Prediction of a clinically effective dose of THY1773, a novel V 1B receptor antagonist, based on preclinical data

Abstract THY1773 is a novel arginine vasopressin 1B (V1B) receptor antagonist that is under development as an oral drug for the treatment of major depressive disorder (MDD). Here we report our strategy to predict a clinically effective dose of THY1773 for MDD in the preclinical stage, and discuss the important insights gained by retrospective analysis of prediction accuracy. To predict human pharmacokinetic (PK) parameters, several extrapolation methods from animal or in vitro data to humans were investigated. The fu correction intercept method and two‐species‐based allometry were used to extrapolate clearance from rats and dogs to humans. The physiologically based pharmacokinetics (PBPK)/receptor occupancy (RO) model was developed by linking free plasma concentration with pituitary V1B RO by the Emax model. As a result, the predicted clinically effective dose of THY1773 associated with 50% V1B RO was low enough (10 mg/day, or at maximum 110 mg/day) to warrant entering phase 1 clinical trials. In the phase 1 single ascending dose study, TS‐121 capsule (active ingredient: THY1773) showed favorable PKs for THY1773 as expected, and in the separately conducted phase 1 RO study using positron emission tomography, the observed pituitary V1B RO was comparable to our prediction. Retrospective analysis of the prediction accuracy suggested that the prediction methods considering plasma protein binding, and avoiding having to apply unknown scaling factors obtained in animals to humans, would lead to better prediction. Selecting mechanism‐based methods with reasonable assumptions would be critical for the successful prediction of a clinically effective dose in the preclinical stage of drug development.


| INTRODUCTION
Dysfunction of the hypothalamic-pituitary-adrenal (HPA) axis activity is observed in a subset of patients with major depressive disorder (MDD) (Holsboer, Haack, Gerken, & Vecsei, 1984). Arginine vasopressin (AVP) plays a crucial role in the regulation of the HPA axis in cooperation with corticotropin-releasing factor (CRF), and excess AVP is hypothesized to be responsible for HPA axis dysfunction in depression (de Winter et al., 2003;Purba, Hoogendijk, Hofman, & Swaab, 1996). AVP regulates HPA axis activity through adrenocorticotropic hormone (ACTH) secretion from the anterior pituitary by binding to its receptor subtype vasopressin 1B (V 1B ) receptor (Saito, Sugimoto, Tahara, & Kawashima, 1995;Wersinger, Ginns, O'Carroll, Lolait, & Young, 2002). Therefore, the V 1B receptor antagonist is expected to ameliorate the abnormalities of the HPA axis observed in depression, and consequently, to improve depressive symptoms in certain subpopulations of depressed individuals with impaired HPA axis function.
To date, several V 1B receptor antagonists have been proven to exhibit antidepressant-like effects in rodent models (Geneste et al., 2018;Griebel et al., 2002;Iijima et al., 2014). Among them, SSR149415, the first orally active V 1B receptor antagonist, was tested in double-blind, placebo-controlled studies for patients with MDD. However, SSR149415 failed to demonstrate clear antidepressant effects in the trials (Griebel, Beeské, & Stahl, 2012).
Although the reason for the failure was not fully understood, it is speculated that dose levels of SSR149415 were insufficient to show efficacy (Griebel , Beeské, & Stahl, 2012). Indeed, in a small scale depression trial (Katz, Locke, Greco, Liu, & Tracy, 2017), another V 1B receptor antagonist, ABT-436, has been reported to be associated with more favorable symptom changes at a dose that inhibits HPA function. Therefore, the therapeutic potential of V 1B receptor antagonists for MDD needs to be further investigated, especially at doses that would show adequate blockade of the V 1B receptor.
THY1773 is a potent and selective V 1B receptor antagonist, which blocks the anterior pituitary V 1B receptor and exhibits antidepressant-like effects in rodents (Kamiya et al., 2020). In rats, THY1773 occupied the anterior pituitary V 1B receptor dosedependently after oral administration and significantly inhibited the increase in plasma ACTH induced by CRF/desmopressin challenge at doses that showed occupancy of the anterior pituitary V 1B receptor of more than 50%. Consistent with these results, THY1773 significantly improved depressive-like behavior induced by repeated corticosterone injection, a model that induces a depressive state by impairing the HPA axis function (Kamiya et al., 2020), suggesting that THY1773 would be effective for patients with HPA axis hyperactivity, and appropriate for testing its efficacy for MDD.
The accurate prediction of a clinically effective dose is critical for improving efficiency and the success rate in drug development (Zou et al., 2012). The pharmacokinetics/pharmacodynamics (PK/PD) modeling and simulation method is a useful tool for the quantitative decision making on the selection of drug candidates, dosing regimens, and the optimization of study designs, particularly by providing a mechanistic rationale and avoiding inaccuracy caused by interspecies differences. For human PK parameter prediction, many methods have been proposed, and the systemic evaluations of these methods using various data sets of drugs have been reported (Fagerholm, 2007;Lombardo et al., 2013a;Ring et al., 2011). However, there is no perfect method that can predict human PK of all drugs accurately. It is important to select the most appropriate prediction method for each human PK parameter of each drug based on its PK properties. For the efficacy prediction, appropriate translation of preclinical dose-efficacy findings to human is critical. A target engaging biomarker, such as receptor occupancy (RO), which correlates with efficacy, plays an important role for this translation (Suhara et al., 2017). Establishing an appropriate quantitative model that links PK and target engaging biomarker dynamics is important for the determination of a dose that is predicted to be clinically effective with confidence.
Here, we report our strategy to predict a clinically effective dose of THY1773 for the treatment of MDD to support decision making in the preclinical stage. We applied the methods proposed by Ring et al. (2011) to predict clearance (CL). In addition, as observed human data have become available, the prediction accuracy was retrospectively evaluated, and important insights gained in selecting appropriate methods to successfully predict a clinically effective dose in the preclinical stage are also discussed.
Male SD rat and male beagle dog liver microsomes were purchased from Corning Gentest. All other chemicals and solvents were of analytical grade and obtained from commercial sources.

| Animals
Male SD rats (7-week old) were purchased from Charles River Laboratories Japan Inc. Experiments involving male beagle dogs (approximately 3 years; LSG Corporation) were conducted at HAMRI Co., Ltd. The animals were maintained under standard laboratory conditions. All animals were maintained on a standard laboratory animal diet and fasted 16-17 h prior to dosing. Feed was provided 4-5 h postdose and water was allowed ad libitum. All of the animal experimental procedures involving animal handling were approved INATANI ET AL.
-205 by the Institutional Animal Care and Use Committee of Taisho Pharmaceutical Co., Ltd (rats and dogs) and HAMRI Co., Ltd (dogs).

| Parallel artificial membrane permeability assay
The membrane permeability of THY1773 was measured using a parallel artificial membrane permeability assay (PAMPA) Evolution instrument (Pion Inc.). A "sandwich" plate consisting of a donor bottom plate with stir bars and an acceptor filter plate was used for the experiment. A donor plate was prepared by the addition of THY1773 to diluted PRISMA HT at pH 6.2 (final concentration of 12.5 μmol/L). An acceptor plate was coated with GIT-0 lipid solution, followed by the addition of acceptor sink buffer (pH 7.4). After incubation for 4 h at room temperature, the concentrations of THY1773 in the reference, donor, and acceptor solutions were measured with a UV plate reader.
The apparent permeability (P app ) was calculated using the PAMPA Evolution software (version 3.1; Pion Inc.).

| Blood-to-plasma ratio
The blood-to-plasma ratios of THY1773 were analyzed using pooled fresh whole blood (anticoagulant: EDTA-2K) and control plasma separated from fresh whole blood in parallel from rats, dogs, and humans. Whole blood and control plasma containing 1 μmol/L THY1773 were incubated at 37°C for 30 min in a shaking water bath.
The incubated whole blood was centrifuged at 2000 � g for 10 min at 4°C to obtain plasma. An aliquot of each plasma sample (i.e., isolated plasma from incubated whole blood and control plasma) was mixed with 8 volumes of acetonitrile/methanol (9:1, vol/vol) containing the IS. After centrifugation at 3974 � g for 10 min at 4°C, the supernatant was subjected to liquid chromatography/tandem mass spectrometry (LC-MS/MS) analysis. The blood-to-plasma ratio was calculated by dividing the THY1773 concentration in control plasma, which is the same as the concentration in whole blood, by that in the separated plasma from incubated whole blood.

| Plasma protein binding
The plasma protein binding of THY1773 in rats, dogs, and humans was evaluated by an equilibrium dialysis method using a 96-well equilibrium dialysis plate (HTDialysis) with a 12-14 kDa cut-off dialysis membrane. THY1773 was dissolved in dimethyl sulfoxide, and spiked with rat, dog, or human plasma at a final concentration of 300 ng/ml. An aliquot of plasma containing THY1773 and sodium phosphate buffer (pH 7.4) was added into the donor side and the receiver side of each designated well, respectively. After incubation for 6 h at a rate of 50 oscillation/min at 37°C in an air incubator, an aliquot of each side of each well was collected and mixed with acetonitrile/methanol (9:1, vol/vol) containing the IS. Each sample was centrifuged at 3974 � g for 10 min at 4°C to remove the precipitated proteins, and the resultant supernatants were subjected to LC-MS/MS analysis. The protein binding (%) was calculated using the Boudinot formula (Boudinot & Jusko, 1984

| Metabolic stability in animal liver microsomes
The oxidative metabolism of THY1773 was evaluated in rat and dog liver microsomes. THY1773 (1 μmol/L) was incubated with pooled liver microsomes (0.25 mg protein/ml) in sodium-potassium phosphate buffer (pH 7.4) consisting of 100 mmol/L sodium phosphate, 100 mmol/L potassium chloride, 2.5 mmol/L magnesium chloride (MgCl 2 ), 1.5 mmol/L glucose-6-phosphate (G-6-P), and 0.18 units/ml G-6-P dehydrogenase (G-6-PDH). The reactions were initiated by the addition of β-nicotinamide-adenine dinucleotide phosphate, oxidized form (NADP + , 0.16 mmol/L), and incubated for 10, 20, 30, 40, 50, and 60 min. The reactions were terminated by the addition of an equal volume of acetonitrile/methanol (9:1, vol/vol) containing the IS, followed by centrifugation at 3974 � g for 10 min at 4°C. An aliquot of the supernatant was subjected to LC-MS/MS analysis. The intrinsic clearance (CL int ) was calculated by the substrate depletion method described by Naritomi et al. (2001 (10 μmol/L) at 37°C for 5 min. The reactions were initiated by the addition of NADPH-generating system (1 mmol/L NADP + , 5 mmol/L G-6-P, 1 unit/ml G-6-PDH, and 5 mmol/L MgCl 2 ) and incubated for 0 or 60 min. The reactions were terminated by the addition of an equal volume of acetonitrile/methanol (9:1, vol/vol), followed by centrifugation at 1650 � g for 10 min at 4°C. An aliquot of the supernatant was subjected to an Agilent 1200 HPLC system (Agilent Technologies, Inc.) equipped with a radiochemical flow detector (Radiomatic 625TR; Perkin Elmer, Inc.), and formation of metabolites was evaluated. Reaction phenotyping study was also conducted using HLMs and selective chemical inhibitors for CYP2D6 and CYP3A, which were assumed to have significant roles in THY1773 metabolism from recombinant metabolic stability. Quinidine (1 μmol/L) for CYP2D6 and ketoconazole (1 μmol/L) for CYP3A were used as selective inhibitors. THY1773 (500 nmol/L) was incubated with pooled HLMs (0.5 mg protein/ml) in potassium phosphate buffer (100 mmol/L, pH 7.4) in the absence or presence of the selective inhibitors. The reactions were initiated by the addition of NADPH generating system (1.3 mmol/L β-NADP + , 3.3 mmol/L G-6-P, 3.3 mmol/L MgCl 2 , and 0.4 units/ml G-6-PDH), and incubated at 37°C for 30 min. The reactions were terminated by the addition of two volumes of acetonitrile/methanol (9:1, vol/vol) containing the IS. After centrifugation at 3974 � g for 10 min at 4°C, an aliquot of the supernatant was subjected to LC-MS/MS analysis. The CL int was calculated by the method as described previously under metabolic stability in animal liver microsomes, and the fraction metabolized (fm) was calculated by Equation (1): where CL int is intrinsic clearance without specific inhibitor and CL int,with inh is intrinsic clearance in the presence of specific inhibitor.

| PK in rats and dogs
The PK and urinary excretion of THY1773 in rats (1 mg/kg) and dogs (

| PK analysis
The plasma concentration-time profiles of THY1773 were analyzed by a non-compartmental analysis with PK analysis software Phoenix WinNonlin® (version 6.1; Certara), and the PK parameters were calculated. The fraction excreted in urine was calculated by dividing the amount of THY1773 excreted in urine by the administered dose.

| CL
Of the many reported methods for the prediction of human CL, the top two methods reported by Ring et al. (2011) were selected based on prediction accuracy; that is, the unbound fraction corrected intercept method (FCIM) and two-species-based allometry (rat and dog) described by Equations (2) and (3), respectively (Tang, Hussain, Leal, Mayersohn, & Fluhler, 2007;Tang & Mayersohn, 2005): Human body weight ðkgÞ ð2Þ where the allometric coefficient (a) is obtained from the intercept of the simple allometry log-log plot between rats and dogs, and Rf u is the ratio of unbound fraction in plasma between rats and humans.
Human body weight was assumed to be 70 kg. The hepatic CL (CL liver ) was also predicted by physiologically based in vitro-in vivo extrapolation (IVIVE) method using physiological parameters listed in Table 1 based on the well stirred model (Yang, Jamei, Yeo, Rostami-Hodjegan, & Tucker, 2007).
Since the CL liver was underestimated in rats and dogs from CL int in liver microsomes, the ratio of observed in vivo CL int and predicted CL int was calculated as the empirical scaling factor (ESF) for the extrapolation of CL liver from in vitro data. Human CL liver was predicted from CL int in HLMs with and without applying the average of ESF derived from rats and dogs. The total plasma clearance (CL total ) was calculated by Equation (4): where Q H is hepatic blood flow, f u is the free fraction in plasma, f u,mic is the fraction unbound in HLMs, R b is blood-to-plasma ratio, and CL renal is renal clearance. The f u,mic value was calculated using the Simcyp population based simulator (Version 16 R1; Simcyp Ltd.). The CL renal was estimated by multiplying f u by glomerular filtration rate (GFR), assuming glomerular filtration only.

| Volume of distribution
Human volume of distribution at steady state (Vd ss ) was predicted by single species scaling from rat assuming that the unbound Vd ss is comparable across species, as described in Equation (5): where f u is the unbound fraction in plasma. Human Vd ss was also predicted by the two methods reported to be the best methods based on the prediction accuracy , which were Øie-Tozer -207 method (Obach et al., 1997) and rat-dog-human proportionality equation described (Wajima, Fukumura, Yano, & Oguma, 2003) by Equations (6) and (7), respectively: where f u,t is the fraction unbound in tissues, R E/I is the ratio of binding protein in the extracellular fluid to that in plasma, V e , V p , and V r is the extracellular, plasma, and remaining fluid volume, respectively. Volumes and plasma protein levels were used as described by Obach et al. (1997).

| PBPK/RO model development
The human PBPK/RO models were developed using Simcyp simulator following the previously reported strategy (Jones, Parrott, Jorga, & Lavé, 2006). Since Vd ss in rat was underpredicted using tissue-to-plasma partition method as proposed by Rodgers and Rowland (2007), the tissue-to-plasma partition coefficient (Kp) scalar of 0.422 was applied to all human tissue compartments to match the predicted human Vd ss by the Øie-Tozer method. The CL int in HLMs were inputted into Simcyp as CYP2D6 and CYP3A4 dependent CL int based on the estimated fraction metabolized in HLMs, with (Model 1) or without (Model 2) applying ESF. The predicted CL renal assuming glomerular filtration only was also inputted. As the relationship between drug concentration and RO typically follow the E max model (Kim et al., 2012), the E max model described by Equation (8) was used as a PD model for THY1773 V 1B RO. As the pituitary gland is outside of the blood-brain barrier, the free concentration of THY1773 around the V 1B receptor in the pituitary gland was assumed to be equal to the free concentration in plasma: where C p,u is free concentration in plasma, E max is the maximum V 1B RO, and EC 50 is plasma free concentration required to produce 50%  (Kamiya et al., 2020). Therefore, the clinically effective dose was assumed to be the dose that maintains RO at more than 50% for the dosing interval (24 h after administration). methods. PK parameters were evaluated following the method as described in Section 2.5 PK analysis.

| Evaluation of prediction accuracy
The fold error of the difference between observed and predicted human PK parameters was calculated by dividing the predicted value by the observed value. The bioavailability of THY1773 was assumed to be 100%, and the volume of distribution at steady state was assumed to be comparable to that during the terminal phase.

| LC-MS/MS conditions
The LC-MS/MS system consisted of a Shimadzu LC-10ADvp, LC-20AD, LC-30AD, or Nexera (Shimadzu) and an API 3000, API 4000, or Triple Quad 5500 mass spectrometer (AB Sciex). The data were collected and processed using Analyst software (version 1.4 or later; AB Sciex). THY1773 was analyzed using a

| Permeability, plasma protein binding, and blood-to-plasma ratio
The passive membrane permeability of THY1773 was measured using PAMPA. The P app of THY1773 in PAMPA at pH 6.2 was 128.9 (10 −6 cm/s), suggesting high permeability in the gut. Plasma protein binding was determined by equilibrium dialysis at a THY1773 concentration of 300 ng/ml, which was roughly estimated as an average concentration in human plasma at an effective dose. The mean plasma protein binding of THY1773 at 300 ng/ml was 89.2% in rats, 89.6% in dogs, and 98.3% in humans. The blood-to-plasma ratio of THY1773 was evaluated in rat, dog, and human pooled whole blood at 1 μmol/L (484 ng/ml). The mean blood-to-plasma ratio was 0.822 in rats, 0.846 in dogs, and 0.540 in humans. THY1773 showed higher plasma protein binding and lower blood-to-plasma ratio in humans than in animals.

| Metabolic stability in animal liver microsomes
The oxidative metabolism of THY1773 in rat and dog liver microsomes was evaluated at 1 μmol/L, at which the reaction was assumed to be linear. The CL int (in units of μL/min/mg protein) in liver microsomes was 22.0 for rats and 32.2 for dogs.

| PK in rats and dogs
The concentration-time profiles in rats and dogs after a single intravenous or oral administration of THY1773 under fasted conditions are shown in Figure 1, and the PK parameters are summarized in Table 4. In rats and dogs, the CL total was moderate (approximately 30% of hepatic blood flow) after single intravenous administration.
The urinary excretion of unchanged THY1773 within 24 h postintravenous dosing was 4.6% in rats and 2.0% in dogs. In both species, the CL renal of THY1773 was comparable to f u � GFR (within twofold), which suggested that active transport was not likely to play an important role in the CL renal . Following oral administration, THY1773 was absorbed rapidly, and oral bioavailability reached 46.4% in rats and 45.0% in dogs. The concentration ratio of brain and pituitary gland to plasma in rats at 1 h after oral administration was 0.2 and 4.7, respectively.

| Prediction of human PK parameters
Of many reported methods for the prediction of human PK parameters, the methods reported to be relatively accurate among various methods using data sets containing lipophilic and highly plasma protein-bound drugs Ring et al., 2011) were selected, as THY1773 was also lipophilic and highly plasma proteinbound. The human CL total was predicted by FCIM, two-speciesbased allometry, or IVIVE method (Table 5). Before conducting IVIVE, the discrepancy between in vitro and in vivo CL int was evaluated in rats and dogs. The in vivo CL int was higher than in vitro CL int in both species, and the ESF for IVIVE was calculated to be 6.5 in rats and 3.6 in dogs. The predicted human CL total values varied greatly depending on the prediction methods, and the difference was over 14-fold. The human Vd ss was predicted by rat single species scaling with Equation (5), Øie-Tozer method, or rat-dog-human proportionality equation (Table 6). The predicted Vd ss values varied depending on the prediction methods, and the difference was over threefold.

| Prediction of a clinically effective dose
The plasma concentration-and RO-time profiles in humans were simulated by two PBPK/RO models considering best and worst case scenarios for human liver CL int . The simulated plasma con- Value is presented as the mean of duplicate determinations.
F I G U R E 1 THY1773 plasma concentration-time profiles (mean ± SD, n = 3) in male rats (left) and dogs (right) after a single intravenous (iv) or oral (po) administration. The plasma concentrations declined below the lower limit of quantification (<0.1 ng/ml) at 24 h except for after intravenous dose to dogs.

T A B L E 4
Pharmacokinetic parameters of THY1773 after a single intravenous or oral administration to male rats (1 mg/kg) and dogs (0.5 mg/kg)

| Evaluation of prediction accuracy
The observed oral blood CL was only 2.6% of the hepatic blood flow, which indicated that hepatic metabolism of THY1773 was not extensive in humans. As THY1773 was mainly metabolized by CYP3A in HLMs, we assumed that gut metabolism, which was known to be mainly mediated by CYP3A, was not extensive in humans either.
Then we assumed that the human bioavailability of THY1773 was 100% to compare the predicted human PK parameters with the observed values obtained after oral administration. The fold errors of the difference between observed and predicted human PK parameters are shown in Table 5 (CL total ) and Table 6

| DISCUSSION AND CONCLUSION
The development of drugs for central nervous system (CNS) diseases is challenging relative to other therapeutic areas (Harrison, 2016;Suhara et al., 2017). Modeling and simulation technology has been used widely as an initiative to overcome the difficulty and to improve success rates of drug discovery and development (Visser, De Alwis, Kerbusch, Stone, & Allerheiligen, 2014). We developed THY1773, a novel selective V 1B receptor antagonist that exhibited antidepressant-like effects in animal models (Kamiya et al., 2020). In the present studies, we established PBPK/RO models of THY1773 fully based on the preclinical data and predicted a clinically effective dose for MDD to support decision making in the preclinical stage of drug development.
In rats and dogs, THY1773 was rapidly and well absorbed with moderate bioavailability, and exhibited moderate CL total and Vd ss .
The renal excretion ratio was less than 5%. As THY1773 had high permeability in PAMPA (a class 2 compound in the extended CL classification system; Varma, Steyn, Allerton, & El-Kattan, 2015), THY1773 was assumed to be primarily eliminated by hepatic metabolism. THY1773 showed species differences in plasma protein binding and blood-to-plasma ratio. THY1773 is a CYP3A4 substrate, and mainly metabolized by CYP3A in HLMs. The CL int in HLMs was lower than that in rats and dogs. From these preclinical PK properties, THY1773 was expected to be well absorbed with good bioavailability and eliminated with moderate CL mainly mediated by CYP3A4 metabolism in humans. Based on these preclinical data, a PBPK/RO model of THY1773 in humans was established for the prediction of a clinically effective dose for MDD.
First, human CL and Vd ss were predicted by several methods.
Human CL was predicted by two allometric and two IVIVE methods, and we found that widely different values were obtained depending on the methods. Of these methods, we selected IVIVE methods for the following reasons: (1) the plasma protein binding of THY1773 was higher in humans than in animals, and it has been reported that allometric methods that did not account for protein binding tended to overpredict human CL (Ring et al., 2011); and (2) as THY1773 is a CYP3A4 substrate, the CL liver , or intestinal F I G U R E 3 The plasma concentration-time profiles (mean ± SD) of THY1773 after oral administration in the phase 1 single ascending dose study. Healthy male or female volunteers received a single oral dose of either TS-121 capsules or placebo in a fasted state.

T A B L E 7
Observed human PK parameters of THY1773 after a single oral administration (30 mg) -213 first pass, was expected to be relatively accurately predicted using CL int in HLMs (Bowman & Benet, 2019b;Gertz, Harrison, Houston, & Galetin, 2010). Since IVIVE methods using HLMs reportedly have a tendency to underpredict human CL liver (Wood, Houston, & Hallifax, 2017), the IVIVE method, with or without ESF obtained from rats and dogs, was used for THY1773 CL liver prediction as the worst or best case scenario, respectively. To predict CL total , CL renal was predicted assuming glomerular filtration only as in animals, and added to the predicted CL liver . The human Vd ss was predicted by two empirical and one semi-mechanistic method. The predicted values were different depending on whether or not the plasma protein binding was considered. The unbound Vd ss was comparable between rats and dogs, suggesting that the assumption in rat single species scaling with Equation (5), (i.e., the unbound Vd ss is comparable across species), was valid, and the predicted human Vd ss would therefore be confident. We ultimately selected the semi-mechanistic Øie-Tozer method as it was reported to be most accurate for many compounds , and the predicted Vd ss was comparable to that predicted by rat single species scaling with Equation (5). with TS-121 in MDD patients at doses of 10 and 50 mg, which would achieve more than 50% RO at steady state (Kamiya et al., 2020).

Parameter AUC 0-∞ (h·ng/ml) C max (ng/ml) t max a (h) t 1/2 (h) CL po (L/h) Vd z /F (L)
Although further investigation of the dose-efficacy relationship is required, these findings support our assumption that a clinically effective dose that maintains RO at more than 50% for the dosing interval is achievable.
A retrospective analysis of the prediction accuracy was conducted after obtaining the observed data. For CL total , only the IVIVE method without ESF could predict human CL total within twofold. This finding was consistent with the previous report that the IVIVE method using human microsomal data provided better prediction than the other methods for compounds eliminated primarily by CYPmediated mechanisms (Hosea et al., 2009 The PBPK model using human liver microsomal CL int without ESF (Model 1) described well the human plasma concentration-time profiles (Figure 4). In the PBPK model, as it was reported that CL int in human intestinal microsomes and HLMs were not significantly different after normalization for tissue-specific CYP3A abundance (Gertz, Harrison, Houston, & Galetin, 2010), CYP3A4 metabolism in the gut was estimated from CYP3A4 CL int in HLMs, accounting for the differences in enzyme abundance between the liver and intestine.
The fraction escaping gut first-pass metabolism was predicted to be 0.97, and the bioavailability was predicted to be 95% by Model 1, which was almost consistent with our assumption for the evaluation of prediction accuracy of PK parameters.
Retrospective analysis of the PBPK/RO model also revealed the limitation of extrapolation of animal data to humans.
According to the results of the phase 1 RO study using PET, the in vivo EC 50 value for V 1B RO was comparable (within twofold) to the in vitro K i value for human V 1B receptor binding inhibition (unpublished data). Based on the results of a rat V 1B RO study using a tritium labeled V 1B receptor ligand (Kamiya et al., 2020), the EC 50 value for V 1B RO of THY1773 in the rat anterior pituitary was estimated to be 4.8 nmol/L, which was much lower than the in vitro K i value of 80.1 nmol/L for rat anterior pituitary membrane V 1B receptor (unpublished data). This discrepancy (17fold higher affinity in vivo) was not likely to be explained by the target concentration because, even if the observed concentration ratio of the pituitary gland to plasma (4.7) was assumed to be applicable for free concentration, it would be much lower than the observed discrepancy. Also, hysteresis (time delay between target concentration and RO) was not likely to explain the discrepancy because hysteresis would result in a higher in vivo EC 50 value than the in vitro K i value, as the RO was evaluated at approximately t max . The discrepancy was possibly due to a more complex biological mechanism in vivo, but such a complex mechanism was difficult to include in the RO model and translate to humans based on the limited data obtained in the preclinical stage. For the prediction of a human effective dose, the discrepancy between in vitro K i and in vivo EC 50 observed in rats was not considered because there was no reasonable hypothesis that the discrepancy would be the same in humans, and we aimed to avoid the risk of lowering the probability of success of clinical trials by underpredicting a clinically effective dose. In addition, in the PET imaging study in rhesus monkeys, as previously reported (Koga et al., 2017), the monkey in vivo EC 50 value of THY1773 for V 1B RO was similar to the in vitro K i value for human recombinant V 1B receptor binding inhibition (unpublished data). Even though the in vitro K i value for monkey V 1B receptor was not evaluated, we assumed that the monkey in vivo EC 50 value for V 1B RO was comparable to the in vitro K i value based on the high degree of V 1B receptor protein homology (96%) between human and monkey, according to the HomoloGene database (https://www.ncbi.nlm. nih.gov/homologene/22678). The aforementioned results provided the rationale for our approach. Retrospective analysis revealed that the discrepancy observed in rats was not observed in humans, and that the unknown scaling factor obtained in preclinical animal studies should not be applied to humans unless the mechanism is clear.
In summary, we established a human PBPK/RO model of THY1773, and succeeded in nearly accurately predicting a clinically effective dose for the treatment of MDD in the preclinical stage. To predict human PK parameters and a clinically effective dose more accurately in the preclinical stage, it is important to identify the most promising prediction method of various available methods. From our studies, it is suggested that mechanism based methods should be selected with reasonable assumptions for the successful predictions.
Our studies also showed the limitations of extrapolation of animal data to humans. Caution should be exercised when applying an in vitro-in vivo scaling factor obtained in the preclinical animal studies INATANI ET AL.
to humans, especially when the reason for the in vitro-in vivo discrepancy is not clear.