Bacterial identification and antimicrobial resistance profiling in a variety of clinical specimens using nanopore metagenomic sequencing

Background: First-line pathogen detection and antimicrobial resistance (AMR) profiling, using culture, requires viable organisms. Metagenomic next generation sequencing (mNGS) can identify and characterize pathogens directly from clinical material without culture. This proof-of-concept study aimed to compare bacterial identification and AMR profiling using mNGS and culture. Methods: To date, 24/48 samples comprising of ascitic fluid, cerebrospinal fluid (CSF), urine, and pleural fluid were tested on routine culture and mNGS (using the Oxford Nanopore Technology (ONT) MinION platform). Results were analysed using EPI2ME wf-metagenomics, following host-depletion using Bowtie2, to characterise pathogens and AMR profiles. Pathogen identification was expressed as relative abundance (RA), and sequencing results were compared to those obtained by routine culture. Results: Culture identified a pathogen in 9/24 (38%) specimens while mNGS identified a potential pathogen in 12/24 (50%). mNGS was able to identify a potential pathogen in 6/15 (40%) culture negative specimens. There was congruence in microbes identified using culture and mNGS in 5/9 (56%) culture positive specimens; however, the species matching culture results was rarely also dominant in terms of abundance (3/9, 30%), complicating causality assignment. Co-detections were also common with mNGS, with ≥5 species detected in the majority of the samples. Further, 9/24 (38%) specimens were interpreted as negative using mNGS due to the exclusion of unknown reads which did not match to any taxonomic group. Phenotypic resistance was seen in 7/24 (29%) samples, with correlation between culture and mNGS in 3/7 (42%) samples. mNGS detected AMR genes in 5/17 (29%) samples without phenotypic resistance. Discussion: mNGS was able to resolve 40% of the culture negative samples, majority of which were urine. The culture negative results may have been due to the fastidious nature of the organisms or prior antibiotic administration. A substantial portion of the reads among the samples could not be classified; this may be attributed to poor sequence quality, resulting from poor DNA, or sequencing library, quality, and low DNA input. Further analysis is needed to decern whether unknown reads could be classified with alternate methods or are true negatives. Co-detections were common in polymicrobial sample types, such as urine, albeit majority were normal flora. There was a low correlation between AMR profiles identified using mNGS and phenotypic drug susceptibility testing, indicating that mNGS alone may not be a suitable approach for describing clinically actionable AMR results. Conclusion: In our proof-of-concept study, mNGS was able to identify a potential pathogen in 50% of the samples, including 40% which were unresolved using culture; however, congruence for pathogen identification with culture was moderate. For future research, a larger sample size should be considered as well as additional measures in conjunction with RA to determine the pathogen of interest and assessing the clinical significance of results.