VLDB Journal

Abstract The time course of biomarkers (e.g., acute phase proteins) are typically described using days relative to events of interest, such as surgery or birth, without specifying the sample time. This limits their use as they may change rapidly during a single day. We investigated strategies to impute missing clock times, using procalcitonin for population modelling as the motivating example. 1275 procalcitonin concentrations from 282 neonates were available with dates but not sample times (Scenario 0). Missing clock times were imputed using a random uniform distribution under three scenarios: (1) minimum sampling intervals (8–12 h); (2) procalcitonin concentrations increase for postnatal days 0–1 then decrease; (3) standard sampling practice at the study hospital. Unique datasets (n = 100) were created with scenario-specific imputed clock times. Procalcitonin was modelled for each scenario using the same non-linear mixed effects model using NONMEM. Scenarios were evaluated by the NONMEM objective function value compared to Scenario 0 (∆OFV) and with visual predictive checks. Scenario 3, based on standard sampling practice at the study hospital, was the best imputation procedure with an improved objective function value compared to Scenario 0 (∆OFV: -62.6). Scenario 3 showed a shorter lag time between the birth event and the procalcitonin concentration increase (average: 12.0 h, 95% interval: 9.7 to 14.3 h) compared to other scenarios (averages: 15.3 to 18.7 h). A methodology for selecting imputation strategies for clock times was developed. This may be applied to other problems where clock times are missing.

Abstract Plasma pharmacokinetic (PK) profiles often do not resemble the PK within the central nervous system (CNS) because of blood–brain-border (BBB) processes, like active efflux by P-glycoprotein (P-gp). Methods to predict CNS-PK are therefore desired. Here we investigate whether in vitro apparent permeability (Papp) and corrected efflux ratio (ERc) extracted from literature can be repurposed as input for the LeiCNS-PK3.4 physiologically-based PK model to confidently predict rat brain extracellular fluid (ECF) PK of P-gp substrates. Literature values of in vitro Caco-2, LLC-PK1-mdr1a/MDR1, and MDCKII-MDR1 cell line transport data were used to calculate P-gp efflux clearance (CLPgp). Subsequently, CLPgp was scaled from in vitro to in vivo through a relative expression factor (REF) based on P-gp expression differences. BrainECF PK was predicted well (within twofold error of the observed data) for 2 out of 4 P-gp substrates after short infusions and 3 out of 4 P-gp substrates after continuous infusions. Variability of in vitro parameters impacted both predicted rate and extent of drug distribution, reducing model applicability. Notably, use of transport data and in vitro P-gp expression obtained from a single study did not guarantee an accurate prediction; it often resulted in worse predictions than when using in vitro expression values reported by other labs. Overall, LeiCNS-PK3.4 shows promise in predicting brainECF PK, but this study highlights that the in vitro to in vivo translation is not yet robust. We conclude that more information is needed about context and drug dependency of in vitro data for robust brainECF PK predictions.

Brain amyloid beta neuritic plaque accumulation is associated with an increased risk of progression to Alzheimer’s disease (AD) [Pfeil, J., et al. in Neurobiol Aging 106: 119–129, 2021]. Several studies estimate rates of change in amyloid plaque over time in clinically heterogeneous cohorts with different factors impacting amyloid plaque accumulation from ADNI and AIBL [Laccarino, L., et al. in Annals Clin and Trans Neurol 6: 1113 1120, 2019, Vos, S.J., et al. in Brain 138: 1327–1338, 2015, Lim, Y.Y., et al. in Alzheimer’s Dementia 9: 538-545, 2013], but there are no reports using non-linear mixed effect model for amyloid plaque progression over time similar to that existing of disease-modifying biomarkers for other diseases [Cook, S.F. and R.R. Bies in Current Pharmacol Rep 2: 221–230, 2016, Gueorguieva, I., et al. in Alzheimer’s Dementia 19: 2253–2264, 2023]. This study describes the natural progression of amyloid accumulation with population mean and between-participant variability for baseline and intrinsic progression rates quantified across the AD spectrum. 1340 ADNI participants were followed over a 10-year period with 18F-florbetapir PET scans used for amyloid plaque detection. Non-linear mixed effect with stepwise covariate modelling (scm) was used. Change in natural amyloid plaque levels over 10 year period followed an exponential growth model with an intrinsic rate of approx. 3 centiloid units/year. Age, gender, APOE4 genotype and disease stage were important factors on the baseline in the natural amyloid model. In APOE4 homozygous carriers mean baseline amyloid was increased compared to APOE4 non carriers. These results demonstrate natural progression of amyloid plaque in the continuum of AD.

Sepiapterin is an exogenously synthesized new chemical entity that is structurally equivalent to endogenous sepiapterin, a biological precursor of tetrahydrobiopterin (BH4), which is a cofactor for phenylalanine hydroxylase. Sepiapterin is being developed for the treatment of hyperphenylalaninemia in pediatric and adult patients with phenylketonuria (PKU). This study employed concentration-QT interval analysis to assess QT prolongation risk following sepiapterin treatment. Data from three phase 1 studies and one phase 3 study were pooled for this analysis. Pediatric and adult PKU patients ≥ 2 years received multiple doses at 60 mg/kg and adult healthy volunteers received a single or multiple doses at 20 or 60 mg/kg. Time-matched triplicate ECG measurements and plasma samples for pharmacokinetic analysis were collected. Prespecified linear mixed models relating ΔQTcF to concentrations of sepiapterin and the major active circulating metabolite BH4 were developed for the analysis. The analysis demonstrated that there is no QTcF prolongation risk in patients with PKU following sepiapterin dosing at the highest therapeutic dose, 60 mg/kg/day. The final model showed a marginal but negligible QTcF reduction with increasing sepiapterin and BH4 concentrations. The effect on ΔQTcF was estimated to -2.72 [-3.72, -1.71] and − 1.25 [-2.75, 0.25] ms at mean baseline adjusted BH4 Cmax of 332 ng/mL (therapeutic exposure) and 675 ng/mL (supratherapeutic exposure) at dose 60 mg/kg, respectively, in PKU patients with food and in healthy volunteers with a high fat diet. Various covariates, such as clinical study ID, age, sex, food effect, race, body weight, and disease status, on QTcF interval were investigated and were found insignificant, except for food effect and age. This study concludes that there is no QTcF prolongation risk in patients with PKU following sepiapterin dosing up to 60 mg/kg/day, and BH4 and sepiapterin concentrations minimally affect ΔQTcF after adjustment for time, sex, and meal.

Alogliptin is a highly selective inhibitor of dipeptidyl peptidase-4 and primarily excreted as unchanged drug in the urine, and differences in clinical outcomes in renal impairment patients increase the risk of serious adverse reactions. In this study, we developed a comprehensive physiologically-based quantitative systematic pharmacology model of the alogliptin-glucose control system to predict plasma exposure and use glucose as a clinical endpoint to prospectively understand its therapeutic outcomes with varying renal function. Our model incorporates a PBPK model for alogliptin, DPP-4 activity described by receptor occupancy theory, and the crosstalk and feedback loops for GLP-1-GIP-glucagon, insulin, and glucose. Based on the optimization of renal function-dependent parameters, the model was extrapolated to different stages renal impairment patients. Ultimately our model adequately describes the pharmacokinetics of alogliptin, the progression of DPP-4 inhibition over time and the dynamics of the glucose control system components. The extrapolation results endorse the dose adjustment regimen of 12.5 mg once daily for moderate patients and 6.25 mg once daily for severe and ESRD patients, while providing additional reflections and insights. In clinical practice, our model could provide additional information on the in vivo fate of DPP4 inhibitors and key regulators of the glucose control system.

Abstract P-glycoprotein (P-gp) is a key efflux transporter and may be involved in drug-drug interactions (DDIs) at the blood-brain barrier (BBB), which could lead to changes in central nervous system (CNS) drug exposure. Morphine is a P-gp substrate and therefore a potential victim drug for P-gp mediated DDIs. It is however unclear if P-gp inhibitors can induce clinically relevant changes in morphine CNS exposure. Here, we used a physiologically-based pharmacokinetic (PBPK) model-based approach to evaluate the potential impact of DDIs on BBB transport of morphine by clinically relevant P-gp inhibitor drugs. The LeiCNS-PK3.0 PBPK model was used to simulate morphine distribution at the brain extracellular fluid (brainECF) for different clinical intravenous dosing regimens of morphine, alone or in combination with a P-gp inhibitor. We included 34 commonly used P-gp inhibitor drugs, with inhibitory constants and expected clinical P-gp inhibitor concentrations derived from literature. The DDI impact was evaluated by the change in brainECF exposure for morphine alone or in combination with different inhibitors. Our analysis demonstrated that P-gp inhibitors had a negligible effect on morphine brainECF exposure in the majority of simulated population, caused by low P-gp inhibition. Sensitivity analyses showed neither major effects of increasing the inhibitory concentration nor changing the inhibitory constant on morphine brainECF exposure. In conclusion, P-gp mediated DDIs on morphine BBB transport for the evaluated P-gp inhibitors are unlikely to induce meaningful changes in clinically relevant morphine CNS exposure. The developed CNS PBPK modeling approach provides a general approach for evaluating BBB transporter DDIs in humans.

Model-informed drug development (MIDD) is an approach to improve the efficiency of drug development. To promote awareness and application of MIDD in Japan, the Data Science Expert Committee of the Drug Evaluation Committee in the Japan Pharmaceutical Manufacturers Association established a task force, which surveyed MIDD applications for approved products in Japan. This study aimed to reveal the trends and challenges in the use of MIDD by analyzing the survey results. A total of 322 cases approved in Japan between January 2020 and March 2022 as medical products were included in the survey. Modeling analysis was performed in approximately half of the cases (47.8% [154/322]) and formed a major basis for the selection or justification of dosage and administration in approximately one-fourth of the cases [24.2% (78/322)]. Modeling analysis/model-based dose selection was frequently conducted in cases involving monoclonal antibodies, first indication, orphan drugs, and multi-regional trials. Moreover, the survey results indicated that modeling analyses contributed to dose optimization throughout the developmental phases, including changing dose levels from phase II to phase III and dose adjustment in special populations. Japanese data were included in most cases in which modeling analysis was used for dosage selection. Thus, modelling analysis may also address ethnic factors introduced in the ICH E5 and/or E17 guidelines. In summary, this survey is useful for understanding the current status of MIDD use in Japan and for future drug development.

In oncology drug development, overall response rate (ORR) is commonly used as an early endpoint to assess the clinical benefits of new interventions; however, ORR benefit may not always translate into a long-term clinical benefit such as overall survival (OS). Most of the work on developing endpoints based on tumor growth dynamics relies on empirical validation, leading to a lack of generalizability of the endpoints across indications and therapeutic modalities. Additionally, many of these metrics are model-based and do not use data from all the patients. The objective of this work is to use longitudinal tumor size data and new lesion information (that is, the same information used by the ORR) to develop novel endpoints that can improve early clinical decision-making compared to the ORR. We investigate in this work multiple candidate novel endpoints based on tumor size ratio that utilize longitudinal tumor size data from all the patients regardless of their follow-up, rely only on tumor size and new lesion information, and are model-free. An extensive simulation study is conducted, exploring a wide spectrum of tumor size data and overall survival outcomes by modulating a variety of trial characteristics such as slow vs fast tumor growth, high vs low drug efficacy rates, variability in patients’ responses, variations in the number of patients, follow-up periods, new lesion rates and survival curve shapes. The proposed novel endpoints based on tumor size ratio consistently outperform the ORR by having a comparable or higher correlation with the OS. Further, the novel endpoints exhibit superior accuracy compared to the ORR in predicting the long-term OS benefit. Retrospective empirical validation on BMS clinical trials confirms our simulation findings. These findings suggest that the tumor size ratio-based endpoints could replace ORR for early clinical decision-making in oncology drug development.

M6495 is a first-in-class NANOBODY® molecule and an inhibitor of ADAMTS-5, with the potential to be a disease modifying osteoarthritis drug. In order to investigate the PK/PD (pharmacokinetic and pharmacodynamic) properties of M6495, a single dose study was performed in cynomolgus monkeys with doses up to 6 mg/kg, with the goal of understanding the PK/PD properties of M6495. The neo-epitope ARGS (Alanine-Arginine-Glycine-Serine) generated by cleavage of aggrecan by ADAMTS-5 was used as a target-engagement biomarker. A long-lasting dose-dependent decrease in serum ARGS could be observed after a single dose of M6495 in cynomolgus monkeys. The serum biomarker ARGS decreased to levels below the limit of quantification of the assay in animals which received doses of M6495 of 6 mg/kg and higher, indicating a strong inhibition of ADAMTS-5. Data from the single-dose PK/PD study was combined with data from a multiple dose study, and a non-linear mixed effects model was used to explore the relationship between plasma concentrations of M6495 and the reduction of serum ARGS. The model was subsequently used to inform the clinical phase 1 study design and was successful in predicting the human clinical pharmacokinetics and pharmacodynamics of M6495. In addition to having enabled a Phase 1 trial with M6495, this is the first PK/PD model describing the pharmacodynamics of the neo-epitope ARGS after ADAMTS5 inhibition. It is expected that in the future, this model can be used or adapted to explore the PK/PD relationship between M6495 serum concentrations and the ARGS serum biomarker.

Antibody drug conjugates (ADC) are a promising class of oncology therapeutics consisting of an antibody conjugated to a payload via a linker. DYP688 is a novel ADC comprising of a signaling protein inhibitor payload (FR900359) that undergoes unique on-antibody inactivation in plasma, resulting in complex pharmacology. To assess the impact of FR inactivation on DYP688 pharmacology and clinical developability, we performed translational modeling of preclinical PK and tumor growth inhibition (TGI) data, accompanied by mechanistic Krogh cylinder tumor modeling. Using a PK-TGI model, we identified a composite exposure-above-tumorostatic concentration (AUCTSC) metric as the PK-driver of efficacy. To underpin the mechanisms behind AUCTSC as the driver of efficacy, we performed quantitative systems pharmacology (QSP) modeling of DYP688 intratumoral pharmacokinetics and pharmacodynamics. Through exploratory simulations, we show that by deviating from canonical ADC design dogma, DYP688 has optimal FR900359 activity despite its transient inactivation. Finally, we performed the successful preclinical to clinical translation of DYP688 PK, including the payload inactivation kinetics, evidenced by good agreement of the predicted PK to the observed interim clinical PK. Overall, this work highlights early quantitative pharmacokinetics as a missing link in the ADC design-developability chasm.

Current treatment recommendations mainly rely on rule-based protocols defined from evidence-based clinical guidelines, which are difficult to adapt for high-risk patients such as those with renal impairment. Consequently, unsuccessful therapies and the occurrence of adverse drug reactions are common. Within the context of personalized medicine, that tries to deliver the right treatment dose to maximize efficacy and minimize toxicity, the concept of model-informed precision dosing proposes the use of mechanistic models, like physiologically based pharmacokinetic (PBPK) modeling, to predict drug regimes outcomes. Nonetheless, PBPK models have limited capability when computing patients’ centric optimized drug doses. Consequently, reinforcement learning (RL) has been previously used to personalize drug dosage. In this work we propose the first PBPK and RL-based precision dosing system for an orally taken drug (benazepril) considering a virtual population of patients with renal disease. Population based PBPK modeling is used in combination with RL for obtaining patient tailored dose regimes. We also perform patient stratification and feature selection to better handle dose tailoring problems. Based on patients’ characteristics with best predictive capabilities, benazepril dose regimes are obtained for a population with features’ diversity. Obtained regimes are evaluated based on PK parameters considered. Results show that the proof-of-concept approach herein is capable of learning good dosing regimes for most patients. The use of a PopPBPK model allowed to account for intervariability of patient characteristics and be more inclusive considering also non-frequent patients. Impact analysis of patients’ features reveals that renal impairment is the main driver affecting RL capabilities.