A pilot study exploring long-read nanopore metagenomics sequencing for heteroresistance detection in clinical specimens

Introduction: Heteroresistance occurs when most cells in a single bacterial population are susceptible to an antibiotic, while a minority exhibit resistance, and may underpin discrepancies between in vitro susceptibility results and clinical responses. The resistant subpopulation may be present at a frequency as low as 1 resistant cell per million. This impedes antibiotic susceptibility testing (AST) and diagnosis, leading to poor treatment outcomes. A potential alternative is a direct detection from clinical specimens using long-read metagenomic sequencing, with Oxford Nanopore Technology (ONT). This study explored ONT sequencing and compared two analysis pipelines for detecting resistant bacterial subpopulations. Methods: Confirmed culture-negative, antibiotic-free, and E. coli DNA-free urine samples were pooled and spiked with well-characterised resistant and susceptible E. coli isolates to simulate 10%, 5%, and 2% heteroresistance. The resistance genes included ciprofloxacin resistance genes (parC, gyrA) gentamycin and amikacin resistant genes (aac(3)IIe, aac(6′)Ib-cr4) B-lactams resistance genes (blaCMY-132, blaCTX–M–15, and blaOXA-1) and Trimethoprim-sulfamethoxazole resistance genes (drfA17, sul1). The spiked specimens were enriched in LB broth by incubating at 37°C for 6 hours, followed by library preparation and sequencing on the MinION platform. Species identification and resistance gene detection were performed using the ONT-based EPI2ME Labs software, wf-metagenomics, and compared to an in-house pipeline using the publicly available assembler, metaFlye, combined with the Resistance Gene Identifier (RGI) tool. To further assess and compare the performance of the pipelines, 5-culture positive urine samples with known phenotypic resistance profiles were included in the study. Results: E. coli was detected by wf-metagenomics in all spiked samples, representing >90% of reads. Fifty percent of expected resistance genes were detected using wf-metagenomics in the 2% and 10% heteroresistant specimens. In contrast, despite similar read counts, only 17% of predicted genes were detected in the 5% specimen. The highest resistance gene detection rate was 50%, using the wf-metagenomics bioinformatics platform. While metaFlye-RGI identified 15% of the expected point mutations in all heteroresistant specimens, wf-metagenomics does not currently enable the detection of point mutations. Discussion: For heteroresistance detection, wf-metagenomics is recommended as taxonomy and resistance detection can be processed simultaneously. It can also detect more resistance genes compared to metaFlye-RGI. There was also a good correlation between resistance genes detected in culture-positive urines and the phenotypic resistance profiles with one phenotypically susceptible isolate having resistance genes detected by wf-metagenomics. Additionally, all species were accurately identified using wf-metagenomics echoing other study findings that were able to speciate using ONT metagenomic sequencing. Conclusion: Since only partial detection of resistance profiles was observed, this pilot study demonstrates the need to optimise the workflow, and pipeline further to effectively detect heteroresistance using long-read nanopore metagenomics sequencing directly from clinical specimens.