henhaobio
Sign in

Microbiology SpectrumPharmacology & ToxicologyPeer-reviewed journal article

A freely accessible, adaptable hollow-fiber setup to reproduce first-order absorption: illustration with linezolid cerebrospinal fluid pharmacokinetic data

Noémie PrébonnaudiD, Alexia ChauzyiD, Nicolas GrégoireiD, Claire Dahyot‐FizelieriD, Christophe Adier, Sandrine MarchandiD, Vincent Aranzana‐ClimentiD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

The main objective of this study was to validate an algorithm and experimental setup to simulate first-order absorption pharmacokinetic profiles without altering the standard in vitro hollow fiber infection model (HFIM). For that, clinical cerebrospinal fluid (CSF) linezolid concentrations after 30-minute infusions at dosing regimens 600 mg q12 h, 900 mg q12 h, and 900 mg q8 h were reproduced in the HFIM over 4 days. To approximate the apparent first-order absorption observed on CSF pharmacokinetic profiles, we split the dosing interval into a series of sub-intervals during which continuous infusions were delivered to the system. During each sub-interval, the same amount of linezolid was delivered but the sub-intervals had different durations and flow rates which were computed by a newly developed algorithm. In addition, we independently reproduced plasma concentrations to validate our system. Samples were collected from the central reservoir and the extracapillary space (ECS) of the cartridge of the HFIM and assayed by liquid chromatography-tandem mass spectrometry. Observed pharmacokinetic parameters and concentrations in the ECS were compared with the target clinical pharmacokinetic parameters and concentrations. Observed pharmacokinetic parameters were within 20 % of target pharmacokinetic parameters for all experiments, thus validating the ability of our experimental setup to reproduce plasma and CSF linezolid pharmacokinetic profiles. The algorithm and setup are available in the open-source web application https://varacli.shinyapps.io/hollow_fiber_app/ to easily design other HFIM experiments.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (23 Dec 2024) · CC BY 4.0

Splitting each dose into equal-amount continuous infusions of varying duration, computed by a new algorithm, lets a standard hollow-fiber model reproduce linezolid CSF pharmacokinetics within 20% of target parameters.

Antibiotic concentrations at infection sites such as cerebrospinal fluid (CSF) often rise later and more slowly than in plasma, a shape a single short infusion in a hollow-fiber infection model (HFIM) cannot mimic, while the existing depot-bottle workaround adds material and medium cost. This setup approximates first-order absorption pharmacokinetics (PK) by adding only a programmable ambulatory infusion pump, with the infusion schedule computed by an algorithm released in a web app.

Key results

  1. ShownThe sub-interval infusion scheme reproduced linezolid CSF profiles, including their delayed absorption phase, for all three dosing regimens without adding a compartment to the HFIM.

    In the cartridge extracapillary space (ECS), Cmax,1, Cmax,ss, t1/2, AUCτ,1 and AUCτ,ss were within 20 % of target for 600 mg q12 h, 900 mg q12 h and 900 mg q8 h (n = 2 - 3 replicates); absolute Tmax,1 bias was at most 0.3 h and Tmax,ss showed no bias. · 4. Our Hollow-Fiber setup can be used to…; Abstract

  2. ShownThe same HFIM, using one 30 min infusion per dose, also reproduced unbound plasma linezolid profiles.

    Across the three regimens (n = 2 - 6), bias versus target was below 10 % for Cmax,1, Cmax,ss, t1/2, AUCτ,1 and AUCτ,ss in the ECS; Tmax was off by an absolute 0.25 h (0.5 h versus 0.75 h). · 3. Our Hollow-Fiber setup can be used to…

  3. ShownDrug moved quickly but not instantly across the hollow-fiber membrane. After short infusions, ECS peaks came later and were slightly lower than in the central reservoir; with the slower CSF-like input there was no difference.

    Estimated Qdiffusion was 6.53 mL/min (0.392 L/h). For plasma-like dosing at 600 mg q12 h (n = 2), ECS Tmax lagged the reservoir by 0.250 h and Cmax was ∼10 % lower; CSF-like dosing showed no difference. AUCτ was the same in both compartments. The other two regimens gave comparable results. · 2. Diffusion across hollow-fiber membrane is…; Estimation of constant diffusion flow rate…; Discussion

  4. SuggestedSplitting each dose into sub-intervals that each deliver the same amount, with a fixed end time for the last one, turns first-order absorption into a step program that a pump limited to discrete rates can run.

    Each interval was split into 12 sub-intervals (8 for q8 h), each infusing close to 2 mL. Worked example: n = 12, Cmax 4.25 mg/L, ka 0.36 h^-1, t1/2 3.15 h, Vtotal 0.360 L, tn = 12 h gives Dose 3.3 mg delivered by tn, which is 98.7 % (fdose) of the total first-order dose. The expected ECS curve closely followed the target CSF curve. · Setup to reproduce linezolid CSF concentrations…; Practical solution; 4. Our Hollow-Fiber setup can be used to…

  5. ShownOne three-compartment model (central reservoir, intracapillary space (ICS), ECS) fitted both reservoir and ECS concentrations in the plasma and CSF setups.

    Checked by comparing observed data with 90 % prediction intervals from 1000 simulations that included residual variability (pass criterion: 90 % of data inside). The text calls the fit good for all regimens but gives no coverage figure. · Validation of the tri-compartmental model; 1. Tri-compartmental model fits well the…

Methods, data and reproducibility

Linezolid plasma and CSF profiles for three regimens were simulated from an earlier ICU population PK model and reproduced for 4 days in an HFIM. CSF absorption was approximated by sub-interval infusions timed by the algorithm. Concentrations in the central reservoir and ECS were measured by LC-MS/MS, fitted with a three-compartment Monolix model and compared with target PK parameters.

  • Simulated unbound plasma and CSF linezolid concentrations from an earlier population PK model of ICU patients with external ventricular drainage (600 mg q12 h, 900 mg q12 h, 900 mg q8 h; 4 days) · human (ICU patients)
  • HFIM linezolid concentrations (central reservoir and cartridge ECS, LC-MS/MS) from experiments reproducing plasma profiles · n = n = 2 - 6 per regimen · 10.57745/NRXPOP
  • HFIM linezolid concentrations (central reservoir and cartridge ECS, LC-MS/MS) from experiments reproducing CSF profiles · n = n = 2 - 3 per regimen · 10.57745/NRXPOP
  • Bacterial inoculum in the cartridge ECS (60 mL cation-adjusted Mueller-Hinton broth) · Staphylococcus aureus · ATCC 29213
Methods & tools
HF-App (R-shiny), R 4.2.2, mrgsolve, Monolix 2024R1, LC-MS/MS (Shimadzu Nexera XR with Sciex TQ3500), XBridge Peptide BEH300 C18 column, FX paed helixone dialyzer cartridge, Mini Rythmic Perf+ ambulatory infusion pump, Masterflex L/S peristaltic pump
Software & environment
R 4.2.2 with the mrgsolve package; Monolix 2024R1.
Reproducibility gaps
Only the hosted Shiny app is linked; the text gives no source repository or version for HF-App. The pump programs and experimental parameters are in supplementary Tables S1 to S3, and the algorithm derivation is in Text S1.
Limitations (5)
  • Target ke and ka for the CSF profiles were adjusted by eye. For 600 mg q12 h, 900 mg q12 h and the first 900 mg q8 h replicate, ke was set by mistake to 0.23 or 0.22 h^-1 instead of 0.19 h^-1. This sped elimination and lowered Cmax and AUCτ, though bias stayed below 20 %. · stated by the authors
  • Fast equilibration between the central reservoir and the cartridge, driven by the high loop flow, was assumed but not demonstrated. · stated by the authors
  • The authors hypothesise that the ICS-to-ECS lag grows with higher infusion rates and shorter half-lives, so other drugs could lose meaningful Cmax. Correcting for this would require estimating Qdiffusion and numerically optimising the HFIM settings. · stated by the authors
  • Validation covers PK only. The ECS was inoculated with S. aureus ATCC 29213, but the text reports no bacterial killing data. · noted in this summary
  • Tested with one drug (linezolid), one cartridge type and 2 to 3 replicates per CSF setup. The targets are simulations from a population PK model approximated by one-compartment fits, not measured patient concentrations. · noted in this summary

Read this if you run hollow-fiber infection models and need to reproduce infection-site or oral-like concentration profiles with an absorption phase without adding a depot compartment.

Citations
3
FWCI
1.52
Field percentile
top 19%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISAyansiji '26Sudarsono '26Herbowski '26Faryami '22Dreyer '24Li '24
56% semantic matchBioengineering
Ayankola O. Ayansiji, Caleb Gardner, Sebastien Dors+4 more·Frontiers in Bioengineering and Biotechnology·Jan 8, 2026· 2 citations