Genomic epidemiology of Neisseria gonorrhoeae with reduced susceptibility to cefixime in the USA: a retrospective observational study.

BACKGROUND: The emergence of Neisseria gonorrhoeae with decreased susceptibility to extended spectrum cephalosporins raises the prospect of untreatable gonorrhoea. In the absence of new treatments, efforts to slow the increasing incidence of resistant gonococcus require insight into the factors that contribute to its emergence and spread. We assessed the relatedness between isolates in the USA and reconstructed likely spread of lineages through different sexual networks.
METHODS: We sequenced the genomes of 236 isolates of N gonorrhoeae collected by the Centers for Disease Control and Prevention's Gonococcal Isolate Surveillance Project (GISP) from sentinel public sexually transmitted disease clinics in the USA, including 118 (97%) of the isolates from 2009-10 in GISP with reduced susceptibility to cefixime (cef(RS)) and 118 cefixime-susceptible isolates from GISP matched as closely as possible by location, collection date, and sexual orientation. We assessed the association between antimicrobial resistance genotype and phenotype and correlated phylogenetic clustering with location and sexual orientation.
FINDINGS: Mosaic penA XXXIV had a high positive predictive value for cef(RS). We found that two of the 118 cef(RS) isolates lacked a mosaic penA allele, and rechecking showed that these two were susceptible to cefixime. Of the 116 remaining cef(RS) isolates, 114 (98%) fell into two distinct lineages that have independently acquired mosaic penA allele XXXIV. A major lineage of cef(RS) strains spread eastward, predominantly through a sexual network of men who have sex with men. Eight of nine inferred transitions between sexual networks were introductions from men who have sex with men into the heterosexual population.
INTERPRETATION: Genomic methods might aid efforts to slow the spread of antibiotic-resistant N gonorrhoeae through augmentation of gonococcal outbreak surveillance and identification of populations that could benefit from increased screening for asymptomatic infections.

Introduction

Neisseria gonorrhoeae causes more than 100 million infections worldwide each year, with an estimated 820 000 in the USA. The high burden of gonorrhoea is a result of its success in transmission and avoidance of host defences and, in the antimicrobial era, the emergence and spread of drug resistance. Since the introduction of antibiotic treatment for gonorrhoea, resistance has emerged to each first-line drug in turn. The global emergence of strains resistant to cephalosporins raises the threat of widespread resistance to affordable, effective, safe, and single-dose treatment. Improved understanding of the epidemiology of resistant gonococcus is essential to our response to this challenge.

In the USA, the Gonococcal Isolate Surveillance Project (GISP)—a sentinel surveillance system supported by the Centers for Disease Control and Prevention (CDC) that monitors gonococcal antimicrobial resistance nationwide—has reported increasing minimum inhibitory concentrations (MICs) for the oral extended-spectrum cephalosporin cefixime since 2009. High MICs have been reported to be associated with increased failure rates of cefixime treatment in Canada. The rising prevalence of isolates with reduced susceptibility to cefixime (cefRS; MIC ≥0·25 μg/mL) in the USA led to the updated CDC recommendation in 2012, that first-line treatment should consist of dual treatment with the injectable extended-spectrum cephalosporin ceftriaxone 250 mg intramuscularly and either azithromycin 1 g orally as a single dose or doxycycline 100 mg orally twice daily for 7 days.

The emergence of cefRS isolates in the USA shows the need for evidence-based strategies to rapidly and accurately identify patients at risk of treatment failure, to provide appropriate treatment, and to use screening and contact tracing to contain or slow the spread of resistance. Such strategies could be strengthened by knowledge of the genetic basis of resistance (for the design of molecular tests), estimates of the relative importance of de-novo emergence of resistance versus transmission of resistant infections, and evidence about the structure of sexual networks through which resistant infections spread so that screening and tracing can be appropriately targeted.

Genomic epidemiology has provided insight into outbreaks and transmission networks for several pathogens, with greater resolution than that achievable by traditional typing methods. Previous observations have suggested an eastward spread of antibiotic resistance in gonococcus among networks of men who have sex with men with occasional introduction into heterosexual networks. We analysed whole-genome sequences to describe the relatedness between isolates, to reconstruct the likely spread of lineages through sexual networks according to geography and sexual orientation, and to identify likely emergence events of resistant phenotypes.

Methods

Specimen collection

We obtained isolates of N gonorrheae from GISP. Inkeeping with GISP protocol, the first 25 urethral isolates each month from men with gonococcal urethritis are submitted from sentinel sexually transmitted disease clinics in 28 cities to regional laboratories for antimicrobial susceptibility testing. Samples were then frozen and stored at −70°C. 236 isolates from 2009–10 were obtained for this study. This set comprises 118 cefRS isolates and 118 isolates susceptible to cefixime (MIC ≤0·06 μg/mL). The 118 cefRS isolates represented 97% of all cefRS isolates identified by GISP from this period, and were from 19 of the 28 GISP sentinel clinic sites. The 118 cefixime-susceptible isolates were selected from a database of GISP isolates maintained at the CDC and matched to the cefRS isolates as closely as possible by location and date of isolation and sexual orientation of the patients (appendix pp 3, 14).

Procedures

DNA was prepared from single colonies and sequenced with the Illumina HiSeq platform according to standard protocols (appendix p 1). Illumina reads were mapped to a reference strain, TCDC-NG08107 (GenBank accession number NC_017511.1), using SMALT. Single nucleotide polymorphisms (SNPs) were called (ie, filtered to determine the working set of SNPs from all candidate sites) according to previous parameters. Each Illumina read set was assembled with velvet (version 1.0.12) and VelvetOptimiser.

We extracted the locus from each of the de-novo genome assemblies for each of the antimicrobial resistance loci of interest (penA, mtrR, porB, pilQ, and ponA). Sequences were visualised in SeaView and aligned with Muscle. We determined phylogenies with PhyML using a generalised time-reversible model of evolution and Γ correction among site rate variation with four categories. To assess variants in the 23S rRNA—of which there are four copies in the gonococcal genome—we mapped reads to a pseudo-reference genome sequence altered to contain only a single copy of the 23S rRNA. The allele copy number for observed variants was inferred from the fraction of reads containing the variant.

Analysis

We identified regions of possible homologous recombination and removed them with the iterative method of Croucher and colleagues. We used RAxML (version 7.0.4) to reconstruct the phylogeny.

To test for a temporal signal indicating significant correlation between date of isolation and genetic distance from the root of the phylogeny, we plotted root-to-tip distance versus date (month and year) of isolation with Path-O-Gen. We used the BEAST package to infer phylogeny, divergence dates, geographical spread, and switching between men who have sex with men and men who have sex with women. We used a continuous-time Markov chain Monte Carlo method with a chain length of 200 million steps with samples taken every 25 000 steps, a GTR+Γ substitution model with site rate variation with four categories, defined tip dates as the month and year of isolation, a strict molecular clock, and a constant size coalescent. Parameters were estimated after removing a 10% burn-in with Tracer (version 1.5). We analysed the city of isolation and sexual orientation as discrete traits, and used BEAST to infer ancestral state of the cities of origin and sexual orientation at the internal nodes.

Role of the funding source

The sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results

We assessed the relationships between the 236 N gonorrhoeae isolates using core-genome SNPs. To minimise the effect of homologous recombination on phylogenetic reconstruction we removed regions of likely homologous recombination identified by anomalous SNP density. The resulting tree is notable for several clusters of very closely related isolates (figure 1A; see appendix for tree without recombination events removed); clinic location and sexual orientation are shown for each isolate (figure 1B). Two of these clusters (cefRS clusters 1 and 2) include most cefRS isolates (n=114; figure 1C); notably, all (n=26) isolates in cefRS cluster 2 are susceptible to ciprofloxacin, by contrast with cluster 1, in which 102 of 104 (98%) isolates are resistant (appendix, pp 5, 14). As expected, whole-genome-based phylogeny yielded greater resolution than did multilocus sequence typing and N gonorrhoeae multiantigen sequence typing (appendix pp 4, 14).

Figure 1

Genetic relatedness, epidemiology, and susceptibility to cefixime of isolates of Neisseria gonorrhoeae

(A) Maximum likelihood phylogeny based on core-genome SNPs of the 236 isolates and the reference genome, TCDC-NG08107. Scale bar represents expected number of changes per site (see methods). (B) Geographical location of the clinic of origin for each isolate (red=men who have sex with men, blue=men who have sex with women). (C) The minimum inhibitory concentration for cefixime (μg/mL). Solid arrows denote branches of the phylogeny predicted to have acquired the mosaic penA XXXIV allele via recombination, dashed arrow indicates the novel mosaic penA. The pale orange rectangles highlight cefRS clusters 1 and 2. A cluster of isolates was predicted to have reverted to cefixime susceptibility within cluster 1, replacing mosaic penA XXXIV with mosaic penA XXXVIII is noted by a black bar. HON=Honolulu. POR=Portland. SFO=San Francisco. SEA=Seattle. LAX=Los Angeles. ORA=Orange County. SDG=San Diego. LVG=Las Vegas. PHX=Phoenix. ALB=Albuquerque. DEN=Denver. MIN=Minneapolis. CHI=Chicago. DTR=Detroit. CLE=Cleveland. GRB=Greensborough. BAL=Baltimore. PHI=Philadelphia. NYC=New York City. CefRS=reduced susceptibility to cefixme.

Analysis of likely regions of recombination in this dataset shows extensive recombination, with 1905 predicted recombination events encompassing 65·5% of the reference genome. The recombination to mutation ratio for the entire dataset was 1·9. Of the polymorphic sites outside of predicted recombinant regions, 48% were homoplasic, suggesting substantial remaining recombinations even after removal of regions with anomalously high SNP density. The recombination to mutation ratio for cefRS cluster 1 (figure 1) was 4·0, and the fraction of homoplasic sites outside of predicted recombinant regions was 11%. Together, the greater homoplasy and lower recombination to mutation ratio for the overall tree compared with recent branches suggest that the deeper branching in the tree is confounded by extensive recombination.

Isolates that are closely related genetically share recent ancestry and are related through transmission events. Inspection of the phylogenetic tree together with the location of isolation and sexual orientation of the infected individual reveals several sets of isolates in cefRS cluster 1 with local transmission, including one in Las Vegas exclusively among men who have sex with women, another in Portland exclusively among men who have sex with men, and a third predominantly in men who have sex with men in Honolulu with spread across geographical and sexual network boundaries (figure 1B).

To infer the spread of isolates across time, space, and sexual networks, we incorporated clinic location and sexual orientation into a discrete phylogeographic diffusion model and used BEAST to assess cefRS cluster 1, which contains temporal signal (appendix p 6). We estimated that the substitution rate was 5·5 sites per chromosome per year (95% highest posterior density interval [HPD] 4·3–6·7), corresponding to 2·5 × 10−6 substitutions per site per year (2·0 × 10−6 to 3·1 × 10−6), similar to rates reported for other bacteria. We estimate that the most recent common ancestor for cefRS cluster 1 dates to 1997 (1993–2000), around the time of the first reports of reduced susceptibility to cefixime and mosaic penA.

Because we did not have samples from other countries to provide broader context, reliable inference of the exact timing, location, and number of introductions of the cluster 1 lineage into the USA is difficult. In keeping with previous epidemiological observations of the spread of resistant gonococcus, cefRS isolates seemed to circulate first on the west coast and in Hawaii (figure 2). A plot of the average longitude of the geographic location of the nodes (both the internal inferred and the sampled leaf nodes) within windows of 2 months shows circulation primarily on the west coast. Regression of the east-most node by date indicates eastward spread (R2=0·54, p<0·0001; figure 3). The phylogenetic reconstruction of cluster 1 shows eight transitions from men who have sex with men to men who have sex with women. Only one transition occurred from men who have sex with women to men who have sex with men, accounting for one case, suggesting that the spread of N gonorrhoeae is predominantly in men who have sex with men, with occasional introductions into networks of men who have sex with women.

Figure 2

Phylogenetic reconstruction of cluster 1 with inferred location and sexual orientation

Tree structure is fixed to the maximum likelihood structure presented in figure 1. The branch and node colours indicate the known (leaves) and predicted (internal branches) location of isolation and the sexual network, respectively. The furthest node right is indexed to the month and year of the most recent isolate (December, 2010). The cluster of isolates that underwent recombination at the dcw locus thereby replacing mosaic penA XXXIV with XXXVIII is noted; the arrow indicates the most recent common ancestor of those isolates. MSM=men who have sex with men. MSW=men who have sex with women.

Figure 3

Longitude of Neisseria gonorrhoeae from samples and inferred locations in reconstructed phylogeny

Nodes from figure 2 are plotted using 2-month windows. Linear regression is for east-most longitude by date.

Although penA alleles are known to confer reduced susceptibility to cephalosporins by reduced binding of cephalosporins to the penA gene product PBP2, the roles of other genes, such as mtrR, porB, pilQ, and ponA—independent of, or in epistasis with, penA—has been uncertain. In our sample—including the reference genome—there were 21 penA alleles. One of these alleles, the mosaic penA XXXIV (termed mosaic because sequence analysis suggests previous recombination between N gonorrhoeae and another Neisseria species), was present in 115 of 118 cefRS isolates; one of the remaining cefRS isolates (GCGS009) contained a distinct novel allele—termed mosaic penA XLI—with a sequence in the transpeptidase domain that most closely matches N meningitidis, suggesting an interspecies mosaic. By contrast, no mtrR, porB, pilQ, or ponA alleles were clearly associated with reduced susceptibility to cefixime (appendix pp 7–11). Of the two isolates initially reported as cefRS but lacking a mosaic penA, retesting showed that these isolates were, in fact, susceptible, with MICs of 0·03 μg/mL and 0·06 μg/mL (appendix p 21). Of the isolates with mosaic penA XXXIV, seven were susceptible; on rechecking, five had minimum inhibitory concentrations of 0·125 μg/mL (one-fold dilution below the cutoff of 0·25 μg/mL we used to define reduced susceptibility), and two were unchanged. An additional isolate possessing mosaic penA XXXIV with a Glu101Asp mutation was also reported susceptible, and on rechecking had an MIC of 0·125 μg/mL (appendix p 21). These results indicate that penA genotype is a reliable indicator of cefRS status for this dataset. That the association is not perfect suggests the possibility of epistasis between penA and other factors in determining cefixime MIC. Comparison of penA alleles and the whole-genome phylogeny revealed three independent acquisitions of mosaic penA XXXIV (figure 1), two of which were associated with large clusters of cefRS isolates and one of which was present only once.

Within cefRS cluster 1, nine isolates within a single clade seemed to revert from cefRS to cefixime susceptibility (eight with an MIC of 0·015 μg/mL, one with an MIC of 0·030 μg/mL; figure 1C). Evidence for a reversion event is based on the presence of an isolate at the base of this clade (figure 2); additional isolates from this part of the tree will help clarify the nature and timing of this event. The proposed reversion is associated with switching from mosaic penA XXXIV to XXXVIII, within a roughly 17 kb recombination of the dcw gene cluster that encompasses the penA locus. The mosaic penA allele XXXVIII has been reported to be associated with cefRS, and also with cephalosporin susceptibility. If this allele is truly associated with both resistance and susceptibility, then there is probably at least one gene or pathway in epistasis with the penA locus that dictates the degee of susceptibility to cephalosporins.

The cefixime-susceptible isolates selected to pair with the cefRS isolates were most commonly selected from GISP for storage at the CDC because they possessed an antimicrobial resistance trait of interest. These isolates therefore offer another opportunity to study gonococcal populations with resistance to other antibiotics (appendix p 5). Among the cefixime-susceptible isolates was a cluster of isolates with high-level resistance to tetracycline; these isolates all carried tetM.

Another cluster of isolates had reduced susceptibility to azithromycin (aziRS; MIC ≥2·0 μg/mL) and were distributed widely across the USA (appendix). 25 of 27 aziRS isolates (including those outside of the cluster) contained a C2599T mutation (N gonorrhoeae numbering)—C2611T (Escherichia coli numbering)—in at least two of the four copies of 23S rRNA. This mutation is associated with high MICs, as was the case for this set of isolates (2·0–16 μg/mL). All of the isolates with MICs of less than 2·0 μg/mL seem to have wild-type 23S rRNA at each of the four loci. The two aziRS isolates without the 23S rRNA C2599T mutation did not share other 23S rRNA mutations.

Discussion

We present the first genomic epidemiology study of gonococcus (panel). Reduced susceptibility to cefixime in the USA during the study period was mainly attributable to the spread and diversification of two lineages, each of which acquired cefRS through a mosaic penA XXXIV allele. We report two other individual cefRS isolates, one of which also has independently acquired the mosaic penA XXXIV allele and the other with a novel mosaic penA allele. How quickly and widely the gonococcal populations represented by these two isolates have propagated in the USA since their introduction is unclear, but can be directly assessed with analysis of isolates from samples after 2010. The larger, more diverse cefRS cluster (cefRS cluster 1) shows a pattern of geographical spread (mainly eastward from California) and predominantly among men who have sex with men, with a small number of introductions into heterosexual networks. This eastward spread is consistent with a model in which resistance is introduced or emerges in the west coast, and then spreads along sexual networks that connect the west coast to the rest of the USA.

The Bayesian model suggests that the most recent common ancestor of the cefRS cluster 1 lineage emerged in the late 1990s. Our data cannot establish whether this emergence occurred in the USA or elsewhere. Our estimate for the most recent common ancestor of cluster 1 coincides with the first clinical identification of cefRS isolates in Japan, although the initial dissemination of cefRS seems mainly to be a result of the spread of isolates with MLST 7363 and penA X. Additional studies of global collections of gonococcal specimens will better define the timing and geography of the origins of the cluster 1 lineage.

We also show that the presence of a mosaic penA XXXIV has a high positive predictive value for increased MIC for cefixime, suggesting that genotyping for penA could be clinically useful. Furthermore, we show that a cluster of isolates with reduced susceptibility to azithromycin is geographically widespread, and that one cluster of cefRS is ciprofloxacin-susceptible (cefRS cluster 2) and one cluster is predominantly ciprofloxacin-resistant (cefRS cluster 1). These findings should be considered when assessing strategies that use azithromycin to contain the spread of cephalosporin-resistant gonococcus and strategies that explore reintroduction of ciprofloxacin treatment based on rapid molecular testing for ciprofloxacin susceptibility.

Slowing or containing the spread of antibiotic-resistant N gonorrhoeae might depend on improved surveillance and identification of the reservoir of asymptomatically infected individuals. Our observations might provide the basis for development of such public health interventions. First, local circulation of isolates—as in the cluster in Los Angeles—suggests that whole-genome sequencing might help to identify and control outbreaks, augmenting and directing contact tracing and active case finding. Second, the identification of populations at risk for resistant gonococcal infection supports increasing screening and surveillance for both symptomatic and asymptomatic disease in these populations.

The possible reversion to cefixime susceptibility from reduced susceptibility raises a question about what is driving the spread of cefRS isolates in the USA and whether it could be a result of a process other than selection for isolates with a reduced susceptibility to extended-spectrum cephalosporins. Studies of the relative biological fitness of these N gonorrhoeae clones are needed to address this question. Detection of a reversion event would not be possible with previous gonococcal typing methods and phylogenetic reconstruction using whole-genome sequencing could provide insights into the dynamics of gonococcal genome rearrangement.

Our study has several limitations. As a sentinel surveillance system, the GISP protocol obtains samples only from men with urethritis at select public sexually transmitted infection clinics across the USA. Bias can be introduced by the populations that visit these clinics. Although the strong positive predictive value of mosaic penA XXXIV in populations with different sexual behaviour suggests that this finding is probably generalisable within the USA, the international spread of a cefRS lineage with MLST 7363 and mosaic penA X, which we did not detect, suggests that an international collection of samples should be assessed. As a sample of the gonococcal populations in men attending public sexually transmitted infection clinics, our data are probably representative. Although extrapolating our findings to the population at large requires caution given the sampling method, future studies that incorporate data from private sexually transmitted infection clinics will reveal how much sampling did affect our findings.

The rise of resistance to extended-spectrum cephalosporins, including the emergence of ceftriaxone-resistant gonorrhoea, warrants consideration of novel approaches to prevent or prepare for the possibility of untreatable gonorrhoea. The use of genomic epidemiology techniques for surveillance, outbreak analysis, and development of diagnostic tests could provide a crucial aid to the design and implementation of evidence-based strategies to slow or contain further spread.