Rationalization and prediction of in vivo metabolite exposures: The role of metabolite kinetics, clearance predictions and in vitro parameters

Abstract

Importance of the field

Due to growing concerns over toxic or active metabolites, significant efforts have been focused on qualitative identification of potential in vivo metabolites from in vitro data. However, limited tools are available to quantitatively predict their human exposures.

Areas covered in this review

Theory of clearance predictions and metabolite kinetics is reviewed together with supporting experimental data. In vitro and in vivo data of known circulating metabolites and their parent drugs was collected and the predictions of in vivo exposures of the metabolites were evaluated.

What the reader will gain

The theory and data reviewed will be useful in early identification of human metabolites that will circulate at significant levels in vivo and help in designing in vivo studies that focus on characterization of metabolites. It will also assist in rationalization of metabolite-to-parent ratios used as markers of specific enzyme activity.

Take home message

The relative importance of a metabolite in comparison to the parent compound as well as other metabolites in vivo can only be predicted using the metabolites in vitro formation and elimination clearances, and the in vivo disposition of a metabolite can only be rationalized when the elimination pathways of that metabolite are known.

1. Introduction

A metabolite can be formed from any enzymatic transformation of a parent drug after the parent is administered in vivo or is incubated as substrate in vitro. Often, these metabolites prove to have in vivo pharmacologic activity. Classic examples of metabolites that have pharmacologic activity are metabolites of tricyclic antidepressants and benzodiazepine anxiolytics, where many of the metabolites are also marketed drugs [1, 2]. Metabolites can also possess toxicological activity. Examples of in vivo toxic metabolites have been well established for many parent drugs, such as carbamazepine, valproic acid and nefazodone [3–5]. Additionally, it has been demonstrated that some in vivo inhibitors have inhibitory metabolites of similar potencies, such as fluoxetine, itraconazole and atomoxetine [6–8]. Due to the realization that metabolites can oftentimes have in vivo activity, it is important to understand the disposition of a metabolite after the administration of a parent drug.

A recent FDA guidance on metabolites in safety testing (MIST) has drawn more attention to identifying and predicting human metabolites [9]. This guidance states that a metabolite found to circulate at equivalent or greater concentrations in at least one pre-clinical animal species when compared to in human has been adequately evaluated for safety and no further non-clinical testing is warranted. If this cannot be demonstrated, any metabolite with exposure > 10% of the parent at steady-state in humans warrants separate non-clinical toxicological and pharmacokinetic studies. In contrast to the MIST guidance, the European guidance states that separate studies are only warranted when a metabolite exposure is > 10 % of the total drug-related material exposure [10]. These guidance pose two important dilemmas in new drug development: 1) how to identify and reliably predict potentially important circulating metabolites sufficiently early in new drug development to allow timely synthesis of reference material, development of validated assays and toxicological evaluation, and 2) how to determine the steady-state area under the plasma concentration versus time curve for the metabolite (AUC_m) relative to the parent (AUC_p), or total drug related material, for relevant metabolites without performing elaborate multiple dose studies with radiolabeled drug.

In vitro metabolism and pre-clinical animal data as well as single dose pharmacokinetic data are often used to predict the in vivo steady-state disposition of new drug candidates, as well as the in vivo metabolite profile of a candidate drug. However, attempts to predict important circulating metabolites in humans from pre-clinical data are qualitative and have met with variable success [11]. Direct translation of the metabolite profile from animal species to humans may be confounded by species differences in enzyme activity and expression, whereas in vitro HLM and hepatocyte studies qualitatively identify the primary metabolites that are likely to be formed in vivo but detection of secondary metabolites remains challenging. In this review, established in vivo metabolite kinetic theory will be discussed and a method for predicting in vivo metabolite disposition from in vitro data will be presented and evaluated for its usefulness in preclinical prediction of metabolite exposure as well as in rationalization of in vivo metabolite exposures.

2. Metabolite Kinetic Theory: In vivo Aspects

During the late 1960’s through to the early 1980’s, much interest was paid to the development of pharmacokinetic theory that describes the in vivo disposition of a metabolite formed after administration of a parent drug. The metabolite plasma concentration (C_m) versus time (t) curve for a metabolite formed after intravenous (IV) administration will exhibit biphasic kinetics and depend on the dose of parent (D), the fraction of parent that is converted to metabolite (f_m), the metabolite volume of distribution (V_d,m) and the formation (k_f) and elimination (k_m) rate constants for the metabolite [12]:

$$C_m=\frac{k_f*f_m*D}{V_{d,m}*(k_m-k_f)}*(e^{-k_f*t}-e^{-k_m*t})$$ (1)

This expression dictates that the slope of the linear terminal portion of a metabolite concentration versus time profile will either be equal to that of the parent, i.e. formation rate limited (FRL, k_f < k_m) kinetics, or less than that of the parent, i.e. elimination rate limited (ERL, k_f < k_m) kinetics [13]. After oral (PO) administration, the metabolite concentration versus time profile will further depend on the fraction of drug absorbed into the body (F_a) and will be either biphasic or triphasic, depending on the efficiency of metabolite formation during first pass metabolism [14].

The AUC_m after either IV or PO administration of the parent was demonstrated to be determined by D, F_a (only after PO administration), f_m and the clearance of the metabolite (Cl_m) [15, 16]. Additionally, the in vivo metabolite-to-parent AUC ratio (AUC_m/AUC_p) was determined to be dependent only on the in vivo formation and elimination clearances for the metabolite [17, 18]. Patel et al [18], demonstrated that after IV administration of a drug the ratio of AUC_m to the AUC_p is:

$$\frac{AU{C}_m}{AU{C}_p}=\frac{f_m*C{l}_p}{C{l}_m}=\frac{C{l}_f}{C{l}_m}$$ (2)

where Cl_p and Cl_m are the total in vivo clearances of the parent and the metabolite, respectively, and Cl_f is the formation clearance of the metabolite. The assumptions that are made in this model are that 1) the kinetics of the parent and metabolite are linear, 2) only one metabolite is formed from the parent (i.e. f_m equals the fraction of drug excreted unchanged subtracted from unity) and 3) all metabolite formed is available to the systemic circulation. It should be noted, that the above expression can be adapted to PO administration if metabolism occurs only in the liver and Cl_p is the apparent oral clearance of parent, i.e. the quotient of the true clearance of parent and the fraction of parent that escapes first pass in the liver (F_h):

$$\frac{AU{C}_m}{AU{C}_p}=\frac{f_m*C{l}_p}{F_h*C{l}_m}$$ (3)

The AUC_m/AUC_p is a primary measure utilized in the MIST guidance and hence, this theory was reexamined to address the first dilemma posed by the guidance. Although the in vivo expression for AUC_m/AUC_p has been available for 25 years, it has only been tested in the norclobazam/clobazam metabolite/parent (M/P) pair [17]. In order to further validate the above kinetic theory, literature data for three M/P pairs, morphine/codeine, morphine-6-glucuronide/morphine and theophylline/caffeine, were reviewed with Equations 2 and 3 to test the existing metabolite kinetic theory in an in vivo-to-in vivo extrapolation of the AUC_m/AUC_p. For these three M/P pairs, the clearance values for both the parents and metabolites were available following IV administration. The literature clearance values utilized for codeine, morphine, morphine-6-glucuronide, caffeine and theophylline were 63, 120, 7.5, 6.7 and 0.59 L/hr, respectively [19–22]. The F_h values utilized for the parents codeine, morphine and caffeine were 0.5, 0.2 and 1, respectively [23, 24]. The f_m values utilized for the formation of the metabolites morphine, morphine-6-glucuronide and theophylline were 0.1, 0.1 and 0.03, respectively [25–27]. Since codeine, morphine and caffeine are highly water soluble and readily absorbed from the gastrointestinal tract, F_a is expected to be unity for these drugs [24, 26, 27]; hence, the bioavailability values utilized for the predictions are considered to exclusively represent the fraction of drug eliminated during first pass hepatic metabolism (F_h). Table 1 summarizes the observed and predicted AUC_m/AUC_p values for these M/P pairs. Using the known in vivo clearance values, the predictions were within a 3-fold error in comparison to the average observed AUC_m/AUC_p for all 3 M/P pairs after both IV and PO administration.

Table 1.

The AUC_m/AUC_p observed in vivo and predicted using observed Cl_p, Cl_m, f_m, and bioavailability (F) values for three M/P pairs for which in vivo clearance data of the metabolite were available.

Parent	Metabolite	Route of Administration	Observed AUCm/ AUCp (Range)	Predicted AUCm/AUCp	Predicted/ Observed (Range)	References
Codeine	Morphine	IV	0.068a	0.053	0.71	1, 2, 3
		PO	0.071 (0.019 – 0.17)	0.11	1.5 (0.64 – 5.8)
Morphine	Morphine-6- glucuronide	IV	1.4a	1.6	1.1	4, 5, 6
		PO	2.8 (2.8 – 2.8)	7.9	2.8 (2.8 – 2.8)
Caffeine	Theophylline	PO	0.12 (0.11 – 0.14)	0.34	2.8 (2.4 – 3.1)	7, 8

Notes: Where multiple data sets exist, the ranges of AUC_m/AUC_p values are provided in parentheses. All observed AUC_m/AUC_p values were determined after single dose administration.

Superscript ‘a’ indicates that only one in vivo study was available. Reference numbers coincide with the below provided references and not with the bibliography references.

References: 1Osborne et al, 1990; 2Yue et al, 1991a; 3Yue et al, 1991b; 4Du et al, 1999; 5Eliot et al, 2002; 6Guay et al, 1988; 7Akinyinka et al, 2000 and 8Rodopoulos et al, 1996.

The second dilemma posed by the MIST guidance is how to determine the steady-state AUC_m/AUC_p for relevant, potentially yet unidentified, metabolites without performing elaborate multiple dose studies with radiolabeled drug. Based on established pharmacokinetic theory [28] single dose data could be used as a measure of steady-state AUC_m/AUC_p as long as clearance is constant (linear kinetics). A classic study demonstrated in nonhuman primates that the AUC_0-∞ ratios of carbamazepine epoxide/carbamazepine and 3-bromocinnamamide/cinromide P/M pairs after single dose are identical to the AUC_0-τ ratios at steady-state [29]. To further examine this relationship, the AUC_m/AUC_p for all M/P pairs with both single dose and steady-state in vivo data were collected. In total, 25 M/P pairs were examined (Figure 1 and Table 2). There was no statistically significant difference between single dose and steady-state AUC_m/AUC_p (Wilcoxon Signed Rank p-value > 0.47) and 84% of the M/P pairs included in the analysis had a single dose AUC_m/AUC_p within 2-fold of the steady-state AUC_m/AUC_p. 56% of the M/P pairs analyzed contained a metabolite that displayed ERL kinetics and hence the metabolite accumulation to steady-state is determined by its elimination half-life rather than the half-life of the parent (Table 2). Therefore parent drug will reach steady state before the metabolite. This presents a fundamental dilemma in study design since it is impossible to know whether a metabolite has reached steady-state, before relevant in vivo metabolites are identified and characterized. The ambiguity of when metabolite steady-state is reached is emphasized by the fact that over half of the metabolites examined between single dose and steady-state, displayed elimination rate limited kinetics. This analysis suggests that single dose AUC_m/AUC_p data can be used as a surrogate for steady-state AUC_m/AUC_p and may be more reliable. However, it is critical to confirm that after single dose administration, samples are collected for 4 to 6 parent or metabolite half-lives, whichever is longer, to capture AUC_0-∞.

Figure 1. Relationship between the single dose and steady-state in vivo AUC_m/AUC_p for 25 M/P pairs.

The solid line indicates no difference and the dashed line indicates a 2-fold difference between single dose and steady-state dosage regimens.

Table 2.

The 25 M/P pairs utilized to examine the difference between single dose and steady-state AUC_m/AUC_p. Included is the AUC_m/AUC_p after a single dose and at steady-state and whether or not the metabolite is elimination rate limited.

Parent	Metabolite	Elimination Rate Limited?	Single Dose AUCm/AUCp	Steady-State AUCm/AUCp	References
Allopurinol	Oxypurinol	Yes	42	33	1, 2
Alprazolam	4-hydroxyalprazolam	No	0.073	0.11	3, 4, 5
	1-hydroxyalprazolam	Yes	0.027	0.034
Atorvastatin	2-hydroxyatorvastatin	Yes	0.69	0.87	6, 7
	4-hydroxyatorvastatin	Yes	0.098	0.88
Clarithromycin	14-hydroxyclarithromycin	Yes	0.67	0.42	8, 9
Ezetimibe	Ezetimibe-O-glucuronide	No	8.8	7.9	10, 11
Hydralazine	Hydralazine pyruvic acid hydrazone	Yes	47	89	12
Imatinib	N-desmethylimatinib	Yes	0.21	0.16	13, 14
Lansoprazole	5-hydroxylansoprazole	No	0.14	0.085	15, 16
	Lansoprazole sulfone	No	0.16	0.61
Modafinil	Modafinil acid	No	0.43	0.37	17, 18
Nelfinavir	Nelfinavir hydroxy-t-butylamide	No	0.45	0.23	19
Nortriptyline	10-hydroxynortriptyline	No	1.3	3.2	20, 21
Omeprazole	5-hydroxyomeprazole	No	0.82	0.92	22, 23
	Omeprazole sulfone	Yes	0.82	0.54
Oxcarbazepine	10-hydroxycarbazepine	Yes	32	27	24, 25
Pentoxifylline	1-(3-carboxypropyl)-3,7- dimethylxanthine	Yes	6.0	8.3	26, 27
	1-(4-carboxybutyl)-3,7- dimethylxanthine	Yes	0.74	0.80
	1-(5-hydroxyhexyl)-3,7- dimethylxanthine	Yes	4.1	4.5
Risperidone	Paliperidone	No	3.5	4.5	28, 29
Sertraline	N-desmethylsertraline	Yes	1.8	1.5	30, 31
Sulfinpyrazone	Sulfinpyrazone sulfone	No	0.069	0.21	32, 33
Terbinafine	N-desmethylterbinafine	No	1.1	0.95	34, 35
Venlafaxine	Desvenlafaxine	Yes	2.7	4.3	36, 37

Notes: The AUC for both parent and metabolite after single dose was assured to represent the AUC_0-∞, and the AUC_0-τ for both parent and metabolite at steady-state were assured to be truly under steady-state conditions. In the studies of hydralazine and nelfinavir, the AUC_m/AUC_p was determined after pairing single dose and multiple doses in the same subjects. Reference numbers coincide with the below provided references and not with the bibliography references.

References: 1Ayalasomayajula et al, 2008; 2Guerra et al, 2001; 3Buch et al, 1993; 4Greene et al, 1995; 5Wennerholm et al, 2005; 6Lau et al, 2007; 7Tekturna New Drug Application, 2007; 8Cheng et al, 1998; 9Hassan-Alin et al, 2006; 10Bergman et al, 2006; 11Reyderman, 2005; 12Shepherd et al, 1980; 13Egorin et al, 2009; 14Van Erp et al, 2007; 15Kim et al, 2002; 16Ieiri et al, 2001; 17Wong et al, 1998; 18Wong et al, 1999; 19Damle et al, 2009; 20Laine et al, 2001; 21Yue et al, 1998; 22He et al, 2003; 23Shirai et al, 2001; 24Keranen et al, 1992; 25Theis et al, 2005; 26Mauro et al, 1992; 27Smith et al, 1986; 28Bialer et al, 2004; 29Bondolfi et al, 2002; 30Allard et al, 1999; 31Hamelin et al, 1996; 32Bradbrook et al, 1982; 33Rosenkranz et al, 1983; 34Kovarik et al, 1992; 35Robbins et al, 1996; 36Lindh et al, 2003 and 37Troy et al, 1995.

3. Metabolite Kinetic Theory: In vitro-to-In vivo Extrapolations

3.1 Qualitative Predictions of In vivo Metabolite Exposure

There has been a considerable amount of discussion on how to identify important circulating metabolites during pre-clinical phases of development of a new drug candidate [11, 30]. Advances in and increased access to analytical technologies have made metabolite identification a routine part of new drug development. Most techniques focus on metabolic incubations of either a radiolabeled (if available) or nonradiolabeled new drug candidate to generate potential metabolites. The products of these incubations are then subjected to ultra performance liquid chromatography, to separate closely chemically related species, and coupled to either accurate mass spectrometry or NMR spectroscopy for structural determination [31–33]. This procedure results in the identification and quantification of potential metabolites formed from the new drug candidate, but the relative abundance of each metabolite formed in vitro often does not agree with its relative abundance in vivo [11].

Intuitively, one would expect that a major metabolite in HLMs or hepatocytes would also be a major metabolite in plasma. However, the clearances of primary metabolites vary, even within closely chemically related species such as two primary metabolites of the same parent. The relative exposure to different metabolites formed from the same parent drug will depend on the rank order of the ratio of formation clearance to elimination clearance for each metabolite (Equation 2). Hence, a major metabolite observed in HLMs or hepatocytes will not be dominant in vivo unless it has sufficiently low elimination clearance in comparison to other metabolites formed and in comparison to the parent drug. Whether this is clinically important was tested using published literature data of all M/P pairs for which both in vivo AUC_m/AUC_p and in vitro metabolite intrinsic formation clearance (Cl_i,f) data for parent drug with at least two metabolites was available. In total, 31 M/P pairs from 14 parent drugs were examined. Table 3 summarizes the in vitro Cl_i,f and the observed in vivo AUC_m/AUC_p ratios for each M/P pair examined for a given parent drug. AUC_m/AUC_p ratios for each M/P pair, for a given parent, were considered to rank correctly with respect to in vitro Cl_i,f if the M/P pair with a > 15% higher Cl_i,f also had a > 15% AUC_m/AUC_p. Only 7 parent drugs (50%) had AUC_m/AUC_p ratios for their respective metabolites that rank ordered correctly. One of the parent drugs examined in the rank order analysis was clomipramine. Clomipramine has three important metabolites: 8-hydroxyclomipramine, N-desmethylclomipramine and 2-hydroxyclomipramine. Based solely on in vitro Cl_i,f, N-desmethylclomipramine would be predicted to be the major metabolite, followed by equal exposures to 8-hydroxy and 2-hydroxyclomipramine. However, 8-hydroxyclomipramine can be detected in vivo in plasma at 40% of the parent whereas 2-hydroxyclomipramine is undetectable in plasma. This can be explained by the greater intrinsic elimination clearance of 2-hydroxyclomipramine when compared to 8-hydroxyclomipramine, 6.5 versus 1.5 µL/min/mg microsomal protein [34]. Incorrect rank ordering of metabolites for a given parent drug for half of the cases examined demonstrates that the consideration of only in vitro metabolite formation clearance is not sufficient for predicting the relative importance of a given metabolite in vivo.

Table 3.

Rank order analysis of the in vitro Cl_i,f and the in vivo AUC_m/AUC_p for M/P pairs of a given parent drug.

Parent	Metabolite	In vitro Cli,f (µL/min/mg)	In vivo AUCm/AUCp	Rank Order Correctly?	References
Alprazolam	1-hydroxyalprazolam	0.51	0.11	No	1, 2, 3
	4-hydroxyalprazolam	4.1	0.034
Atorvastatin	2-hydroxyatorvastatin	20	0.87	Nob	4, 5
	4-hydroxyatorvastatin	9.1	0.88
Caffeine*	Theophylline	0.072	0.14	Yes	6, 7
	Paraxanthine	1.4	0.48
	Theobromine	0.019	0.080
Clomipramine	2-hydroxyclomipramine	0.98	NDa	Nob	8, 9
	8-hydroxyclomipramine	0.94	0.36
	N-desmethylclomipramine	16	2.2
Codeine*	Morphine	0.23	0.029	No	10, 11, 12
	Norcodeine	1.1	0.25
	Codeine-6-glucuronide	0.45	9.2
Imipramine*	Desipramine	8.3	0.54	Yes	13, 14
	2-hydroxyimipramine	2.8	0.37
Lansoprazole	5-hydroxylansoprazole	19	0.085	No	15, 16
	Lansoprazole sulfone	10	0.61
Morphine*	Morphine-3-glucuronide	8.4	23	Yes	17, 18
	Morphine-6-glucuronide	1.8	2.8
Nefazodone	Hydroxynefazodone	239	0.36	Yes	2, 19
	Meta-chlorophenylpiperazine	190	0.090
Omeprazole	5-hydroxyomeprazole	24	0.92	Yes	20, 21
	Omeprazole sulfone	20	0.54
Propafenone	5-hydroxypropafenone	530	0.50	Yes	22, 23
	N-desalkylpropafenone	3.5	0.064
Quinidine	3-hydroxyquinidine	15	0.24	No	24, 25
	Quinidine N-oxide	2.9	0.37
Tamoxifen	4-hydroxytamoxifen	2.3	0.034	Yes	26, 27
	N-desmethyltamoxifen	31	2.1
Theophylline*	3-methylxanthine	0.012	0.035	Nob	28, 29
	1,3-dimethyluric acid	0.16	0.032

Notes: The in vitro Cl_i,f and in vivo AUC_m/AUC_p values are provided. Additionally, whether or not the metabolites of a given parent rank ordered correctly with respect to these two values is provided. All in vivo AUC_m/AUC_p values were determined under steady-state conditions except for those parent drugs marked with an asterisk, which were determined after single dose administration.

Superscript ‘a’ indicates a value not determined because the observed in vivo metabolite concentrations were not detectable.

Superscript ‘b’ indicates a set of M/P pairs for a given parent drug that were considered to not rank order correctly because the in vitro Cl_i,f values exhibited a > 15% difference but the AUC_m/AUC_p values did not or the AUC_m/AUC_p values exhibited a > 15% difference but the in vitro Cl_i,f values did not. Reference numbers coincide with the below provided references and not with the bibliography references.

References: 1Buch et al, 1993; 2Greene et al, 1995; 3Hirota et al, 2001; 4Park et al, 2008; 5Tekturna New Drug Application, 2007; 6Labedzki et al, 2002; 7Akinyinka et al, 2000; 8Kramer-Nielsen et al, 1996; 9Linnoila et al, 1982; 10Soars et al, 2001; 11Yue et al, 1997; 12Yue et al, 1991b; 13Skelbo et al, 1992; 14Koyama et al, 1994; 15Ieira et al, 2001; 16Kim et al, 2003; 17Morrish et al, 2006; 18Eliot et al, 2002; 19VonMoltke et al, 1999; 20Shirai et al, 2001; 21Shu et al, 2000; 22Dilger et al, 1999; 23Hemeryck et al, 2000; 24Nielsen et al, 1999; 25Schellens et al, 1991; 26Desta et al, 2004; 27Lien et al, 1990; 28Tjia et al, 1996 and 29Rodopoulos et al, 1996.

3.2 Quantitative Predictions of In vivo Metabolite Exposure

The in vivo metabolite kinetic theory developed by Pang et al, Houston et al and others laid the foundation of in vivo metabolite pharmacokinetics but these theories have not yet been applied to in vitro-to-in vivo extrapolation, a useful tool in anticipating the in vivo pharmacokinetics of a parent drug during new drug development. The prediction of in vivo clearance of drugs based on in vitro metabolism data is well established, although predictions have varying degrees of accuracy [35, 36]. In an extensive analysis of scaling in vitro HLM clearance values to in vivo clearance using multiple hepatic clearance models and plasma protein binding considerations, 29 drugs with varying physicochemical properties were predicted with a 2.14 to 4.39 average fold error [37]. Another study of scaling in vitro Cl_i values from human hepatocyte data for 50 drugs, obtained a 2.5 average fold error to the observed in vivo clearance, with outliers having up to 15-fold error [38]. Although quantitative in vitro-to-in vivo clearance prediction for a parent drug is now commonplace, little attention has been paid to the prediction of the in vivo disposition of a metabolite from in vitro metabolism data.

The MIST guidance requires the evaluation of absolute steady-state AUC_m between pre-clinical animal species and human for major metabolites. If similar exposure is not obtained in animals, additional safety testing of the metabolite may be required. This requirement generates a need to predict, prior to clinical studies, what metabolites will be quantitatively important in humans. Predicting absolute AUC_m values in humans poses a significant challenge because the result will depend on the dose of the parent, the fraction of the dose absorbed after PO administration, and the overall clearance of the parent drug. In addition, the AUC_m will depend on the metabolite specific parameters, such as the fraction of the dose converted to the metabolite of interest and the metabolite clearance. Within the MIST guidance, the secondary qualification of the relative exposure to the metabolite in human, i.e. AUC_m/AUC_p > 0.1, appears more conducive to prediction. The AUC_m/AUC_p is independent of the parent dose and fraction absorbed after PO administration and the predicted AUC_m/AUC_p can be utilized as a proportionality constant for anticipating the absolute levels of the metabolite of interest when a desired AUC or steady-state concentration of parent is ascertained. This can be illustrated via a review of the existing data on desipramine as a metabolite of imipramine, which demonstrates that within a 4-fold range of in vivo doses of imipramine, the AUC_m/AUC_p remains constant (Table 4) in cytochrome P450 (CYP) 2D6 extensive metabolizers (EMs). The absolute steady-state concentration of desipramine could be predicted by use of the steady-state concentration of imipramine and the AUC_m/AUC_p ratio. Normalizing the predicted metabolite exposure to that of the parent also provides valuable insight into whether the metabolite will be quantitatively important in vivo, regardless of the parent dose.

Table 4.

The effect of dose and CYP2D6 phenotype on the imipramine to desipramine AUC_m/AUC_p ratio after PO administration of imipramine.

Imipramine Dose (mg)	CYP2D6 Phenotype	Imipramine AUC (nmol*h/mL)	Desipramine AUC (nmol*h/mL)	Observed AUCm/AUCp	References
25	EM	0.68	0.37	0.54	1, 2, 3
25	PM	1.2	3.4	2.9
50	EM	1.4	0.79	0.58
100	EM	3.1	1.6	0.50

Notes: All AUC values were determined after single dose administration. Reference numbers coincide with the below provided references and not with the bibliography references.

References: 1Albers et al, 2000; 2Koyama et al, 1994 and 3Kurtz et al, 1997.

The prediction of in vivo AUC_m/AUC_p from in vitro parameters relies on methods for clearance predictions of both parent and metabolite. This is because an important principle of metabolite kinetics is that the in vivo disposition of a metabolite is dependent not only on its formation clearance, but also its elimination. Based on this principle, predicting relative exposure to human metabolites can only be done if the formation and elimination clearances for the metabolite are predicted.

To adapt Equation 2 to in vitro-to-in vivo extrapolation of AUC_m/AUC_p after intravenous (IV) or oral (PO) administration, four assumptions were made: 1) the kinetics of both parent and metabolite are linear, 2) all metabolite formed is available to the systemic circulation, 3) parent and metabolite elimination is via metabolism only and 4) metabolism occurs only in the liver which can be represented by the well-stirred model [39]:

$$C{l}_h=\frac{Q*f_u*C{l}_i}{Q+f_u*C{l}_i}$$ (4)

where CL_h is the hepatic clearance, Q is the hepatic blood flow, f_u, is the plasma fraction unbound and Cl_i is the hepatic intrinsic metabolic clearance. This clearance model can be applied to either the metabolite or the parent drug. In addition to this well-stirred model of the liver, two other prominent hepatic clearance models, the parallel tube and dispersion model, could be used for predictions [40]. At present the potential advantages of the alternative hepatic clearance models for predicting metabolite formation and elimination clearances are unknown and require further study. The well-stirred model was chosen for this analysis because it is the most commonly utilized hepatic clearance model and easiest to adapt for predictions. Furthermore, a general metabolic scheme based on the scheme of Houston and Taylor [14] can be considered for both IV and PO administration (Figure 2). In this scheme, D_g, D_h and D_s are the amounts of drug in the gut lumen, liver during first pass and systemic circulation, respectively. F_a and F_h are the fraction of drug absorbed from the gut lumen into the enterocytes and fraction of drug that escapes first pass elimination in the liver, respectively, and M_h and M_s refer to the amount of metabolite formed from first pass in the liver and from systemic elimination, respectively. The in vivo fraction of parent converted to the metabolite of interest, when the parent is cleared only through hepatic metabolism was previously defined as the fraction of hepatic parent drug clearance that results in the metabolite of interest [41]. This definition was adapted to in vitro parameters and defined as f_m,h:

$$f_{m,h}=\frac{C{l}_{i,f}}{C{l}_{i,p}}$$ (5)

where Cl_i,f and Cl_i,f are the intrinsic formation clearance of the metabolite in vitro and intrinsic elimination clearance of the parent in vitro, respectively. When the parent is cleared entirely via hepatic metabolism, the in vivo f_m for the metabolite of interest is equal to the in vitro f_m,h. Utilizing this metabolic scheme, the AUC_m after PO administration can be defined as:

$$AU{C}_m=\frac{f_{m,h}*F_a*(1-F_h)*D}{C{l}_m}+\frac{f_{m,h}*F_a*F_h*D}{C{l}_m}$$ (6)

and the AUC_p after PO administration can be defined as:

$$AU{C}_p=\frac{F_a*F_h*D}{C{l}_p}$$ (7)

Figure 2. Schematic representation of the metabolic fate of a parent drug after PO or IV administration.

In this scheme, D_g, D_h and D_s are the amounts of drug in the gut lumen, liver during first pass and systemic circulation, respectively. F_a and F_h are the fractions of drug that are absorbed into the enterocytes and that escape first pass elimination in the liver, respectively. The f_m,h term is the fraction of hepatic metabolism that results in the metabolite of interest and M_h and M_s refer to the amount of metabolite of interest, formed from first pass in the liver and systemic elimination, respectively.

By definition [42], the hepatic bioavailability (F_h) is a function of the extraction ratio of the parent (ER_p):

$$F_h=1-E{R}_p=\frac{Q}{Q+f_{u,p}*C{l}_{i,p}}$$ (8)

Substituting for F_h and the well-stirred model for Cl_p and Cl_m (as defined by Equations 9 and 5, respectively) into the quotient of Equations 7 and 8 yields:

$$\frac{AU{C}_m}{AU{C}_p}=\frac{f_{u,p}*C{l}_{i,f}}{f_{u,m}*C{l}_{i,m}}*\frac{Q+f_{u,m}*C{l}_{i,m}}{Q}$$ (9)

After intravenous administration, by substituting f_m,h for f_m (Equation 6) and the well-stirred model for Cl_m and Cl_p (Equation 5), Equation 2 can be defined as:

$$\frac{AU{C}_m}{AU{C}_p}=\frac{f_{u,p}*C{l}_{i,f}}{f_{u,m}*C{l}_{i,m}}*\frac{Q+f_{u,m}*C{l}_{i,m}}{Q+f_{u,p}*C{l}_{i,p}}$$ (10)

Utilizing the common technique of evaluating the limits of pharmacokinetic models with respect to high ER (Q ≪ f_u*Cl_i) or low ER (Q ≫ f_u*Cl_i), the above two models presented result in three pharmacokinetic outcomes: 1) the relative exposure to the metabolite (AUC_m/AUC_p) will be different after IV and PO administration when the parent drug has a high ER, but the AUC_m/AUC_p is independent of route of administration when ER_p is low, 2) changes in the intrinsic clearance of a metabolite with a low ER (ER_m) will alter the exposure to the metabolite resulting in changes in the AUC_m/AUC_p and 3) the relative exposure to the metabolite depends on the ratio between its formation and elimination clearances, not on the absolute value of either of these two terms. To determine whether the relative exposure to a metabolite is dependent on route of administration when the parent drug has a high ER, the exposure to morphine after PO or IV administration of codeine and the exposure to morphine-6-glucuronide after PO or IV administration of morphine was revisited (Table 1). Based on literature in vivo clearance values, formation of morphine from codeine is classified as high ER_p, high ER_m, while formation of morphine-6-glucuronide from morphine is classified as high ER_p, low ER_m. The observed AUC_m/AUC_p values are shown in Table 1. Indeed, as the model suggests, the relative exposure to the metabolite was dependent on route of administration for morphine-6-glucuronide, where it was always greater after PO administration than IV administration. The relative exposure to morphine after IV administration of codeine ranged from 3.5-fold less to 2.4-fold greater than after PO administration. Given the variability in AUC_m/AUC_p for the morphine/codeine pair after PO administration, it was not possible to define a clear relationship between route of administration and AUC_m/AUC_p.

It has been theoretically demonstrated that changes in the elimination clearance of a low ER metabolite will alter the relative exposure to that metabolite [13]. This can be illustrated by considering the exposure to desipramine as a metabolite of imipramine. If the clearance pathway of the metabolite is subject to genetic polymorphisms, the AUC_m/AUC_p will depend on the individual’s genotype for that elimination pathway. The formation of desipramine from imipramine is mediated primarily by CYP2C19, whereas desipramine elimination is mediated by CYP2D6. The desipramine/imipramine AUC_m/AUC_p was 5.4-fold higher in CYP2D6 poor metabolizers (PMs) in comparison to EMs making desipramine the major circulating species in CYP2D6 PMs (Table 4). This increase in AUC_m/AUC_p was due to a 1.7- and 9.1-fold increase in the AUC of imipramine and desipramine, respectively.

To test the developed in vitro-to-in vivo extrapolation model (Equations 10 and 11 for PO and IV administration, respectively), the AUC_m/AUC_p of seven M/P pairs: morphine/codeine, theophylline/caffeine, desipramine/imipramine, nortriptyline/amitriptyline, N-desmethylclomipramine/clomipramine, 8-hydroxyclomipramine/clomipramine and 2-hydroxyclomipramine/clomipramine were predicted. The predictions were accomplished using literature in vitro metabolite Cl_i,f and Cl_i,m obtained in human liver microsomes (HLM), literature plasma fraction unbound values for both parent and metabolite and common methods of clearance scaling. The Cl_i values (in µL/min/mg microsomal protein) were first scaled to grams of liver, using the value of 54.7 mg microsomal protein per g liver, and then to kg body weight, using the value of 21.43 g liver per kg body weight [43, 44]. An average body weight of 70 kg was considered. Summary of the obtained predicted and observed AUC_m/AUC_p values are shown in Table 5. Four of the seven M/P pairs were accurately predicted (< 2-fold error in comparison to the average observed AUC_m/AUC_p) and two of the seven were predicted within < 5-fold error (Table 5). 2-hydroxyclomipramine had undetectable metabolite levels after clomipramine administration and hence the prediction of very low AUC_m/AUC_p for this pair is considered in agreement with the in vivo finding.

Table 5.

The in vivo observed and in vitro predicted AUC_m/AUC_p values for seven M/P pairs.

Parent	Metabolite	Predicted AUCm/AUCp	Observed AUCm/ AUCp (Range)	Predicted/ Observed (Range)	References
Codeine	Morphine	0.051a	0.068a,b	0.75	1, 2, 3
		0.042	0.071 (0.019 – 0.17)	0.60 (0.26 – 2.2)
Caffeine	Theophylline	0.50	0.12 (0.11 – 0.14)	4.0 (3.6 – 4.6)	4, 5
Imipramine	Desipramine	1.43	0.89 (0.50 – 1.6)	1.6 (0.89 – 2.9)	6, 7, 8, 9, 10, 11, 12
Clomipramine	N-desmethylclomipramine	7.49	1.7 (1.2 – 2.2)	4.3 (3.4 – 6.1)	13, 14
	8-hydroxyclomipramine	0.22	0.40 (0.36 – 0.44)	0.55 (0.49 – 0.61)
	2-hydroxyclomipramine	0.05	NDc	ND
Amitriptyline	Nortriptyline	0.45	0.47 ( 0.43 – 0.53)	0.91 (0.81 – 1.0)	15, 16, 17

Notes: All values are for after PO administration unless otherwise noted. All observed AUC_m/AUC_p ratios were determined after single dose administration, save for the two in vivo studies of clomipramine. ND indicates a value not determined.

Superscript ‘a’ indicates a predicted or observed AUC_m/AUC_p after IV administration.

Superscript ‘b’ indicates that only one in vivo study was available.

Superscript ‘c’ indicates a value not determined because the metabolite was not detected in vivo. Reference numbers coincide with the below provided references and not with the bibliography references.

References: 1Guay et al, 1988; 2Yue et al, 1991a; 3Yue et al, 1991b; 4Akinyinka et al, 2000; 5Rodopoulos et al, 1996; 6Abernethy et al, 1984; 7Albers et al, 1995; 8Albers et al, 2000; 9Bergstrom et al, 1992; 10Koyama et al, 1994; 11Kurtz et al, 1997; 12Wells et al, 1986; 13Linnoila et al, 1982; 14Shimoda et al, 1995; 15Jiang et al, 2002; 16Liedholm et al, 1998 and 17Venkatakrishnan et al, 2001.

It is unlikely that the major inaccuracies in the above AUC_m/AUC_p predictions are a result of in vitro to in vivo scaling, since the use of in vivo parameters did not yield more accurate predictions (Tables 1 and 5). The morphine/codeine AUC_m/AUC_p could be predicted accurately from both in vitro and in vivo parameters (< 2-fold error in comparison to the average observed AUC_m/AUC_p) after both IV and PO administration and the theophylline/caffeine M/P pair yielded similar prediction accuracies when using either in vitro values or in vivo (4-fold versus 3-fold error). Additionally, by utilizing a ratio as the primary predicted measure, any systematic error made in the scaling of in vitro-to-in vivo clearances is negated although random error in the prediction will propagate in the AUC ratio.

The model is designed to predict the nonparametric outcome of AUC_m/AUC_p. Oftentimes, the pharmacologic or toxicologic effect is a function of the maximum concentration (C_max) and not the total body exposure. Predictions of metabolite C_max would require additional parametric information about the input and disposition rates of both the parent and the metabolite only obtainable after in vivo administration of the compounds.

One obvious limitation to this model is that it can only address the disposition of primary metabolites formed after parent administration. Theoretically, the same necessity of predicting both formation and elimination are relevant for subsequent downstream metabolites making quantitative predictions of downstream metabolites very complicated. However, if primary metabolites are used in in vitro incubation experiments, the likelihood of qualitatively identifying downstream metabolites is greatly increased.

Renal clearance as well as biliary excretion and gut metabolism are often important elimination pathways for xenobiotics. These pathways were not considered for either the parent or metabolite in this review and it is likely that to be fully applicable in new drug development, prediction of the total clearance of the metabolite (sum of predicted hepatic and renal clearances) will be necessary. Unfortunately in vitro-to-in vivo extrapolation models for prediction of renal clearance, transport and gut metabolism are currently not as well established as hepatic clearance predictions. Nonetheless, the preliminary success of the predictions indicates that in vitro-to-in vivo extrapolation of AUC_m/AUC_p, after further development and validation, could prove to be a useful tool in addressing metabolite-related concerns in new drug development.

4. Conclusions

This review of the available data on metabolite disposition shows that as predicted by original metabolite kinetic theory, for any quantitative or semi-quantitative prediction of metabolite abundance or relative importance in vivo, the in vivo elimination clearance of the metabolite has to be predicted or rationalized in addition to basic metabolite profiling in vitro. This is shown by the fact that the in vivo abundance of metabolites of a given drug are no more likely to rank order correctly based on in vitro formation clearances than when left to random probability. An interesting outcome of the literature review is that in comparison to parent clearance predictions, the AUC_m/AUC_p predicts with similar accuracy suggesting that important human metabolites can be quantitatively predicted using in vitro data. This accuracy was achieved despite the fact that in vitro Cl_i,f and Cl_i,m were usually not determined in the same study. Finally, the fact that AUC_m/AUC_p ratios measured after single dose administration were not significantly different from multiple dose AUC_m/AUC_p ratios, but ERL kinetics of metabolites were common suggests that early single dose studies for metabolite identification may be justified.

5. Expert Opinion

Based on available pharmacokinetic theory and literature data, the AUC_m/AUC_p is most appropriate value to be used as the relevant outcome measure of metabolite exposure in in vitro-to-in vivo extrapolation. Since this value does not depend on dose or the F_a of the parent drug, it is a more robust and generally applicable measure of metabolite exposure than the dose-dependent measure of absolute AUC_m. Predicting absolute AUC_m would require knowledge of the clinical dose of the parent drug and prediction of its F_a. Additionally, the predicted AUC_m/AUC_p, when multiplied by the expected clinically effective average steady state concentration, can be used to determine prior to human studies how likely the need for additional safety evaluation is.

The results of this review clearly show that the formation clearance of a metabolite is not sufficient for understanding and predicting its in vivo relative exposure (AUC_m/AUC_p) or importance in comparison to other metabolites. Although the equally important role of metabolite elimination and metabolite formation in metabolite disposition in vivo is generally known, this concept has yet to be applied to in vitro-to-in vivo extrapolation in preclinical new drug development. Inclusion of metabolite formation and elimination clearances in predictions allows prediction of in vivo AUC_m/AUC_p or, if AUC_p is known, the absolute AUC_m from in vitro data. It is interesting that after development of the early metabolite kinetic theory it has not been thoroughly reexamined in light of modern experimental approaches. It is likely that better understanding of metabolite disposition can be obtained by further testing in vitro-to-in vivo extrapolation models applied to metabolites even by using a relatively simplistic model detailed in this review. It would also be beneficial if such models would be further developed and validated. This is important not only to address the MIST guidance, but to improve our current understanding of the in vivo pharmacokinetics of probe metabolic ratios that are used in drug-drug interaction studies as well as in pharmacogenetic studies.

The developed model and future models addressing the same primary outcome may be useful in lead compound selection and toxicology stages of new drug development, allowing early attention on potential quantitatively important metabolites. Since chemical synthesis of metabolites can be time consuming, expensive and difficult, it is likely that predictions of metabolite clearance need to be obtained from minimal amounts of primary metabolites generated in in vitro systems and isolated using chromatographic techniques to justify investment of resources to synthesis of reference materials. However, the obtained predictions can be used to guide prioritization of synthetic efforts of metabolite standards. It is noteworthy, that intrinsic metabolic elimination clearance of a metabolite can be predicted from a substrate depletion experiment conducted below K_m (Michaelis-Menten affinity constant) concentrations of the metabolite [45, 46]. Additionally, pre-clinical animal studies can be utilized not only to determine the absolute abundance of a metabolite after parent administration, but also to examine the overall pharmacokinetics of the metabolite after administration of the metabolite or the parent. This data can be leveraged for in vivo human metabolite kinetic prediction and may provide further confidence in the in vivo predictions of metabolite clearance and AUC_m/AUC_p from in vitro data.

The in vivo AUC_m/AUC_p of only seven M/P pairs were predicted from in vitro data in this review. The unfortunate limitation was the dearth of literature on in vitro metabolite Cl_i values, and this review suggests that there is a great need to generate more metabolite relevant in vitro kinetic data. Even M/P pairs that are commonly utilized as in vivo CYP probes possess metabolite elimination pathways that are not kinetically characterized. For example, the urinary or plasma ratio of dextrorphan and dextromethorphan is a common probe for phenotyping CYP2D6, yet the major elimination pathway of dextrorphan is via glucuronidation, a pathway that has never been kinetically characterized in vitro [47, 48]. Additionally, the plasma ratios of 5-hydroxyomeprazole and omeprazole or omeprazole sulfone and omeprazole are common probes for CYP2C19 and CYP3A4 drug interactions, respectively, yet neither the kinetics of 5-hydroxyomeprazole nor omeprazole sulfone metabolism are characterized in vitro [49, 50]. This raises some concerns of the validity of these ratios, as genetic factors or drug-drug interactions affecting the unknown elimination pathways of the metabolite could result in skewed data.

Increasingly, M/P ratios are being utilized as specific in vivo CYP markers [51]. When a metabolite is considered pharmacokinetically relevant, the enzymes responsible for the elimination of that metabolite should be identified for proper interpretation of drug-drug interaction and genetic polymorphism studies. When a metabolite is considered pharmacologically relevant, again, these secondary metabolic pathways should be identified in order to understand the therapeutic impact of said drug-drug interaction or genetic polymorphism.

ARTICLE HIGHLIGHTS

• There is an increased interest in qualitative and quantitative prediction of in vivo circulating metabolites and in rationalization of metabolite exposures.

• The relative importance of metabolites in vivo is measured as the ratio between the metabolite’s and parent drug’s area under the plasma concentration versus time curve (AUC_m/AUC_p). This ratio can be predicted using in vitro-to-in vivo scaling of formation and elimination clearances of the metabolite of interest.

• Single dose AUC_m/AUC_p data can be utilized as a surrogate for steady-state AUC_m/AUC_p when the compounds have linear kinetics. 56% of reviewed metabolites underwent elimination rate limited kinetics and hence time to reach steady-state for the metabolites is governed by the half-life of the metabolite, not the half-life of the parent drug.

• When multiple metabolites are formed from the same parent drug, the rank order of importance observed in in vitro systems correctly predicts the rank order importance in vivo for only half of the drugs. This discrepancy is most likely due to great variability in the clearances of the metabolites. Prediction of the identity of major metabolites in contrast to minor metabolites in circulation is likely to require determination of the clearance of the metabolites as well. There is no significant correlation between the formation clearance of a given metabolite in vitro and the relative abundance (AUC_m/AUC_p) of that metabolite in vivo.

• If a metabolite is used as a probe of specific enzyme activity, the elimination pathways of that metabolite should be characterized.