Model-based design of innovative treatment strategies to suppress antimicrobial resistance using collateral sensitivity
Linda B. S. Aulin(1), Apostolos Liakopoulos(2), Daniel E. Rozen(2), J. G. Coen van Hasselt(1)
1: Leiden Academic Centre for Drug Research, Leiden University, Leiden, The Netherlands. 2: Department of Microbial Biotechnology and Health, Institute of Biology, Leiden University, Leiden, The Netherlands
Objectives: Antimicrobial resistance (AMR) is a serious threat to public health. To alleviate this threat, innovative treatment strategies using available antibiotics are urgently needed. The phenomenon of collateral sensitivity (CS) may be utilized to achieve this goal. CS occurs when resistance to one antibiotic increases the sensitivity to another antibiotic [1] and is observed through a reduction in the minimum inhibitory concentration (MIC) of the second antibiotic. In vitro, CS has been shown to occur for multiple bacterial species and antibiotics [2–4]. As a result, it has been proposed that CS can be used to design combination treatments that can suppress resistance [3], but it remains unclear how CS-based treatment should be designed. For instance, it is unclear if CS should occur in a two-directional reciprocal fashion between two antibiotics, and what effect size is relevant to achieve clinical utility. To this end, we use a model-based approach to assess the potential of CS-based treatments to suppress AMR, integrating pharmacokinetic-pharmacodynamic (PK-PD) and principles of evolutionary dynamics. Specifically, we aimed to i) perform a systems analysis to derive drug- and pathogen-specific factors aiding optimal design of CS-based dosing schedules; ii) apply our modelling strategy to multiple isogenic S. pneumoniae strains with distinct fluoroquinolone (FQ) resistance mutations, integrating experimental fitness and MIC data, and clinical PK profiles.
Methods: Development of the modelling framework: A modelling framework was developed including components accounting for antibiotic PK-PD and population dynamics of bacterial growth and evolution of resistance. Bacterial population dynamics was described using a four-state stochastic hybrid ordinary differential equation model, where each state represents a bacterial subpopulation, including a wild type (WT) population, two single mutant subpopulations, and a double mutant subpopulation. Resistance evolution was modelled according to a stochastic process based on a binomial distribution informed by a mutation rate. Each subpopulation had unique antibiotic sensitivity based on the MIC and collateral effect size for a second antibiotic. Antibiotic sensitivities were incorporated in sigmoidal concentration-effect relationships[5]. The framework was implemented using the RxODE package[6].
Systems analysis of CS-based dosing schedules: We used the framework to systematically study how certain pathogen- and drug-specific parameters influence AMR evolution. We simulated different combination dosing regimens with sequential, cyclic, or simultaneous administration using two antibiotics, ABA and ABB, which were assumed to have identical PK and additive bacterial killing effects. Here, in the bacterial model, the two single mutant subpopulations represent resistance (10xMICWT) to ABA and ABB, respectively, while the double mutant subpopulation was resistant to both ABA and ABB. To understand how antibiotic with different types of PD, i.e., bactericidal vs bacteriostatic, and concentration- or time-dependent killing, impact AMR evolution we performed simulations varying the drug-specific parameters relating to the maximal effect and shape of the concentration-effect relationship. We studied the importance of CS reciprocity, effect magnitude (0-90% reduction of MIC), and how pathogen-specific factors including fitness cost (0-50% fitness cost) and mutation rate (10-9-10-6 mut./bac./h), influence the ability of CS to suppress the resistance for different dosing regimens.
Application for AMR evolution in S. pneumoniae: We applied our modelling framework to study the evolution of FQ resistance in S. pneumoniae. Here, the two single mutants represent FQ-resistant subpopulations through mutation in gyrA and parC, respectively, while the FQ-resistant double mutant harbours a mutation in both gyrA and parC. The framework was informed by experimentally derived pathogen-specific parameters of fitness cost and MIC, and drug-specific PD bacterial kill dynamics parameters fitted using nlmixr [7], and published PK models. PD interactions were included where relevant. We simulated and evaluated clinical ciprofloxacin (CIP) treatments, including monotherapy and combination therapies using antibiotics showing collateral effects (erythromycin, linezolid, or penicillin). All treatments were simulated for different combinations of specific gyrA and parC mutations, resulting in eight unique evolutionary trajectories, with corresponding unique sets of CS effect and fitness profiles.
Results: Systems analysis of CS-based dosing schedules: We find that the impact of reciprocal CS relationships on the probability of resistance (PoR) at end of treatment is dependent on the drug PD type and dosing regimen. Simultaneous or one-day cycling treatment schedules were most effective dosing regimens to suppress resistance. For these treatments, a CS effect of 50% fully suppressed the resistance when concentration-dependent antibiotics were used (-ΔPoR 5.2-42.8%). The effect of antibiotic concentration shows that CS-based treatments are most clinically relevant for antibiotics with a narrow therapeutic window. One-directional CS relationships, and not only reciprocal relationships, can be utilized in the design of CS-based treatment schedules. Overall, we found that both drug- and pathogen-specific characteristics must be considered when designing CS-based treatments.
Application for AMR evolution in S. pneumoniae: By simulating eight common evolutionary trajectories of FQ treated S. pneumoniae we could assess the variability in AMR development under clinically relevant treatments. CIP monotherapy failed to eradicate the WT bacteria and promoted resistance for either or both of the single mutants (gyrA and parC), depending on the trajectory. Treatments combining CIP with either linezolid or penicillin were successful for all trajectories. When CIP was administrated with erythromycin, the WT was eradicated but the PoR of the single mutants varied between trajectories (PoR 0-77 % and 0-80.4 % for gyrA and parC mutants, respectively). This suggests that the specific mutation, and the competition between mutants, can have a pronounced impact on resistance development during treatment.
Conclusions: In this analysis, we used modelling and simulation to systematically unravel drug- and pathogen-specific factors influencing optimal design of CS-based treatment strategies to suppress AMR. Our modelling approach addresses important open questions around the topic of CS that are not easily tested experimentally, and provides new insights regarding key design aspects of CS-based treatments, contributing to the unmet need toward innovative strategies to alleviate the threat of AMR. Finally, we show how our modelling framework can be used to translate experimental data to design CS-based antibiotic dosing schedules, as was demonstrated for FQ resistance in S. pneumoniae.
References:
[1] Baym M, Stone LK, Kishony R. Multidrug evolutionary strategies to reverse antibiotic resistance. Science (80- ) [Internet]. 2016;351(6268):aad3292–aad3292.
[2] Barbosa C, Römhild R, Rosenstiel P, Schulenburg H. Evolutionary stability of collateral sensitivity to antibiotics in the model pathogen Pseudomonas aeruginosa. Elife [Internet]. 2019 Oct 29;8:1–22.
[3] Imamovic L, Sommer MOA. Use of collateral sensitivity networks to design drug cycling\nprotocols that avoid resistance development. Sci Transl Med. 2013;5(204):204ra132.
[4] Gonzales PR, Pesesky MW, Bouley R, Ballard A, Biddy BA, Suckow MA, et al. Synergistic, collaterally sensitive β-lactam combinations suppress resistance in MRSA. Nat Chem Biol. 2015;11(11):855–61.
[5] Regoes RR, Wiuff C, Zappala RM, Garner KN, Baquero F, Levin BR. Pharmacodynamic Functions: a Multiparameter Approach to the Design of Antibiotic Treatment Regimens. Antimicrob Agents Chemother. 2004;48(10):3670–6.
[6] Fidler M, Hallow M, Wilkins J WW. RxODE: Facilities for Simulating from ODE-Based Models. R package version 1.0.6 [Internet]. 2021. Available from: https://cran.r-project.org/package=RxODE
[7] Fidler M, Wilkins JJ, Hooijmaijers R, Post TM, Schoemaker R, Trame MN, et al. Nonlinear Mixed-Effects Model Development and Simulation Using nlmixr and Related R Open-Source Packages. CPT Pharmacometrics Syst Pharmacol. 2019;8(9):621–33.
