Prediction of human pharmacokinetics for low-clearance compounds using pharmacokinetic data from chimeric mice with humanized livers.

Development of low-clearance (CL) compounds that are slowly metabolized is a major goal in the pharmaceutical industry. However, the pursuit of low intrinsic CL (CLint ) often leads to significant challenges in evaluating the pharmacokinetics of such compounds. Although in vitro-in vivo extrapolation is widely used to predict human CL, its application has been limited for low-CLint compounds because of the low turnover of parent compounds in metabolic stability assays. To address this issue, we focused on chimeric mice with humanized livers (PXB-mice), which have been increasingly reported to accurately predict human CL in recent years. The predictive accuracy for nine low-CLint compounds with no significant turnover in a human hepatocyte assay was investigated using PXB-mouse methods, such as single-species allometric scaling (PXB-SSS) approach and a novel physiologically based scaling (PXB-PBS) approach that assumes that the CLint per hepatocyte is equal between humans and PXB-mice. The percentages of compounds with predicted CL within 2- and 3-fold ranges of the observed CL for low-CLint compounds were 89% and 100%, respectively, for both PXB-SSS and PXB-PBS approaches. Moreover, the predicted CL was mostly consistent among the methods. Conversely, the percentages of compounds with predicted CL within 2- and 3-fold ranges of the observed CL for low-CLint compounds were 50% and 63%, respectively, for multispecies allometric (MA) scaling. Overall, these PXB-mouse methods were much more accurate than conventional MA scaling approaches, suggesting that PXB-mice are useful tools for predicting the human CL of low-CLint compounds that are slowly metabolized.

WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

It is important to identify low‐clearance (CL) compounds that are slowly metabolized during the drug discovery process, but the ability of in vitro–in vivo extrapolation to predict human CL decreases as the value of intrinsic CL (CLint) decreases. Although chimeric mice with humanized livers (PXB‐mice) have been reported to be useful for predicting human CL, their applicability to low‐CLint compounds with no significant turnover in human hepatocyte assays has not been clarified.

WHAT QUESTION DID THIS STUDY ADDRESS?

This study examined the predictive accuracy of PXB‐mouse methods for low‐CLint compounds.

WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

In addition to the previously reported single‐species allometric scaling approach, we proposed a novel physiologically based scaling approach. Both methods displayed much greater predictive accuracy for low‐CLint compounds than conventional multispecies allometric scaling approaches.

HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

The findings should improve the accuracy of human pharmacokinetic prediction and enable efficient and safe first‐in‐human studies.

INTRODUCTION

The accurate prediction of human pharmacokinetics using preclinical data is essential at the drug discovery stage because it provides important insights into drug candidate selection and first‐in‐human study design. Clearance (CL) is a key determinant of human pharmacokinetics, and several approaches, such as in vitro–in vivo extrapolation (IVIVE) and multispecies allometric (MA) scaling, have been widely used to predict human CL. Furthermore, the proportion of low‐CL compounds in drug discovery portfolios has increased in the pharmaceutical industry to approximately 30%. , This is largely attributable to the tendency to extract metabolically stable compounds during high‐throughput absorption, distribution, metabolism, and excretion (ADME) screening to achieve adequate exposure at a lower dosage in humans. In other words, as a result of the selection of low intrinsic CL (CLint) compounds, it has become increasingly common that IVIVE cannot be applied because of the low turnover of the parent compound in microsomal or hepatocyte stability assays. In these in vitro assays, it is difficult to estimate the in vitro CLint of low‐CLint compounds because of the limited incubation time for which enzymatic activity can be maintained.

Chimeric mice with humanized livers (PXB‐mice) were generated from urokinase‐type plasminogen activator‐cDNA/severe combined immunodeficiency mice injected with human hepatocytes. A wide range of drug‐metabolizing enzymes and transporters were expressed in the PXB‐mice liver, this is considering the fact that greater than 80% of mouse hepatocytes are replaced by human hepatocytes. , Therefore, PXB‐mice represent a useful animal model for predicting human CL. Furthermore, the single‐species allometric scaling (PXB‐SSS) approach has been reported to provide high predictive accuracy , and to be applicable to compounds that undergo hepatic organic anion‐transporting polypeptide‐mediated transport and compounds with long half‐lives. However, the predictive accuracy has not been investigated for low‐CLint compounds with no significant turnover of the parent compound in human hepatocyte assays.

This study evaluated the usefulness of PXB‐mouse methods for predicting the human CL of low‐CLint compounds compared with the conventional MA scaling approach. Furthermore, we proposed a novel physiologically based scaling (PXB‐PBS) approach that assumes that in vivo CLint per hepatocyte is equal between humans and PXB‐mice based on the concept that mouse‐derived hepatocytes in PXB‐mice are mostly replaced by human hepatocytes. Whereas the PXB‐SSS approach is an empirical, simple equation using body weight and exponents, the PXB‐PBS approach has physiological significance and aids in understanding the prediction of CL using PXB‐mice, which can lead to improved confidence in its application in drug discovery. In addition, the PXB‐PBS approach can potentially provide more accurate CL predictions than the PXB‐SSS approach because it can incorporate more detailed information, such as physiological parameters and compound‐dependent pharmacokinetic parameters.

METHODS

Definition of low‐CLint compounds

A total of 16 commercially available compounds were selected, including those expected to have low CLint based on clinical data (<5 ml/min/kg ) and those metabolized by cytochrome P450 (CYP) and non‐CYP enzymes. Because most of these compounds were metabolized in the liver, this study assumed that total CL (CLt) is equal to hepatic CL. An in vitro metabolic stability assay using cryopreserved human hepatocytes purchased from BioIVT (ZOL, 10‐donor mixed gender pool, Baltimore, MD, USA) was conducted to confirm whether the compounds exhibited low turnover. Tested compounds at a final concentration of 0.1 or 1 µmol/l were incubated at a cell density of 0.5 million cells/ml for 4 h with human hepatocytes suspended in Williams’ E medium containing 0.125% bovine serum albumin, 15 mmol/l HEPES, and 2 mmol/l GlutaMAX‐1. Low‐CLint compounds in this study were defined as those that did not display significant turnover (<20% ; Table 1). By contrast, compounds that exhibited significant turnover in human hepatocyte assays were classified as moderate‐ to high‐CLint compounds in this study. The details of chemicals and reagents used in this study and the detailed procedures of the human hepatocyte assay are provided in the Supporting information.

TABLE 1

Summary of drug disposition and in vitro and in vivo parameters to predict human CLt for all compounds

Compounds	Disposition	In vitro parameters					In vivo parameters
		% remaining after a 4‐h incubation	CLint per hepatocyte (µl/min/1 × 106 cells)	R b a	f u,p		CLt (ml/min/kg)
			Humans	Humans	Humans	Rats	PXB‐mice	Rats	Monkeys	Dogs
Antipyrine	P450	96	ND	1	0.990	0.690	4.60	7.86	11.5	7.22
Bosentan	P450, OATP	69	12.3	0.48	0.023	0.012	8.78	16.40	20.6	1.72
Carbazeran	AO	0	96.9	1	0.148	0.169	99.67	42.22	87.87	11.5
Dapsone	P450, NAT	94	ND	1.04	0.381	0.280	4.40	6.87	5.08	1.21
Diazepam	P450	62	3.9	0.71	0.036	0.171	42.147	61.51	17.90	46.71
Disopyramide	P450	71	3.2	1.2	0.161	0.610	28.86	179.46	19	29
Doxazosin	P450	11	41.5	1	0.066	0.050	34.245	30.00	15.47	11.21
Ranitidine	P450, FMO	83	ND	1	1.000	0.900	131.02	99.88	40.23	10.4
Reboxetine	P450	35	13.9	1	0.041	0.253	17.658	61.51	14.94	22.39
(S)‐Naproxen	P450, UGT	102	ND	0.55	0.007	0.008	0.556	0.41	0.83	0.04
(S)‐Warfarin	P450	80	ND	0.55	0.013	0.005	0.484	0.21	0.102	1.49
Tenoxicam	P450	96	ND	0.67	0.015	0.030	0.2878	0.49	0.061	0.104
Theophylline	P450	96	ND	0.85	0.580	0.400	4.70	1.91	1.08	1.73
Timolol	P450	85	ND	0.84	0.715	0.760	130.05	137.02	13.6	–b
Tolbutamide	P450	89	ND	0.55	0.039	0.049	0.58	0.39	0.0456	0.142
UCN−01	–b	4	83.5	1	0.003	0.0175	0.0369	77.37	3.36	10.27

References are provided in Table S1.

Abbreviations: CLint, intrinsic clearance; R b, blood‐to‐plasma concentration ratio; f u,p, fraction unbound in plasma; CLt, total clearance; PXB‐mice, chimeric mice with humanized livers; ND, not determined because of the absence of significant turnover (<20%) during a 4‐h incubation in the human hepatocyte assay; OATP, organic anion‐transporting polypeptide; AO, aldehyde oxidase; NAT, N‐acetyltransferase; FMO, flavin‐containing monooxygenase; UGT, UDP‐glucuronosyltransferase.

R b was assumed to be 1 for carbazeran, doxazosin, ranitidine, reboxetine, and UCN‐01 because of a lack of data in the literature.

Not available in the literature.

Data collection

In vitro and in vivo data for all compounds to predict human CL are summarized in Table 1. CLt in humans, PXB‐mice, rats, monkeys, and dogs and fraction unbound in plasma (f u,p) in humans and rats were obtained from experiments or the literature. References are provided in Table S1. In vitro CLint,human per hepatocyte was calculated by correcting the compound disappearance rate in human hepatocyte assays with fraction unbound in hepatocytes and medium. Further, f u,p in humans and rats was measured by the equilibrium dialysis method using a Rapid Equilibrium Dialysis Device purchased from Thermo Fisher Scientific. The detailed methods for the determination of in vitro CLint and f u,p are provided in the Supporting information. At various sampling points, the plasma samples after intravenous administration to PXB‐mice (N = 3) at a cassette dose of 0.1 mg/kg were collected. The plasma samples were extracted with acetonitrile containing the internal standard verapamil, and then the plasma concentration was determined using a liquid chromatography/tandem mass spectrometry method. The details of tandem mass spectrometry are provided in Table S2. Moreover, cassette doses of 0.1 mg/kg were administered intravenously to rats and monkeys (N = 3 each), and the plasma concentration was determined in a similar manner. CLt was calculated via noncompartmental analysis using Phoenix WinNonlin version 6.3 (Certara) based on the plasma concentration–time profiles following intravenous dosing. The sources of animals were as follows: 3 male PXB‐mice weighing 18.5–21.5 g and aged 17–19 weeks at dosing were supplied by PhoenixBio Co., Ltd., and the replacement rate of mouse hepatocytes with human hepatocytes was 89%–91%; 3 male Sprague–Dawley rats weighing 242–261 g and aged 6 weeks at dosing were supplied by Charles River Laboratories Japan, Inc.; and 3 male cynomolgus monkeys weighing 2.9–3.6 kg and aged approximately 3.6–3.8 years at dosing were supplied by Hamri Co., Ltd. All in vivo experiments were approved by the Institutional Animal Care and Use Committee of Mitsubishi Tanabe Pharma Corporation (Kanagawa, Japan).

In vitro–in vivo extrapolation

In vitro CLint,human per hepatocyte was converted to scaled CLint,human using human scaling factors, such as liver weight (LW), body weight (BW), and hepatocellularity, which are presented in Table 2, according to Equation 1. CLt,human was calculated from scaled CLint,human via a dispersion model (Equation 2) using hepatic blood flow (Q h), which is presented in Table 2, and fraction unbound in blood (f u,b) was calculated using the human blood‐to‐plasma concentration ratio (R b) obtained from the literature.

TABLE 2

Scaling factors for the IVIVE and PXB‐PBS approaches

Species	Q h (ml/min/kg)	Liver weight (g)	Body weight (kg)	Hepatocellularity (1 × 106 cells/g liver)
Humans	20	1470	70	120
PXB‐mice	91.3	1.977	0.02	168

Abbreviations: IVIVE, in vitro–in vivo extrapolation; PXB‐mice, chimeric mice with humanized livers; PXB‐PBS, physiologically based scaling using PXB‐mice; Q h, hepatic blood flow.

In Equation 2, , and .

MA scaling from nonclinical animal species

CLt,human was predicted using the rule of exponent (ROE) and f u corrected intercept method (FCIM). According to ROE, CLt values in rats, monkeys, and dogs were plotted against BW on a log–log scale as simple allometry (SA), and then CLt,human was predicted by an allometric equation (Equation 3, 4, or 5) that was based on the exponent values obtained from SA (Equation 3). SA was used when the exponents of SA ranged from 0.55 to 0.70. Maximum life‐span potential (MLP) correction was applied when SA was 0.71–1.0; that is, the product of MLP and CLt for each animal was plotted as a function of BW on a log–log scale using Equation 4. When SA greater than or equal to 1.0, brain weight (BrW) correction was applied; that is, the product of BrW and CLt for each animal was plotted in a similar manner using Equation 5. In these equations, a, b, and c were the coefficients of the allometric equations, and x, y, and z were the exponents. BW, BrW, and MLP were set at 0.25 kg, 0.00174 kg, and 4.4 years, respectively, for rats; 3.75 kg, 0.0424 kg, and 18.5 years, respectively, for monkeys; and 12 kg, 0.0754 kg, and 20.5 years, respectively, for dogs. MLP in years was calculated as a function of BW and BrW for each animal according to Equation 6.

According to FCIM, CLt,human was predicted using Equation 7. In this equation, a was the intercept obtained from the log–log plot of CLt versus BW, and Rf u,p was the ratio of f u,p for rats and humans.

PXB‐SSS

CLt,human was predicted by the PXB‐SSS approach using Equation 8. BW was set at 70 and 0.02 kg for humans and PXB‐mice, respectively.

PXB‐PBS

CLt,human was predicted using the PXB‐PBS approach following several steps (Figure 1). Based on the scaling factors of PXB‐mice presented in Table 2, scaled CLint,PXB was calculated from CLt,PXB via a dispersion model (Equation 2) using Q h and f u,b, and then in vivo CLint,PXB per hepatocyte was calculated (Equation 1). The estimation or measurement methods of LW and hepatocellularity in PXB‐mice, which were used as scaling factors, are provided in the Supporting information. Based on the concept that mouse‐derived hepatocytes in PXB‐mice are largely replaced by human hepatocytes, in vivo CLint per hepatocyte was assumed to be equal between humans and PXB‐mice (Equation 9). The same assumption was applied for f u,p and R b. Based on the assumption that Q h is equal in normal and PXB‐mice, the Q h values presented in Table 2 were used. The inverse conversion was performed by a dispersion model using human scaling factors, and then CLt,human was estimated. For compounds for which CLt,PXB exceeds Q h, CLt was set at 90% of Q h.

FIGURE 1

Scheme for predicting human clearance (CL) for low intrinsic CL compounds with no significant turnover in human hepatocyte assays. PXB‐mice, chimeric mice with humanized livers that were repopulated with human hepatocytes. PXB‐SSS, prediction method based on single‐species allometric scaling using PXB‐mice. PXB‐PBS, prediction method based on physiologically based scaling using PXB‐mice. CLt, CLint, and BW, total clearance, intrinsic clearance, and body weight, respectively

Evaluation of predictive accuracy

Observed CLt,human was compared with the predicted value for the IVIVE, PXB‐SSS, PXB‐PBS, ROE, and FCIM approaches to calculate the fold error. For each low‐CLint compound, moderate‐ to high‐CLint compounds, and all compounds, the percentages predicted within 2‐ and 3‐fold ranges of the observed CLt,human were calculated. Moreover, the geometric mean of the ratio between the predicted and observed values, which was frequently used as the absolute average fold error (AAFE), was calculated according to Equation 10, and subsequently the predictability of each method was compared. For low‐CLint compounds, the PXB‐SSS, PXB‐PBS, ROE, and FCIM approaches were compared, and IVIVE was additionally included in the comparison for moderate‐ to high‐CLint compounds and all compounds.

Sensitivity analysis

To fully understand the PXB‐PBS approach, sensitivity analyses were conducted using a dataset of (S)‐naproxen as a representative of low‐CLint compounds and diazepam as a representative of moderate‐ to high‐CLint compounds. The impact of parameters such as f u,p, R b, and Q h, which included assumptions in the PXB‐PBS approach in this study, on predicted CLt,human was investigated. Simulated CLt,human was defined as the CL generated by changing each parameter, and the magnitude of change from the predicted CLt,human was evaluated according to Equation 11 for an increased predicted value of CLt,human and Equation 12 for a decreased predicted value of CLt,human. These equations were primarily used to express the changes in predicted CLt,human on a similar magnitude for both positive and negative value and to evaluate the compounds on the same scale on a 3D plot, regardless of the observed CLt,human. Sensitivity analysis was conducted for f u,p in humans and PXB‐mice within a 3‐fold range of human f u,p, for R b in humans and PXB‐mice within the range of 0.5–2, for Q h in PXB‐mice within the range of 90–180 ml/min/kg, and for Q h in humans within the range of 18–23 ml/min/kg, which were frequently used as physiological parameters in literature.

RESULTS

Determination of low‐CLint compounds based on human hepatocyte assays

As the purpose of this study was to verify whether the prediction method using PXB‐mice was useful for low‐CLint compounds, we first conducted an in vitro metabolic stability assay using cryopreserved human hepatocytes to identify low‐CLint compounds. The proportion of the parent compound remaining after 4 h of incubation is presented in Table 1. Antipyrine, dapsone, ranitidine, (S)‐naproxen, (S)‐warfarin, tenoxicam, theophylline, timolol, and tolbutamide met the criteria for low‐CLint compounds; that is, they did not display significant turnover (<20%) during a 4‐h incubation. Conversely, bosentan, carbazeran, diazepam, disopyramide, doxazosin, reboxetine, and UCN‐01 (7‐hydroxystaurosporine) were classified as moderate‐ to high‐CLint compounds because they exhibited significant turnover. The plots of the proportion of each parent compound remaining after 4 h of incubation are summarized in Figure S1.

Prediction of human CL by IVIVE, MA, and PXB‐mouse methods

For low‐CLint compounds, CL was predicted using conventional MA scaling approaches and PXB‐mouse methods, whereas for moderate‐ to high‐CLint compounds, CL was predicted by all methods including IVIVE. The relationships between observed and predicted CLt,human are presented in Figure 2. The observed and predicted CLt,human and the fold errors are summarized in Table 3. To compare the predictability of each method, both the percentages predicted within 2‐ and 3‐fold ranges of the observed value and AAFE are summarized in Table 4.

FIGURE 2

Relationships between observed and predicted total human clearance (CLt,human) for low intrinsic clearance (CLint) compounds and moderate‐ to high‐CLint compounds. Panels (a), (b), (c), (d), and (e) presents the results of single‐species allometric scaling from chimeric mice with humanized livers (PXB‐mice), a physiologically based scaling using PXB‐mice, in vitro–in vivo extrapolation (IVIVE), rule of exponent, and the f u corrected intercept method, respectively. Solid and dotted lines represent the unity and 3‐fold error, respectively. PXB‐SSS, prediction method based on single‐species allometric scaling using PXB‐mice; PXB‐PBS, prediction method based on physiologically based scaling using PXB‐mice; ROE, rule of exponent; FCIM, f u corrected intercept method

TABLE 3

Comparisons of observed and predicted CLt,human for all compounds

Category of compounds	Compounds	Observed CLt,human (ml/min/kg)	Predicted CLt,human (ml/min/kg)					Fold error (predicted/observed)
			PXB‐SSS	PXB‐PBS	IVIVE	ROE	FCIM	PXB‐SSS	PXB‐PBS	IVIVE	ROE	FCIM
Low‐CLint compounds	Antipyrine	0.64	0.59	0.71	ND	3.32	3.31	0.92	1.10	ND	5.19	5.17
	Dapsone	0.48	0.60	0.67	ND	0.91	2.01	1.24	1.40	ND	1.90	4.19
	Ranitidine	9.6	16.9	16.4	ND	5.4	11.1	1.76	1.70	ND	0.56	1.16
	(S)‐Naproxen	0.11	0.072	0.085	ND	0.051	0.189	0.65	0.77	ND	0.46	1.71
	(S)‐Warfarin	0.055	0.062	0.074	ND	0.417	0.330	1.13	1.34	ND	7.58	5.99
	Tenoxicam	0.03	0.039	0.044	ND	0.029	0.085	1.29	1.46	ND	0.98	2.83
	Theophylline	0.86	0.60	0.72	ND	0.68	0.91	0.70	0.84	ND	0.79	1.06
	Timolol	8.5	16.8	13.7	ND	NC	NC	1.98	1.62	ND	NC	NC
	Tolbutamide	0.21	0.074	0.088	ND	0.040	0.107	0.35	0.42	ND	0.19	0.51
Moderate‐ to high‐CLint compounds	Bosentan	2.1	1.18	1.39	0.68	1.68	5.37	0.56	0.66	0.32	0.80	2.56
	Carbazeran	37.6	12.8	16.4	15.5	8.6	7.6	0.34	0.44	0.41	0.23	0.20
	Diazepam	0.38	5.68	7.52	0.69	12.46	2.53	14.94	19.80	1.82	32.79	6.67
	Disopyramide	0.9	3.72	4.63	1.24	7.36	4.56	4.14	5.15	1.38	8.17	5.06
	Doxazosin	1.6	4.61	5.65	5.62	3.93	6.20	2.88	3.53	3.51	2.46	3.88
	Reboxetine	0.82	2.39	2.79	1.37	9.52	1.82	2.91	3.40	1.67	11.61	2.22
	UCN−01	0.0037	0.0050	0.0056	0.6186	1.5640	1.3765	1.35	1.52	167.20	422.70	372.03

References are provided in Table S1.

Abbreviations: CLint, intrinsic clearance; CLt,human, total human clearance; FCIM, f u corrected intercept method; IVIVE, in vitro–in vivo extrapolation; NC, not calculated because of a lack of data; ND, not determined because of the absence of significant turnover (<20%) during a 4‐h incubation in the human hepatocyte assay; PXB‐PBS, physiologically based scaling using chimeric mice with humanized livers; PXB‐SSS, single‐species allometric scaling using chimeric mice with humanized livers; ROE, rule of exponents.

TABLE 4

Comparative evaluation of various prediction methods

	PXB‐SSS	PXB‐PBS	IVIVE	ROE	FCIM
Low‐CLint compounds
Number of compounds	9	9	ND	8	8
Within 2‐fold error (%)	89	89	ND	50	50
Within 3‐fold error (%)	100	100	ND	63	63
AAFE	1.51	1.46	ND	2.58	2.50
Moderate‐ to high‐CLint compounds
Number of compounds	7	7	7	7	7
Within 2‐fold error (%)	29	29	43	14	0
Within 3‐fold error (%)	71	43	57	29	29
AAFE	3.23	3.50	4.07	10.84	7.52
All compounds
Number of compounds	16	16	7	15	15
Within 2‐fold error (%)	63	63	43	33	27
Within 3‐fold error (%)	88	75	57	47	47
AAFE	2.11	2.14	4.07	5.04	4.18

Abbreviations: CLint, intrinsic clearance; PXB‐SSS, single‐species allometric scaling using chimeric mice with humanized livers; PXB‐PBS, physiologically based scaling using chimeric mice with humanized livers; IVIVE, in vitro–in vivo extrapolation; ROE, rule of exponents; FCIM, f u corrected intercept method; ND, not determined because of the absence of significant turnover (<20%) during a 4‐h incubation in the human hepatocyte assay; AAFE, absolute average fold error.

The percentage predicted within 2‐fold ranges of the observed CLt,human for low‐CLint compounds was 89% for both the PXB‐SSS and PXB‐PBS approaches, and the AAFEs were 1.51 and 1.46, respectively. Furthermore, the percentage predicted within 3‐fold ranges of the observed CLt,human was 100% for both methods. Conversely, according to ROE and FCIM approaches, some compounds displayed large discrepancies exceeding 3‐fold between the observed and predicted values, leading to higher AAFEs, such as 2.58 and 2.50. When the predictability of each method was compared among moderate‐ to high‐CLint compounds, the percentages predicted within 3‐fold ranges of observed CLt,human values and AAFEs were 71% and 3.23, respectively, for the PXB‐SSS approach, 43% and 3.50, respectively, for the PXB‐PBS approach, 57% and 4.07, respectively, for IVIVE, 29% and 10.84, respectively, for the ROE approach, and 29% and 7.52, respectively, for the FCIM approach. It was suggested that both PXB‐SSS and PXB‐PBS approaches were more accurate than the MA scaling approach, but they had comparable accuracy as IVIVE. Thus, even for moderate‐ to high‐CLint compounds, PXB‐mouse methods displayed relatively high predictive accuracy, but large discrepancies were noted between the observed and predicted values for diazepam in PXB‐mouse methods, leading to large AAFEs. IVIVE also provided highly accurate predictions overall, but AAFE was large because of the considerably greater overprediction in UCN‐01. For all compounds, the percentages predicted within 3‐fold ranges of observed CLt,human and AAFE were 88% and 2.11, respectively, for the PXB‐SSS approach, 75% and 2.14, respectively, for the PXB‐PBS approach, 47% and 5.04, respectively, for the ROE approach, and 47% and 4.18, respectively, for the FCIM approach, suggesting that PXB‐mouse methods exhibited the best accuracy.

Predictive performance of the PXB‐PBS approach and investigation for critical parameters by sensitivity analysis

As illustrated in Table 3 and Figure S2, CLt,human predicted using the PXB‐PBS approach for all compounds was mostly consistent with that predicted using the PXB‐SSS approach, revealing the high predictive accuracy as described previously despite having several assumptions that in vivo CLint per hepatocyte, f u,p, and R b were equal between humans and PXB‐mice and that Q h was equal between normal and PXB‐mice. Sensitivity analysis was conducted using the (S)‐naproxen and diazepam dataset to examine the impact of f u,p, R b, and Q h on predicted CLt,human, and the results are summarized in Figure 3. When f u,p was larger in humans than in PXB‐mice, predicted CLt,human increased, but in the opposite case, predicted CLt,human decreased. The magnitude of this change was greater for (S)‐naproxen than for diazepam. Conversely, when there was no significant difference in f u,p between humans and PXB‐mice, predicted CLt,human was constant regardless of the value of f u,p for both compounds. R b in humans and PXB‐mice and Q h in humans and PXB‐mice did not significantly affect predicted CLt,human within the range examined for both compounds.

FIGURE 3

Sensitivity analysis of total human clearance (CLt,human) predicted using the PXB‐PBS approach. The impact of fraction unbound in plasma (f u,p), blood‐to‐plasma concentration ratio (R b), and hepatic blood flow (Q h) on predicted CLt,human value was examined using a dataset of (S)‐naproxen as a representative of low intrinsic clearance (CLint) compounds and diazepam as a representative of moderate‐ to high‐CLint compounds. Sensitivity analysis was conducted for each f u,p value in humans and PXB‐mice within a 3‐fold range of human f u,p, for each R b value in humans and PXB‐mice within a 0.5–2 range, and for each Q h value within the ranges of 90–180 ml/min/kg in PXB‐mice and 18–23 ml/min/kg in humans. The simulated CLt,human was defined as the clearance generated by changing each parameter, and the impact of these parameters on predicted CLt,human was evaluated as the percent change (%) calculated according to the following equations: ((simulated CLt,human −predicted CLt,human)/predicted CLt,human) ×100% for an increase in predicted CLt,human and ((simulated CLt,human −predicted CLt,human)/simulated CLt,human) ×100% for a decrease in predicted CLt,human. Panels (a), (b), and (c) present the results obtained using f u,p, R b, and Q h for (S)‐naproxen as variables, respectively. Panels (d), (e), and (f) present the results obtained using f u,p, R b, and Q h for diazepam as variables, respectively. PXB‐mice, chimeric mice with humanized livers; PXB‐PBS, physiologically based scaling using PXB‐mice

DISCUSSION

The discovery and development of low‐CLint compounds is one of the most important tasks common to all pharmaceutical companies. It is an important mission of drug metabolism and pharmacokinetics scientists to efficiently assess drug metabolism and accurately predict CLt,human during the drug discovery process. In suspended hepatocyte assays, low‐CLint compounds frequently display no significant turnover within the incubation period in which enzymatic activity can be maintained. Therefore, an in vitro metabolic stability assay to detect slow turnover for a longer period, such as the Hepatopac, HμREL, and relay methods, have been developed, whereas in vivo prediction methods specific for low‐CLint compounds have not been developed.

In this study, we used PXB‐mice to construct an alternative approach to the suspended hepatocyte assay. In fact, although it is difficult to accurately compare the predictive accuracy of these in vitro assays with PXB‐mouse methods because of the limited number of overlapping tested compounds, we propose two main advantages of PXB‐mouse methods. First, in vitro assays successfully maintained not only CYP activity but also non‐CYP activity for approximately 1 week. However, how a longer incubation period affects activity remains unclear, particularly for non‐CYP enzymes and transporters. In addition, impact of longer incubations on IVIVE predictive accuracy requires further elucidation. Conversely, PXB‐mice are not limited to the in vitro specific time‐dependent loss of activity. Therefore, it is easy to estimate CLt if appropriate blood sampling time points can be set. Second, it has been reported that CLint may lead to underprediction if compound binding to the culture ware or fibroblasts is not considered, whereas PXB‐mouse methods do not have such limitations. Overall, PXB‐mouse methods can provide high predictive accuracy, ease of use, and ease of interpretation, resulting in efficient drug discovery and development.

For the nine compounds defined in this study as low‐CLint compounds, the predictive accuracy of PXB‐mouse methods was evaluated, and the higher predictive accuracy was confirmed. By contrast, antipyrine, (S)‐warfarin, and tenoxicam exhibited more than 3‐fold discrepancies between the observed and predicted values in both the ROE and FCIM approaches. Although these MA scaling approaches using pharmacokinetic data from nonclinical animal species have been commonly used, their use in CL prediction for compounds with large interspecies difference in drug disposition has been limited. In addition, there are some compounds for which CL is overpredicted for unknown reasons. Antipyrine and (S)‐warfarin are already known to follow vertical allometry as compounds for which CL is overpredicted, and the present results again indicated the limitation of MA scaling approaches. Furthermore, the predictive accuracy of PXB‐mouse methods for moderate‐ to high‐CLint compounds was also higher than those of MA scaling approaches, but AAFE was similar to that of IVIVE. There have been some reports of overprediction for diazepam, which leads to a high AAFE in PXB‐SSS approach. , These reports suggest that the metabolic activity derived from mouse hepatocytes remaining in the livers of PXB‐mice may contribute to the overprediction, but these findings have not been clearly elucidated. Concurrently, PXB‐mouse methods were relatively less accurate for moderate‐ to high‐CLint compounds than for low‐CLint compounds. Sawada et al. noted a case of benzydamine, in which the flavin‐containing monooxygenase (FMO) enzymes that were expressed in the mouse kidney were attributed for causing the overprediction of CLt, and proposed that interspecies differences other than hepatic metabolism may lead to poor predictive accuracy. They may also have caused poor predictive accuracy in moderate‐ to high‐CLint compounds in this study; however, there is no evidence that such a trend is observed more than low‐CLint compounds, thereby requiring further investigation. Furthermore, despite having renal CL mechanism, ranitidine, which is a low‐CLint compound, successfully predicted CL by PXB‐SSS approach, but it was not successful for disopyramide, which is a moderate‐ to high‐CLint compound. Further analysis after measuring renal CL in PXB‐mice is warranted for more accurate predictions. On the contrary, overprediction of UCN‐01 also led to a high AAFE for the IVIVE method. Because UCN‐01 is known to bind to α1‐acid glycoprotein more strongly than to albumin and α1‐acid glycoprotein was not contained in this hepatocyte assay, such a difference between in vitro and in vivo conditions may have resulted in high CLint and a large overprediction. It is noteworthy that PXB‐mouse methods exhibited high predictive accuracy for UCN‐01, and IVIVE displayed high predictive accuracy for diazepam. Therefore, the prediction strategy of selecting PXB‐mouse methods or IVIVE according to the compound profile is most effective for moderate‐ to high‐CLint compounds. In IVIVE, a strategy for increasing confidence in the accuracy of the prediction by confirming the in vitro–in vivo correlation in at least two preclinical species has been proposed, but there is currently no clear method for prospectively recognizing whether PXB‐mouse methods are adaptive. Overall, it is expected that factors related to overprediction and the characteristics of the compound will be clarified and appropriate prediction strategies will be devised, or a prediction method exhibiting high predictive accuracy for all compounds will be found.

It has been reported that in vivo CLint per LW is correlated between humans and PXB‐mice, and cases of prediction in which the values were assumed to be equal have been reported. On the contrary, based on the concept that mouse‐derived hepatocytes in PXB‐mice are largely replaced by human hepatocytes, we proposed the PXB‐PBS approach on the assumption that in vivo CLint per hepatocyte was equal between humans and PXB‐mice. It has been proven that hepatocellularity should be considered to improve the predictive accuracy for CL ; thus, we also incorporated interspecies differences in hepatocellularity into the PXB‐PBS approach. This approach and sensitivity analysis of certain parameters provided several new findings. First, CLt,human predicted using the PXB‐PBS approach was mostly consistent with that predicted using the PXB‐SSS approach for all compounds. The PXB‐SSS approach is a simple allometric equation using the BW ratio and exponent. This exponent describes the relationship of physiological parameters between humans and PXB‐mice, and it might feature the same concepts as the various assumptions in the PXB‐PBS approach. Second, the results of sensitivity analysis demonstrated that a small interspecies difference in f u,p had little effect on predicted CLt,human, whereas a large interspecies difference in f u,p greatly affected predicted CLt,human. Because the PXB‐PBS approach exhibited high predictive accuracy for many compounds, it was suggested that f u,p might be similar between humans and PXB‐mice. In fact, human albumin and human α1‐acid glycoprotein involved in the plasma protein binding are secreted into the plasma of PXB‐mice , and Miyamoto et al. have reported that interspecies differences in f u,p are within 3‐fold for many compounds. However, the differences exceeded 3‐fold for diazepam, (S)‐naproxen, and UCN‐01, and f u,p was larger in PXB‐mice than in humans for each compound. As revealed in the sensitivity analysis, predicted CLt,human decreased when this interspecies difference was considered, which may have resulted in a relaxation of overprediction. Therefore, this interspecies difference in f u,p might be one of the causes of overprediction for diazepam. By contrast, for (S)‐naproxen and UCN‐01, considering these interspecies differences resulted in a less accurate CL prediction, and as Miyamoto et al. also stated, the factors responsible for these interspecies differences should be further investigated. Third, the result of sensitivity analysis illustrated that if f u,p was equal between humans and PXB‐mice, predicted CLt was constant regardless of the value of f u,p. This finding appeared useful for compounds with high plasma protein binding, for which it is difficult to accurately measure f u,p in terms of the sensitivity and robustness of the analytical method or necessity of setting appropriate experimental conditions. Finally, the results of sensitivity analysis indicated that R b or Q h did not significantly affect predicted CLt,human. In fact, there were no cases in which R b or Q h was accurately estimated in PXB‐mice, and this finding appeared meaningful.

In conclusion, PXB‐mouse methods predicted CLt,human for 89% of low‐CLint compounds within a 2‐fold range of the observed values and 100% of low‐CLint compounds within a 3‐fold range of the observed values. These prediction methods displayed much greater predictive accuracy than the conventional MA scaling approaches, indicating the usefulness of PXB‐mice as novel predictive tools for low‐CLint compounds with no significant turnover in human hepatocyte assays. Furthermore, CLt,human predicted using a novel PXB‐PBS approach was mostly consistent with that predicted using the PXB‐SSS approach for all 16 tested compounds. This finding and the results of sensitivity analysis suggested that the high predictive accuracy of the PXB‐SSS approach may be attributable to the similarity of CLint per hepatocyte and f u,p between humans and PXB‐mice. Overall, a translational understanding from a physiological perspective has increased the confidence in the application of PXB‐mice to drug discovery.

CONFLICT OF INTEREST

C.T. is an employee of PhoenixBio Co., Ltd. and S.S. received financial support from PhoenixBio Co., Ltd. in collaboration study. All other authors declared no competing interests for this work.

AUTHOR CONTRIBUTIONS

K.Y., T.N., T.W., C.T., S.S., and Y.K. wrote the manuscript. K.Y., N.I., H.S., and T.N. designed the research. K.Y., Y.D., N.I., H.Y., Y.I., and C.T. performed the research. K.Y. and Y.D. analyzed the data. K.Y. and Y.D. contributed new reagents/analytical tools.

Supplementary Material

Click here for additional data file.

Fig S1

Click here for additional data file.

Fig S2

Click here for additional data file.

Table S1

Click here for additional data file.

Table S2

Click here for additional data file.