Understanding the spread of de novo and transmitted macrolide-resistance in Mycoplasma genitalium

Epidemiological data

We searched Pubmed up to May 4, 2018. We used the medical subject headings Mycoplasma genitalium AND drug resistance, bacterial and found 67 publications. From these, two authors independently selected six studies for three countries that met the following criteria: country with multiple studies that reported on M. genitalium and macrolide resistance mutations, data for more than three years from the same region or the entire country, and use of different strategies to test and treat M. genitalium. For each country, we recorded the testing strategy and treatment regimen, year in which azithromycin was introduced for M. genitalium treatment, numbers of specimens with positive results for M. genitalium and the number with macrolide resistance mutations. We contacted study authors for additional information. For each year, we calculated the proportion (with 95% CI) of azithromycin-resistant M. genitalium.

Transmission model

We developed a deterministic, population-based compartmental model that describes the spread of drug resistant M. genitalium (Fig. 1, Table 1). The model consists of four compartments: susceptibles (S), people infected with a drug-sensitive strain of M. genitalium (I_S), and people infected with a drug-resistant strain of M. genitalium that was either acquired during treatment (I_A) or transmitted (I_T). Assuming a homogenous population without demography, the transmission dynamics can be described by the following set of ordinary differential equations (ODEs): (1)${\displaystyle \frac{dS}{dt}=-\beta S\left(I_S+I_A+I_T\right)+\gamma \left(I_S+I_A+I_T\right)+\left(1-\mu \right)\tau I_S,}$ (2)${\displaystyle \frac{d{I}_S}{dt}=\beta S{I}_S-\gamma I_S-\tau I_S,}$ (3)${\displaystyle \frac{d{I}_A}{dt}=\mu \tau I_S-\gamma I_A,}$ (4)${\displaystyle \frac{d{I}_T}{dt}=\beta S\left(I_A+I_T\right)-\gamma I_T,}$ where β is the transmission rate, which is assumed to be the same for both strains of M. genitalium. Both types of infections can clear naturally at rate γ. Patients receive treatment at rate τ. The treatment rate is defined as all occasions of treatment with a single 1g dose of azithromycin in a person infected with M. genitalium, either with or without symptoms. µdenotes the probability of de novo resistance emergence during treatment. The de novo emergence of resistance also implies that the treatment failed. We used the point estimate of the probability of de novo resistance emergence of 12% from Horner et al. (2018). For simplicity, we assumed that resistant infections only clear naturally, with no second-line treatment.

In the transmission model, drug-sensitive (I_S) and drug-resistant (I_A and I_T) M. genitalium strains compete for the same resource, i.e., the susceptible hosts (S). The rate at which the resistant strain replaces the sensitive strain can be expressed by the difference in their net growth rates (Δϕ) (Bonhoeffer, Lipsitch & Levin, 1997; Fingerhuth et al., 2016): (5)${\displaystyle \Delta \varphi ={\varphi }_{A+T}-{\varphi }_S=\frac{\frac{d{I}_A}{dt}+\frac{d{I}_T}{dt}}{I_A+I_T}-\frac{\frac{d{I}_S}{dt}}{I_S}=\left(\beta S-\gamma +\frac{\mu \tau I_S}{I_A+I_T}\right)-\left(\beta S-\gamma -\tau \right)=\tau \left(1+\frac{\mu I_S}{I_A+I_T}\right)=\tau \left(1+\frac{\mu \left(1-p\right)}{p}\right),}$ where p denotes the proportion of resistant infections among all infections. Note that Δϕ does not depend on the transmission rate β or the natural clearance rate γ, i.e., is unaffected by the overall prevalence of M. genitalium.

Model parameters

We set the natural clearance rate (γ) of M. genitalium to 0.8 y⁻¹ (Smieszek & White, 2016). We calibrated the transmission rate β to 0.816 person⁻¹ y⁻¹, which results in an equilibrium prevalence of 2% in the absence of treatment and is consistent with estimates of the prevalence of M. genitalium in sexually active adults in high-income countries (Baumann et al., 2018). The values for the transmission rate and the natural clearance rate, and correspondingly the initial prevalence, do not govern the relative growth rate of the drug-resistant proportion (Δϕ), so they do not influence the relative prevalence of resistant infections or the model fits and parameter estimates. We did not find any published evidence of the effect of macrolide resistance on the fitness of M. genitalium strains, so we assumed that any fitness reduction is negligible and that resistant and wild-type strains have the same infectivity. The probability of emergence of de novo resistance during treatment (µ) was set to 12%, as reported in the meta-analysis by Horner et al. (2018).

Model fitting and simulations

We fitted the transmission model to country-specific resistance data to obtain maximum likelihood estimates of the treatment rate of infected people, τ, and the time point T for the introduction of azithromycin. Given a model-predicted proportion of resistant strains $p_i=\frac{I_A\left(i\right)+I_T\left(i\right)}{I_S\left(i\right)+I_A\left(i\right)+I_T\left(i\right)}$ in year i (Table 2), the binomial log-likelihood to find k_i resistant samples in N_i tested individuals is (6)${\displaystyle L\left(\tau ,T\right)=\sum \left(log\frac{N_i}{k_i}+k_{i}log{p}_i+\left(N_i-k_i\right)log\left(1-p_i\right)\right).}$

Simulations start at time T with 98% uninfected people, 2% people with drug-susceptible infections and no drug-resistant infections. We used log-transformed parameters for the estimation and stipulated that the lower and upper limits of T could not be before 1990 or after the time point when resistance was first observed. We derived simulation-based 95% CIs for the model curve from 10,000 bootstrap samples from the multivariate normal distribution of the two parameters using the R package mvtnorm. We used the ode function from the package deSolve to solve the ODEs, and the mle2 function from the package bbmle using the Nelder–Mead method for log-likelihood optimization.

To investigate the influence of the level of de novo resistance emergence on the rapid rise in the proportion of resistant infections, we simulated two alternative scenarios. In these scenarios, we kept the model-derived maximum likelihood estimates of τ and T but set the probability of de novo resistance emergence to lower values (µ= 1% and µ= 0.1%).

Description of the data

We included six studies that provided data about the proportion of azithromycin-resistant M. genitalium infections over time and the management of M. genitalium infection in France (Chrisment et al., 2012; Touati et al., 2014; Le Roy et al., 2016; Le Roy et al., 2017), Denmark (Salado-Rasmussen & Jensen, 2014), and Sweden (Anagrius, Loré & Jensen, 2013) (Table 2). Study authors provided additional information from Denmark (data disaggregated by year) and Sweden (numbers of patients per year and unpublished data for 2012 and 2013).

In France, we included four studies with data from 542 samples from 2003 to 2016 (Chrisment et al., 2012; Touati et al., 2014; Le Roy et al., 2016; Le Roy et al., 2017). None of 17 M. genitalium positive specimens from 2003 to 2005 contained macrolide resistance mutations. From 2006 onwards, mutations were detected in 8% to 17% of specimens tested in each year. In France, azithromycin was introduced for first line treatment of NGU in the 1990s (Joly-Guillou & Lasry, 1999). For Denmark, one study reported nationwide data from 1,008 patients with M. genitalium detected from 2006 to 2010, with 27% to 42% of specimens containing macrolide resistance mutations (Salado-Rasmussen & Jensen, 2014). In Denmark, 1g single dose azithromycin is routinely prescribed for treatment of NGU; erythromycin was the first-line treatment before azithromycin became available. An extended azithromycin regimen is prescribed if a M. genitalium infection was diagnosed and NAAT for detection of M. genitalium infections have been available since 2003 (Salado-Rasmussen & Jensen, 2014). In Sweden, we analyzed one study with data about macrolide resistance mutations from 408 samples obtained from 2006 to 2013 from patients at a single clinic in Falun (Anagrius, Loré & Jensen, 2013). Macrolide resistance mutations were first detected in a single specimen in 2008 and increased to 16% of 95 specimens in 2011. In Sweden, doxycycline is used as first line treatment for NGU (Björnelius, Magnusson & Jensen, 2017). Azithromycin is used only when M. genitalium is identified as the cause, with testing introduced in the 2000s (Anagrius, Loré & Jensen, 2013).

Mathematical modeling

The transmission model fitted the increase in M. genitalium resistance in France, Denmark and Sweden well (Figs. 2A–2C). The model estimated treatment rates of infected people and dates of introduction of azithromycin were: France, treatment rate of 0.04 y⁻¹ (95% CI [0.03–0.04] y⁻¹), introduction of azithromycin in 1990 (95% CI [1990–2006]); Denmark, treatment rate of 0.13 y⁻¹ (95% CI [0.05–0.34] y⁻¹), introduction of azithromycin in 1995 (95% CI [1990–2006]); Sweden, treatment rate of 0.14 y⁻¹ (95% CI [0.11–0.18] y⁻¹), introduction of azithromycin in 2006 (95% CI [2005–2007]). A treatment rate of 0.14 y⁻¹, such as in Sweden, corresponds to a proportion of 1 − e^−0.14 ≈ 13% of infected individuals that will have received treatment after one year. If treatment with single-dose azithromycin continues at the estimated rates, macrolide-resistant M. genitalium infections will reach 25% (95% CI [9–30]%) in France, 84% (95% CI [36–98]%) in Denmark and 62% (95% CI [48–76]%) in Sweden by 2025.

The importance of de novo resistance emergence for the early spread of macrolide-resistant M. genitalium becomes apparent in the alternative scenarios. Lower probabilities of de novo resistance, at the same estimated treatment rates and time points for the introduction of azithromycin as in the main model, would have resulted in considerably lower proportions of resistant infections (Figs. 2A, 2B and 2C). The influence of de novo resistance emergence on the rate of resistance spread can be explained by Eq. (5) (Fig. 3). As long as the proportion of resistant infections (p) is low, the contribution of de novo resistance emergence (µ) to the rate at which the resistant strain replaces the susceptible strain (Δϕ) is high. With increasing levels of the resistant strain, its growth advantage diminishes and slowly approaches Δϕ = τ, i.e., the spread of resistant infections will mainly be driven by transmitted resistance. This transition is depicted in Figs. 2D, 2E and 2F. At the time of introduction of azithromycin, the proportion of de novo resistance started at 100% and subsequently dropped in France, Denmark and Sweden. Since around 2015, the proportion of de novo resistance among all circulating macrolide-resistant M. genitalium infections has been low in all three countries.