Our results provide support for both the bimodal (type A) and the general shift (type B) of MIC distributions over time. The changes in MIC distribution for ceftriaxone fit best with a type B shift where the key change in MIC distribution has been an increase in the prevalence of isolates with higher MICs and a commensurate decrease but not elimination of isolates with low MICs. This is most clearly seen in South Africa where in 1995, 100% of isolates had a MIC of £0.007 mg/L, but by 2013 this had been reduced to 77%. In all three countries, where this could be assessed, the gonococci with the lowest MIC assessed were not eradicated over time.
The changes in MIC distribution for benzylpenicillin fit better with a type A shift. In Boston, for example, the pre-1947, pre-penicillin isolates are approximately normally distributed around a MIC of 0.003 mg/L. By 1949, a new sub-population of more resistant isolates has emerged centering on a MIC of 0.015 mg/L (type A, stage 1). By 1987, the USA distribution is approximately normal once again but now centered around a MIC of 0.125 mg/L (type A, stage 2). Only 1.5% of the 1987 isolates had a MIC of £0.012 mg/L.
Similarly, in Denmark, the introduction of penicillin is followed by a new bimodal distribution (type A, stage 1). The earliest benzylpenicillin MIC distributions from South Africa (1976) and the UK (1957) are from the penicillin period and demonstrate bimodal distributions. The later distributions from these countries are right shifted compared to the earlier period and in the case of South Africa, exhibit a new bimodal distribution including a new population centered on a MIC of 8 mg/L (type A, stage 3). The earliest data from Japan (1968) do not demonstrate a bimodal peak but a unimodal distribution centered around a high MIC of 0.5 mg/L. By 2009, similar to South Africa, bimodality has emerged as a result of the emergence of a new population with MICs around 1 mg/L (type A, stage 3).
The results for azithromycin are more mixed. South Africa’s trajectory fits best with type A, stage 2 – the whole MIC distribution is right shifted. In the UK and USA on the other hand, the distribution has been stretched to the right and thus best characterized as type B shifting.
The different trajectories of azithromycin, ceftriaxone and benzylpenicillin may be due to the different length of time they were observed for. This hypothesis is compatible with the fact that benzylpenicillin which has the longest follow-up data available exhibited the most marked changes in MIC distribution. As longer follow-up data become available for ceftriaxone and azithromycin they may follow a similar trajectory to benzylpenicillin. Formally defining type A and B shifts was beyond the scope of this analysis. We therefore acknowledge that our classification of MICs shifts into these types will remain open to debate until larger datasets are analyzed that enables a more formal classification of how MICs change over time per species. A further considerable limitation of these analyses is that changes in MIC distributions between studies may reflect differences in ascertainment of susceptibility. An example of this is a change of supplier of the agar used for susceptibility testing in the UK’s GRASP in 2015, which was shown to lead to slight increases in MICs for benzylpenicillin, ceftriaxone and azithromycin . These limitations would not however apply to the benzylpenicillin analyses from Denmark where the isolates from both periods were tested in parallel using the same methodology. Further limitations include the fact that we used the susceptibility data as they were reported and did not assess if publication or selection biases were present or absent. In particular we did not assess if changes in the population sampled may have influenced the results. For example, if subpopulations with higher rates of AMR were more likely to be included in the later dates this may lead to a right shifting of MIC distributions. The potential for this bias is particularly relevant in three of our comparisons: 1. Benzylpenicillin, Japan 1968 versus 2009-2010. In 2009 the samples were form men attending an STI clinic whereas the samples from 1968 were from navy personnel who may have been more likely to have obtained their infections from sex workers who have been shown to have more resistant strains in some studies , 2. Benzylpenicillin, the USA 1945-1949 versus 1987-onwards. The 1940s isolates were from men and women versus the post 1986 isolates which were all from men. 3. Azithromycin, the UK, 2001 versus 2015. The proportion of samples that were from MSM was considerably higher in 2015 which may been responsible for the right shifting of the MIC distribution [34, 35]. We were also unable to assess if unreported quality issues may have affected the accuracy of the data. Other limitations include that our literature search may have missed published reports, particularly in the early period when the indexing of articles was suboptimal. A further problem was that our analysis was limited to three antimicrobials in four countries, which limits generalizability. Finally, our analysis is descriptive and not quantitative.
Our findings from the pre-penicillin period are however congruent with those from 16 other studies which found that gonococci in this period were highly susceptible to penicillin (reviewed in ). Likewise, our findings of a bimodal penicillin MIC distribution emerging soon after significant penicillin exposure are similar to those in multiple populations around the world [36-38]. There is also evidence from other populations not included in our review of the emergence of a second bimodal peak in the penicillin MIC distribution in more recent decades [39, 40]. These studies thus provide some support for our finding that reductions in susceptibility to penicillin, at least in certain populations, are best characterized by type A shifts that evolve through up to 3 stages. By the third stage, benzylpenicillin MIC distributions such as those from Japan and South Africa were so right-shifted that no or very few isolates have MICs £0.03 mg/L. Their distribution no longer overlaps with that of the pre-penicillin population. How can we explain this?
What mechanisms underpin the decline/extinction of the lowest MIC populations?
We propose that 3 mechanisms may be relevant: 1) The zero-sum nature of changes in distributions means that an increase in the proportion with high MICs necessarily entails a commensurate decrease in the proportion with lower MICs; 2) Sustained antimicrobial exposure may have led to a selective sweep of N. gonorrhoeae whereby highly sensitive strains were eliminated from the gene pool/local pangenome; 3) Sustained antimicrobial pressure may have selected for antimicrobial resistance in commensal bacteria, including commensal Neisseria spp. If this change is profound enough and there is sufficient DNA exchange with these commensals then this could result in the decline or elimination of low MIC strains in both commensal and pathogenic species.
The first mechanism is able to explain a decline but not the elimination of low MIC isolates. Support for the second mechanism comes from studies that have found associations between the intensity of antimicrobial consumption in the general population and homologous gonococcal AMR [6, 9]. Furthermore, studies have found that particular gonococcal genogroups can emerge and disseminate rapidly displacing other genogroups. For example, genogroup 1407, which is strongly associated with reduced cephalosporin susceptibility, emerged explosively in Europe around 2007 and rapidly displaced other more susceptible genogroups . In support of the third mechanism, recent receipt of ciprofloxacin, ceftriaxone or cefixime has been established as an independent risk factor for homologous AMR in oropharyngeal commensal Neisseria . The relative contributions of mechanisms 2 and 3 is likely to vary between different classes of antimicrobials. For example, horizontal gene transfer of resistance conferring genes from commensal Neisseria has been shown to have played an important role in cephalosporin and macrolide resistance but not for fluoroquinolone resistance .
The major relevance of this study is showing the utility of longer-term analyses of MIC distribution. Contemporary analyses of gonococcal AMR are frequently limited to the last decade or so [6, 9, 43]. These analyses will likely miss changes in MIC distribution that occur slowly over longer periods. In other fields (mainly environmental conservation) this short-term focus has led to what has been termed the shifting baseline syndrome, where in the absence of sufficient knowledge of historical conditions, members of each new generation accept the situation in which they are raised as normal . An important consequence of this syndrome is that it increases an individual’s tolerance of man-made environmental damage.
It is important to remember that the emergence of AMR in N. gonorrhoeae is not inevitable. Populations in countries such as the Netherlands with low rates of consumption of macrolides, quinolones and cephalosporins have correspondingly lower rates of gonococcal AMR than high consumption countries . The treatment of choice for gonorrhoea in the rural part of the Northern Territories, Australia, remains penicillin, because the prevalence of penicillin resistance here is still less than 1% .
These findings indicate that it is possible to contain the emergence of gonococcal AMR.
The fact that a unimodal distribution can re-emerge in stage 2 of a type A shift may compound the shifting baseline problem. Observers looking at a stage 2 MIC distribution in isolation may conclude that this represents a wild type distribution. Indeed, contemporary datasets are favored in the construction of epidemiological cutoffs (ECOFFs) for MICs . If one analyzes EUCAST’s ‘Antimicrobial wild type distributions of microorganisms’ gonococcal dataset with the EUCAST recommended software “ECOFFinder”, this generates an ECOFF for benzylpenicillin of 8 mg/L . Gonococcal isolates with MICs of over 1 have a high likelihood of non-response to benzylpenicillin therapy  which illustrates why it is not meaningful to try to find penicillin ECOFFs for contemporary gonococcal populations that have experienced right shifting of their MICs. Evaluating ECOFFs from the pre-penicillin populations would however be more appropriate. Following our historical methodology, the EUCAST ECOFF MIC distribution would be characterized as having undergone a type A, stage 3 shift. Likely because more recent datasets are favored in the EUCAST dataset, the older MIC distributions, such as those from 1945 (Fig. 3), are not included in this distribution.
Further studies could analyze what mechanisms underpin the elimination of highly susceptible strains, what the correlates are of populations that have undergone a stage 3 as opposed to a stage 1 or 2, type A shift in penicillin/other antimicrobial AMR. Can this all be explained by differential antimicrobial exposure? If so are there thresholds of consumption that populations should aim to not exceed ? Which gonococcal genotypes/genogroups have been driven extinct in populations with stage 2 and 3 shifts? Is this mirrored by a similar extinction in susceptible strains in commensal species? Finally, it may be useful to investigate if this process has played a role in the reduction in diversity of the gut microbiome in Western populations [46, 47].