Evaluation of an Optimal Epidemiological Typing Scheme for Legionella pneumophila with Whole-Genome Sequence Data Using Validation Guidelines

Abstract

Sequence-based typing (SBT), analogous to multilocus sequence typing (MLST), is the current "gold standard" typing method for investigation of legionellosis outbreaks caused by Legionella pneumophila However, as common sequence types (STs) cause many infections, some investigations remain unresolved. In this study, various whole-genome sequencing (WGS)-based methods were evaluated according to published guidelines, including (i) a single nucleotide polymorphism (SNP)-based method, (ii) extended MLST using different numbers of genes, (iii) determination of gene presence or absence, and (iv) a kmer-based method. L. pneumophila serogroup 1 isolates (n = 106) from the standard "typing panel," previously used by the European Society for Clinical Microbiology Study Group on Legionella Infections (ESGLI), were tested together with another 229 isolates. Over 98% of isolates were considered typeable using the SNP- and kmer-based methods. Percentages of isolates with complete extended MLST profiles ranged from 99.1% (50 genes) to 86.8% (1,455 genes), while only 41.5% produced a full profile with the gene presence/absence scheme. Replicates demonstrated that all methods offer 100% reproducibility. Indices of discrimination range from 0.972 (ribosomal MLST) to 0.999 (SNP based), and all values were higher than that achieved with SBT (0.940). Epidemiological concordance is generally inversely related to discriminatory power. We propose that an extended MLST scheme with ∼50 genes provides optimal epidemiological concordance while substantially improving the discrimination offered by SBT and can be used as part of a hierarchical typing scheme that should maintain backwards compatibility and increase discrimination where necessary. This analysis will be useful for the ESGLI to design a scheme that has the potential to become the new gold standard typing method for L. pneumophila.

FIG 1

Box plots with pairwise differences between typing panel isolates as calculated by each of the WGS-based methods. Included are sequencing replicates (6 pairs), “definitely related” isolates (10 isolates comprising 4 sets), “probably related” isolates (34 isolates comprising 13 sets), and 79 epidemiologically “unrelated” isolates.

FIG 2

(A) Index of discrimination (D) and epidemiological concordance (E) of each of the current and WGS-based methods based on the 106 typing panel isolates. Isolates were classified as the same type if they shared no differences and a different type if they shared 1 or more differences, except using the kmer-based method, where isolates were categorized into types using single-linkage clustering with a threshold equal to the maximum difference detected between sequencing replicates. (B) D and E values of each of the current and WGS-based methods when single-linkage clustering was used for some methods with a threshold that maintains the E of at least definitely related isolates at 1. The threshold is one allele difference using the cgMLST schemes with 100 or more genes, one SNP using the SNP-based method, and 0.065 using the kmer-based method. Using the rMLST scheme, the 50-gene cgMLST scheme, and the gene presence/absence scheme, isolates were classified as different types if they shared 1 or more differences (as in panel A).

FIG 3

Neighbor-net tree of the 106 typing panel isolates constructed using the 50-gene cgMLST scheme. All isolates are colored by their epidemiological relatedness as indicated in the key. Isolates belonging to the same type (i.e., with no allele differences) are enclosed in a red box. The ST1 cluster, comprising both ST1 isolates and isolates derived from ST1, is shown at a higher resolution on the right. The scale bars indicate the number of allelic differences.

FIG 4

Histograms showing pairwise SNP differences between epidemiologically unrelated and related isolates belonging to some of the major disease-associated sequence types (STs): ST1 (A), ST37 (B), ST42 (C), ST62 (D), and ST47 (E). In panels A to D, the top histogram shows pairwise SNP differences up to 20 only, while the bottom portion presents the full range. The maximum pairwise SNP difference within the ST47 isolates is <20 SNPs, and thus, only one illustration is shown (E). Left and right y axes represent the frequency of epidemiologically unrelated and related isolates, respectively. The epidemiologically unrelated and related isolates are colored as indicated in the key at the bottom right.

FIG 5

Maximum likelihood tree of 74 ST37 isolates constructed using 8,648 variable positions. Isolates are colored by their epidemiological relatedness as indicated in the key. The total number of SNPs identified between isolates of each epidemiologically related set is indicated. The scale shows the number of SNPs per variable site.