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
Resistant and sensitive strains compete for the same hosts
Resistant and sensitive strains compete for the same hosts competition
sensitive bacteria resistant bacteria Resistant and sensitive strains compete for the same hosts frequency of resistance 1.0 0.8 0.6 0.4 0.2 0.0 weak cost strong cost 2000 2005 2010 2015 Time
Frequency of resistance is intermediate and stable Streptococcus pneumoniae, PEN frequency resistance PEN 0.0 0.1 0.2 0.3 0.4 0.5 France Spain United Kingdom Germany Italy 2000 2005 2010 Year ECDC data
2000 2005 2010 0.0 0.1 0.2 0.3 0.4 0.5 Streptococcus pneumoniae, ERY Year frequency resistance ERY France Spain United Kingdom Germany Italy Frequency of resistance is intermediate and stable ECDC data
Frequency of resistance is intermediate and stable Escherichia coli, CTX frequency resistance CTX 0.0 0.1 0.2 0.3 0.4 0.5 France Spain United Kingdom Germany Italy 2000 2005 2010 Year ECDC data
Frequency of resistance is intermediate and stable Escherichia coli, CIP frequency resistance CIP 0.0 0.1 0.2 0.3 0.4 0.5 France Spain United Kingdom Germany Italy 2000 2005 2010 Year ECDC data
Frequency of resistance is intermediate and stable Streptococcus pneumoniae 2014 frequency resistance to ERY 0.0 0.2 0.4 0.6 0.8 1.0 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 0 1 2 3 4 5 consumption of macrolides in Community ECDC data
Frequency of resistance is intermediate and stable Escherichia coli 2014 frequency resistance to CIP 0.0 0.2 0.4 0.6 0.8 1.0 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 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 consumption of quinolones in Community ECDC data
Frequency of resistance is intermediate and stable Escherichia coli 2014 frequency resistance to CIP 0.0 0.2 0.4 0.6 0.8 1.0 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 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 consumption of quinolones in Community ECDC data
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
Short-term fluctuations in the frequency of resistance Israel, 1401 bacterial isolates from children with acute otitis media 1.0 resistance penicillin 0.8 0.6 0.4 0.2 0.0 1999 2000 2001 2002 2003 2004 year Dagan et al. JID 2008
Short-term fluctuations in the frequency of resistance total > 200,000 prescriptions (~ 20% prescriptions in the area) monthly antibiotic prescription per 1000 children 250 200 150 100 50 0 amoxicillin amoxicillin clavulanate cephalosporins azithromycin 1999 2000 2001 2002 2003 2004 year Dagan et al. JID 2008
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 1.0 107 0.8 0.6 108 0.4 0.8 0.6! =!!!! + 0.2 0.0 0.4 0.2 0.0 1999 2000 109 250 rewrite equation (2) as: resistance penicillin resistance penicillin 1.0 monthly antibiotic prescription per 1000 children!" 106!,! =!!!!! +!" 2001 2002 2003 2004!" where! =!" year 200 < 0. F amoxicillin amoxicillin clavulanate cephalosporins azithromycin 150!!!!!! 0 1999 50!!,!!! 100 2000 2001 2002 2003 2004!"2004 is positive, and!! =!! 1999 2000 2001 year year 2002 2003. The paramete
No stabilising force monthly antibiotic prescription per 1000 children 0 50 100 150 200 250 maximal increase at maximal use = 3 months lag 1999 2000 2001 2002 0 0.25 0.5 0.75 1 frequency of penicillin resistance n year
No stabilising force Strong stabilising force monthly antibiotic prescription per 1000 children 0 50 100 150 200 250 maximal increase at maximal use = 3 months lag 1999 2000 2001 2002 0 0.25 0.5 0.75 1 frequency of penicillin resistance monthly antibiotic prescription per 1000 children 0 50 100 150 200 250 equilibrium between direct and stabilising force = fluctuations are in phase 1999 2000 2001 2002 n 0 0.25 0.5 0.75 1 frequency of penicillin resistance year year
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 0.025 0.015 0.005 0.005 0.010 penicillin resistance erythromycin resistance multi drug resistance amo ceph azi amo ceph azi amo ceph azi antibiotic prescribed
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 0.025 0.015 0.005 0.005 0.010 penicillin resistance erythromycin resistance multi drug resistance amo ceph azi amo ceph azi amo ceph azi antibiotic prescribed
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 0.025 0.015 0.005 0.005 0.010 penicillin resistance erythromycin resistance multi drug resistance amo ceph azi amo ceph azi amo ceph azi antibiotic prescribed
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 0.025 0.015 0.005 0.005 0.010 penicillin resistance erythromycin resistance multi drug resistance amo ceph azi amo ceph azi amo ceph azi antibiotic prescribed
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 0.6 0.4 0.2 0.0 1999 2000 2001 2002 2003 2004 year
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 2002-2007 0.6 reduction in consumption -26.5% frequency of resistance 0.4 0.2 0.0 2000 2005 2010 2015 year ECDC data Sabuncu et al. PLOS Med 2009
What explains the coexistence of R and S strains?
What explains the coexistence of R and S strains?
What explains the coexistence of R and S strains? colonised by S colonised by R treated host untreated host
What explains the coexistence of R and S strains? on treatment
What explains the coexistence of R and S strains?
Computing the invasion fitness of each strain = exponential growth rate when rare (with scheme) Escherichia coli, CTX log10 frequency resistance CTX 0.0 0.2 0.4 0.6 0.8 1.0 France Italy 2000 2005 2010 Year
Expressions for the invasion fitness λ(s) = natural clearance+colonisation(untreated)+antibiotic clearance λ(r) = natural clearance+colonisation(untreated)+colonisation(treated)
Pathogen adaptation to a structured host population inter-class transmission = 0.001 average treatment rate per month 0.15 0.1 0.05 R only coexistence S only 0. 0. 0.05 0.1 0.15 transmission cost of resistance frequency of resistance 1 0.8 0.6 0.4 0.2 0 0. 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 average treatment rate per month
What explains the coexistence of R and S strains?
Pathogen adaptation to a structured host population inter-class transmission = 0.1 average treatment rate per month 0.15 0.1 0.05 R only coexistence S only 0. 0. 0.05 0.1 0.15 transmission cost of resistance frequency of resistance 1 0.8 0.6 0.4 0.2 0 0. 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 average treatment rate per month
Pathogen adaptation to a structured host population full inter-class transmission average treatment rate per month 0.15 0.1 0.05 R only τ low < τ < τ high S only 0. 0. 0.05 0.1 0.15 transmission cost of resistance frequency of resistance 1 0.8 0.6 0.4 0.2 0 0. 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 average treatment rate per month
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
Pathogen adaptation to a country-structured host population: example of erythromycin resistance in S. pneumoniae under macrolide consumption frequency 0.0 0.2 0.4 0.6 0.8 1.0 simulation data linear model Each year, inference of the two model parameters - cost of resistance - cross-country transmission rate median cost = 0.011 per month median migration rate = 0.0013 per month 0.000 0.005 0.010 0.015 0.020 treatment rate per month
Pathogen adaptation to a country-structured host population: example of erythromycin resistance in S. pneumoniae under macrolide consumption coefficient of determination 1.0 0.8 0.6 0.4 0.2 R2 simulation R2 linear model 2-parameters evolutionary model explains the data as well as the 2- parameters linear model drop in correlation in 2005-2010 -> impact of the PCV vaccine? 0.0 2000 2005 2010 2015 year
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
Acknowledgements Christophe Fraser Sonja Lehtinen Marc Lipsitch Ron Dagan