Targeted surveillance strategies for efficient detection of novel antibiotic resistance variants

Abstract

Genotype-based diagnostics for antibiotic resistance represent a promising alternative to empiric therapy, reducing inappropriate antibiotic use. However, because such assays infer resistance based on known genetic markers, their utility will wane with the emergence of novel resistance. Maintenance of these diagnostics will therefore require surveillance to ensure early detection of novel resistance variants, but efficient strategies to do so remain undefined. We evaluate the efficiency of targeted sampling approaches informed by patient and pathogen characteristics in detecting antibiotic resistance and diagnostic escape variants in Neisseria gonorrhoeae, a pathogen associated with a high burden of disease and antibiotic resistance and the development of genotype-based diagnostics. We show that patient characteristic-informed sampling is not a reliable strategy for efficient variant detection. In contrast, sampling informed by pathogen characteristics, such as genomic diversity and genomic background, is significantly more efficient than random sampling in identifying genetic variants associated with resistance and diagnostic escape.

Keywords: Neisseria gonorrhoeae; antibiotic resistance; diagnostic; epidemiology; global health; infectious disease; microbiology; surveillance.

Figure 1.. The impact of demography-, niche-, and geography-aware sampling on the detection efficiency of genetic resistance variants.

Dot plots showing the detection efficiency (with lines indicating the mean and 95% confidence intervals from 100 simulations) for resistance variants RplD G70D (A–B), 23S rRNA C2611T (C–D), and penA XXXIV (E–F) in datasets 1 and 2. In datasets 1 and 2, targeted sampling was informed by demographic (gender and sexual behavior) and anatomical site of isolate collection (niche) information (A, C, and E), and in datasets 3 and 4, targeted sampling was informed by country or prefecture of sample collection (B, D, and F). Dot colors indicate the sampling approach, and asterisks indicate a significant difference (p<0.05 by Mann-Whitney U test) in detection efficiency between the demography-, niche- or geography-aware approach compared to random sampling (*p<0.05, **p<0.01, ***p<0.001; red asterisks indicate significantly lower detection efficiency of demography- or geography-aware approaches compared to random sampling). Note that sampling simulations were not performed for RplD G70D in datasets 1 and 4 or for penA XXXIV in dataset 3 as prevalence of the variants in these datasets was >10%. n.s., not significant at α = 0.05; M, men; W, women; MSM, men who have sex with men; MSW, men who have sex with women; WSM, women who have sex with men.

Figure 1—figure supplement 1.. Isolates from patients with travel-associated gonorrhea are associated with longer terminal branches compared to patients with locally-acquired gonorrhea.

Maximum-likelihood phylogeny produced from the pseudogenome alignment (with predicted regions of recombination removed) of isolates from dataset 2 (A). Patient travel history is indicated by the colored ring in A. Scatter dot plots showing the terminal branch lengths (with lines indicating the mean and 95% confidence intervals) associated with isolates from patients with travel-associated gonorrhea compared to patients with locally-acquired gonorrhea with asterisks indicating a significant difference (p<0.001 by Mann-Whitney U test) in terminal branch lengths between the two groups (B).

Figure 2.. The impact of phylogeny-aware sampling on the detection efficiency of genetic resistance and diagnostic escape variants.

Scatter dot plots showing the detection efficiency (with lines indicating the mean and 95% confidence intervals from 100 simulations) for resistance variants RplD G70D (A), 23S rRNA C2611T (B), and penA XXXIV (C) in datasets 1–5. Note that sampling simulations were not performed for RplD G70D in datasets 1 and 4 or for penA XXXIV in dataset 3 as prevalence of the variants in these datasets was >10%. Maximum-likelihood phylogenies produced from pseudogenome alignments (with predicted regions of recombination removed) of isolates from dataset 4 (D) and dataset 2 (E). Presence or absence of the 23S rRNA C2611T mutation (in at least 2/4 alleles) and the mosaic penA XXXIV allele is indicated by colored rings. Scatter dot plots showing the detection efficiency (with lines indicating the mean and 95% confidence intervals from 100 simulations) for diagnostic-associated variants 16S rRNA C1209A (F), N. meningitidis-like porA (G), cppB deletion (H), and DR-9A G168A (I) in all datasets in which the variant was present. Dot colors in A–C) and F–I) indicate the sampling approach, and asterisks indicate a significant difference (p<0.05 by Mann-Whitney U test) in detection efficiency between the phylogeny-aware approach compared to random sampling (*p<0.05, **p<0.01, ***p<0.001; red asterisks indicate significantly lower detection efficiency of the phylogeny-aware approach compared to random sampling, and green asterisks indicate significantly higher detection efficiency of the phylogeny-aware approach compared to random sampling). n.s., not significant at α = 0.05.

Figure 2—figure supplement 1.. Detection efficiency of clonal group sampling across different similarity thresholds.

Scatter dot plots showing the detection efficiency (with lines indicating the mean and 95% confidence intervals from 100 simulations) for resistance variants RplD G70D (A), 23S rRNA C2611T (B), and penA XXXIV (C) in datasets 1–5 and for diagnostic-associated variants 16S rRNA C1209A (D), N. meningitidis-like porA (E), cppB deletion (F), and DR-9A G168A (G) in all datasets in which the variant was present. Note that sampling simulations were not performed for RplD G70D in datasets 1 and 4 or for penA XXXIV in dataset 3 as prevalence of the variants in these datasets was >10%. Dot colors indicate the sampling approach, and asterisks indicate a significant difference (p<0.05 by Mann-Whitney U test) in detection efficiency between the phylogeny-aware approach compared to random sampling (*p<0.05, **p<0.01, ***p<0.001; red asterisks indicate significantly lower detection efficiency of the phylogeny-aware approach compared to random sampling, and green asterisks indicate significantly higher detection efficiency of the phylogeny-aware approach compared to random sampling). n.s., not significant at α = 0.05.

Figure 3.. The impact of genomic background-aware sampling on the detection efficiency of phenotypic resistance variants.

Bar charts showing the proportions of ceftriaxone reduced susceptibility (CRO-RS) isolates, ceftriaxone susceptible (CRO-S) isolates, cefixime resistant (CFX-R) isolates, and cefixime susceptible (CFX-S) isolates with GyrA S91F and GyrA S91 wild-type alleles (A) and with PorB G120 and/or A121 mutations and PorB G120 and A121 wild-type alleles (B) across datasets 1–5. Bar charts showing the number of (C) CRO-RS and (D) CFX-R isolates with each haplotype, along with heatmaps showing the presence or absence of the GyrA S19F mutation, the PorB G120 and/or A121 mutations, and other alleles at loci previously associated with extended spectrum cephalosporin resistance. Bar colors in (C) and (D) indicate the dataset from which the isolates were derived. Scatter dot plots showing the detection efficiency (with lines indicating the mean and 95% confidence intervals from 100 simulations) for CRO-RS (E) and CFX-R (F) in all datasets in which the variant was present. Dot colors in E–F) indicate the sampling approach, and asterisks indicate a significant difference (p<0.05 by Mann-Whitney U test) in detection efficiency between the phylogeny-aware approach compared to random sampling (*p<0.05, **p<0.01, ***p<0.001; green asterisks indicate significantly higher detection efficiency of the genomic background-aware approach compared to random sampling).