A novel in vitro allometric scaling methodology for aldehyde oxidase substrates to enable selection of appropriate species for traditional allometry

Abstract

1. Failure to predict human pharmacokinetics of aldehyde oxidase (AO) substrates using traditional allometry has been attributed to species differences in AO metabolism.

2. To identify appropriate species for predicting human in vivo clearance by single species scaling (SSS) or multispecies allometry (MA), we scaled in vitro intrinsic clearance (CL_int) of five AO substrates obtained from hepatic S9 of mouse, rat, guinea pig, monkey, and minipig to human in vitro CL_int.

3. When predicting human in vitro CL_int, average absolute fold-error was ≤ 2.0 by SSS with monkey, minipig, and guinea pig (rat/mouse >3.0), and was <3.0 by most MA species combinations (including rat/mouse combinations).

4. Interspecies variables, including fraction metabolized by AO (F_m,AO) and hepatic extraction ratios (E) were estimated in vitro. SSS prediction fold-errors correlated with the animal:human ratio of E (r² =0.6488), but not F_m,AO (r² =0.0051).

5. Using plasma clearance (CL_p) from the literature, SSS with monkey was superior to rat or mouse at predicting human CL_p of BIBX1382 and zoniporide, consistent with in vitro SSS assessments.

6. Evaluation of in vitro allometry, F_m,AO, and E, may prove useful to guide selection of suitable species for traditional allometry and prediction of human pharmacokinetics of AO substrates.

Introduction

Aldehyde oxidase (AO) is a molybdo-flavoenzyme known to oxidize aldehydes and aromatic azaheterocycles, as well as reduce N- or S-oxides, nitro groups and some heterocycles. Interest in AO-mediated drug metabolism has increased in recent years as new generations of drug candidates increasingly contain AO-susceptible aromatic azaheterocyles (Pryde et al., 2010). Moreover, the termination of several promising development programs during clinical trial assessment due to unrecognized or underestimated AO metabolism highlights the necessity for innovative approaches to predict human pharmacokinetics (PK) and disposition where AO is the primary clearance mechanism. For example, discontinuation of BIBX1382 (Dittrich et al., 2002), FK3453 (Akabane et al., 2011), and RO1 (Zhang et al., 2011) during clinical trials resulted from unexpectedly poor oral bioavailability attributed to AO-mediated clearance that went unidentified in preclinical and in vitro studies. The use of microsomes (lacking cytosol, thus lacking AO) and preclinical species with decreased AO activity (e.g. rat), or in species altogether missing an AO metabolism gene (e.g. dog), ultimately resulted in a lack of clinical translation in PK and anticipated exposure of these candidate drugs (and their metabolites).

A challenge in predicting human drug disposition where AO-mediated metabolism predominates has been attributed to species differences in AO expression and activity. While humans express only one functional gene, AOX1, AO expression in other species commonly used to model human PK ranges anywhere from 2–4 isoforms (e.g. rat) to a complete absence of the gene in dog (dog does express non-hepatic, non-drug metabolizing isoforms) (Garattini and Terao, 2012). Human liver S9, cytosolic fractions, and hepatocytes have proven useful in identifying AO metabolism, although at times these systems have resulted in under-prediction or variable activity, proposed to be associated with possible single nucleotide polymorphisms (SNPs) in the AOX1 gene, instability of the dimer, or deficiency of the essential molybdenum cofactor (Hartmann et al., 2012, Hutzler et al., 2012, Fu et al., 2013, Hutzler et al., 2014b), as well as extra-hepatic expression of AO (Kurosaki et al., 1999, Moriwaki et al., 2001, Nishimura and Naito, 2006, Terao et al., 2016), although the contribution of these mechanisms are poorly understood. Consequently, promising drug candidates are routinely discarded or structurally modified to eliminate AO metabolism, often at the expense of pharmacological potency and/or selectivity.

While research has been conducted to evaluate human in vitro methods for direct scaling of AO-mediated clearance (Zientek et al., 2010, Hutzler et al., 2012, Hutzler et al., 2014b), studies investigating allometric scaling approaches are limited. Allometric scaling of in vivo plasma clearance in nonclinical species is commonly used to predict human total body clearance of drugs eliminated renally and/or via hepatic metabolism (Mahmood, 2007). However, the aforementioned differences observed in AO expression and activity of species commonly employed in allometric scaling (particularly dog) have resulted in decreased confidence in the utility of this method to predict AO-mediated clearance. Interestingly, similar to AO, metabolism mediated by uridine diphosphate-glucuronosyltransferases (UGTs) has been found to vary with species, yet success in predicting human clearance of UGT-metabolized drugs with multispecies allometry (MA) and single-species scaling (SSS) has been demonstrated (Deguchi et al., 2011). Furthermore, while rat underestimated the human plasma clearance of the AO substrate BIBX1382, a report by Hutzler et al. demonstrated comparable BIBX1382 plasma clearance (as a percentage of liver blood flow) between cynomolgus monkey and human, indicating SSS with monkey may be useful in predicting the clearance of drugs subject to AO metabolism (Hutzler et al., 2014a). Choughule et al., however, reported that relative hepatic intrinsic clearance (CL_int) mediated by AO in human, rhesus monkey, and guinea pig cytosol was substrate-dependent, suggesting that no single species could be reliably employed to consistently predict human clearance (Choughule et al., 2013b). Given that the ability to successfully predict human clearance of AO substrates by allometry using a particular species (or combination of species) is likely to be substrate-dependent in nature, we sought to investigate the utility of scaling in vitro CL_int of AO substrates with allometry to human in vitro CL_int as a tool to determine which species (or combination of species) may be suitable for conducting traditional allometric scaling of in vivo clearance. Specifically, the in vitro CL_int of five compounds (Figure 1) known to be cleared either predominantly or partially by AO in human was determined in hepatic S9 from multiple nonclinical species known to express hepatic AO (mouse, rat, guinea pig, cynomolgus monkey, rhesus monkey, and minipig). Subsequently, CL_int obtained from multispecies hepatic S9 incubations was subjected to MA or SSS to predict human in vitro hepatic CL_int, and a folderror analysis was used to evaluate which species may be most suitable for estimating human in vivo clearance. In addition, we estimated F_m,AO (fraction of metabolism mediated by AO) and hepatic extraction ratio (E) in each species (in vitro) to understand the potential influence of these variables on the accuracy of prediction by the scaling methods presented herein.

Figure 1.

Structures of AO substrates subjected to in vitro allometric scaling. Arrows indicate site of AO oxidation.

Materials and Methods

Materials

Potassium phosphate, formic acid, NADPH, MgCl₂, zaleplon, O⁶-benzylguanine, and hydralazine hydrochloride were purchased from Sigma-Aldrich (St. Louis, MO). Zoniporide dihydrochloride, BIBX1382 dihydrochloride, and SGX523 were purchased from Tocris Bioscience (R&D Systems, Minneapolis, MN). Pooled human hepatic S9 (150-donor, mixed gender pool) was obtained from BD Biosciences (San Diego, CA), and male Sprague-Dawley rat (n=36 pool), cynomolgus monkey (n=2, pool), and CD-1 mouse (n=170 pool) hepatic S9 were obtained from Corning Inc. (Tewksbury, MA). Male rhesus monkey (n= 6 pool) and Hartley guinea pig (n=50 pool) hepatic S9 were purchased from XenoTech (Lenexa, KS), and male Gottingen minipig (n= 7 pool) hepatic S9 was purchased from BioreclamationIVT (Baltimore, MD). All solvents used for bioanalysis were purchased from Sigma-Aldrich or Fisher Scientific (Waltham, MA) and were of high-performance liquid chromatography (HPLC) grade.

Incubations with Multispecies Hepatic S9 Fractions

Substrates (1 μM) were incubated at 37°C for 60 minutes in a potassium phosphatebuffered reaction (100 mM, pH 7.4) containing hepatic S9 from human (mixed gender) or male mouse, rat, guinea pig, cynomolgus monkey, rhesus monkey, or minipig (2.5 mg/mL; +/− NADPH, 1 mM) and MgCl₂ (3 mM), with preincubation in the presence or absence of hydralazine (50 μM) for estimation of fraction metabolized by AO (F_m,AO). Reactions were initiated with addition of substrate, and at designated times (t = 0, 7, 15, 30, 45, and 60 min), aliquots were removed and precipitated with ice-cold acetonitrile containing an internal standard (carbamazepine, 50 nM). The mixture was centrifuged at 3500 rcf for 5 min and resulting supernatants diluted with water in preparation for LC/MS/MS analysis of substrate depletion, by monitoring the analyte/internal standard peak area ratio.

Liquid Chromatography-Mass Spectrometry Analysis

The extent of substrate depletion in S9 fractions was determined employing LC/MS/MS analysis with an electrospray ionization enabled Sciex API-4000 triple quadrupole instrument (Sciex, Foster City, CA) that was coupled to LC-10AD pumps (Shimadzu, Columbia, MD) and a CTC PAL autosampler (Leap Technologies, Carrboro, NC). Analytes were separated by gradient elution using a Fortis C18 column (3 × 50 mm, 3 μm; Fortis Technologies Ltd., Cheshire, UK) warmed to 40°C. Mobile phase A was 0.1% formic acid in water (pH unadjusted); mobile phase B was 0.1% formic acid in acetonitrile, and the flow rate was 0.5 mL/min. Mass spectral analyses were performed using multiple reaction monitoring (MRM), with transitions and voltages specific for each analyte using a Turbo Ion Spray source (source temp 500°C) in positive ionization mode (5.0 kV spray voltage). MRM transitions were the following: zaleplon (m/z 306→236), O⁶-benzylguanine (m/z 242→91), zoniporide (m/z 321→262), BIBX1382 (m/z 388→98), SGX523 (m/z 360→160), and carbamazepine (m/z 237 → 194). Data were analyzed using Sciex Analyst 1.5.1 software.

Data Analysis

Intrinsic Clearance (CL_int)

In vitro hepatic intrinsic clearance (CL_int, mL/min/kg) for each species was extrapolated from hepatic S9 using the substrate depletion method and eq. 1 (Zientek et al., 2010):

$$C{L}_{int}=\frac{ln2}{t_{1/2}(min)}\times \frac{mL}{2.5mgprotei{n}_{S9}}\times \frac{120.7mgprotei{n}_{S9}}{gliverweight}\times \frac{(A)gliverweight}{kgbodyweight}$$ (1)

The scaling factor, A, for each species is listed in Supplemental Table 1. CL_int was not calculated if the mean ln[C] versus time slope was not significantly different from zero (determined using GraphPad Prism version 5.04 by an F test with a significance level of p < 0.05) or if the slope was exclusively dependent on the terminal time point in order to be considered different from zero.

Hepatic Clearance (CL_HEP)

Hepatic clearance (CL_HEP, mL/min/kg) was estimated using eq. 2, according to the wellstirred model, uncorrected for fraction unbound in plasma (Obach, 1999):

$$C{L}_{HEP}=\frac{Q_H\times C{L}_{int}}{Q_H+C{L}_{int}}$$ (2)

Where Q_H is species specific hepatic blood flow (Supplemental Table 1) and CL_int is the intrinsic clearance calculated from eq. 1. Protein binding was purposely not incorporated into CL_HEP estimations, as the intent of the CL_HEP estimations was not to scale in vitro CL_int to in vivo clearance, but rather to compare the relative extent of metabolism-mediated clearance across species (via estimation of the hepatic extraction ratio, eq. 3 below).

Hepatic Extraction Ratio (E)

The estimated hepatic extraction (E) was calculated using eq 3:

$$E=\frac{C{L}_{HEP}}{Q_H}$$ (3)

Estimation of fraction metabolized by AO (F_m,AO)

The F_m,AO of the five compounds was estimated for each species. Hydralazine has been reported as a specific AO inhibitor suitable for use in determination of F_m,AO (Strelevitz et al., 2012). Utilizing this method in hepatic S9 fractions, incubations were fortified with NADPH in the presence or absence of hydralazine, and eq. 4 (method A) was applied to estimate the F_m,AO:

$$F_{m,AO(A)}=\frac{C{L}_{int}-C{L}_{int(+Hyd)}}{C{L}_{int}}$$ (4)

where CL_int is the intrinsic clearance in S9 fortified with NADPH and CL_{int (+Hyd)} is the intrinsic clearance in S9 containing both NADPH and hydralazine. Alternatively, the F_m,AO can also be estimated with eq. 5 (method B), utilizing S9 in the absence of NADPH (AO catalytic activity is NADPH-independent):

$$F_{m,AO(B)}=\frac{C{L}_{int(noNADPH)}-C{L}_{int(noNADPH+Hyd)}}{C{L}_{int}}$$ (5)

where CL_{int (no NADPH)} is the intrinsic clearance in S9 without NADPH and CL_{int (no NADPH+Hyd)} is the intrinsic clearance in S9 containing hydralazine without NADPH. In some cases, low turnover prevented an estimation of CL_int in S9 incubations containing NADPH and hydralazine (CL_{int (+Hyd)} = 0) or in S9 incubations absent NADPH (CL_{int (no NADPH)} = 0), which results in an F_m,AO estimation equal to 1 by method A or 0 by method B, respectively. Due to this limitation, both methods A and B were used to approximate F_m,AO. In addition, agreement between the two methods provides more confidence in the estimate. In some cases low turnover prevented a calculation of F_m,AO by both methods A and B.

Multi-species Allometry (MA) and Single-Species Scaling (SSS)

For allometric scaling of in vitro hepatic intrinsic clearance (CL_int), the in vitro values obtained from incubations of hepatic S9 from preclinical species were allometrically scaled to predict human hepatic S9 CL_int. The simple allometric equation (eq. 6) was applied for predictions using 3 or 4 species:

$$C{L}_{int}(mL/min)=a\times W^b$$ (6)

where CL_int is the predicted intrinsic clearance_, W is body weight, and a and b are the allometric coefficient and exponent, respectively. The coefficient and exponents a and b were obtained from a plot of CL_int versus body weight (W) (Supplemental Figure 1). A standard body weight for each species was used in the analyses (Supplemental Table 1). Following the same principle, eq. 7 was applied in single-species scaling (SSS) analyses, using a fixed exponent of 0.75, which has been proposed for use in SSS of PK parameters based on the understanding that many physiological factors including basal metabolic rate and passive renal clearance may be scaled using this exponent (Hosea et al., 2009):

$$C{L}_{int(human)}(mL/min)=C{L}_{int(animal)}\times \frac{W_{human}{}^{0.75}}{W_{animal}}$$ (7)

Success Criteria

The success of each prediction method was assessed by calculation of the average absolute fold-error (AAFE) and the average fold-error (AFE), described by eq. 8 and 9, respectively (Obach et al., 1997, Tang et al., 2007),

$$AAFE={10}^{\frac{\sum |log(folderror)|}{N}}$$ (8)

$$AFE={10}^{\frac{\sum log\left(folderror\right)}{N}}$$ (9)

where N equals the total number of compounds, and fold-error (eq. 10) is the ratio of the predicted clearance (CL_pred) to the observed clearance (CL_obs),

$$folderror=\frac{C{L}_{pred}}{C{L}_{obs}}$$ (10)

where CL_obs represents the human CL_int (obtained using human hepatic S9, eq. 1) and CL_pred equals the predicted human hepatic S9 CL_int scaled from either MA or SSS of CL_int obtained using hepatic S9 of preclinical species. AFE is equal to the geometric mean of the fold-error and represents a measurement of the overall bias in both directions (above or below the reference value of 1), whereas the AAFE gives both over-predictions and under-predictions equal value. Therefore, the overall bias of the prediction method towards under- or over-prediction is represented by the AFE, and the AAFE is an unbiased representation of the fold-error. An AAFE of ≤ 3 was considered successful; this criteria is in line with a convention reported in a recent PhRMA consensus paper by Ring et.al., identifying a < 2-, < 3- and <10-fold criteria in the identification of successful scaling methods (>10-fold criteria considered unsuccessful) (Ring et al., 2011). The percentage of compounds within 2-fold-error (fold-error = 0.5–2.0) and 3-folderror (fold-error = 0.33–3.0) was also considered when assessing each method.

SSS Correlation with F_m or E

The fold-error for in vitro SSS CL_int predictions was plotted against the animal:human ratio of either F_m,AO or E, with correlation coefficients (r²) determined using GraphPad Prism version 5.04. F_m,AO estimates calculated by method B were used for the analysis (except F_m,AO of zaleplon in mouse and minipig because method B resulted in F_m,AO = 0 due to low turnover; estimates calculated from method A were instead used in these two instances). F_m,AO of zaleplon in rat, O⁶-benzylguanine in mouse, rat, and minipig, and BIBX1382 in mouse could not be determined by either method and were thus excluded from the analysis. F_m,AO estimations > 1, were assumed to be equal to 1. No data were excluded from the analysis involving E.

For comparison to another non-cytochrome P450 pathway, similar correlation analyses were performed using in vivo data derived from a report by Deguchi et al. of SSS to predict human CL_p of uridine diphosphate-glucuronosltransferase (UGT) substrates (Deguchi et al., 2011). For twelve UGT substrates, Deguchi et al. reported CL_p in each species (mouse, rat, monkey, and dog), human predictions of CL_p obtained from SSS, and F_m,UGT in each species (estimated from in vivo production of glucuronide metabolites excreted into the bile and urine). For our correlation analyses using UGT substrate data from Deguchi et al., fold-errors of the SSS predictions were calculated as described above in eq. 10 from the observed human CL_p values and SSS CL_p predictions reported by Deguchi et al. F_m,UGT was not reported by Deguchi et al for imipramine in mouse, nor was it reported for levofloxacin and telmisartan in human, resulting in exclusion of levofloxacin and telmisartan from the F_m,UGT analysis. Similar to the correlation analysis involving E, an analysis was conducted using CL_p as a percentage of Q_H obtained from Deguchi et al’s report. Eq. 3 was adapted to calculate CL_p as a percentage of Q_H using CL_p values reported by Deguchi et al. and species-specific hepatic blood flow, resulting in eq. 11:

$$C{L}_{p}asapercentageof{Q}_H=\frac{C{L}_p}{Q_H}$$ (11)

When CL_p is exclusively mediated by hepatic elimination, this value (CL_p as a percentage of Q_H) will be equal to E. If extra-hepatic elimination is present, this value will be greater than E. Therefore, levofloxacin and furosemide, which are predominantly excreted unchanged in the urine, were excluded from this analysis.

Results

Intrinsic clearance in hepatic S9 fractions and relationship to hepatic extraction ratio

Considering the test articles employed in the present investigation are metabolized by multiple drug metabolizing enzymes, namely P450 enzymes and aldehyde oxidase (AO), we limited the sub cellular fraction employed in the in vitro pharmacokinetic assays to S9 fractions (containing both cytosol/AO and microsomes/P450). Importantly, the ability to attenuate the contribution(s) of P450 enzymes by limiting co-factor fortification (i.e., NADPH) was of utmost importance in the determination of the total contribution of these individual enzymes in the clearance of the test articles in vitro, and towards the determination of a fraction-metabolized by AO (F_m,_AO; vide infra). The intrinsic clearance (CL_int) estimates measured in hepatic S9 fractions of human and nonclinical species are summarized in Table 1. These data represent estimates from incubations in the presence of NADPH, thus encompassing clearance mediated by NADPH-independent (e.g. AO) as well as any NADPH-dependent (e.g. P450) pathways. CL_int was converted to hepatic clearance (CL_HEP) according to equation 2 for the purpose of estimating the hepatic extraction ratio (E), according to equation 3 (data summarized in Table 2). CL_int estimates for zaleplon were low for all species, with E estimated to be ≤ 0.32 in each species. O⁶-benzylguanine CL_int estimates were moderate in human, monkey, and guinea pig and lower in rat, mouse and minipig, resulting in E estimates that ranged from approximately 0.1 (rat, mouse, minipig) to approximately 0.5 (human, cynomolgus, and guinea pig). Conversely, zoniporide was moderately cleared in human, monkey, guinea pig, and minipig S9 incubations (E = 0.41 – 0.59), while it was rapidly cleared in incubations with rat and mouse S9 (E = 0.87 and 0.79, respectively). Estimated CL_int of BIBX1382 was high in human, monkey, and minipig (E = 0.72 – 0.83), moderate in guinea pig (E = 0.54), and low in rat and mouse (E = 0.30 and 0.27, respectively). SGX523 exhibited low-moderate clearance in all species, with E ranging from 0.25 – 0.50. Consistent with the observations of others (Choughule et al., 2013a), no single species was fully representative of human when considering E for each compound, but rather substratedependence was observed. However, E estimated in cynomolgus and rhesus monkey was most similar to human overall, followed by guinea pig and minipig; rat and mouse generally provided a poorer representation of human E. These data highlight the need to conduct an interspecies evaluation for selection of appropriate species for human PK prediction

Table 1.

Multispecies intrinsic clearance (CL_int, mL/min/kg) of zaleplon, O⁶-benzylguanine, zoniporide, BIBX1382, and SGX523 in incubations with hepatic S9 (in the presence of NADPH).

	Zaleplon	O6-benzylguanine	Zoniporide	BIBX1382	SGX523
Human	5.8 ± 1.02	17.8 ± 1.94	17.9 ± 1.77	80.2 ± 9.61	7.1 ± 2.28
Mouse	24.4 ± 5.28	20.7 ± 7.10	332 ± 11.3	33.2 ± 7.10	39.4 ± 7.19
Rat	15.6 ± 3.05	8.3 ± 2.03	466 ± 27.0	30.7 ± 3.56	35.3 ± 6.39
Guinea Pig	14.2 ± 1.89	60.3 ± 4.29	79.9 ± 7.47	72.7 ± 4.05	42.4 ± 4.71
Cynomolgus Monkey	20.3 ± 2.45	42.5 ± 8.40	62.8 ± 8.34	181 ± 11.5	44.4 ± 5.60
Rhesus Monkey	15.0 ± 2.06	22.9 ± 3.98	30.2 ± 3.29	114 ± 7.46	29.6 ± 7.22
Minipig	5.55 ± 1.34	4.43 ± 1.44	23.4 ± 2.07	134 ± 5.69	25.4 ± 1.46

Data represent means of triplicate determinations from 2–3 experiments (± SD).

Table 2.

Multispecies hepatic clearance (CL_HEP, mL/min/kg) of zaleplon, O⁶-benzylguanine, zoniporide, BIBX1382, and SGX523 in incubations with hepatic S9 (in the presence of NADPH) and the estimated hepatic extraction ratio (E).

	Zaleplon		O6-benzylguanine		Zoniporide		BIBX1382		SGX523
	CLHEP	E	CLHEP	E	CLHEP	E	CLHEP	E	CLHEP	E
Human	4.55 ± 0.625	0.22	9.62 ± 0.593	0.46	9.64 ± 0.503	0.46	16.6 ± 0.417	0.79	5.22 ± 1.15	0.25
Mouse	19.0 ± 3.35	0.21	16.6 ± 4.51	0.18	70.8 ± 0.517	0.79	24.1 ± 3.85	0.27	27.2 ± 3.66	0.30
Rat	12.7 ± 2.10	0.18	7.40 ± 1.61	0.11	60.8 ± 0.485	0.87	21.3 ± 1.73	0.30	23.3 ± 2.85	0.33
Guinea Pig	11.5 ± 1.25	0.19	30.3 ± 1.11	0.50	34.5 ± 1.40	0.57	33.1 ± 0.838	0.54	24.9 ± 1.64	0.41
Cynomolgus Monkey	13.9 ± 1.16	0.32	21.4 ± 2.28	0.49	25.8 ± 1.43	0.59	35.4 ± 0.434	0.80	22.0 ± 1.39	0.50
Rhesus Monkey	11.2 ± 1.14	0.25	15.0 ± 1.75	0.34	17.9 ± 1.16	0.41	31.7 ± 0.593	0.72	17.5 ± 2.69	0.40
Minipig	4.60 ± 0.934	0.16	3.79 ± 1.04	0.14	12.7 ± 0.620	0.45	23.1 ± 0.170	0.83	13.3 ± 0.401	0.48

Data, calculated using CL_int values from Table 1, represent means of triplicate determinations from 2–3 experiments (± SD).

Estimation of F_m,AO in hepatic S9 fractions

Table 3 summarizes F_m,AO estimated for each compound in each species using methods A and B (eq. 4 and 5, respectively). F_m,AO obtained for O⁶-benzylguanine, zaleplon, zoniporide, and BIBX1382 in human is consistent with previous reports (Hutzler et al., 2012, Strelevitz et al., 2012). In addition, with a few exceptions, the two methods used to estimate human F_m,AO generally produced similar values, providing more confidence in these estimates. It is important to note that the lower the rate of depletion, the less accurate the estimation of CL_int, (and thus, F_m,AO) using the substrate depletion method due to difficulty in distinguishing legitimate metabolism-mediated depletion from biological or bioanalytical variability in detection (Di and Obach, 2015, Hutzler et al., 2015). For example, in isolated incidences, low turnover resulted in an F_m,AO estimation of 1 or 0, by method A or B, respectively (e.g. estimation of zaleplon F_m,AO in rat), in which case, F_m,AO was not reported. Ideally, under circumstances of low turnover, CL_int could be estimated via metabolite formation and determination of Michaelis-Menten parameters (CL_int = V_max /K_m); however, this method requires authentic standards of each major metabolite (e.g. AO and P450-mediated metabolites) contributing to the overall clearance.

Table 3.

Multispecies intrinsic clearance (CL_int) estimated from S9 fractions containing NADPH, containing NADPH and hydralazine, without NADPH, or containing hydralazine without NADPH, and the estimated fraction metabolized by AO (F_m,AO) calculated from the CL_int data using methods A and B.

Species	CLint (mL/min/kg)				Fm,AO
	+NADPH	+NADPH +Hyd	−NADPH	−NADPH +Hyd	Method A	Method B
Zaleplon
Human	5.57 ± 1.01	n/c	3.96 ± 1.35	n/c	1	0.71
Mouse	28.0 ± 2.81	22.0 ± 6.14	n/c	n/c	0.22	0
Rat	18.0 ± 0.698	n/c	n/c	n/c	n/a	n/a
Guinea Pig	13.5 ± 1.90	n/c	6.73 ± 3.63	n/c	1	0.50
Cynomolgus Monkey	18.3 ± 1.41	n/c	12.8 ± 1.34	n/c	1	0.70
Rhesus Monkey	16.7 ± 1.31	9.61 ± 0.964	10.8 ± 0.743	n/c	0.42	0.65
Minipig	6.70 ±0.621	5.54*	n/c	n/c	0.17	0
O6-benzylguanine
Human	17.8 ± 3.05	5.37 ± 1.16	14.8 ± 1.98	n/c	0.70	0.83
Mouse	15.5 ± 2.48	n/c	n/c	n/c	n/a	n/a
Rat	6.59 ± 0.664	n/c	n/c	n/c	n/a	n/a
Guinea Pig	58.7 ± 4.65	15.0 ± 2.14	48.7 ± 5.15	n/c	0.74	0.83
Cynomolgus Monkey	35.6 ± 5.10	n/c	33.5 ± 3.84	n/c	1	0.94
Rhesus Monkey	19.5 ± 2.05	n/c	21.3 ± 1.52	n/c	1	1#
Minipig	4.68 ± 2.20	n/c	n/c	n/c	n/a	n/a
Zoniporide
Human	17.0 ± 1.49	4.00 ± 0.442	17.1 ± 1.53	4.80 ±0.685	0.77	0.72
Mouse	331 ± 17.5	31.7 ± 1.12	297 ± 16.0	n/c	0.90	0.90
Rat	482 ± 13.9	125 ± 11.0	357 ± 25.1	9.70 ± 1.90	0.74	0.72
Guinea Pig	73.6 ± 2.61	11.6 ± 2.06	84.4 ± 2.75	12.9 ± 2.54	0.84	0.97
Cynomolgus Monkey	55.2 ± 1.73	5.74 ± 1.16	55.3 ± 1.63	n/c	0.90	1
Rhesus Monkey	27.8 ± 2.03	6.47 ± 3.97	31.2 ± 2.60	n/c	0.77	1#
Minipig	22.1 ± 1.87	3.67 ± 1.55	21.5 ± 1.01	n/c	0.83	0.97
BIBX1382
Human	71.6 ± 1.92	5.83 ± 1.22	69.6 ± 2.10	3.93 ± 1.75	0.92	0.92
Mouse	21.4 ± 10.9	n/c	n/c	n/c	n/a	n/a
Rat	27.6 ± 0.995	25.1 ± 6.29	19.0 ± 3.17	n/c	0.09	0.69
Guinea Pig	69.4 ± 1.66	24.5 ± 2.92	50.4 ± 1.13	n/c	0.65	0.73
Cynomolgus Monkey	172 ±1.26	n/c	209 ± 19.6	n/c	1	1#
Rhesus Monkey	108 ± 5.43	6.03 ± 3.34	92.7 ± 1.21	n/c	0.94	0.86
Minipig	133 ± 6.98	20.0 ± 2.92	127 ± 7.08	n/c	0.85	0.96
SGX523
Human	7.34 ± 3.34	4.70 ± 2.40	4.51 ± 0.310	n/c	0.36	0.61
Mouse	35.9 ± 10.3	24.2 ± 5.04	24.4 ± 9.31	16.5 ± 4.03	0.33	0.22
Rat	36.9 ± 4.23	21.4 ± 0.664	10.4 ± 3.20	n/c	0.42	0.28
Guinea Pig	38.2 ± 1.13	29.4 ± 3.05	15.8 ± 6.15	14.6 ± 6.05	0.23	0.03
Cynomolgus Monkey	39.4 ± 1.25	16.6 ± 3.42	26.9 ± 4.35	n/c	0.58	0.68
Rhesus Monkey	23.6 ± 4.13	15.6 ± 2.43	11.3 ± 0.942	7.97 ± 3.59	0.34	0.14
Minipig	26.0 ± 1.20	21.4 ± 1.98	3.20 ± 0.574	n/c	0.18	0.12

CL_int data represent means of triplicate determinations (±SD); F_m,AO calculated using mean CL_int values

n/c = CL_int not calculated; mean ln[C] versus time slope not significantly different from zero

n/a = insufficient CL_int data to calculate F_m,AO

^*mean of duplicate determinations

^#calculations resulting in an F_m,AO > 1 were assumed to be equal to 1

Zaleplon F_m,AO.

CL_int of zaleplon in human S9 incubations with hydralazine could not be measured; however, a value of 0.71 was determined using method B, consistent with previous reports (Strelevitz et al., 2012). Cynomolgus monkey, rhesus monkey, and guinea pig demonstrated similar F_m,AO to human, with estimates ranging between 0.42–0.70, while mouse and minipig estimates were ≤ 0.22. Turnover of zaleplon in rat S9 was only measurable in incubations with NADPH absent hydralazine, preventing an F_m,AO estimate from being obtained.

O⁶-benzylguanine F_m,AO.

A high F_m,AO (≥ 0.70) was estimated for O⁶-benzylguanine in all species except rat, mouse, and minipig, which could not be determined due to low turnover of the compound. As was the case for zaleplon in rat S9 incubations, a lack of measurable turnover of O⁶benzylguanine in rat, mouse, and minipig S9 incubations in the presence of both NADPH and hydralazine as well as incubations absent NADPH prevented estimation of F_m,AO.

Zoniporide F_m,AO.

A high F_m,AO (> 0.70) for zoniporide was estimated in all species. In human, rat, and guinea pig, NADPH-independent clearance of zoniporide was not completely inhibited by hydralazine. Literature reports indicate hydrolysis as a secondary metabolism pathway of zoniporide (Dalvie et al., 2010, Strelevitz et al., 2012), which may account for this observation.

BIBX1382 F_m,AO.

F_m,AO for BIBX1382 was estimated to be ≥ 0.70 by both methods in all species except rat and mouse. BIBX1382 reportedly undergoes some degree of P450 2D6-mediated metabolism (Dittrich et al., 2002), and it has been reported that hydralazine may exert mild human 2D6 inhibition (Strelevitz et al., 2012, Zientek and Youdim, 2015), in which case method A could potentially estimate an inflated F_m,AO; however, both methods A and B resulted in the same F_m,AO estimation in human, suggesting that either P450 2D6-mediated clearance was insignificant, or hydralazine did not inhibit this pathway. This result is in agreement with a previous report phenotyping BIBX1382 clearance in human hepatocytes, where substrate depletion was predominantly mediated by AO (Hutzler et al., 2012). Interestingly, CL_int in rat S9 incubations fortified with NADPH was only slightly inhibited by hydralazine (decreased from 29 to 25 mL/min/kg), suggestive of predominantly NADPH-dependent clearance; however, in incubations absent NADPH, a CL_int of 19 mL/min/kg still remained. This finding indicates the possibility that the cytosolic enzyme xanthine oxidase (XO) may mediate part of the NADPH-independent clearance in rat S9 since XO is not inhibited by hydralazine; however, no measurable substrate depletion was observed in incubations containing hydralazine without NADPH. Minor depletion of BIBX1382 CL_int was observed in human S9 incubations containing hydralazine without NADPH, but no other species exhibited measurable depletion under these conditions.

SGX523 F_m,AO.

Considerable variability between the two methods was observed in the F_m,AO of SGX523 in some species (e.g. guinea pig), but overall a low-moderate F_m,AO was estimated in all species (range of 0.03 – 0.68). In some species (mouse, rhesus, and guinea pig), not all NADPHindependent activity was inhibited by hydralazine, indicating potential involvement of XO; Diamond et al. reported an absence of XO metabolism in formation of the SGX523 oxidative metabolite M11 in human and cynomolgus S9, but rather that the metabolite was solely attributed to AO (Diamond et al., 2010). It is possible that the challenges associated with low substrate turnover, as described previously, contributed to our observations.

In general, the F_m,AO calculated by the two different methods were in better agreement for compounds/species exhibiting more rapid clearance, and thus, more confidence can be placed in the accuracy of these estimations. Monkey and guinea pig demonstrated F_m,AO values most similar to human. However, because low turnover prevented an F_m,AO calculation for some compounds in mouse, rat, and minipig, it is unclear how closely these species replicate human F_m,AO.

Prediction of human hepatic S9 clearance by multi-or single species allometry

Intrinsic clearance estimates from hepatic S9 incubations from mouse, rat, guinea pig, cynomolgus, rhesus, and minipig were employed in various combinations of 3 or 4 species for allometric scaling (MA) as well as individually for direct extrapolation from a single-species (SSS). Human CL_int values predicted from MA and SSS were compared to CL_int measured in incubations with human hepatic S9. The overall performance of each method is summarized in Supplemental Table 2 (MA) and Table 4 (SSS), and plots of the observed CL_int for each compound versus the human CL_int predicted by each method are displayed in Supplemental Figure 2 (MA) and Figure 2 (SSS) (see also tabulated data in Supplemental Tables 3 and 4, along with the allometric exponents obtained for each MA method).

MA of CL_int.

An AAFE of ≤ 2.0 was obtained from all of the 4-species combinations and from the rhesus/rat/mouse combination, with rhesus/rat/mouse yielding the lowest AAFE (1.7) and 80% of the five compounds predicted within 2-fold of the experimentally measured human S9 CL_int (Supplemental Table 2). Only two combinations, cynomolgus/rat/mouse and cynomolgus/guinea pig/mouse, resulted in AAFE of > 3.0 (3.3-fold and 3.5-fold respectively). SGX523 is the only compound of the five for which a fold-error of < 3 could not be obtained by at least one of the species combinations; however, the minipig/rhesus/guinea pig combination yielded a fold-error of 3.0 (Supplemental Figure 2, Supplemental Table 3). Interestingly, all but one of the species combinations (minipig/rat/mouse) had an AFE of ≥ 1.0, indicating that the predictions were biased towards over-prediction rather than under-prediction. This clearly was not the case in every instance, however, particularly for O⁶-benzylguanine and zoniporide (Supplemental Figure 2, Supplemental Table 3).

SSS of CL_int.

Single-species scaling (SSS) has previously been reported to be as accurate or better than MA in predicting human pharmacokinetics, regardless of clearance mechanism, including P450 and non-P450 metabolism (Hosea et al., 2009). Therefore, we chose to investigate this method in addition to MA for prediction of human S9 CL_int. SSS predictions yielded AAFEs of ≤ 2.0 when scaling from cynomolgus and rhesus monkey, guinea pig, and minipig S9 CL_int, with 80% of compounds predicted within 3-fold-error for cynomolgus, guinea pig, and minipig, and 100% for rhesus (Table 4; Figure 2). The compound falling outside of 3-fold-error differed for each species—SGX523 for cynomolgus, BIBX1382 for guinea pig, and O⁶-benzylguanine for minipig (Figure 2, Supplemental Table 4). AAFEs for rat and mouse were 4.1 and 3.8, respectively, with only 40% and 60% of compounds predicted within 3-fold-error, respectively. A tendency towards under-prediction via SSS was exhibited in all species except monkey, which had AFEs of 1.6 (cynomolgus) and 1.2 (rhesus).

Table 4.

Average absolute fold-error (AAFE), average fold-error (AFE), and percentage of compounds predicted within 2 or 3 fold-error of observed CL_int measured in human S9, as predicted by single-species scaling.

Single species scaling	AAFE	AFE	% within 2 fold	% within 3 fold
Cynomolgus Monkey	1.6	1.6	80%	80%
Rhesus Monkey	1.4	1.2	80%	100%
Rat	4.1	0.56	40%	40%
Mouse	3.8	0.38	40%	60%
Minipig	2.0	0.76	60%	80%
Guinea Pig	1.7	0.76	80%	80%

Figure 2.

Plots of observed human S9 CL_int vs that predicted from single-species scaling. Inner dotted line represents unity, solid line represents 2-fold-error, and outer dashed line represents 3fold-error. Zaleplon (•), O⁶-benzylguanine (■), zoniporide (○), BIBX1382 (□), and SGX523 (*).

Relationship of F_m or E to SSS prediction accuracy

Based on their studies examining allometric scaling of UGT substrates, Deguchi et al. reported that overall F_m,UGT values in monkey were more similar to human than other species evaluated, concluding that this likely contributed to a higher rate of prediction accuracy from monkey SSS relative to the other species investigated (mouse, rat, and dog) (Deguchi et al., 2011). Likewise, we observed similar F_m,AO values between monkey and human, as well as better overall success with predictions from monkey SSS. However, a similar F_m,AO between animals and human did not always translate to a more accurate CL_int prediction. Furthermore, in Deguchi’s report it can also be observed when comparing individual F_m,UGT of each compound for each species with predicted CL_p by SSS, that a species exhibiting a similar F_m,UGT to human did not always yield a more accurate prediction versus another species displaying a F_m,UGT substantially different from human (Deguchi et al., 2011). In some cases however we observed a similar E between human and a given species despite a divergence in F_m,AO, suggesting that other metabolism pathways are contributing to compensate for the lacking AO pathway, resulting in a reasonable human CL_int prediction. To evaluate the relationship between F_m,AO or E and the accuracy of prediction by SSS, the CL_int prediction fold-error by SSS was plotted against the animal:human ratio of either F_m,AO or E (Figure 3A and C) A correlation was not observed between the animal:human ratio of F_m,AO and the fold-error in the CL_int predicted from SSS (r² = 0.0051). However, a positive correlation was observed between SSS CL_int prediction fold-error and E (r² = 0.6488). To determine if similar trends existed among the 12 UGT substrates evaluated by Deguchi et al., data were obtained from this report (Deguchi et al., 2011) to plot the CL_p prediction fold-error by SSS against the animal:human ratio of F_m,UGT (Figure 3B) or against the animal:human ratio of CL_p as a percentage of Q_H (Figure 3D). Because CL_p as a percentage of Q_H will be equal to E when clearance is mediated by hepatic elimination, we excluded UGT substrates that are predominantly cleared renally. Once again, correlation with F_m,UGT was poor (r² = 0.00034), but was strong with CL_p as a percentage of Q_H (r² = 0.9573). Overall, these data suggest that the fold-error in clearance prediction by SSS is more closely associated with the overall hepatic extraction efficiency than the F_m between human and a given species.

Figure 3.

Correlation of SSS fold-error and animal/human ratios of F_m E, or CL_p as a percentage of liver blood flow for in vitro data with AO substrates (A and C) or in vivo data with UGT substrates reported by Deguchi et al. (B and D). (A) fold-error in SSS of CL_int vs animal/human ratio of F_m,AO (inset, axes magnified)_, (B) fold-error in SSS of CL_p vs animal/human ratio of F_m,UGT (inset, axes magnified), (C) fold-error in SSS of CL_int vs animal/human ratio of E (inset, axes magnified), and (D) fold-error in SSS of CL_p vs animal/human ratio of CL_p as a percentage of liver blood flow (inset, axes magnified)_. Zaleplon (•), O⁶-benzylguanine (■), zoniporide (○), BIBX1382 (□), and SGX523 (*). UGT substrates are collectively represented by (▲).

SSS of zoniporide and BIBX1382 CL_p

In vivo plasma clearance (CL_p) reported in the literature for zoniporide and BIBX1382 in mouse, rat, and cynomolgus monkey (Dalvie et al., 2013, Hutzler et al., 2014a) were subjected to SSS to predict human CL_p. These data are displayed in Table 5, alongside in vitro data for comparison. When comparing zoniporide in vitro data (E, F_m,AO, and SSS human CL_int prediction) of mouse, rat, cynomolgus monkey, and human, the species exhibiting the most similarity to human across all data is cynomolgus monkey. Accordingly, SSS of CL_p using cynomolgus monkey resulted in a more accurate human zoniporide CL_p prediction (15.2 mL/min/kg) versus mouse and rat (39.2 and 62.7, respectively) relative to the observed human CL_p (21 mL/min/kg). Comparison of in vitro data for BIBX1382 also demonstrated greater similarities between human and cynomolgus monkey relative to mouse and rat. Likewise, SSS of CL_p using cynomolgus monkey yielded a better human BIBX1382 CL_p prediction (56 mL/min/kg) versus mouse and rat (7.2 and 13.4, respectively) relative to the observed human CL_p (25–55 mL/min/kg).

Table 5.

Interspecies comparison of hepatic extraction ratio (E), fraction metabolized by AO (F_m,AO), and SSS predictions using in vitro CL_int (mL/min/kg) or in vivo CL_p (mL/min/kg) for zoniporide (top) and BIBX1382 (bottom).

Species	E*	Fm,AO#	S9 CLint	SSS Human Predicted S9 CLint	CLp	SSS Human Predicted CLp
Zoniporide
Human	0.46	0.72–0.77	17.9	---	21§	---
Mouse	0.79	0.90	332	43.1	298§	39.2
Rat	0.87	0.72–0.74	466	114	237§	62.7
Cynomolgus Monkey	0.59	≥ 0.90	62.8	30.7	31§	15.2
BIBX1382
Human	0.79	0.92	80.2	---	25–55¶	---
Mouse	0.27	n/a	33.2	4.3	55¶	7.2
Rat	0.30	≤ 0.69	30.7	7.5	55¶	13.4
Cynomolgus Monkey	0.80	1.0	181	88.7	118¶	56.0

^*Data from Table 2

^#Data from Table 3

^§Obtained from: Dalvie D, Zhang C, Chen W, et al. (2010). Drug Metab Dispos 38:641–654.

^¶Obtained from: Hutzler JM, Cerny MA, Yang YS et al. (2014). Drug Metab Dispos 42:1751–1760.

Discussion

The general assumption in the utility of traditional allometric scaling to predict human clearance is that it requires conserved drug elimination mechanisms across species. Accordingly, where AO-mediated clearance exists, confidence in this approach is lacking due to differences in AO expression and activity between human and preclinical species traditionally employed (e.g. mouse, rat, dog), resulting in limited studies examining allometry to predict human clearance of AO substrates. In the present investigation, human in vitro hepatic CL_int of five AO substrates was successfully predicted by MA and SSS with certain species, indicating the possibility that allometry may be useful for predicting human in vivo clearance of AO substrates when the appropriate species are utilized. Others have proposed that species expressing only the AOX1 isoenzyme in the liver (e.g. guinea pig, monkey) may serve as better species to estimate human AO-mediated metabolism versus species expressing both the AOX1 and AOX3 isoenzymes (e.g. rat, mouse) (Garattini and Terao, 2012). Consistent with this proposal, our in vitro data indicate that the hepatic CL_int of AO substrates may be scaled from preclinical species to human by SSS with monkey, guinea pig, and minipig with reasonable accuracy and precision, while this relationship does not appear to be consistent (highly substrate-dependent) when directly scaling from rat or mouse. However, even SSS with guinea pig, monkey, and minipig exhibited some degree of substrate-dependency, which was similarly reported by Choughule et al. with regard to guinea pig and rhesus monkey cytosolic CL_int of AO substrates DACA and phthalazine (Choughule et al., 2013b). Given these observations, we propose that allometric scaling of in vitro hepatic CL_int may be beneficial for guiding selection of suitable species to be utilized for prediction of human in vivo clearance via allometry. In addition, we propose that species selection may be further aided by interspecies comparison of F_m,AO and E estimated from in vitro CL_int. Table 5 illustrates an example of this approach, where comparison of in vitro data (E, F_m,AO, and human CL_int prediction by SSS) indicates that SSS with cynomolgus monkey would be more appropriate than SSS with mouse or rat to estimate human CL_p of zoniporide and BIBX1382. Accordingly, when CL_p obtained from the literature was scaled to human CL_p via SSS with cynomolgus monkey, rat, or mouse (Table 5), SSS with cynomolgus monkey did in fact yield the most accurate estimate of human CL_p for both zoniporide and BIBX1382. These examples together support the proposed in vitro allometry approach as a potentially useful method to identify suitable nonclinical species for estimating human in vivo clearance via traditional allometry. Figure 4 depicts a flow chart summarizing this approach, which may prevent unnecessary PK studies in species that are not likely to reflect human PK. Furthermore, though beyond the scope of the present investigation, this approach may also benefit (in combination with in vitro biotransformation experiments) selection of appropriate species for toxicity testing.

Figure 4.

Flow chart integrating current strategies to identify and predict human AO-mediated metabolism with a novel in vitro approach to guide selection of appropriate species to be employed in traditional (in vivo) allometric scaling for projection of human in vivo clearance.

A potential limitation to the utility of this in vitro allometry approach is the use of hepatic S9, which only represents clearance mediated by hepatic metabolism. Though the contribution of extrahepatic metabolism to total body clearance of AO-metabolized compounds is currently poorly understood, extrahepatic expression of AO has been demonstrated in human as well as nonclinical species (Kurosaki et al., 1999, Moriwaki et al., 2001, Nishimura and Naito, 2006, Terao et al., 2016). Furthermore, differences in the AOX1 mRNA levels observed in various tissues between humans and mice (Terao et al., 2016) indicate tissue-specific expression patterns may not parallel across species. For example, Hutzler et al. demonstrated metabolism of BIBX1382 in lung and kidney S9 fractions of human and cynomolgus monkey (Hutzler et al., 2014a), although species differences in the relative rates of elimination of BIBX1382 from these two tissues were noted. Indeed, human CL_p for four of the five substrates we evaluated are reported in the literature and were each under-represented by our CL_HEP estimates from human S9 incubations, including BIBX1382 (Table 6). This is not an uncommon observation, which may be attributed to extrahepatic metabolism, among other possibilities, such as SNPs or other sources of donor variability, ex vivo protein instability, or procedural differences in tissue procurement that may yield lot-to-lot variability in AO activity from commercial sources of S9 (Hartmann et al., 2012, Hutzler et al., 2012, Fu et al., 2013, Hutzler et al., 2014b). While our in vitro studies do not account for potential extrahepatic clearance which may occur in vivo, they were conducted under the assumption that hepatic AO-mediated metabolism is largely responsible for drug clearance, such that hepatic CL_int measurements will permit suitable species selection, even though the in vitro CL_int may underrepresent the total body clearance occurring in vivo. Accordingly, despite differences observed in elimination from extrahepatic S9 fractions of cynomolgus and human (Hutzler et al., 2014a), SSS of cynomolgus CL_p of BIBX1382 still predicted the rapid CL_p observed in clinical pharmacokinetic studies (Table 5), and our in vitro assessments with hepatic S9 successfully identified cynomolgus as an appropriate species. Future research to establish species-specific tissue expression patters, mechanisms regulating AO expression, and importantly, to develop standardized in vitro scaling factors that can be used to estimate total organ clearance from in vitro CL_int in extra-hepatic tissues will all be critical steps towards understanding the potential contribution of extra-hepatic metabolism.

Table 6.

Human i.v. plasma clearance of zoniporide, O⁶-benzylguanine, zaleplon, and BIBX1382 reported in the literature.

Compound	Human Plasma Clearance (mL/min/kg)
Zoniporide	21*
O6-Benzylguanine	14.5#
Zaleplon	16§
BIBX1382	22–55¶
SGX523	N/A

^*Dalvie D, Zhang C, Chen W, et al. (2010). Drug Metab Dispos 38:641–654.

^#Dolan ME, Roy SK, Fasanmade AA, et al. (1998). J Clin Oncol 16:1803–1810.

^§Rosen AS, Fournie P, Darwish M, et al. (1999). Biopharm Drug Dispos 20:171–175.

^¶Dittrich C, Greim G, Borner M, et al. (2002). Eur J Cancer 38:1072–1080.

In addition to our in vitro allometry assessments, we evaluated the relationship between interspecies F_m,AO or E and the fold-error of human CL_int prediction by SSS. Interestingly, comparison of prediction fold-errors by SSS with the animal:human ratio of either F_m,AO or E revealed little correlation with F_m,AO, but a positive correlation with E; similar trends were also observed for UGT substrates when plotting SSS prediction fold-error of CL_p versus the animal:human ratio of F_m,UGT or CL_p as a percentage of Q_H. We observed some examples where a discrepancy in F_m,AO between human and animal did not preclude a similar E, which suggests a non-AO metabolism pathway may compensate for the lacking AO pathway to result in an overall similar hepatic extraction efficiency. For example, the F_m,AO obtained for zaleplon was substantially higher in human (≥ 0.71) versus minipig (≤ 0.17), while the E between the two species was similar (human = 0.22, minipig = 0.16). Consequently, these data may indicate that compounds containing a mixed AO/P450 metabolism phenotype, or possessing P450 favorable sites in addition to the AO metabolism site, could help to enable allometric scaling approaches if alternate metabolism pathway(s) in certain species compensate for reduced AO-mediated clearance. Nonetheless, these observations suggest that there is not a strong relationship between F_m and the prediction fold-error by SSS or that methods to obtain F_m are not sufficiently accurate to observe this relationship. With regard to the present study, it should be noted that our F_m,AO calculations assume that 50 μM hydralazine is adequate to selectively and completely inhibit AO metabolism, whereas the potency and selectivity of hydralazine is not known for all species studied. In addition, our F_m,AO calculations are dependent on the ability to measure substrate depletion when turnover may be low, which presents an additional challenge in obtaining an accurate F_m,AO estimate. Importantly, the challenge of low substrate turnover is not unique to the present AO investigation, as the P450 literature is replete with similar data generated for the purposes of scaling a particular PK parameter (such as clearance), and subsequently resulting in a successful model or simulation. Current research efforts among the drug metabolism community are focused on the development of novel in vitro models towards the resolution of this issue (Di and Obach, 2015, Hutzler et al., 2015). Finally, four of the five compounds exhibited an F_m,AO in human of ≥ 0.70. A larger data set consisting of compounds exhibiting a broader range of F_m,AO would help to better understand the importance of this value to obtaining an accurate human hepatic clearance prediction by SSS.

For our in vitro allometry assessment, we included data obtained from minipig, which is gaining popularity in preclinical development (particularly safety testing) due to similarities in anatomy and physiology to humans, as well as advantages related to regulatory acceptability and animal welfare (Bode et al., 2010, Ellegaard et al., 2010, van der Laan et al., 2010). With regard to drug metabolism, minipigs hold an advantage over dogs (concerning nonrodent/large animals) when AO is involved, as dogs are essentially devoid of AO activity in the liver (Dalgaard, 2015, Terao et al., 2016). In addition, a recent report found minipig to be useful in allometric scaling of drugs mostly cleared by P450 or glucuronidation (Yoshimatsu et al., 2016). For the present investigation, we chose to replace dog, which is commonly used for allometric scaling, with minipig, a species of similar body weight. Indeed, the use of minipig for MA enabled human S9 CL_int to be predicted with an AAFE of < 3, and SSS with minipig predicted CL_int within 3-fold for four out of five substrates. However, as noted previously, the F_m,AO of zaleplon was higher in human versus minipig, indicating that a non-AO metabolism pathway enabled the accurate human CL_int prediction (unlike monkey and guinea pig which demonstrated similar F_m,AO to human). Along with substantial underprediction of O⁶-benzylguanine human CL_int by minipig SSS, this observation highlights the potential substrate-dependency associated with this species. Interestingly, based on the AAFE and individual evaluation of fold-error for each of the five compounds, MA methods utilizing four species (including minipig) may prove to be more reliable than 3-species combinations, even though rat and mouse (which yielded poorer predictions by SSS) were included in the 4-species combinations (Supplemental Tables 2 and 3; Supplemental Figure 2). Evaluation of these species combinations in vivo will be important to fully understand this approach.

Conclusions

In summary, our data support prior postulations that guinea pig and monkey represent better models of AO-mediated drug clearance in human versus commonly employed nonclinical models such as rat or mouse (Garattini and Terao, 2012, Hutzler et al., 2013, Hutzler et al., 2014a), while reiterating that no single species should be expected to reflect human clearance of all AO substrates (Choughule et al., 2013b, Hutzler et al., 2013). Moreover, the minipig represents a large animal species to consider when investigating AO metabolism, particularly when employed in multispecies allometry and drug safety assessments of NCEs (Bode et al., 2010, Ellegaard et al., 2010, van der Laan et al., 2010) Collectively, our data support the need for a multiple species assessment when gauging the intrinsic lability of new chemical entities (NCEs) towards AO metabolism and the projection of drug clearance in human, and we have offered a potentially useful in vitro approach to aid in selecting nonclinical species that may provide the best estimates of human clearance when evaluated in vivo (potentially reducing the number of unnecessary in vivo PK studies conducted in inappropriate species). While mechanisms behind variable AO activity and contributions of extra-hepatic metabolism remain important unanswered questions towards the implementation of standardized methods pertaining to AO-mediated drug disposition, application of the methodology presented herein would likely reduce the risk of encountering unexpected rapid AO-mediated clearance in clinical trials due to the use of inappropriate species for preclinical assessments.