Body composition and drug exposure in adipose tissue are key for antibiotic dose-individualisation in obese individuals
David Busse (1,2), Philipp Simon (3,4), Robin Michelet (1), Niklas Hartung (5), Lisa Schmitt (1,2), Christoph Dorn (6), Hermann Wrigge (3,4,7), Wilhelm Huisinga (5), Charlotte Kloft (1)
(1) Dept. of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universitaet Berlin, (2) and Graduate Research Training program PharMetrX, (3) Department of Anaesthesiology and Intensive Care, University of Leipzig Medical Center, (4) Integrated Research and Treatment Center (IFB) Adiposity Diseases, University of Leipzig, (5) Institute of Mathematics, University of Potsdam, (6) University of Regensburg, Institute of Pharmacy, Regensburg, (7) Department of Anaesthesiology, Intensive Care and Emergency Medicine, Pain Therapy, Bergmannstrost Hospital Halle
Objectives: The introduction of model-informed precision dosing (MIPD) promises a shift from the “one-size-fits-all” paradigm in antimicrobial dosing. Yet, the lack of pharmacokinetic (PK) models, suitable to scale over a wide covariate range, like body mass, limits extrapolation to special patient populations, such as the obese [1]. Currently, alternative body size descriptors based on height and weight are used in pharmacometric analyses of antibiotics in obese patients. However, these descriptors lack a clear mechanistic justification, their use in covariate models is oftentimes based on small trials and different descriptors have been identified for the same drugs [2].
To allow for standardised allometric scaling, physiologically-motivated approaches based on lean body weight (LBW [3]) and a drug- and structural PK parameter-specific estimated fraction of FM have been proposed: (i) “Normal fat mass” (NFM) to scale volumes (V) and flows [4] and (ii) “LBW/FM” to scale flows and the central V with LBW, and a peripheral V with LBW and fat mass (FM=total body weight-LBW) [5]. Despite theoretical advantages of both approaches over other body size descriptors, they have found limited use in covariate model development and are absent in clinical guidances on antibiotic therapy.
This analysis leveraged densely sampled exposure data of 4 antibiotics in plasma and interstitial fluid (ISF) of adipose tissue from the largest clinical microdialysis trial to date (EudraCT No. 2012-004383-22). The aim was to identify if physiologically-motivated scaling approaches can explain PK differences between obese and nonobese individuals using pharmacometric approaches.
Methods:
Clinical study data
Exposure data in plasma and ISF of the adipose tissue collected via microdialysis (0-8 h) were available from 30 obese (BMImean±SD=47±9 kg/m2) and 30 non-obese patients (BMImean±SD=25±3 kg/m2) receiving a 30-min infusion of 1000 mg meropenem/600 mg linezolid or 8000 mg fosfomycin/4000 mg piperacillin + 500 mg tazobactam before abdominal surgery [6].
PBPK model development and model reduction by lumping
A whole-body permeability-limited PBPK model previously implemented in MATLAB (R2020b) [7] was expanded based on parameters extracted from patient anatomy and physiology, and compound-specific in vivo/in vitro experiments. Interindividual variability in anatomical and physiological PBPK parameters was based on [5]. Lumping of extra- and intracellular spaces per organ into compartments was based on inspecting the simulated antibiotic concentrations over time normalised by predicted concentration in the quasi-steady state elimination phase (t=8 h, Cmin) to identify kinetically comparable distribution spaces.
NLME model development
Plasma and ISF data were analysed via a nonlinear mixed-effects (NLME) approach. NFM and LBW/FM, respectively, were implemented via theory based allometric scaling (exponents 0.75 for flows and 1 for volumes) as exemplarily shown for the population peripheral volume (Vperi, pop) scaled to the individual i (Vperi, i) via individual and population median LBW and FM:
(1) Vperi, i=Vperi, pop·((LBWi+Ffat·FMi)/(LBWpop+Ffat·FMpop))1
(2) Vperi, i=Vperi, pop·((1-R)·LBWi/LBWpop+R·FMi/FMpop)1
In NFM (1) the scaling factor Ffat was estimated (i) separately for V and flows and (ii) simultaneously for all PK parameters. In LBW/FM (2) the scaling factor R was only estimated for Vperi and all other PK parameters were scaled by LBW.
Datasets were prepared in R 3.6.0 and analyses were performed in NONMEM 7.4.3 using the FOCE+I method, with assistance of PsN 4.6.0 and Pirana 2.9.6. Micro-/retrodialysis data were integrated via an integral-based modelling approach [8]. Sampling importance resampling was used to obtain estimates of parameter uncertainty [9].
Results: The a priori PBPK model predictions vs observations showed adequate absolute average fold error (AAFE) in plasma (0.831≤AAFE≤1.52) and ISF of adipose tissue (0.832≤AAFE≤1.77) for all patients. In PBPK model reduction via lumping, plasma and adipose tissue showed distinctly different shapes of predicted Cmin-normalised concentration-time profiles for all antibiotics. Whereas for the high-permeability drug linezolid adipose tissue was lumped with intra- and extracellular concentrations of muscle and bone tissue, low-permeability antibiotics (piperacillin, meropenem, fosfomycin) showed lumping of extracellular concentrations of adipose tissue with intracellular concentrations of liver tissue only.
Informed by PBPK model lumping, in the NLME model ISF data were attributed to a peripheral and plasma data to the central compartment. Implementation of physiologically-motivated scaling approaches (i) decreased interindividual variability in central (6.85%-35.8% relative reduction) and peripheral volume of distribution (9.28%-60.0% relative reduction) for all 4 drugs and (ii) reduced the deviation from the slope of population predictions vs observations from 1 (line of unity; slopeobese=0.653-1.041, slopenonobese=1.09-1.70) upon covariate implementation (slopeobese=1.01-1.13, slopenonobese=0.956-1.09). AIC decreased by 18.0-274 compared to the covariate-free base model.
Precision of the FM scaling factor for all drugs was acceptable (RSE≤45.1%) for LBW/FM and NFM. Yet, LBW/FM showed overall higher precision of the FM scaling factor (2.06-20.7% lower RSE) compared to NFM. When only plasma data were integrated in model development, neither of the modelling approaches showed adequate precision of the FM scaling parameters R and Ffat (RSE=46.0%-172%).
Conclusions: An effect of FM on Vperi alone in LBW/FM represented a physiologically plausible approach supported by PBPK model-based lumping. Additionally, higher precision was achieved by LBW/FM compared to NFM. This suggested that dosing of antibiotics should be governed by the physiologically-motivated LBW/FM, allowing extension of MIPD to a wide body composition range.
The high imprecision of FM scaling parameters with only plasma data might explain the hitherto limited use of NFM or LBW/FM in covariate model development. It was demonstrated that only the availability of ISF data in adipose tissue resulted in precise estimates of the impact of FM on PK parameters. Moreover, exposure data in ISF are needed to overcome the current limitations of purely empirical antimicrobial PK/PD approaches based on plasma samples alone: Target site-based models should be used to (i) quantitatively compare target-site penetration between patient populations and (ii) link target-site exposure to mechanistic PD models to evaluate current dosing regimens.
The LBW/FM approach is envisioned to facilitate physiologically-motivated allometry in other therapeutic areas with available exposure data in adipose tissue, especially in the understudied obese population.
References:
[1] R.J. Keizer et al., Model-Informed Precision Dosing at the Bedside: Scientific Challenges and Opportunities. CPT:PSP 12: 785-787 (2018).
[2] M.P. Pai et al., Antimicrobial Dosing in Specific Populations and Novel Clinical Methodologies: Obesity. CPT 109: 942-951 (2021).
[3] S. Janmahasatian et al., Quantification of lean bodyweight. Clin Pharmacokinet. 44: 1051-1065 (2005).
[4] B.J. Anderson et al., Mechanistic basis of using body size and maturation to predict clearance in humans. Drug Metab Pharmacokinet. 24: 25-36 (2009).
[5] W. Huisinga et al., Modeling Interindividual Variability in Physiologically Based Pharmacokinetics and Its Link to Mechanistic Covariate Modeling. CPT:PSP 1: 1-10 (2012).
[6] P. Simon et al., Measurement of soft tissue drug concentrations in morbidly obese and non-obese patients – A prospective, parallel group, open-labeled, controlled, phase IV, single center clinical trial. Contemp Clin Trials Commun 10: 1-6 (2019).
[7] N. Hartung et al., 28th Population Approach Group Europe (PAGE), Stockholm, Sweden: 9082 (2019).
[8] D. Busse et al., Which Analysis Approach Is Adequate to Leverage Clinical Microdialysis Data? A Quantitative Comparison to Investigate Exposure and Response Exemplified by Levofloxacin. Pharm Res 38: 381-395 (2021).
[9] A. Dosne et al., An automated sampling importance resampling procedure for estimating parameter uncertainty. J Pharmacokinet Pharmacodyn 44: 509-520 (2017).
[10] M. Mueller et al., Issues in Pharmacokinetics and Pharmacodynamics of Anti-Infective Agents: Kill Curves versus MIC. AAC 48: 1441-1453 (2004).