Lessons from 5 Years of Routine Whole-Genome Sequencing for Epidemiologic Surveillance of Shiga Toxin-Producing Escherichia coli, France, 2018-2022

Abstract

Whole-genome sequencing (WGS) is routine for surveillance of Shiga toxin-producing Escherichia coli human isolates in France. Protocols use EnteroBase hierarchical clustering at <5 allelic differences (HC5) as screening for cluster detection. We assessed current implementation after 5 years for 1,002 sequenced isolates. From genomic distances of serotypes O26:H11, O157:H7, O80:H2, and O103:H2, we determined statistical thresholds for cluster determination and compared those with HC5 clusters. Thresholds varied by serotype, 5-16 allelic distances and 15-20 single-nucleotide polymorphisms, showing limits of a single-threshold approach. We confirmed validity of HC5 screening for 3 serotypes because statistical thresholds had limited effect on isolate clustering (high sensitivity and specificity). For O80:H2, results suggest that HC5 is less reliable, and other approaches should be explored. Public health officials should regularly assess WGS used for Shiga toxin-producing E. coli surveillance to account for serotype and genomic evolution and to interpret WGS-linked isolates in light of epidemiologic data.

Keywords: France; STEC; Shiga toxin–producing Escherichia coli; bacteria; cluster detection; epidemiologic surveillance; hierarchical clustering; whole-genome sequencing.

Figure 1

Characteristics from 5 years of routine whole-genome sequencing for epidemiologic surveillance of Shiga toxin–producing Escherichia coli, France, 2018–2022. A) Distribution of pairwise allelic distances; B) SNP distances, censured at 50. Shiga toxin–producing Escherichia coli serotypes are shown for each panel. SNP, single-nucleotide polymorphism.

Figure 2

Mixture of distributions model applied to allelic distance data from 5 years of routine whole-genome sequencing for epidemiologic surveillance of Shiga toxin–producing Escherichia coli, France, 2018–2022. A) Number of components fit to the data distribution; B) threshold represented as the probability of belonging to the first distribution. Shiga toxin–producing Escherichia coli serotypes are shown for each panel. Black line indicates global estimated density; black circles, probability of belonging to first distribution for each observed allelic or single-nucleotide polymorphism distance; red line, largest allelic or single-nucleotide polymorphism distance that has a 50% probability of belonging to the first distribution. Comp, component.

Figure 3

Mixture of distributions model applied to SNP distance data from 5 years of routine whole-genome sequencing for epidemiologic surveillance of Shiga toxin–producing Escherichia coli, France, 2018–2022. A) Number of components fit to the data distribution; B) threshold represented as a probability of belonging to the first or second distribution. Shiga toxin–producing Escherichia coli serotypes are shown for each panel. Comp, component; SNP, single-nucleotide polymorphism.

Figure 4

Regression from hierarchical clustering at a threshold of 5 allelic differences from 5 years of routine whole-genome sequencing for epidemiologic surveillance of Shiga toxin–producing Escherichia coli, France, 2018–2022. A) Allelic distance; B) SNP distance. Distances calculated as a function of time in days by multivariable fractional polynomial linear regression. Black circles indicate estimated allelic or SNP distance for each observed temporal distance in days; blue, red, green, and black vertical lines, 95% CIs of the estimated genomic distances for each observed temporal distance in days. SNP, single-nucleotide polymorphism.

Figure 5

Single-nucleotide polymorphism–based maximum likelihood phylogenetic tree of 226 080:H2 isolates from 5 years of routine whole-genome sequencing for epidemiologic surveillance of Shiga toxin–producing Escherichia coli, France, 2018–2022. Tree was built based on the sequence alignment of 3,949 single-nucleotide variant sites of the recombination-free core genome of E. coli strain MOD1-EC6881 (GenBank accession no. GCF_002520045.1). Tree was midpoint-rooted and visualized with iTOL (https://itol.embl.de). Bootstrap support values >90% are indicated with red dots on the branches. Branch lengths and corresponding scale bar indicate numbers of single-nucleotide polymorphisms per base of the final alignment. HC5, hierarchical clustering at a threshold of 5 allelic differences.