Hadija Marchiori (1), Michele Schiavon (1), Chiara Roversi (2), Simone Zannoni (2), Massimiliano Fonsi (2), Maria Teresa di Martino (3), Alessia Tagliavini (2), Chiara Dalla Man (1)
(1) Department of Information Engineering, University of Padova, Padova, Italy; (2) Evotec (SE); (3) Department of Experimental and Clinical Medicine, Magna Graecia University, Catanzaro, Italy.
Objectives:
MicroRNAs (miRs) are small, noncoding regions of single-stranded RNA that have recently garnered major attention as key regulators of several biological pathways. Deregulated miRs have been proven to be related to cancer onset and progression and, consequently, they have been deeply investigated as potential targets of RNA-based therapeutics [1]. Among miRs, miR-221 has been the most studied, both for its oncogenic potential and as a promising therapeutic target. Its inhibition can be obtained using the locked nucleic acid (LNA) LNA-i-miR-221 [1], a chemically modified antisense oligonucleotide. Indeed, it efficiently downregulates miRNA-221 while also upregulating its targets both in-vitro and in-vivo [2]. Although many studies have been conducted to assess the kinetics and safety of LNA-i-miR-221 by means of non-compartmental modeling, a population modeling analysis of its kinetics has not been performed yet. The aim of this work is the development of a Nonlinear Mixed Effect (NLME) model of LNA-i-mir-221 kinetics in Sprague Dawley rats. Such model can then be used for the prediction of LNA-i-miR-221 key pharmacokinetic parameters in humans through single species allometric scaling methods.
Methods:
The database consists of LNA-i-miR-221 pharmacokinetic data gathered from 42 specimens of Sprague Dawley rats (21 females and 21 males, age = 6 +/- 1 weeks, mean weight = 0.256 kg at the beginning of the study). A bolus of LNA-i-miR-221 was intravenously injected during 2 administration cycles, each lasting 4 consecutive days, separated by 10 days of washout. Three dosing regimens were tested (5 mg/Kg/day, 12.5 mg/Kg/day, and 125 mg/Kg/day). Blood samples were collected right after each administration and following the end of the administration cycles.
The model was developed within the NLME estimation framework. For the choice of the structural model, two and three compartment models, parametrized with clearance (CL), central (V1) and peripheral distribution volumes (V2, V3) and intercompartmental clearances (BCL2, BCL3), were tested. Moreover, a population time lag (Tlag) was introduced to correct for a possible mismatch between dose injection and appearance of the drug in the accessible compartment. All parameters were assumed to be log-normally distributed. Finally, the error model describing the unexplained variability was assumed to be a gaussian random noise with standard deviation modeled as the sum of a constant component, a, and a term proportional, through the coefficient b, to the measured concentration of LNA-i-miR-221.
Model identification was performed using the commercially available software Monolix (Version 2023R1, ©Lixoft, Antony, France [3]). Correlations between random effects and covariates were introduced only if they improved model performance. In particular, correlations between random effects were investigated using a forward inclusion and backward elimination procedure; whereas the inclusion of meaningful covariates, such as body weight and sex, was tested using the “Stepwise Approach”, a Monolix algorithm based on Correlation test (COSSAC) [3]. Of note, no inter-subject variability was associated to the parameter Tlag. Thereafter, models were assessed considering the precision of the estimated parameters, expressed in terms of the relative standard error (RSE) of the estimates, and evaluating their ability to fit the data, that can be assessed from the residual distribution. In particular, models providing at least one parameter with a RSE > 100% or nonnormally distributed residuals were discarded. Models that performed well were compared using the corrected Bayesian Information Criterion (BICc): the model that scored the lowest BICc was selected as the best one.
Different single species allometric scaling methods, such as fixed exponent simple allometry, with and without correction factors accounting for known physiological differences between humans and rats (e.g., the brain weight and the maximum lifespan potential), were applied to animal clearance and distribution volume obtained with the best model to predict the kinetics of LNA-i-miR-221 in humans [4] [5] [6]. Finally, human data [7] were used to assess the different allometric scaling methods comparing the predicted pharmacokinetics variables with the observed human parameters through the fold error, defined as the ratio between the predicted parameter and its observed value. This comparison also allowed for the selection of the best scaling algorithm for each pharmacokinetic parameter.
Results:
The distribution of the residuals was always acceptable. Some models provided at least one parameter with a RSE > 100% and were therefore discarded from further investigations. Among the remaining models, the one providing the lowest BICc (8143.89) was the delayed three-compartment model including the correlation between the random effects of CL and V2, CL and BCL3, and V2 and BCL3. The population estimates were: CL=0.11 L/h (RSE= 7%), V1=0.047 L (9%), V2=0.15 L (9%), V3=1.85 L (27%), BCL2=0.22 L/h (15%), BCL3= 0.03 L/h (9%), and Tlag=0.04 h (22%). Standard deviations of the random effects were: ωCL=0.38 (13%), ωV1=0.27 (18%), ωV2=0.43 (16%), ωV3=0.61 (37%), ωBCL2=0.13 (48%), and ωBCL3=0.32 (37%). Correlations were: ρCL,V2=0.87 (9%), ρCL,BCL3=0.8 (17%), and ρV2,BCL3=0.86 (14%). The coefficients of the error model resulted a=3.51 ng/mL (37%), and b=0.24 (5%).
Among the allometric scaling algorithms that were tested, simple allometry with exponent fixed to 0.85 performed best for the estimation of human clearance, with a fold error of 1.07. Scaling of the distribution volume was more cumbersome; indeed, all the tested methods performed poorly with respect to the fold error metric.
Conclusions:
In this work, a NLME model describing the kinetics of LNA-i-miR-221 in Spragye Dawley rats has been developed. This represents an improvement in the understanding of LNA-i-miR-221 kinetics, which, so far, had only been studied through non-compartmental analysis. Moreover, allometric scaling methods were applied to antisense oligonucleotides, suggesting that single species simple allometry can be employed to successfully predict the clearance but not the volume of distribution of LNA-i-miR-221 in humans.
References:
[1] Di Martino MT et al., mir-221/222 as biomarkers and targets for therapeutic intervention on cancer and other diseases: A systematic review. Molecular Therapy – Nucleic Acids, 27:1191–1224, 2022.
[2] Di Martino MT et al., In vitro and in vivo activity of a novel locked nucleic acid (lna) inhibitor-mir-221 against multiple myeloma cells, PLoS ONE, 9: e89659, 2014.
[3] LixoftSAS. Documentation of monolix, version 2020r1. Antony, France: http://monolix.lixoft.com/, 2019.
[4] Mahmood I et al., Interspecies scaling of antibody–drug conjugates (adc) for the prediction of human clearance. Antibodies, 10(1):1, 2021.
[5] Patel D et al., Single-species allometric scaling: A strategic approach to support drug discovery, Journal of Pharmaceutical Research International, 22(3):1–7, 2018.
[6] Huang Q et al., The application of allometric scaling principles to predict pharmacokinetic parameters across species, Expert Opin Drug Metab Toxicol., Sep;10(9):1241-53, 2014.
[7] Tassone T et al., Safety and activity of the first-in-class locked nucleic acid (LNA) miR-221 selective inhibitor in refractory advanced cancer patients: a first-in-human, phase 1, open-label, dose-escalation study. J Hematol Oncol, 16:68, 2023.
Reference: PAGE 32 (2024) Abstr 10790 [www.page-meeting.org/?abstract=10790]
Poster: Drug/Disease Modelling - Other Topics