Projecting the development of antimicrobial resistance in Neisseria gonorrhoeae from antimicrobial surveillance data: a mathematical modelling study

Neisseria gonorrhoeae has developed resistance to multiple antibiotic classes, leading to cases of ineffective treatment and the potential for untreatable gonorrhoea [1‐3]. The World Health Organization (WHO) estimates that there were more than 82 million new gonorrhoea cases among adults aged 15–49 years in 2020, globally, mostly in low and middle-income countries [4]. Untreated gonorrhoea can cause pelvic inflammatory disease, infertility, and ectopic pregnancy in women, epididymo-orchitis in men, and neonatal ophthalmia [5, 6]. Empirical treatment of people with symptoms suggestive of gonorrhoea, regardless of knowledge of the aetiological agent or antimicrobial resistance (AMR) profile, is encouraged to limit the further spread of N. gonorrhoeae and occurrence of complications. WHO suggests that the first-line regimen for empirical treatment in a given country should have a microbiological cure rate > 95% and should be changed when ≥ 5% of cases fail treatment or ≥ 5% of tested gonococcal isolates are resistant [7]. In many countries, AMR levels for previously and currently recommended first-line antimicrobials exceed 5% [8].

The injectable extended spectrum cephalosporin (ESC), ceftriaxone, is recommended for empirical treatment of suspected gonorrhoea in most countries [9]. The development of resistance to ceftriaxone and other extended-spectrum cephalosporins (ESC) is a complex process, most frequently involving multiple mechanisms (‘multi-step’ resistance) [10‐12]. Sporadic cases of treatment failure due to ceftriaxone resistance have been reported since 2009 from Japan, Australia, Sweden, Slovenia, Austria and the UK [13‐15]. However, surveillance systems, such as the Gonococcal Resistance to Antimicrobials Surveillance Programme (GRASP) in England and Wales, show that isolates above the cut-off for resistance to ceftriaxone remain rare and are may have been imported in association with travel [16]. The distribution of ceftriaxone minimum inhibitory concentrations (MICs) drifted upwards until 2018, but might have stabilised or reversed since 2019 [10, 16]. The overall number of reported gonorrhoea cases and number of GRASP isolates declined in 2020, during the coronavirus disease 2019 (COVID-19) pandemic [16]. MIC drift can be interpreted as a progressive accumulation of mutations in response to the selection pressure exerted by the continued use of ESCs, including the suboptimal use of oral ESCs, such as cefixime, which is expressed phenotypically as an increase in MICs [17]. There is no licensed antimicrobial available yet that could replace ceftriaxone once AMR levels exceed the 5% threshold.

Projections about the probable time when the 5% threshold for resistance might be reached could help public health agencies to improve strategies for dealing with AMR in N. gonorrhoeae. Mathematical modelling studies provide insights into the emergence and spread of AMR and have been used to model the emergence of and future spread of resistance in N. gonorrhoeae [17‐19] and Mycoplasma genitalium, and have also shown that higher levels of antibiotic treatment for N. gonorrhoeae result in faster development of AMR [17, 20]. The modelling studies of N. gonorrhoeae focused on resistance to ciprofloxacin, in which the presence of a single point mutation in the gyrA gene (GyrA 91F, often seen together with GyrA 95) can result in clinical resistance and treatment failure, so-called single-step resistance (2). The consideration of the development of AMR as a binary property [17‐19] may be an appropriate simplification for the study of resistance to antimicrobials like ciprofloxacin, but may be inadequate for antimicrobial classes such as ESCs with more complicated patterns of AMR development [10, 17, 21]. AMR surveillance systems that report the MIC for each tested isolate, and the antibiotic dispensed, provide more information about decreases in susceptibility than binary data about isolates with full resistance. The overall aim of this study was to model the development of AMR in N. gonorrhoeae by linking data from a national surveillance system that records MIC values and antibiotic treatment. The main outputs are projections for when the proportion of N. gonorrhoeae that is resistant to ceftriaxone will reach the 5% threshold, should current trends continue.

The GRASP data from 2000–2018 included results for 15,699 individual N. gonorrhoeae isolates from HMW and 12,368 isolates from MSM. Of these, we retained a total of 26,437 isolates with information on MIC for ciprofloxacin, 25,626 for azithromycin, 19,403 for cefixime and 19,424 for ceftriaxone. Additional file 1 shows detailed data about proportions of resistant isolates and MIC distributions and additional file 2 shows the use of each antibiotic.

In this study, we developed a mechanistic model to analyse and extrapolate detailed N. gonorrhoeae surveillance data about MIC drift and antibiotic treatment, linking the accumulation of resistance mutations to the selective pressure exerted by antibiotics. The multi-step model captured the progressive shift in the MIC values for ceftriaxone well. The drift in MIC values towards decreasing ceftriaxone susceptibility is a longstanding global concern [7, 10, 25]. The GRASP data included only one ceftriaxone-resistant isolate by 2018, so in the single-step model, which relies on the proportions, the projected level did not exceed the 5% AMR threshold at all. Extrapolating from the trend in MICs in the multi-step model predicts that ceftriaxone resistance will likely reach a proportion of 5% in MSM by 2030, if current trends continue, but with a wide credibility interval.

To our knowledge, this is the first mathematical modelling study that describes the sequential accumulation of resistance mutations and integrates data about MIC drift. The Bayesian approach allowed us to propagate the considerable uncertainty in model parameters and to provide meaningful projections of the expected growth in antibiotic resistant N. gonorrhoeae. There are limitations to the study. First, the approach assumes that N. gonorrhoeae resistance is solely driven by selection pressure from antibiotic usage within the country [17]. Whether de novo acquisition of resistance mutations due to treatment pressure is the main driver of MIC drift remains unknown. N. gonorrhoeae can acquire resistance mutations through horizontal gene transfer from reservoirs of AMR genes in commensal Neisseria spp. in the pharynx [1‐3]. The resulting mosaic penA allele, and other mutations e.g. in mtrR and porB1b confer intermediate resistance to N. gonorrhoeae and can together lead to higher level resistance [1, 2, 12]. There is some evidence from studies of Streptococcus pneumoniae, demonstrating the evolution of MIC drift to full resistance [26]. Second, the model does not incorporate the importation of resistant strains of N. gonorrhoeae from regions with higher frequencies of AMR, such as south east Asia [2, 27]. It can be argued, however, that intermittent import of resistant N. gonorrhoeae strains that result in treatment failure and spread rapidly would produce a sudden, rather than the observed gradual shifts in MIC, which would then decrease due to enhanced gonorrhoea control measures, clonal replacement or fitness costs. Increasing then decreasing ESC resistance has occurred, including the circulation of a multi-drug resistant clone, NG-MAST ST1407, originating in Japan in 2004 and subsequently spreading widely internationally [27‐31]. Introduction of new strains might thus add to the underlying, progressive accumulation of resistance mutations due to domestic treatment pressure. The temporary upward shift in MIC observed around 2009 and 2010 in MSM in the GRASP data supports this interpretation. In the sensitivity analysis in our model, removing observations from 2009 and 2010 led to an even earlier projection of increasing resistance in MSM because the 2009–2010 peak lowered the estimated slope of resistance. Further model developments could account for international spread, but the lack of representative data about AMR in many other countries makes this complicated. Third, our analysis includes GRASP data up to 2018 and the study was delayed by the COVID-19 pandemic. GRASP data collection continued and in 2020 there were no ceftriaxone-resistant samples and MIC values had decreased slightly [16]. The representativeness of the GRASP data might also have been affected by the pandemic conditions, given that the number of isolates and total number of diagnosed cases declined. Including data up to 2018 is therefore reasonable but, if a downward trend in MICs continues, the projected probability of AMR > 5% in the multi-step model would be reduced. Fourth, there are more GRASP collaborating centres in London than in other regions. If levels of gonorrhoea and AMR are higher in London, extrapolation of the rate of projected spread of AMR might be overestimated at the national level.

Our results rely upon several additional assumptions. First, increasing levels of AMR are assumed to lead to a linear decrease in treatment efficacy, although some non-linear effects are likely. A more abrupt decrease in treatment efficacy at the resistance threshold would lead to higher future levels of resistance. Second, we did not incorporate the effects of fitness costs and epistasis, which might be especially important when considering long-term resistance to antibiotics that are no longer in use such as ciprofloxacin. For this reason, we only propose long-term forecasts for ceftriaxone, assuming continuous use, as it is the net effect of treatment pressure and fitness costs that matters in this case. Third, we only considered antibiotics used to treat gonorrhoea; antibiotics may exert selective pressure when prescribed for other infections, which could partly explain persistent high levels of ciprofloxacin resistance [32]. Finally, we used deterministic models that ignore the role of stochastic events but accounted for the additional uncertainty by using over-dispersed probability distributions (beta-binomial and dirichlet-multinomial).

Our study and its findings have implications for future strategies for gonococcal prevention, treatment and surveillance. Our projections suggest a window of ten or more years before resistance to ceftriaxone exceeds 5% in England and Wales, which provides some time for planning strategies for continued antimicrobial stewardship and developing and evaluating new antimicrobials. New diagnostic tests might allow resistance-guided therapy for the 50–70% of isolates that are still susceptible to ciprofloxacin [33]. Phase 3 randomised clinical trials are evaluating newly developed antibiotics, including zoliflodacin, and existing antibiotics [34, 35]. Development of a gonococcal vaccine as primary prevention could delay the emergence of AMR if uptake is high enough, although the technical development pathway is even less advanced than that for new antimicrobials [36]. Our models could also be extended to explore the potential impact of new diagnostic and treatment strategies such as point-of-care tests or introduction of new antibiotics on the progression of AMR [37]. Considering the time required to develop, evaluate and approve new antibiotics and vaccines, our results highlight the urgent need for strategies that can delay the emergence and spread of AMR. Our models, which show the utility of long-term representative surveillance of national and international gonococcal AMR surveillance networks based on isolation and culture of Neisseria gonorrhoeae for susceptibility testing [38, 39], can be adapted for use in, and for comparison with, other countries.