Targeted sequencing of Enterobacterales bacteria using CRISPR-Cas9 enrichment and Oxford Nanopore Technologies

Abstract

Sequencing DNA directly from patient samples enables faster pathogen characterization compared to traditional culture-based approaches, but often yields insufficient sequence data for effective downstream analysis. CRISPR-Cas9 enrichment is designed to improve the yield of low abundance sequences but has not been thoroughly explored with Oxford Nanopore Technologies (ONT) for use in clinical bacterial epidemiology. We designed CRISPR-Cas9 guide RNAs to enrich the human pathogen Klebsiella pneumoniae, by targeting multi-locus sequence type (MLST) and transfer RNA (tRNA) genes, as well as common antimicrobial resistance (AMR) genes and the resistance-associated integron gene intI1. We validated enrichment performance in 20 K. pneumoniae isolates, finding that guides generated successful enrichment across all conserved sites except for one AMR gene in two isolates. Enrichment of MLST genes led to a correct allele call in all seven loci for 8 out of 10 isolates that had depth of 30× or more in these regions. We then compared enriched and unenriched sequencing of three human fecal samples spiked with K. pneumoniae at varying abundance. Enriched sequencing generated 56× and 11.3× the number of AMR and MLST reads, respectively, compared to unenriched sequencing, and required approximately one-third of the computational storage space. Targeting the intI1 gene often led to detection of 10-20 proximal resistance genes due to the long reads produced by ONT sequencing. We demonstrated that CRISPR-Cas9 enrichment combined with ONT sequencing enabled improved genomic characterization outcomes over unenriched sequencing of patient samples. This method could be used to inform infection control strategies by identifying patients colonized with high-risk strains.

Importance: Understanding bacteria in complex samples can be challenging due to their low abundance, which often results in insufficient data for analysis. To improve the detection of harmful bacteria, we implemented a technique aimed at increasing the amount of data from target pathogens when combined with modern DNA sequencing technologies. Our technique uses CRISPR-Cas9 to target specific gene sequences in the bacterial pathogen Klebsiella pneumoniae and improve recovery from human stool samples. We found our enrichment method to significantly outperform traditional methods, generating far more data originating from our target genes. Additionally, we developed new computational techniques to further enhance the analysis, providing a thorough method for characterizing pathogens from complex biological samples.

Keywords: CRISPR-Cas9 enrichment; Enterobacterales; Klebsiella; Oxford Nanopore; metagenomics.

Fig 1

Library preparation differences between unenriched and CRISPR-Cas9 enriched sequencing. During unenriched sequencing, sequencing adapters are ligated to native terminal phosphate groups on DNA molecules to allow for sequencing. During CRISPR-Cas9 enrichment, native phosphate groups are removed from all DNA molecules so that adapters cannot ligate. CRISPR-Cas9 is then used to cleave molecules of interest, exposing their terminal phosphate groups and allowing for specific adapter ligation and sequencing.

Fig 2

Conservation of tRNA guides across Enterobacterales. (A) Neighbour-joining tree of representative genomes from all genera in GTDB R95 classified as Enterobacterales (one genome per genus, n = 250 genera). The color spectrum of the heatmap shows the proportion of genomes matched to guide sequences in a dereplicated version of the full GTDB database for each genus (n = 11,339 total Enterobacterales genomes, genome count for each family shown in brackets). The color bar to the left of heatmap shows the GTDB-defined family of each genus. (B) Guide conservation in notable Enterobacterales pathogens.

Fig 3

Guide pair performance in CRISPR-Cas9 enriched libraries of KpSC isolates. (A) Sequencing depth of all target contigs, for example, K. pneumoniae isolate INF157 following CRISPR-Cas9 enrichment. Gene target locations are shown as colored rectangles across the genome. Depth is shown relative to median depth of off-target alignments. The inner bar denoted as “10×” shows regions where relative depth is greater than 10 (shown in black). GenBank accessions of the target contigs are CP024528.1, CP024529.1, and CP024531.1 (B) Summary of guide performance across all 20 KpSC isolates. A successful target is defined as when the number of on-target reads is equal to or greater than 10× median depth of off-target reads. Run 1 refers to the initial sequencing with all 20 isolates, while run 2 refers to a repeat validation run on five randomly selected isolates.

Fig 4

Allele call accuracy of target AMR and MLST genes following enriched and unenriched sequencing of human fecal samples spiked with K. pneumoniae strain INF298 (Refseq accession GCA_904864465.1) at 4 × 10⁶–4 × 10⁷ CFU/g. Median depth was calculated from the total number of reads aligning to the target gene. (A) Allele call accuracy following CRISPR-Cas9 enriched sequencing. (B) Allele accuracy following unenriched sequencing.

Fig 5

Depth of sequencing of K. pneumoniae strain INF298 plasmid (GenBank accession CP110595.1) following sequencing of three human fecal samples spiked with INF298 at 4 × 10⁷ CFU/g. The bottom panel is labeled with the regions of all AMR genes on the plasmid, with the commonly observed class 1 integron labeled.

Fig 6

Relationship between guide conservation and enrichment performance following CRISPR-Cas9 enriched sequencing of a mock microbial community with varying amount of K. pneumoniae strain INF298 (Refseq accession GCA_904864465.1) DNA. (A) Taxonomic distribution of sequencing output based on an alignment-based approach (see Materials and Methods). (B) Linear regression analysis between the total guide targets and total on-target reads for isolates of a given genus following sequencing of the mock community with no K. pneumoniae included.