Physiologically-based pharmacokinetic modeling of nafamostat to support dose selection for treatment of pediatric patients with COVID-19.

Pediatric patients with coronavirus disease 2019 (COVID-19) are increasing, and severe cases such as multisystem inflammatory syndrome are being reported. Nafamostat, a repurposing drug, is currently being explored for the treatment of COVID-19 in adults. However, the data supporting its exposure in pediatrics remains scarce. Physiologically-based pharmacokinetic (PBPK) modeling enables the prediction of drug exposure in pediatrics based on ontogeny of metabolic enzymes and age dependent anatomical and physiological changes. The study aimed to establish a PBPK model of nafamostat in adults, then scale the adult PBPK model to children for predicting pediatric exposures of nafamostat and an optimal weight-based nafamostat dose in pediatric population. The developed model adequately described adult exposure data in healthy volunteers following i.v. administration with three doses (10, 20, and 40 mg). Scaling adult PBPK models to five pediatric groups predicted that as age advances from neonate to adult, the exposure of nafamostat slightly increased from neonate to infant, steadily decreased from infant to child, and then increased from child to adult after the administration of 0.2 mg/kg/h for 14 days, a dosing regimen being conducted in a clinical trial for COVID-19. Based on the fold change of predicted area under the curve for the respective pediatric group over those of adults, weight-based dosages for each pediatric group may be suggested. The novel PBPK model described in this study may be useful to investigate nafamostat pharmacokinetics in a pediatric subgroup further.

INTRODUCTION

The current coronavirus disease 2019 (COVID-19) pandemic has emerged as a critical global health crisis. This disease is caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). As of November 2021, there have been over 250 million cases and greater than 5 million deaths worldwide (https://coronavirus.jhu.edu/map.html). Although the percentage of COVID-19 diagnosis in children is likely to be lower than that of adults, the incidence of COVID-19 in a pediatric population is not well-known. In South Korea, around 15% of the reported COVID-19 cases occurs in individuals less than 19 years of age (http://ncov.mohw.go.kr/). The disease severity generally appears to be milder and the hospitalization rate is also prone to be lower in pediatric patients in comparison with adults [1]. However, the Center for Disease Control and Prevention (CDC) of the United States reported multisystem inflammation syndrome in children (MIS-C) associated with COVID-19 from a minor group of pediatric patients, which arose from the recent report investigating clusters of children manifesting with multi-organ involved severe inflammation and recent SARS-CoV-2 infection [23456789]. Therefore, a preliminary case definition for MIS-C has been developed by WHO reflecting the clinical and laboratory characteristics and identifying suspected or confirmed cases [10]. Treatment for MIS-C consists of immune modulating drugs such as corticosteroids and immunoglobulin and supportive care for pneumonia, respiratory failure, and sepsis [78911].

Medication to treat COVID-19 has not met a demand yet and repurposing drugs with well-established safety are an appealing option. However, several repurposing drugs in particular chloroquine, hydroxychloroquine and lopinavir-ritonavir were investigated for the treatment of COVID-19, and only remdesivir obtained approval from the US Food and Drug Administration for inpatients including children ≥ 12 years old and weighing ≥ 40 kg as well as adults [1213141516171819].

Nafamostat, a potent inhibitor of various serine proteases, was initially authorized as a short-acting anticoagulant and used for the treatment of pancreatitis in Japan and Korea for more than 20 years with a well-established safety profile [20]. Previous studies have established that serine protease inhibitors targeting TMPRSS2, such as nafamostat, can block SARS-CoV-2 entry and has been demonstrated in vitro and using animal models [212223]. It is currently in the clinical trials for the treatment of COVID-19, including a phase 3 clinical trial. In particular, the result of an open-label, randomized phase 2 clinical trial exhibited that in high-risk patients requiring oxygen treatment, nafamostat had a considerably higher recovery and a lower mortality rates in comparison to standard care alone. However, there was no significant discrepancy in time to clinical improvement between nafamostat and standard care [24]. A main route of elimination of nafamostat represents the hydrolytic metabolism to the inactive metabolites such as 6-amino-2-nphthol (AN) and p-guanidinobenzoic acid (pGBA), mainly through arylesterases and carboxylesterase 2, and thus, the unaltered form of nafamostat is less detectable in urine and feces in in vitro or in vivo assessment [25262728].

Generally, since pediatric clinical researches are difficult to perform and occur with various ethical issues, pharmacokinetic (PK) modeling and simulation are useful tools to define the risk- benefit ratio of a new drug in pediatric drug development and to allow investigators to determine the personalized precision dosing schemes [29]. Furthermore, the models maximize the utilization of available data and simultaneously minimize the need for unnecessary clinical trials by crossing gaps from adults and supporting efficient clinical trial design [30313233]. Especially as the physiologically-based pharmacokinetic (PBPK) modeling approach in pediatrics takes into account ontogeny of metabolic enzymes and age dependent anatomical and physiological changes, it provides more reliable predictions of plasma drug concentrations for the optimization of the clinical trial design for the recommendation of initial doses in children than traditional allometric scaling [3034].

This study aims to establish a whole body intravenous PBPK model of nafamostat in adults and to scale down the adult PBPK model to children and provide pediatric exposures of nafamostat for different age groups and suggest an optimal weight-based nafamostat dose in pediatric populations.

METHODS

Software used

PK-Sim software of the Open Systems Pharmacology Suite, version 8.0 was used to develop the PBPK models. Model parameter optimization was achieved using the Monte Carlo algorithm executed in PK-Sim. Clinical data from the publication were extracted and digitized by means of WebPlotDigitizer web-based tool (version 4.5; Ankit Rohatgi, Oakland, CA, USA). Data analysis and graphics were performed with the R programming language version 3.6.2 (R Foundation for Statistical Computing, Seoul, Korea) and R Studio version 1.2.5033 (R Studio, Inc, CA, USA).

Data collection

Owing to the limited availability of relevant clinical data for nafamostat, the PK data was from Chinese healthy volunteers that had been administered three doses (10, 20, and 40 mg) of nafamostat intravenously (i.v.) over 2hr and was only used for model development [35]. There was a specific consideration in the corresponding PBPK model in terms of the PK collection of data, anatomical and physiological features of the subject and the study design as well, i.e., dose and administration intervals. For population simulations, mean patient PBPK models were utilized. By altering anatomical and physiological parameters for 1,000 individuals, a simulated population of a mean PBPK model was generated [36].

PBPK model development in adults

The dose, physicochemical properties, and in vitro metabolic elimination that were utilized for the final nafamostat PBPK model in adults are listed in Table 1. For the development of adult PBPK model of nafamostat, the information about a) physicochemical properties, b) distribution, metabolism and excretion processes, and c) clinical design for the above mentioned i.v. infusion of nafamostat was obtained and used not only to implement relevant metabolic enzymes but also to inform drug-specific input parameters. Since collected data could not be adequately obtained from the literature, parameter estimation was conducted by fitting the model to observed PK data to acquire valuable input parameters for model. Observed, computed, or assumed parameters were integrated as follows. 1) physicochemical properties of nafamostat such as molecular weight, cLogP, fraction unbound, and blood/plasma ratio values were calculated based on chemical structure (DrugBank, https://go.drugbank.com/). 2) distribution estimation of tissue partition ratio was needed for accurate reproduction of disposition PK profile. Nafamostat typically exhibits two compartment distribution profile. However, The PK profile implemented by the initial estimated parameter showed one compartment distribution. To fit to the observed PK profile, we modified and optimized the partition coefficient and permeability of the muscle, one of the major large organs which can affect distribution of PK but is not the infection site to influence to efficacy of the drug. In addition, the parameters for distribution between blood cell and plasma, i.e., partition coefficient (blood cells/plasma) and permeability from blood cell to plasma, were also modulated to finetune to the PK profile, which the final estimates of them were not considerably deviated. 3) in vitro kinetic data for metabolism of nafamostat via carboxylesterase2 and arylesterase was inputed to recover observed nafamostat area under the concentration-time curve (AUC). The PK Sim expression data was used to inform tissue expression distribution of the implemented metabolic enzymes [37]. Since PK-Sim does not provide information about tissue specific expression for subtype of carboxylesterase, we incorporate the kinetic parameters into the integrated carboxylesterase process. The literature did not report the subtype of arylesterase, i.e., paraoxonase 1 and 2 (PON1 and 2), involved in metabolism of nafamostat. With this limitation, the kinetic parameters were evenly incorporated into two subtypes of arylesterase. 4) nafamostat is transported by organic cation transporters (OCTs) in the basolateral membrane of proximal tube. However, it harbors a short plasma half-life owing to high hydrolytic metabolism and activities of OCTs for nafamostat in kidney are not expected to result in significant changes in nafamostat plasma exposure. Because of this, OCTs transport for nafamostat was not incorporated into the model and the exclusion of this disposition mechanism are not anticipated to considerably affect predicted pediatric nafamostat exposure. 5) As only negligible levels of the parent nafamostat can be detected in urine and feces, arbitrary low values of renal or biliary excretion parameters were incorporated into the model.

Table 1

Nafamostat key parameters for adult PBPK development

Parameters		Initial estimate, unit	Final estimate, unit	Reference/Comment
Basic Physico-chemistry
	Molecular weight	378.38 g/mol	378.38 g/mol	Drug Bank
	Lipophilicity (logP)	2.75	2.59	Drug Bank and fitted
	Fraction unbound	0.73	0.50	Drug Bank and fitted
	pKa	11.32	11.32	Drug Bank
	Solubility	0.03 mg/mL	0.03 mg/mL	Drug Bank and fitted
Distribution
	Specific organ permeability	0.004 cm/min	0.004 cm/min	PK-Sim standard
	Partition coefficient (blood cells/plasma)	3.85	4.45	Calculated and fitted by PK-Sim
	Permeability from blood cell to plasma	0.0047 cm/min	0.0068 cm/min	Calculated and fitted by PK-Sim
	Partition coefficient (interstitial/plasma) in muscle	9.03	110.94	Calculated and fitted by PK-Sim
	Permeability between plasma and interstitial in muscle	0.02 cm/min	9.97 cm/min	Calculated and fitted by PK-Sim
Metabolism
	Carboxylesterase 2, Km	1,790 μM	1,790 μM	[28]
	Carboxylesterase 2, Vmax	26.9 nmol/min/mg protein	26.9 nmol/min/mg protein	[28]
	Arylesterase, Km	628 μM	628 μM	[28]
	Arylesterase, Vmax	140 nmol/ml/min specific enzyme	140 nmol/ml/min specific enzyme	[28]
Excretion
	Renal clearance: GFR fraction	0.01	0.01	Arbitrary low value
	Biliary clearance	0.01 1/min	0.01 1/min	Arbitrary low value

GFR, glomerular filtration.

Pediatric scaling and model application

The model was scaled to adolescent, child, young child, infant, and neonate for prediction of PK in the respective populations after the establishment of the adult PBPK model. Each age group was defined as follows: Neonate (–4 weeks postnatal age), Infant (4 weeks–2 years postnatal age), Young child (2–6 years), Child (6–12 years), Adolescent (12–18 years), Adult (18–45 years). To build the pediatric PBPK model, both anthropometric and physiologic parameters, as well as tissue concentration of metabolic enzymes, were scaled to values of the corresponding key population explaining age-dependent changes. In particular, the composition and size of tissue compartments, maturation and protein binding of elimination process are all factors to consider. Since there is no ontogeny function currently presenting for arylesterases and carboxylesterase 2 in PK-Sim, the ontogeny patterns for activities of the enzymes was found in published literature and applied to the PK-Sim platform [3839]. Plasma concentration-time profiles in pediatric populations was simulated subsequently by using the extrapolated PBPK model. To compare the effect of the PK prediction using the pediatric PBPK models, a typical allometric scaling approach was used. Here, the clearance was scaled by allometry from adults to the pediatric populations with exponent of 0.75 and normalized to the body weight of 56.4 kg, the mean body weight in simulation data used for the adult PBPK model development, as follows.

PBPK model evaluation

Due to the scarcity of PK data from pediatric population, only adult PBPK models were evaluated using the following approaches. In goodness-of-fit plots, the predicted plasma concentrations and their corresponding observed values were compared. Furthermore, the observed plasma concentration-time profiles from adult were visually compared to the predicted plasma PK profile from PBPK models. To predict the variability of plasma PK profile, virtual populations of 100 individuals representing the respective clinical trial population were created. The predictions of population were plotted as median with 95% prediction interval. The geometric mean fold errors (GMFE) of AUC and maximum plasma concentration (Cmax) ratios were calculated. As a reference, a two-fold error range from the observed values for model predictions was taken. Such a range is considered appropriate for a predicted model and also commonly reported by other investigators [404142].

RESULTS

Adult PBPK model development and evaluation

Following a thorough review of the literature, two PK studies in healthy adults with i.v. administration of the same dosages were identified [4344]. However, their exposures were different, up to around two-fold, due to a different quantitative method. We selected the more recently published PK study, which was performed in 30 Chinese healthy subjects and quantified by liquid chromatography-mass spectrometry (LCMS). A Japanese population model of PK-Sim was used for simulating adult PK profiles and exposure as the Japanese were considered a more closely related ethnic group for the respective real population, i.e., Chinese, in the population model provided by PK-Sim. To match the observed clinical study, 100 virtual subjects with ages ranging from 20 to 26 years and an equal portion of males and females were simulated. The whole-body adult PBPK model adequately predicted plasma concentration-time profiles of nafamostat following i.v. administration. Visual comparisons of the predicted to observed plasma PK profiles are depicted in Fig. 1. The goodness-of-fit of predicted versus observed plasma concentrations is illustrated in Fig. 2. The GMFE values for the adult PBPK model were 1.65 and 1.21 for AUC and Cmax, respectively. Furthermore, 100% of all predicted plasma concentrations fall within two-fold of the respective observed concentrations. The PBPK model predictions for the fraction of nafamostat metabolism was ~99% and the unchanged fraction excreted in urine or feces was negligible, which is in concordance with the literature [25262728]. Two factors for tissue distribution of nafamostat were estimated in order to more accurately describe observed disposition profile of it.

Figure 1

The predicted and observed plasma concentration-time profile after intravenous nafamostat administration in adults with semi-log scale (A-C) and linear scale (D-F): a single dose of intravenous administration of 10, 20, and 40 mg, respectively. Population simulations (n = 100) are shown as line with shaded areas, which represents median and 95% prediction interval, respectively. Observed data are shown as points.

iv, intravenous.

Figure 2

The goodness-of-fit plot after intravenous nafamostat administration in adults. Predicted versus observed plasma concentrations are displayed. Each point represents a single plasma concentration. Red line indicates the line of identity. Yellow and blue line mark 2-fold and 3-fold deviation, respectively.

Pediatric PBPK model development by scaling the adult model

The adult PBPK model was scaled to predict pediatric exposure of nafamostat after i.v. administration of the highest weight-based dosing regimen being used in an adult phase II and III study, 0.2 mg/kg/h during 14 days. This is because 0.2 mg/kg/h is the approved dose of nafamostat for disseminated intravascular coagulation (DIC) and acute pancreatitis and has been broadly well tolerated in clinical trials [2024]. The model predicted the exposure of nafamostat according to six different age groups as depicted in Fig. 3. As age advances from neonate to adult, AUC and Cmax of nafamostat slightly increased from neonate to infant, continuously decreased to child with nadir, and then increased from child to adult. The fold change of predicted AUC for the respective pediatric groups over those of adults after the administration of the same weight based dosing regimen were 0.88, 0.92, 0.85, 0.78, and 0.92 for neonate, infant, young child, child, and adolescent subjects, respectively. The fold change of predicted Cmax for the respective pediatric group over those of adults after the administration of the same weight-based dosing regimen were 0.87, 0.92, 0.84, 0.78, and 0.92 for neonate, infant, young child, child, and adolescent subjects, respectively

Figure 3

The predicted exposure after intravenous nafamostat administration of 0.2 mg/kg/h during 14 days in six different age groups. AUC (A) and Cmax (B) predicted by the physiologically-based pharmacokinetic model scaled from the adult population. Gray closed circles are predicted exposure values of simulated individual virtual subjects. Horizontal lines represent the median, with the top and the bottom of the boxes representing the first and third quantiles (IQR), whiskers indicate the extreme data within 1.5 × IQR.

AUC, area under the curve; Cmax, maximum plasma concentration; IQR, interquartile range.

Comparison of pediatric exposures predicted by PBPK and allometry

A classical allometric scaling approach was used (body weight exponent = 0.75) to predict pediatric AUC and Cmax of nafamostat and directly compare with those predicted by the PBPK model. As shown in Fig. 4, we found that both methodologies predicted similarly for adult, adolescent, and child subjects, whereas there is a remarkable deviation between PBPK and allometry prediction in neonate, infant, and young child subjects.

Figure 4

The comparison between AUC (A) and Cmax (B) predicted by allometry and the PBPK model after intravenous nafamostat administration of 0.2 mg/kg/h during 14 days in six different age groups. Red boxes and points indicate AUC (A) and Cmax (B) predicted by allometry, blue boxes and points indicate AUC (A) and Cmax (B) predicted by the PBPK model. Open circles are predicted AUC (A) and Cmax (B) values of individual simulated virtual subjects. Horizontal lines represent the median, with the top and the bottom of the boxes representing the first and third quantiles (IQR), whiskers indicate the extreme data within 1.5 × IQR.

AUC, area under the curve; Cmax, maximum plasma concentration; IQR, interquartile range.

DISCUSSION

As the COVID-19 pandemic is rapidly outspreading and the number of pediatric patients is gradually growing, severe cases for children such as MIS-C are being reported [2345678945]. However, studies that have investigated pharmacotherapy for the COVID-19 treatment have been performed mostly in adults [46]. Nafamostat is currently being explored for the treatment of COVID-19 in adults and the data that support its exposure in children remain limited. The unmet medical need for providing nafamostat as a therapeutic option in pediatrics demand the use of PBPK modeling and simulation to optimize pharmacotherapies.

In this study, PBPK models of nafamostat for an adult population have been successfully developed. The model provides a consistent representation of dose-exposure relationship following i.v. administration of a dose range and adequately describes plasma concentration-time profiles of nafamostat. More importantly, by scaling the adult PBPK model to pediatrics, we delineate the potential of PBPK modeling approach to more reasonably predict PK in children.

Nafamostat is unstable in plasma and a highly polar drug with few practical methods for its in vivo quantification and only an insufficient number of PK studies with valid nafamostat concentration measurements available [434447]. In the same context, the recent randomized controlled study exploring safety and PK/pharmacodynamics reported that the majority of patients exhibited undetectable levels of nafamostat [unpublished data]. We selected a PK study quantified using LCMS rather than a radioisotope-labeling method which often overestimates the parent drug because the metabolites also consist of radioisotope, particularly for unstable compounds. The adult PBPK model was first established using a PK study of i.v. administration of three doses (10, 20, and 40 mg). The kinetic parameters for nafamostat metabolism were input using in vitro experimental data from the literature and several parameters were fitted utilizing population means of the PK study. The resulting adult PBPK model was able to predict nafamostat exposure reliably over a dose range, indicating that the crucial processes driving nafamostat PK were adequately captured. The initial establishment of the model in adults offered a modeling strategy that served as a solid foundation for age extrapolation to improve accuracy of the pediatric model predictions.

Based on the concept of defining absorption, distribution, metabolism and excretion as a function of anatomy, physiology, and biochemical reaction, PBPK modeling and simulation provides the chance of reasonable scaling between adults and pediatrics. Such a strategy is prevalent in the establishment of models in children and is already utilized by other investigators [414248]. This study defined six different age groups for predictions of pediatric clearance. Based on the information, including size change and ontogeny of metabolic enzyme of nafamostat, i.e., arylesterases and carboxylesterase 2, the exposures foreach pediatric group was predicted using a virtual population of 100 individuals from that age category. As PK-Sim does not currently equip an ontogeny function for arylesterases and carboxylesterase 2, it was user-provided from the literature and integrated into PK-Sim [3839]. In the case of microsomal carboxylesterase 2, its expression increases across the three consecutive age groups, which are children from birth to 3 weeks, between 3 weeks and 6 years, and over 6 years [38]. In the case of arylesterases, its activity continuously increases from birth to 7 years [39]. As age increases, the predicted nafamostat exposures are slightly raised from neonate to infant followed by a steady decrease from infant to child, and then increased from child to adult, as shown in Fig. 3. Whereas the predicted exposures of child subjects were 0.78 fold lower than those of adults, the predicted exposures of all other pediatric groups fell within 0.85 fold when compared to those of adults, indicating the pediatric exposures predicted by PBPK model are not substantially different from those of adults. Based on these finding, we cautiously may recommend a weight based initial dosing regimen for each age group. Assuming 0.2 mg/kg/h currently approved for DIC and acute pancreatitis and being conducted in a clinical trial for COVID-19 is suitable for adults, for neonate, infant, young child, child, and adolescent subjects, the adult dosage regimen is likely to be increased with 0.23 mg/kg/h, 0.22 mg/kg/h, 0.24 mg/kg/h, 0.25 mg/kg/h, and 0.22 mg/kg/h, respectively. For child subjects, the largest dose changes relative to the adult dosage regimen may be considered. Considering nafamostat is primarily eliminated by metabolism, pediatric clearance can be determined by size as well as maturation of the metabolic enzyme, i.e., arylesterases and carboxylesterase 2. In pediatric PBPK models, most efforts have been contributed to the incorporation of age-dependent changes in metabolic clearance. Especially for the enzyme cytochrome P450s (CYP) developmental patterns and clearance of drugs mainly metabolized by these enzymes are relatively well documented [49]. Generally, because of differences in enzyme levels, drugs highly metabolized are administered at a higher mg/kg dose in young child compared with newborns, which was consistent in the case of nafamostat [50]. The hepatic clearance of the drugs can be higher in infants and child as liver blood flow is increased in comparison with adults due to the increased ratio of liver to total body mass in the preceding group [51]. Thus, drugs primarily metabolized in the liver are likely to exhibit a lower exposure rate compared to adults. Similarly, in nafamostat, the predicted exposure of infants, young child, and child were lower than that of adults.

The selection of dose in pediatric populations is generally obtained from adult PK data through two common approaches, PBPK and allometric scaling [525354]. In this study, we delineated nafamostat AUC in pediatric patients based on a PBPK model and compared the results from allometric scaling. The comparison analysis indicated that there was a notable separation in predicted pediatric exposure between the methods in neonate, infant, and young child populations, while those of adult, adolescent, and child subjects coincided between them. Generally, PBPK models explain enzyme ontogeny and age dependent alteration in organ development and function. Therefore, they can provide more reliable prediction of plasma drug concentrations [3052]. Allometric scaling does not account for drug specific disposition mechanisms, instead extrapolating exposure based on body size and fixed exponent (usually 0.75), which can result in large overestimations of metabolic clearance in very young children due to their immature enzymes, which is in line with our finding [3034]. When compared to simple allometry, PBPK models tend to predict higher exposures in younger children. Therefore, PBPK models present more conventional predictions with respect to safety.

This study had several limitations. First, since available PK studies with valid nafamostat concentrations were limited, we utilized only one study using three dosages for establishing the PBPK model and did not validate the model. Second, our recommended pediatric dosing regimen assumes that the exposure-response relationship in pediatric populations will be similar to that of adults and can be rationally extrapolated from adults. Third, owing to the scarcity of available observed pediatric PK data of nafamostat, the predictive performance of the PBPK model could not be verified.

Our study provides reasonable evidence to recommend a nafamostat weight-based dosing regimen with efficacious exposure in pediatric COVID-19 patients. A PBPK model has been established that adequately captures the observed PK profile of nafamostat in adult healthy volunteers. By scaling this model to pediatric populations, the pediatric exposures of nafamostat were predicted and reasonable pediatric doses were cautiously recommended, aiding future investigations of nafamostat PKs in pediatric populations, including the design of clinical trials and precision dosing.