Background: Describing the kinetics of cytokines involved as biomarkers of sepsis progression could help to optimise interventions in septic patients. This work aimed to quantitively characterise the cytokine kinetics upon exposure to live E. coli by developing an in silico model, and to explore predicted cytokine kinetics at different bacterial exposure scenarios.

Methods: Data from published in vivo studies using a porcine sepsis model were analysed. A model describing the time courses of bacterial dynamics, endotoxin (ETX) release, and the kinetics of TNF and IL-6 was developed. The model structure was extended from a published model that quantifies the ETX-cytokines relationship. An external model evaluation was conducted by applying the model to literature data. Model simulations were performed to explore the sensitivity of the host response towards differences in the input rate of bacteria, while keeping the total bacterial burden constant.

Results: The analysis included 645 observations from 30 animals. The blood bacterial count was well described by a one-compartment model with linear elimination. A scaling factor was estimated to quantify the ETX release by bacteria. The model successfully described the profiles of TNF, and IL-6 without a need to modify the ETXcytokines model structure. The kinetics of TNF, and IL-6 in the external datasets were well predicted. According to the simulations, the ETX tolerance development results in that low initial input rates of bacteria trigger the lowest cytokine release.

Conclusion: The model quantitively described and predicted the cytokine kinetics triggered by E. coli exposure. The host response was found to be sensitive to the bacterial exposure rate given the same total bacterial burden.

Objectives

Applying physiologically-based pharmacokinetic (PBPK) modelling in sepsis could help to better understand how PK changes are influenced by drug- and patient-related factors. We aimed to elucidate the influence of sepsis pathophysiology on the PK of meropenem by applying PBPK modelling.

Methods

A whole-body meropenem PBPK model was developed and evaluated in healthy individuals, and renally impaired non-septic patients. Sepsis-induced physiological changes in body composition, organ blood flow, kidney function, albumin, and haematocrit were implemented according to a previously proposed PBPK sepsis model. Model performance was evaluated, and a local sensitivity analysis was conducted.

Results

The model-predicted PK metrics (AUC, C_max, CL, V_ss) were within 1.33-fold-error margin of published data for 87.5% of the simulated profiles in healthy individuals. In sepsis, the model provided good predictions for literature-digitised average plasma and tissue exposure data, where the model-predicted AUC was within 1.33-fold-error margin for 9 out 11 simulated study profiles. Furthermore, the model was applied to individual plasma concentration data from 52 septic patients, where the model-predicted AUC, C_max, and CL had a fold-error ratio range of 0.98–1.12, with alignment of the predicted and observed variability. For V_ss, the fold-error ratio was 0.81, and the model underpredicted the population variability. CL was sensitive to renal plasma clearance, and kidney volume, whereas V_ss was sensitive to the unbound fraction, organ volume fraction of the interstitial compartment, and the organ volume.

Conclusions

These findings may be extended to more diverse drug types and support a more mechanistic understanding of the effect of sepsis on drug exposure.

In vitro time-kill curve (TKC) experiments are an important part of the pharmacokinetic- pharmacodynamic (PKPD) characterisation of antibiotics. Traditional TKCs use Mueller-Hinton broth (MHB), which lacks specific plasma components that could potentially influence the bacterial growth and killing dynamics, and affect translation to in vivo. This study aimed to evaluate the impact of plasma on the PKPD characterisation of two antibiotics; cefazolin and clindamycin. TKC experiments were conducted in pure MHB, and MHB spiked with 20% and 70% human plasma. Plasma protein binding (PPB) data were available, and a linear model described cefazolin's PPB, while clindamycin's PPB was best described by a second-order polynomial model. PKPD models were developed based on pure MHB and described drug effects using an E_max model, with consideration of adaptive resistance for cefazolin. The observed bacterial growth and killing in the plasma-spiked MHB TKC data was insufficiently described when applying the developed PPB and PKPD models. In plasma spiked MHB, a growth delay was observed, estimated to 0.25 h (20% plasma), or 2.90 h (70% plasma) for cefazolin, and 0.64 h (20% plasma), or 1.40 h (70% plasma) for clindamycin. Furthermore, the drug effect was higher than expected in plasma-spiked MHB, with bacterial stasis and/or killing at unbound concentrations below MIC, necessitating drug effect parameter scaling (C₅₀ for cefazolin, Hill coefficient for clindamycin). The findings highlight significant differences in bacterial growth and killing dynamics between pure MHB and plasma-spiked MHB and exemplify how PKPD modelling may be used to improve the translation of in vitro results.