitivity was comparatively analysed against standard mNGS. Using phenotypic AST as reference, we then evaluated the clinical predictive value of capture mNGS-detected ARGs in a retrospective cohort from Sun Yat-sen Memorial Hospital (SYSMH) and an external cohort from Liuzhou Worker's Hospital (LWH). In addition, a prospective cohort from SYSMH was used to explore the clinical utility of ARG detection by mNGS. FINDINGS: Compared to standard mNGS, capture mNGS significantly enhanced ARG detection sensitivity, achieving a 44-fold increase in sequencing depth. In our retrospective cohort, key resistance genes detected by capture mNGS accurately predicted phenotypic resistance: blaCTX-M achieved a sensitivity of 1.00 (95% CI: 0.86, 1.00) and specificity of 1.00 (95% CI: 0.59, 1.00) for ceftriaxone resistance prediction, with an area under the receiver operating characteristic curve (AUC) of 0.93 (95% CI: 0.87, 0.99). BlaKPC demonstrated a sensitivity of 0.94 (95% CI: 0.73, 1.00) and specificity of 1.00 (95% CI: 0.95, 1.00) for carbapenem resistance (AUC = 0.97, 95% CI: 0.92, 1.00). Similarly, blaOXA-23 exhibited a sensitivity of 0.95 (95% CI: 0.82, 0.99) and specificity of 1.00 (95% CI: 0.69, 1.00) for carbapenem resistance (AUC = 0.97, 95% CI: 0.94, 1.00), which was externally validated in the LWH cohort. In addition, mecA showed a sensitivity of 0.94 (95% CI: 0.71, 1.00) and specificity of 0.94 (95% CI: 0.81, 0.99) for oxacillin resistance (AUC = 0.94, 95% CI: 0.87, 1.00). Whereas blaTEM/blaSHV showed higher false-positive rates for cephalosporin resistance and ErmB/ErmC showed lower sensitivity (0.6, 95% CI: 0.32, 0.84) for macrolide-lincosamide-streptogramin (MLS) resistance. Capture mNGS reported results (median turnaround time (TAT): 24.71 h (IQR 22.74-41.00)) were shorter than AST (median TAT: 73.16 h (IQR 54.19-93.42)). In a prospective cohort, the time to guide antibiotic therapy based on reported positive ARGs was significantly shorter than that based on reported resistant phenotypes from AST. INTERPRETATION: These results highlight that ARGs can be leveraged to rapidly and accurately predict bacterial resistance phenotypes with high sensitivity and specificity, thereby guiding antibiotic management in clinical practice. FUNDING: The National Natural Science Foundation of China, the Guangdong Science and Technology Department, Science and Technology Projects in Guangzhou.