Peter Gennemark (1), Ulf Andersson (2), Marie Elebring (1)
(1) Cardiovascular, Renal and Metabolism, IMED Biotech Unit, AstraZeneca, Gothenburg, Sweden, (2) Drug Safety and Metabolism, IMED Biotech Unit, AstraZeneca, Gothenburg, Sweden
Objectives: To estimate the tissue half-life of an antisense oligonucleotide (ASO) accurately and precisely, by properly designing and analysing a cynomolgus monkey study. The half-life determines tissue accumulation upon chronic dosing and is pivotal in the human dose-response prediction.
Methods: ASO pharmacokinetics (PK) is largely sequence-independent and characterized by rapid (hours) distribution to tissue and slow (weeks) terminal half-life dependent on re-distribution from tissue to blood [1]. Therefore, one expects accumulation of drug in tissue for standard (such as once weekly or monthly) dosing intervals. Determination of tissue half-life is of general interest to interpret safety aspects, and to predict efficacy if target is situated in tissue. Here, the liver PK of an ASO was to be measured in 28 cynomolgus monkeys using four monthly subcutaneous (SC) injections (2, 8 or 32 mg/kg) plus one extra loading dose after one week. To investigate toxicology aspects of the ASO, 6 monkeys per dose group (18 in total) were to be terminated two days after the last dose. We sought a suitable termination schedule for the remaining 10 monkeys to estimate tissue half-life. To this end, we used Yu et al.’s cynomolgus monkey population model describing plasma and liver PK of an ASO [2]. The model contains three compartments, representing plasma, liver and all other tissues. We considered a space of 20 experimentally plausible designs, plus one rich design with >100 monkeys to serve as baseline. The search space is hence not exhaustive but represent designs that could realistically be achieved in practice. For each design we simulated 1000 studies. A one-compartment model with dose-dependent volume of distribution was simultaneously fitted to data from all dose groups in each simulated study. The non-linearity represents saturated liver uptake at high doses, and a similar non-linearity was used by Yu et al. The absorption rate was fixed to 1 1/day. The secondary parameter thalf and its precision and its bias with respect to the estimate of the rich design was calculated for each design.
Results: The two best proposed designs (bias 4-11%; 90% confidence of thalf within 14-17% of point estimate) included one satellite group (either 2 or 8 mg/kg) that was observed during washout 3 months after the last dose (N=4) and the remaining 6 animals assigned to the other two dose groups, three to each. The two following best proposed designs (bias 10-29%; 90% confidence of thalf within 17-24% of point estimate) included a satellite single dose group (2 or 8 mg/kg) that was observed in time-series during days 3 to 57 (N=6), and another satellite group (32 mg/kg) was observed during washout 3 months after the last dose (N=4). Repeating the analysis with a linear PK model gave qualitatively similar results. For robustness, it was decided to go for the latter designs to ensure that the half-life could be determined without a modelling approach in case the tested ASO would exhibit significantly different PK from the ASO considered in Yu et al. The design with a dose of 8 mg/kg, and not 2 mg/kg, of the first satellite group was run in vivo to reduce the risk of obtaining data below the limit of quantification. The liver exposure half-life was estimated to 17.6 (15, 21; 5th and 95th percentiles) days based on mathematical modelling taking all liver exposure data simultaneously into account. General learnings from this analysis include: (1) ASO toxicology studies can inform the human dose-response prediction, (2) the best designs sample the majority of animals at steady state, and a few animals in time-series, (3) robustness of the thalf estimate may increase by sampling a larger fraction of the animals in time-series, to avoid non-conclusive data in case accumulation turns out to be negligible, (4) the lowest dose is often of greatest interest from an efficacy point of view but carefully consider the risk of obtaining data below the limit of quantification, and (5) if longitudinal target engagement biomarker data are available, consider dose-response-time modelling to infer the tissue half-life [3]. Concerning (5), naturally, robustness increases if several approaches are combined.
Conclusion: Modelling and reasoning were used to design a cynomolgus monkey study to determine the tissue half-life of an ASO with predicted bias <20% and good precision. From the resulting data, the tissue half-life could be inferred as 17.6 (15, 21; 5th and 95th percentiles) days.
References:
[1] Crooke ST, Witztum JL, Bennett CF, Baker BF. RNA-Targeted Therapeutics. Cell Metab. 27(4):714-739, 2018.
[2] Yu RZ, Gunawan R, Post N, Zanardi T, Hall S, Burkey J, Kim TW, Graham MJ, Prakash TP, Seth PP, Swayze EE, Geary RS, Henry SP, Wang Y. Disposition and Pharmacokinetics of a GalNAc3-Conjugated Antisense Oligonucleotide Targeting Human Lipoprotein (a) in Monkeys. Nucleic Acid Ther. 26(6):372-380, 2016.
[3] Jansson-Löfmark R, Gennemark P. Inferring Half-Lives at the Effect Site of Oligonucleotide Drugs. Nucleic Acid Ther. 28(6):319-325, 2018.
Reference: PAGE 28 (2019) Abstr 8966 [www.page-meeting.org/?abstract=8966]
Poster: Study Design