Model-based characterisation of antibiotic activity in the presence of immune response: towards improved preclinical-to-clinical translation
Raphaël Saporta1, Elisabet I. Nielsen1, Annick Menetrey2, Natália Tassi3, David R. Cameron2, Veronica Biordi3, Olga Ticha3, Aghavni Ginosyan1, Irena Loryan1, Isabelle Bekeredjian-Ding3, Bernhard Kerscher3, Valérie Nicolas-Metral2, Lena E. Friberg1
1Department of Pharmacy, Uppsala University, 2Translational Medicine Department, Debiopharm International SA, 3Division of Infectiology, Paul-Ehrlich-Institut
Objectives The increasing burden of antimicrobial resistance (AMR), with an estimated 4.71 million deaths associated with bacterial AMR in 2021 [1], calls for developing new antibiotics and optimising the use of existing ones. Dosing regimen selection in antibiotic development relies on comprehending antibiotic pharmacokinetic-pharmacodynamic (PKPD) properties through in vitro and in vivo systems [2], notably neutropenic mouse thigh and lung infection models [3]. Since preclinical antibiotic PKPD studies are generally conducted in neutropenic conditions, interactions between the immune system and antibiotics and their quantitative contribution to bacterial killing in immunocompetent conditions have seldom been evaluated [4-6]. Translation of antibiotic activity across preclinical systems and to clinical efficacy is typically performed using PK/PD indices. However, this approach has limitations, such as its reliance on summary PK and PD metrics [7]. Model-informed drug development approaches, considering disease mechanisms and describing the time course of antibiotic effects, have gained attention as a potential alternative [8]. This work comprises analyses for two antibiotics: the FabI-inhibitor afabicin and the carbapenem meropenem. These analyses aimed to characterise the antibiotic PKPD in preclinical settings and quantify the contribution of the immune response and antibiotics to bacterial killing in immunocompetent conditions. Additionally, the capacity of model-based approaches to translate antibiotic activity from in vitro to in vivo and subsequently to clinical settings was evaluated for afabicin. Methods In analyses for both antibiotics, the base bacterial model structure assumed a drug-susceptible and growing state, and a non-susceptible and dormant state [9]. Preclinical data originating from drug development, evaluating afabicin against Staphylococcus aureus strains, were analysed in a PKPD modelling framework. Data from different settings were integrated step by step as follows. First, bacterial count data from 48-hour in vitro static time-kill curves (n=162, 21 strains) were used for PKPD model development. Strain differences in drug effect were assessed by scaling parameters with the minimum inhibitory concentrations (MICs). The PKPD model was applied jointly with an afabicin PK model for mice to predict bacterial dynamics in 74-hour neutropenic mouse thigh infection studies (n=952 mice, 9 strains). In vitro time-kill and neutropenic mouse data were then analysed jointly to quantify any differences between experimental settings. Subsequently, immunocompetent mouse thigh infection data (n=819 mice, 4 strains) were included, and the model was expanded to characterise the bacterial killing by the immune response and afabicin in immunocompetent mice. Predictions of bacterial dynamics in immunocompetent patients receiving dosing regimens tested in clinical trials [10] were performed from the final model combined with an afabicin PK model developed from data in humans. Meropenem was investigated against one Klebsiella pneumoniae strain (MIC=0.032 mg/L) in a mouse lung infection model. Meropenem PK following a subcutaneous (SC) administration of 40 or 300 mg/kg was monitored over 4 hours in immunocompetent and neutropenic mice (n=60). Bacterial counts were measured over 24 hours during SC meropenem treatment of 40 or 300 mg/kg every 4 hours in neutropenic, intermediately immunosuppressed, or immunocompetent mice (n=180). A PK model for meropenem was developed, exploring differences in parameters based on the immune status. The bacterial killing mediated by the immune response and meropenem was then quantified in a PKPD model. Dose-fractionation studies were simulated using the final model to investigate meropenem dose-response across immune states. Results Afabicin activity in vitro was implemented as a killing rate with a sigmoid Emax model, with compartments for bacterial adaptation under drug exposure to describe regrowth. Differences in drug effect between strains were estimated as a scaling of EC50 with MICs. The model for in vitro data could adequately translate to in vivo and across strains, predicting bacterial counts in neutropenic mice within 1 log10 of most observations. The same model structure was applied in neutropenic mice, with a 39-45% lower EC50 estimate than in vitro. The bacterial killing by the immune system was characterised as a phagocytosis process followed by the digestion of phagocytosed bacteria, linked with neutrophil dynamics fixed from the literature [4]. Phagocytosis was quantified as saturable, with an increase in the ratio of phagocytosed bacteria to neutrophils reducing the phagocytosis rate. Differences in afabicin activity in immunocompetent conditions were explained by a reduced Emax and a 70% lower EC50 compared with neutropenic mice. Predictions from the final model suggested bacterial killing in >90% of immunocompetent patients for the tested afabicin regimens, in agreement with the early clinical response rate observed in a Phase II trial [10]. A one-compartment model with first-order absorption described meropenem PK. Observed differences in concentration-time profiles were best characterised by a higher volume of distribution in immunocompetent mice (1.10 L/kg) as compared with neutropenic mice (0.694 L/kg). As found for afabicin data, bacterial killing by the immune system in intermediately immunosuppressed and competent mice was implemented as a phagocytosis and digestion process. Meropenem activity was described by the same Emax model (Emax=0.953 h-1, EC50=5.10 mg/L) for all immune states. However, the relative contribution of meropenem to bacterial killing was reduced in immunocompetent conditions, explained in the model by a lower fraction of bacteria affected by meropenem. Simulations showed a reduced meropenem dose-response gradient in immunocompetent mice, indicating that, similarly to afabicin, a near-maximal bacterial killing effect would be reached at lower doses in immunocompetent conditions. Conclusion Model-based approaches were developed to leverage the insights gained from preclinical studies, simultaneously quantifying antibiotic activity and the immune response over time. The PKPD model for afabicin exhibited translational capacity across strains and experimental settings, underscoring the potential of modelling to inform study design, decision-making, and ultimately increase the efficiency of antibiotic drug development. Using modelling methods, differences in PK and/or effects of antibiotics in immunocompetent conditions were identified, with a relatively reduced contribution of antibiotics to bacterial killing and lower doses required to attain near-maximal efficacy. By quantifying the antibiotic-driven bacterial killing over time while considering the immune system's role, model-based approaches could improve translation and dose selection for antibiotics.