An optimised method for bacterial nucleic acid extraction from positive blood culture broths for whole genome sequencing, resistance phenotype prediction and downstream molecular applications
BackgroundA prerequisite to rapid molecular detection of pathogens causing bloodstream infections is an efficient, cost effective and robust DNA extraction solution. We describe methods for microbial DNA extraction direct from positive blood culture broths, suitable for metagenomic sequencing and the application of machine-learning based tools to predict antimicrobial susceptibility. MethodsProspectively collected culture-positive blood culture broths with Gram-negative bacteria, were directly extracted using various commercially available kits. We compared methods for efficient inhibitor removal, avoidance of DNA shearing or degradation, to achieve DNA of high quality and purity. Bacterial species identified via whole-genome metagenomic sequencing (Illumina, MiniSeq) from blood culture extracts were compared to conventional methods from cultured isolates (Vitek MS). A machine-learning algorithm (AREScloud) was used to predict susceptibility against commercially available antibiotics, compared to susceptibility testing (Vitek 2) and other commercially available rapid diagnostic instruments (Accelerate Pheno and BCID). ResultsA two-kit method using a modified MolYsis Basic kit (for host DNA depletion) and extraction using Qiagen DNeasy UltraClean microbial kits resulted in optimal extractions appropriate for multiple molecular applications, including PCR, short-read and long-read sequencing. DNA extracts from 40 blood culture broths were included. Taxonomic profiling by direct metagenomic sequencing matched species identification by conventional methods in 38/40 (95%) of samples, with two showing agreement to genus level. In two polymicrobial samples, a second organism was missed by sequencing. Whole genome sequencing antimicrobial susceptibility testing (WGS-AST) models were able to accurately infer profiles for 6 common pathogens against 17 antibiotics. Overall categorical agreement (CA) was 95%, with 11% very major errors (VME) and 3.9% major errors (ME). CA for WGS-AST was >95% for 5/6 of the most common pathogens (E. coli, K. pneumoniae, P. mirabilis, P. aeruginosa and C. jejuni) while it was lower for K. oxytoca (66.7%), likely due to the presence of inducible cephalosporinases. Performance of WGS-AST was sub-optimal for uncommon pathogens (e.g. Elizabethkingia) and some combination antibiotic compounds (e.g. ticarcillin-clavulanate). Time to pathogen identification and resistance gene detection was fastest with BCID (1 h from blood culture positivity). Accelerate Pheno provided a rapid MIC result in approximately 8 h. While Illumina based direct metagenomic sequencing did not result in faster turn-around times compared conventional methods, use of real-time nanopore sequencing may allow faster data acquisition. ConclusionsThe application of direct metagenomic sequencing from positive blood culture broths is a feasible approach and solves some of the challenges of sequencing from low-bacterial load samples. Machine-learning based algorithms are also accurate for common pathogen / drug combinations, although additional work is required to optimise algorithms for uncommon species and more complex resistance genotypes, as well as streamlining methods to provide more rapid sequencing results.