The evolutionary epidemiology of antibiotic resistance evolution

Transcription

1 The evolutionary epidemiology of antibiotic resistance evolution François Blanquart, CNRS Stochastic Models for the Inference of Life Evolution CIRB Collège de France Quantitative Evolutionary Microbiology IAME UFR de médecine Paris Diderot

2 Resistant and sensitive strains compete for the same hosts

3 Resistant and sensitive strains compete for the same hosts competition

4 sensitive bacteria resistant bacteria Resistant and sensitive strains compete for the same hosts frequency of resistance weak cost strong cost Time

5 Frequency of resistance is intermediate and stable Streptococcus pneumoniae, PEN frequency resistance PEN France Spain United Kingdom Germany Italy Year ECDC data

6 Streptococcus pneumoniae, ERY Year frequency resistance ERY France Spain United Kingdom Germany Italy Frequency of resistance is intermediate and stable ECDC data

7 Frequency of resistance is intermediate and stable Escherichia coli, CTX frequency resistance CTX France Spain United Kingdom Germany Italy Year ECDC data

8 Frequency of resistance is intermediate and stable Escherichia coli, CIP frequency resistance CIP France Spain United Kingdom Germany Italy Year ECDC data

9 Frequency of resistance is intermediate and stable Streptococcus pneumoniae 2014 frequency resistance to ERY Sweden Finland Lithuania Slovenia Iceland Spain Denmark Netherlands Norway Latvia Germany Estonia Portugal Croatia France Luxembourg Belgium Hungary Austria Malta Bulgaria Poland United Kingdom Czech Republic Ireland Italy Slovakia consumption of macrolides in Community ECDC data

10 Frequency of resistance is intermediate and stable Escherichia coli 2014 frequency resistance to CIP Slovakia Ireland United Kingdom Denmark Netherlands Norway Sweden Poland Slovenia Germany Austria Czech Republic Latvia Estonia Finland Lithuania Iceland Croatia France Portugal Spain Greece Hungary Belgium Luxembourg Bulgaria Malta Italy consumption of quinolones in Community ECDC data

11 Frequency of resistance is intermediate and stable Escherichia coli 2014 frequency resistance to CIP Slovakia Ireland United Kingdom Denmark Netherlands Norway Sweden Poland Slovenia Germany Austria Czech Republic Latvia Estonia Finland Lithuania Iceland Croatia France Portugal Spain Greece Hungary Belgium Luxembourg Bulgaria Malta Italy model consumption of quinolones in Community ECDC data

12 The coexistence problem Lipsitch et al. Epidemics 2009 Colijn et al. JRSI 2009 Lehtinen et al PNAS 2017 Cobey et al JRSI 2017 Davies et al biorxiv 2018

13 Short-term fluctuations in the frequency of resistance Israel, 1401 bacterial isolates from children with acute otitis media 1.0 resistance penicillin year Dagan et al. JID 2008

14 Short-term fluctuations in the frequency of resistance total > 200,000 prescriptions (~ 20% prescriptions in the area) monthly antibiotic prescription per 1000 children amoxicillin amoxicillin clavulanate cephalosporins azithromycin year Dagan et al. JID 2008

15 103 The linearized system is given by the first order Taylor series ap!"!"!!!of!,resistance!)antibiotic + prescriptions!! + (!! to Model linking the104 frequency!!!!!,!!!!!!"! =!!!! +!,!!!!!,!!! Emerges from linearisation of any dynamical model around the overall average resistance!!!!!,! Two main drivers of the change in resistance Stabilising force 105(phenomenological) where! is the vector of the rates of antibiotic use. Becau Antibiotic selection!!"!!!!"!,! = 0;!! because!!,!!! this equilibrium is stable, Jewish children ! =!!!! rewrite equation (2) as: resistance penicillin resistance penicillin 1.0 monthly antibiotic prescription per 1000 children!" 106!,! =!!!!! +!" !" where! =!" year 200 < 0. F amoxicillin amoxicillin clavulanate cephalosporins azithromycin 150!!!!!! !!,!!! !"2004 is positive, and!! =!! year year The paramete

16 No stabilising force monthly antibiotic prescription per 1000 children maximal increase at maximal use = 3 months lag frequency of penicillin resistance n year

17 No stabilising force Strong stabilising force monthly antibiotic prescription per 1000 children maximal increase at maximal use = 3 months lag frequency of penicillin resistance monthly antibiotic prescription per 1000 children equilibrium between direct and stabilising force = fluctuations are in phase n frequency of penicillin resistance year year

18 Model linking the frequency of resistance to antibiotic prescriptions Strong stabilising force Inference of bj / c = change in the frequency of resistance caused by a one unit increase in the prescription of antibiotic j increase in frequency of resistance per additional prescription unit penicillin resistance erythromycin resistance multi drug resistance amo ceph azi amo ceph azi amo ceph azi antibiotic prescribed

19 Model linking the frequency of resistance to antibiotic prescriptions Strong stabilising force Inference of bj / c = change in the frequency of resistance caused by a one unit increase in the prescription of antibiotic j Amoxicillin selects for penicillin resistance only increase in frequency of resistance per additional prescription unit penicillin resistance erythromycin resistance multi drug resistance amo ceph azi amo ceph azi amo ceph azi antibiotic prescribed

20 Model linking the frequency of resistance to antibiotic prescriptions Strong stabilising force Inference of bj / c = change in the frequency of resistance caused by a one unit increase in the prescription of antibiotic j Amoxicillin selects for penicillin resistance only Azithromycin selects for penicillin, erythromycin and multidrug resistance increase in frequency of resistance per additional prescription unit penicillin resistance erythromycin resistance multi drug resistance amo ceph azi amo ceph azi amo ceph azi antibiotic prescribed

21 Model linking the frequency of resistance to antibiotic prescriptions Strong stabilising force Inference of bj / c = change in the frequency of resistance caused by a one unit increase in the prescription of antibiotic j Amoxicillin selects for penicillin resistance only Azithromycin selects for penicillin, erythromycin and multidrug resistance Cephalosporin counter-selects penicillin, erythromycin and multidrug resistance -> most cephalosporin-resistant strains are penicillin intermediate, a prediction verified in the data (Dagan personal communication) increase in frequency of resistance per additional prescription unit penicillin resistance erythromycin resistance multi drug resistance amo ceph azi amo ceph azi amo ceph azi antibiotic prescribed

22 The model explains 18 to 43% of temporal fluctuations in antibiotic resistance 1.0 training dataset prediction 0.8 R 2 = 0.38 resistance penicillin year

23 The stabilising force implies that any reduction in the consumption of antibiotic has immediate effect on resistance Frequency of penicillin resistance in France following nationwide prevention campaign in reduction in consumption -26.5% frequency of resistance year ECDC data Sabuncu et al. PLOS Med 2009

24 What explains the coexistence of R and S strains?

25 What explains the coexistence of R and S strains?

26 What explains the coexistence of R and S strains? colonised by S colonised by R treated host untreated host

27 What explains the coexistence of R and S strains? on treatment

28 What explains the coexistence of R and S strains?

29 Computing the invasion fitness of each strain = exponential growth rate when rare (with scheme) Escherichia coli, CTX log10 frequency resistance CTX France Italy Year

30 Expressions for the invasion fitness λ(s) = natural clearance+colonisation(untreated)+antibiotic clearance λ(r) = natural clearance+colonisation(untreated)+colonisation(treated)

31 Pathogen adaptation to a structured host population inter-class transmission = average treatment rate per month R only coexistence S only transmission cost of resistance frequency of resistance average treatment rate per month

32 What explains the coexistence of R and S strains?

33 Pathogen adaptation to a structured host population inter-class transmission = 0.1 average treatment rate per month R only coexistence S only transmission cost of resistance frequency of resistance average treatment rate per month

34 Pathogen adaptation to a structured host population full inter-class transmission average treatment rate per month R only τ low < τ < τ high S only transmission cost of resistance frequency of resistance average treatment rate per month

35 Pathogen adaptation to a country-structured host population Evolution of resistance under three forces 1. local selection for resistance under local antibiotic use 2. cost of resistance 3. cross-country transmission flights gravity model

36 Pathogen adaptation to a country-structured host population: example of erythromycin resistance in S. pneumoniae under macrolide consumption frequency simulation data linear model Each year, inference of the two model parameters - cost of resistance - cross-country transmission rate median cost = per month median migration rate = per month treatment rate per month

37 Pathogen adaptation to a country-structured host population: example of erythromycin resistance in S. pneumoniae under macrolide consumption coefficient of determination R2 simulation R2 linear model 2-parameters evolutionary model explains the data as well as the 2- parameters linear model drop in correlation in > impact of the PCV vaccine? year

38 Conclusions Antibiotic resistance is under balancing selection / negative frequencydependent selection and we do not fully understand why Host population structure helps maintain coexistence Fitness interactions with loci themselves under balancing selection also helps Lehtinen et al. PNAS 2017 Blanquart et al. Proc B 2017 Blanquart et al. J. Roy Soc Interface 2018

39 Acknowledgements Christophe Fraser Sonja Lehtinen Marc Lipsitch Ron Dagan