In vitro and in silico parameters for precise cgMLST typing of Listeria monocytogenes

Background Whole genome sequencing analyzed by core genome multi-locus sequence typing (cgMLST) is widely used in surveillance of the pathogenic bacteria Listeria monocytogenes. Given the heterogeneity of available bioinformatics tools to define cgMLST alleles, our aim was to identify parameters influencing the precision of cgMLST profiles. Methods We used three L. monocytogenes reference genomes from different phylogenetic lineages and assessed the impact of in vitro (i.e. tested genomes, successive platings, replicates of DNA extraction and sequencing) and in silico parameters (i.e. targeted depth of coverage, depth of coverage, breadth of coverage, assembly metrics, cgMLST workflows, cgMLST completeness) on cgMLST precision made of 1748 core loci. Six cgMLST workflows were tested, comprising assembly-based (BIGSdb, INNUENDO, GENPAT, SeqSphere and BioNumerics) and assembly-free (i.e. kmer-based MentaLiST) allele callers. Principal component analyses and generalized linear models were used to identify the most impactful parameters on cgMLST precision. Results The isolate's genetic background, cgMLST workflows, cgMLST completeness, as well as depth and breadth of coverage were the parameters that impacted most on cgMLST precision (i.e. identical alleles against reference circular genomes). All workflows performed well at >= 40X of depth of coverage, with high loci detection (> 99.54% for all, except for BioNumerics with 97.78%) and showed consistent cluster definitions using the reference cut-off of <= 7 allele differences. Conclusions This highlights that bioinformatics workflows dedicated to cgMLST allele calling are largely robust when paired-end reads are of high quality and when the sequencing depth is >= 40X.