General Guidelines for Enzyme Reaction Phenotyping

Comprehensive metabolite profiling in complete human liver systems such as hepatocytes to capture major metabolic pathways should be obtained before conducting detailed f_m studies. Whether a NME is cleared via oxidation, reduction, hydrolysis, or direct conjugation guides the selection of metabolizing enzymes to be prioritized for phenotyping studies. After major metabolic pathways have been identified, it is recommended to determine metabolite formation kinetics over a range of NME concentrations to establish K_m,apparent (apparent/observed Michaelis-Menten constant) value for the metabolic pathways of interest, particularly if using chemical inhibition method. Once K_m,apparent has been established, f_m experiments are conducted maintaining NME/substrate concentration at approximately 1/10th value of K_m,apparent to capture accurate contribution of the physiologically relevant high-affinity enzymes toward the metabolic pathway of interest. Because most of the P450 substrates have K_m,apparent values in 1–100 μM range, 0.1–10 μM is an appropriate concentration range to investigate metabolite formation kinetics while maintaining linear first-order kinetics. Investigators are encouraged to consider nonspecific binding of a NME to microsomes (and hepatocytes) in incubation mixture. For NMEs with high nonspecific binding (as in cases of lipophilic basic drugs), although the V_max values are unaffected, the observed K_m,apparent can be substantially higher than K_m,unbound(K_m,apparent = f_{unbound, incubation} *K_m,unbound), which represents the true unbound/pharmacologic drug concentration at the enzyme active site and is key determinant of in vitro CL_int. Challenges in assay sensitivity of metabolite monitoring at such low K_m,unbound concentration is a practical hurdle, and most in vitro experiments are conducted at approximately 1/10th value of K_m,apparent or in cases where K_m,apparent has not been fully characterized at three concentrations, typically 0.1, 1, and 10 μM. Generally, under optimal/low protein (e.g., microsomal) concentration (ideally 0.05–0.2 mg/ml), linear first-order kinetics are still maintained in these K_m,apparent concentration ranges. Graphical analysis of metabolite formation kinetics (using Eadie-Hofstee or Lineweaver-Burke plots) may allow one to determine involvement of multiple enzymes and saturable high affinity pathway in some cases. Enzyme-selective inhibitor or RAF/ISEF scaling methods (Supplemental Materials) are then applied to quantitatively determine f_m for each of the major metabolic pathways tested. This overall process is best done quantitatively with radiolabeled NME or with authentic metabolite standards if they are available. During early stages of drug discovery, radiolabeled NME or metabolite standards are rarely available and reliance on identifying metabolic pathways using unlabeled NME is commonly used for decision making and early victim DDI risk assessment.

When unlabeled NME is used, “relative” contribution of metabolic pathways is estimated using either UV-absorption or LC/MS/MS methods. Underlying assumption is that either the UV-absorption (when using UV) or the ionization efficiency (in LC/MS/MS) of the metabolite(s) is similar to the parent NME. Both the assumptions warrant cautious evaluation (Dahal et al., 2011) depending on the nature of the biotransformation of the NME.

When radiolabeled NME is available, the contribution of various metabolic pathways can be quantitatively assessed when radiolabeled NME is available. Metabolites can be separated and quantified via HPLC-radiometric detection device, whereas LC/MS techniques can be used to confirm the identity of the metabolites. If several minor metabolites are formed, the sensitivity of an inline radiochemical detector may not be sufficient to provide a quantitative measurement of all the metabolites, and a scintillation counter is often used after fraction collection from the HPLC separation. For extremely low turnover NMEs, a combined approach of radiometric detector and LC/MS/MS can be used (Yi and Luffer-Atlas, 2010)(discussed further in Special Considerations: Low Turnover, Extrahepatic Metabolism, Inhibition of Parallel Metabolic Pathways).

When biotransformation pathways have not been fully characterized, monitoring total parent NME disappearance is also commonly done. Enzyme-selective inhibitors or RAF/ISEF scaling methods (Supplemental Materials) are then applied to determine semiquantitatively f_m for overall NME disappearance. The parent disappearance approach is limited by the challenge to monitor accurately depletion of NME, especially when percent depletion is low with assays usually being conducted under initial rate conditions (NME depletion is maintained at ≤15–20% to ensure metabolite formation is linear with protein concentration and incubation time).When using enzyme-selective inhibitors, at least 40% of parent NME disappearance has been reported to be desirable (Lu et al., 2008) to capture maximum sensitivity of inhibition of NME metabolism. In comparison, when using the RAF or ISEF approaches, ∼15–20% of parent NME disappearance has been demonstrated to give acceptable results (Uttamsingh et al., 2005). These are in contrast to the level of sensitivity achieved when monitoring metabolites that can be measured at ∼5% of parent turnover with reasonable accuracy. The parent disappearance approach provides an advantage that allows the researcher to easily monitor the total NME metabolism rather than individual metabolite-formation kinetics, which is significantly more labor intensive and usually not done until a candidate is identified for clinical development. However monitoring NME disappearance does not yield a complete picture of metabolic pathways and enzymes involved in each of the pathways that could have important implications in victim DDI risk predictions. This can be demonstrated with an example of a NME metabolized predominantly through a major metabolic pathway, which has significant contribution from two distinct isoforms CYP3A4 and CYP2C9, each with different kinetic behaviors (K_mCYP2C9 = 5 μM, V_maxCYP2C9 = 18 pmol min⁻¹ mg⁻¹; K_mCYP3A4 = 100 μM, V_maxCYP3A4 = 170 pmol min⁻¹ mg⁻¹). When overall NME disappearance is monitored, the apparent observed kinetic parameters obtained for the NME will be a result of hybrid of the kinetic parameters for both isoforms. At low NME concentration (usually therapeutically relevant concentrations), contribution of the high-affinity isoform CYP2C9 will be predominant toward NME metabolism. At high NME concentration due to higher V_max of CYP3A4-mediated pathway, contribution of CYP3A4 will be primary. Several drugs demonstrate such biphasic kinetics (bufuralol, dextromethorphan, diazepam, omeprazole, lansoprazole, voriconazole) (Griffin and Houston, 2004; Pelkonen et al., 2008) where in vivo metabolism is dominated by the high-affinity/low-capacity enzymes (e.g., 2D6 or 2C19) although in vitro, contribution of low-affinity/high-capacity enzyme (e.g., 3A4) may be predominant at high substrate concentrations. It is important to understand such mechanistic aspects in DDI risk assessment at clinically efficacious doses of a NME so as to not underestimate or miss contribution of the physiologically relevant high-affinity enzyme under supratherapeutic in vitro assay conditions. In the next few sections, summary of current tools and approaches for f_m determination of P450 and non-P450 enzymes will be provided. The approaches can be evaluated by monitoring CL_int of a NME via either major metabolite formation or overall NME depletion, depending on the sensitivity of assays available to investigators that imparts data interpretation with high level of confidence. Technical details of typical incubation conditions for P450- and non-P450-mediated in vitro metabolism assays are beyond the scope of this manuscript. Interested readers are referred to past reviews of this topic (Johnson and Waterman, 1996, Venkatakrishnan et al., 2001; Kramer and Tracy, 2008, Ogilvie et al., 2008; Sensenhauser, 2014; Korzekwa, 2014) and citations made throughout this document.

Early Enzyme Mapping: Identifying P450 and Non-P450 Enzymes Involved in NME Metabolism

P450 and non-P450 enzymes such as flavin-containing monooxygenase (FMO), monoamine oxidase (MAO), and aldehyde oxidase (AO)/xanthine oxidase (XO), all catalyze oxidative metabolism of xenobiotics with differential preference based on the chemical nature of their substrates. AO/XO preferentially oxidize electron-deficient carbons (Beedham, 1985; Pryde et al., 2010) compared with the P450s that prefer electron rich C-oxidations. Hence aldehyde, imine, or heteroaromatic functional groups are typical substrates of AO/XO. FMO is another NADPH-dependent oxidative enzyme like P450 enzymes but preferentially mediates N- and S- oxidations, typical peroxy acid reactions (e.g., Baeyer-Villiger oxidation) but not C-oxidation or heteroatom-dealkylations like P450 enzymes (Hines et al., 1994; Cashman, 1995; Ziegler, 2002; Harper and Brassil, 2008; Lang and Kalgutkar, 2003). MAO-mediated metabolism often occurs on compounds containing basic amine (Geha et al., 2001). Glucuronidation and sulfation commonly occurs on phenol and aliphatic alcohol groups forming O-glucuronides and O-sulfates, respectively. Glucuronidation may also occur on amines and carboxylic acids forming N-glucuronides and acyl glucuronides, respectively, whereas NAT has been observed to acetylate both aromatic and aliphatic amines. Carboxylesterase (CES) is usually involved in ester hydrolysis, although some amide and ketones with α-CF3 have been reported to be substrates of CES (Wadkins et al., 2007).

Knowledge of the NME structure combined with information about major metabolic pathways (oxidation, reduction, hydrolysis, or direct conjugation) guides the nature of enzyme mapping assays. High level outlines of these studies are described in Figs. 2–4. If major metabolic pathways are identified to be phase 1 in nature then involvement of NADPH-dependent enzymes such as P450 and FMO (Fig. 3, Scheme A) and the NADPH-independent enzymes such as MAO or AO/XO (Fig. 4, Scheme B) are evaluated. When the NME undergoes direct conjugation, involvement of enzymes such as UGT, SULT, NAT, or GST is evaluated depending on the major metabolites identified. For a NME that primarily undergoes hydrolysis, enzymes such as carboxylesterases (CES) and amidases are further investigated. These studies may be conducted in conjunction with preliminary assessments in recombinant enzymes when available to identify better the enzymes responsible for major metabolic pathways. In the case of AO/XO, because commercial availability of recombinant enzymes is very limited, assessments are best made in cytosol (details in the following section of “Estimating Fractional Contribution of Aldehyde Oxidase/Xanthine Oxidase”).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Overall enzyme mapping strategy to identify class of DME involved in metabolism of NME.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Scheme A: Distinguishing P450- and FMO-mediated oxidation toward overall NME metabolism.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Scheme B: Distinguishing AO-, XO-, and MAO-mediated oxidation toward overall NME metabolism.

Preliminary assessment of relative contribution of FMO and P450 enzymes toward N- or S- oxidation (Fig. 3) can be assessed by 1) treating HLM with a nonionic detergent (e.g., Triton X-100) or pan-P450 inhibitor (e.g., 1-aminobenzotriazole) (both of which selectively suppress the majority of P450 activities but not FMO activity); or 2) incubating HLM at 50°C for 2–3 minutes without NADPH. FMO is heat inactivated in the absence of NADPH but P450 activity is not significantly compromised. Although none of these methods is ideal (e.g., P450 and FMO activity may not be fully inhibited with 1-aminobenzotriazole and heat inactivation, respectively), results obtained from combined studies are acceptable for early estimates of the relative contribution of P450 versus FMO to then initiate detailed studies described in sections below for “Determining Fractional Metabolism by P450: P450 Reaction Phenotyping”.

Identification of NADPH-independent oxidative enzymes AO/XO and MAO is outlined in Fig. 4. Significant oxidative metabolism of a NME in human liver cytosol is indicative of metabolism by AO or XO. Inhibition of overall metabolism in cytosol by hydralazine or allopurinol would implicate AO or XO, respectively, in the metabolism of the NME. Inhibition of metabolism in the mitochondrial fraction with tranylcypromine or pargylline is indicative of MAO involvement, although if the NME does not possess a primary, secondary, or tertiary amine functional group, there is a very high probability that MAO will not be involved.

Teasing out the contribution of P450 versus non-P450 oxidative enzymes (AO/XO, FMO, and MAO) can be challenging in cases where the same oxidative metabolite may be formed via multiple enzymes (e.g., P450 and FMO, P450 and AO/XO, or P450 and MAO). Based on the reported incidence and magnitude of victim DDIs in the clinical setting (Yu et al., 2014), as well as the continued trend of P450 being the major enzyme involved in metabolic clearance, P450 stands out as the high-risk enzyme class compared with others. Careful assessment of contribution of individual P450 isoforms toward overall metabolism of a NME is therefore crucial to accurately assess the victim DDI liability of a NME in the clinic.

Strategies for f_m determination discussed for the non-P450 enzymes are based on literature reported cases or current practices within member companies. Because known cases of drugs where non-P450 enzymes are the primary metabolizing enzyme is low, experience in determination of f_m and extrapolation to in vivo outcome for non-P450 enzymes is not extensive. It is therefore recommended to thoroughly assess and optimize conditions for f_m of non-P450 enzymes when using the strategies described for the non-P450s as good starting guidelines.

Determining Fractional Metabolism by P450: P450 Reaction Phenotyping

Relative contributions of individual P450 enzymes to total human hepatic microsomal clearance can be assessed in vitro by some commonly used techniques (Rodrigues, 1999; Wienkers and Stevens, 2003; Zhang et al., 2007; Ogilvie et al., 2008; Korzekwa, 2014; Zientek and Youdim, 2015). 1) Recombinant P450 kinetics scaled to HLM: Commonly referred to as the RAF/ISEF method (Supplemental Materials), this involves determination of enzyme kinetic parameters for the metabolism of the NME in a panel of recombinant human (rh) P450 (with predetermined specific activity and normalized for protein content) and scaling the rhCYP intrinsic clearance (CL_int or V_max/K_m) to HLM CL_int via a RAF/ISEF approach. When the contributions of specific P450 enzymes are scaled to HLM CL_int, then the percent contribution of each P450 enzyme toward the total HLM CL_int can be estimated. It is not uncommon that the absolute value of the HLM CL_int and rhCYP-scaled HLM CL_int may differ. Several factors could contribute to the differences including potential involvement of enzymes other than P450 enzymes in HLM in the formation of the same metabolite(s), lack of 100% specificity of probe substrates used to calculate the scaling factors, and the potential for compound-specific (probe versus NME) differences in metabolism in HLM versus rhCYP enzymes. However, with respect to determining the relative contributions of individual P450 enzymes to the overall P450-mediated oxidative clearance in HLM, RAF/ISEF methods are deemed very useful. 2) Chemical inhibition method: This approach involves the use of P450-isoform selective chemical inhibitors. After major metabolic pathways are identified and their kinetics evaluated, the effect of P450-isoform-specific chemical inhibitors on a particular metabolic pathway (in HLM) is assessed. Percent inhibition of a metabolic pathway by a known isoform-selective chemical inhibitor reflects contribution of the isoform toward that pathway. Ideally chemical inhibitors should be potent, selective, and metabolically stable. Sometimes the balance between potency and selectivity can be challenging to attain, because in reaction phenotyping assays the concentration of inhibitors should be such that they inhibit the desired isoform >90%. This is usually achieved at approximately 10-fold of inhibitor K_i values. However, at such high inhibitor concentrations some P450 inhibitors may demonstrate nonselectivity where slight inhibition toward other unintended P450 isoforms may exist and hence overestimate percent contribution via intended P450 pathway. A mathematical correction has been proposed using P450 probe substrates to correct for the cross reactivity and partial inhibition toward the target P450 (Lu et al., 2007), but this has not been widely qualified. Nonspecific binding of chemical inhibitors (f_u,incubation) also needs to be carefully evaluated (Waters et al., 2014), especially in cases where phenotyping studies are conducted at high microsomal protein concentration >0.5 mg/ml, where certain chemical inhibitors (e.g., α-napthaflavone, ketoconazole, benzylnirvanol, quinidine) show substantial nonspecific binding (Nirogi et al., 2015). Because of high nonspecific binding, unbound concentration of an inhibitor at the enzyme active site may be suboptimal to attain the desired >90% inhibition of an intended enzyme. Furthermore, when using irreversible inhibitors, it is recommended that the experimental conditions including preincubation times be carefully optimized. Based upon a literature search of reported K_i or IC₅₀ values for the inhibition of individual P450 enzymes and compilation of the information, a recommended concentration range of inhibitors to use is shown in Table 1. This range is intended to maximize the selectivity of the inhibitor for the specific P450 isoforms listed. The extent of inhibition by the P450-selective inhibitor provides information regarding the contribution of the inhibited enzyme to the total HLM oxidative metabolism. It is also recommended to maintain the NME concentrations below their K_m,apparent values, so the IC₅₀ value that is produced would be closer to the actual K_i value for the specific enzyme inhibition. Hence it is a good practice to establish K_m,apparent values of a NME when using the chemical inhibition method.

Combined use of P450-selective inhibitors in HLM and rhCYP scaling methods (RAF/ISEF) provides investigators with high level of confidence in estimating the primary P450 isoforms involved in the hepatic oxidative clearance of a NME. In majority of cases of P450-mediated oxidative metabolism, f_m values obtained via either of these two methods are not expected to be substantially different, but on occasions differences may arise. Potential reasons include: 1) involvement of uncommon P450 isoforms (2J2, 4A11) or non-P450 enzymes; 2) nonideal experimental conditions where suboptimal inhibition is obtained from chemical inhibitors. Scenarios include high protein concentration resulting in high nonspecific binding of inhibitor resulting in <90% inhibition of the intended P450—this underestimates f_m of intended P450; use of excessive inhibitor concentration resulting in substantial cross-reactivity, which leads to overestimation of f_m of intended P450; high substrate/ NME concentration > >K_m, resulting in overwhelming the inhibitory effect of reversible inhibitors and underestimating f_m of intended P450; 3) the lack of absolute specificity of the probe substrate and/or inhibitor for P450; 4) potential difference in P450 binding site for the probe substrate and the NME, where the NME may interact with the P450 at a binding site, different from that of the probe substrate (Mathur et al., 2013). Large discrepancies in f_m values obtained from chemical inhibition and rhCYP scaling methods, often result in considerable differences in the magnitude of the predicted victim DDI liability for a NME and warrant a systematic look into factors discussed above

Correlation analysis is another method that has been assessed for the P450 isoforms (and infrequent cases of UGT; Kamdem et al., 2010); however, review of current industry practices among the working group member companies did not advocate this to be a method of choice for phenotyping of drug candidates, especially when there is discrepancy in the f_m values from RAF/ISEF and chemical inhibitor methods. A reason may be that a prerequisite for correlation analysis is the maintenance of a consistent collection of individual human liver microsomes, with a wide range of activity characterized for the metabolizing enzymes of interest. This is probably less practical/attractive for industry laboratories, which routinely use commercially available pooled human liver microsomes from 100–150 donors, for RAF/ISEF and chemical inhibitor studies. Correlation analysis is also substantially more labor intensive than RAF/ISEF and chemical inhibitor methods, and when multiple P450 isoforms are involved, data interpretation may not be straight forward and often yields only qualitative information on f_m.

Successful examples of predictions of exposure change of victim drug in the clinic based on experimentally determined in vitro f_m,CYP have been reported. These examples use the methods described in the P450 phenotyping section as the standard approach (Youdim et al., 2008; Lu et al., 2008) and in theory can also be applied to non-P450 enzymes when appropriate tools are available.

Identification of Flavin-Containing Monooxygenases FMO 1, 3, 5

The flavin-containing monooxygenase (FMO) family comprises a group of flavin adenine dinucleotide-containing microsomal enzymes that catalyze oxidation of compounds containing soft nucleophilic groups. Typical FMO-catalyzed reactions are monooxygenation of heteroatoms such as nitrogen and sulfur (Hines et al., 1994; Cashman, 1995; Ziegler, 2002; Lang and Kalgutkar, 2003). Because the key intermediate of the FMO catalytic cycle is a reactive peroxide intermediate C(4α)-hydroperoxyflavin, the FMO-mediated biotransformation is typically consistent with chemical reactions that involve peroxides or peroxyacids. Similar to P450 enzymes, NADPH and oxygen are required for FMO-mediated reactions. However, unlike P450 enzymes, FMO enzyme activities are generally greater at higher pH and can be inactivated by a brief heat treatment in the absence of the cofactor NADPH (Cashman, 2008). These methods are often applied in vitro to indicate and assess potential involvement of FMO in drug metabolism. The commercial availability of recombinant FMO1, 3, 5 isoforms has facilitated rapid identification of FMO-mediated pathways, because negative results from properly conducted recombinant enzyme experiments generally allows exclusion of FMO involvement in metabolism. Because of the lack of selective inhibitors and limited information on tissue abundance for various FMO isoforms, quantitative f_m determination via RAF/ISEF or chemical inhibitors for FMO has not been established. Nevertheless, a set of properly designed in vitro experiments may still provide a good estimate of the relative contribution of FMO to overall oxidative metabolism (Fig. 5).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Overall strategy to confirm role of FMO 1, 3, and 5 in NME metabolism.

FMO3 and FMO5 are both expressed at high levels in adult human liver (Overby et al., 1997; Zhang and Cashman, 2006; Cashman, 2012). Methimazole (Tables 2, 3, 4, 5) is a well-characterized substrate/inhibitor of FMO1 and FMO3, but displays very low affinity for FMO5 (Overby et al., 1995). Therefore methimazole can be used to assess the relative contributions of FMO3 and FMO5 in HLM to overall FMO-mediated metabolism (Fig. 5). However, this approach must be used in conjunction with the thermal inactivation approach because methimazole is also known to inhibit several P450 isoforms (Guo et al., 1997). FMO1 is highly expressed in human adult kidney. Therefore incubations in human kidney microsomes in the presence or absence of FMO1 inhibitor methimazole can be used to assess involvement of FMO1 (Fig. 5). Unlike human liver microsomal systems, heat inactivation of FMO has not been well studied or documented in human kidney microsomes.

In contrast to the P450 enzymes, FMO activity is not susceptible to induction or inhibition (Cashman, 2012) and known instances of drugs whose metabolism is primarily mediated by FMO is scarce (ranitidine, nicotine). Hence from the viewpoint of potential DDIs, FMO-mediated pathways are generally of less concern relative to P450-mediated pathways.

Estimating Fractional Contribution of Aldehyde Oxidase/Xanthine Oxidase

AO and XO are cytosolic molybdoflavoproteins that catalyze NADPH-independent oxidation of a wide range of substrates. Typical reactions include oxidation of aldehydes to the corresponding carboxylic acids and hydroxylation of aromatic N-heterocycles at the electron-deficient carbon atom adjacent to the heteroatom (Beedham, 1985; Pryde et al., 2010). Unlike P450 enzymes, the source of the oxygen atom that is added by AO or XO is from water. Therefore incubations with appropriate matrices can be carried out under ¹⁸O₂ or in the presence of H₂O¹⁸ to further investigate the enzymes involved, if necessary.

The involvement of AO or XO in oxidative metabolism of a NME can be determined directly by monitoring the metabolism of a NME in liver cytosol or S9 fraction in the absence of NADPH (Hutzler et al., 2013; Barr et al., 2014a). It is important that the source of liver tissue used in these studies was not derived from liver that had been perfused with solutions containing allopurinol, a common practice after full hepatectomy, because that tissue would be expected to have little or no XO activity due to residual allopurinol, an inhibitor of XO (Barr et al., 2014b).

If oxidative metabolism is observed in human liver cytosol, the enzyme(s) involved can be elucidated by the use of chemical inhibitors. Hydralazine and allopurinol (∼50 μM) have been identified as selective inhibitors of AO and XO (Tables 2–5), respectively, and, therefore, inhibition of oxidative metabolism by either of these inhibitors would implicate the corresponding enzyme (Johnson et al., 1985; Elion, 1988). Raloxifene or menadione can also be used to inhibit AO in incubations with liver cytosol, but these inhibitors are less suitable for studies in liver S9 fractions or hepatocytes due to either their instability or potential to inhibit P450 enzymes (Obach, 2004; Strelevitz et al., 2012).

Fractional metabolism by AO can be determined using hydralazine in human hepatocytes as has been established (Strelevitz et al., 2012). The difference in intrinsic clearance with and without hydralazine (Tables 2–5) divided by intrinsic clearance in the uninhibited state gives the fractional metabolism via AO. Conceptually this same approach could be applied to determine the fractional metabolism via XO by using allopurinol in place of hydralazine, but the authors are not aware of examples where this has been done.

Few marketed drugs are metabolized primarily by, or potently inhibit, either AO or XO, so examples of drug interactions are rare. One notable example is allopurinol, an XO inhibitor designed to treat gout, which is contraindicated for coadministration with azathioprine, an immunosuppressant that is cleared via XO (Gearry et al., 2010). To date there is no example of drug interaction that occurs via inhibition of AO but the potential exists. Famciclovir is an antiviral prodrug that is converted to its active form penciclovir by AO (Clarke et al., 1995). Coadministration of famciclovir with a potent inhibitor of AO such as raloxifene could potentially lead to a reduction in antiviral efficacy (Obach, 2004), but studies to determine the magnitude of this interaction have not been conducted. It was recently demonstrated that inhibition of AO may be substrate dependent and occur by mixed modes of reversible and irreversible inhibition, so care should be exercised in predicting drug interactions based on in vitro data (Barr and Jones, 2011, 2013a; Barr et al., 2014b).

Because of species differences in AO isoforms and activities, good preclinical models for human AO activity have yet to be identified (Choughule et al., 2013), so caution should be exercised in extrapolating AO-mediated overall metabolic clearance from animals to humans. In vitro-in vivo correlation (IVIVC) assessed for set of 11 drugs (Zientek et al., 2010) (Supplemental Materials) predominantly metabolized by this AO suggested that the IVIVC is typically poor and the AO-mediated in vivo clearance is usually underestimated. Reasons for this poor correlation have been attributed to the extrahepatic contribution of AO to total clearance and the potential lability of AO in liver preparations used for in vitro studies. Recently the protein content of AO has been quantified in human liver cytosol (Barr and Jones, 2013; Barr et al., 2013; Fu et al., 2013). Use of these methods and their extension to extrahepatic tissues may help to define better our ability to predict contribution of human metabolic clearance via AO in the future.

Identification of Monoamine Oxidases MAO-A and MAO-B

Monoamine oxidase (MAO) is a flavin-containing protein located on the outer mitochondrial membrane of a wide range of mammalian cells in various tissues (Saura et al., 1996). MAO catalyzes the oxidative deamination of biogenic amines as well as basic amine-containing xenobiotics. MAO exists in two forms encoded by separate genes with 70% sequence homology (Weyler et al., 1990). MAO-A is found in the liver, gastrointestinal tract, and placenta, whereas MAO-B is found primarily in blood platelets.

When a NME contains a primary, secondary, or tertiary amine, evaluation of the NME as a substrate of MAO-A or MAO-B is important for victim DDI risk assessment, especially if the NME is intended to be used concomitantly with MAO inhibitors in the clinic. The liver mitochondrial fraction where MAO activity is the highest may be used as an appropriate system to evaluate the contribution of MAO-A or MAO-B. The overall strategy to identify MAO-A or MAO-B is outlined in Fig. 6. When a basic amine-containing NME exhibits NADPH-independent metabolism in a mitochondrial fraction, it is warranted to further evaluate its metabolism in recombinant MAO-A and MAO-B (both are commercially available). If no metabolism is observed in either of the recombinant MAO isoforms, the probability of MAO involvement is low and no further evaluation is warranted. MAO-A preferentially oxidizes serotonin and is inhibited by low concentrations of clorgyline (Tables 2–5), whereas MAO-B preferentially oxidizes β-phenylethylamine and is inhibited by low concentrations of (−)deprenyl (Kalgutkar and Castagnoli, 1995; Geha et al., 2001; Youdim et al., 2006). Therefore the relative contribution of MAO-A and MAO-B toward the overall oxidative metabolism of a NME may theoretically be assessed in hepatocytes using these isoform-selective inhibitors. However, use of chemical inhibition to assess contribution of MAO-A and MAO-B is not routinely done among companies, and literature cases of this approach are rare (Erickson et al., 2007). A RAF approach similar to that for P450s has been reported (Pybus et al., 2012) to evaluate the relative contribution of MAOs toward overall NME metabolism. But because the presence of MAO in microsomes is due to contamination during preparation, levels of MAO vary significantly between microsomal lots and microsomes are not recommended for MAO f_m determination, thereby limiting the utility of this approach.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Confirming role of MAO-A or MAO-B toward NME metabolism.

MAO enzymes play a vital role in the inactivation of neurotransmitters (serotonin, noradrenaline, epinephrine, norepinephrine, dopamine), and a number of psychiatric and neurologic disorders are attributed to MAO dysfunction. MAO inhibitors are one of the major classes of drugs prescribed for the treatment of depression and anxiety (MAO-A inhibitors) and are used alone or in combination to treat Alzheimer’s and Parkinson’s disease (MAO-B inhibitors). Several potent MAO inhibitors are marketed drugs, primarily to treat depression [such as Azilect (Teva Pharmaceuticals Ltd, Essex UK), Nardil (Pfizer Inc, New York, NY), Parnate (Glaxo Smith Klein, Ontario, Canada), Eldepryl (Somerset Pharmaceuticals, Inc., Tampa, FL), Marplan (Validus Pharmaceuticals, Parsippany, NJ), and Zelapar (Valeant Pharmaceuticals, Redwood City, CA)]. It is therefore useful to assess whether MAO-A/B has a major role in the metabolic clearance of an amine-containing NME to better understand potential victim DDI risk when coadministered with known MAO inhibitors.

Determination of Fractional Metabolism by Uridine 5′-diphospho-Glucuronosyltransferase

Commonly used methods for UGT reaction phenotyping are similar to those of P450s (albeit less well established for several UGTs still) and include the use of human recombinant enzymes and selective chemical inhibitors in human liver microsomal incubations (Court, 2005; Miners et al., 2010a,b).

Recombinantly Expressed UGT Isoforms

Advances in the availability of human recombinant UGT isoforms, identification of selective probe substrates for several of the major UGT isoforms, and information of UGT protein quantification in various organs has allowed for f_m estimation of UGT via RAF/ISEF methods. Commercially available human recombinant UGT isoforms include the major hepatic UGTs, such as UGT1A1, 1A3, 1A4, 1A6, 1A9, 2B7, and 2B15, as well as UGTs that are expressed predominantly in the intestine such as UGT1A8 and 1A10. Selective substrates for the major hepatic UGT isoforms have been identified (Tables 2–5) that may be suitable for reaction phenotyping studies and include 17β-estradiol (3-glucuronide formation) for 1A1, trifluoperazine for 1A4, 5-hydroxytryptophol for 1A6, propofol for 1A9 and zidovudine for 2B7 (Lepine et al., 2004; Krishnaswamy et al., 2004; Court, 2005; Uchaipichat et al., 2006a,b; Itaaho et al., 2008; Manevski et al., 2011; Walsky et al., 2012). Availability of selective probe substrates for the different UGT isoforms has allowed the determination of RAFs that have been useful in determination of relative contribution of different UGT isoforms in drug’s metabolic clearance, as exemplified by recent publications on haloperidol (Kato et al., 2012) and laropiprant (Gibson et al., 2013). Information on protein quantification of several UGT isoforms has enabled determination of ISEF values to allow improved prediction of UGT reaction phenotyping. Protein abundance of various UGT isoforms including 1A1, 1A3, 1A4, 1A5, 1A6, 1A7, 1A8, 1A9, 1A10, 2B7, 2B10, 2B15, and 2B17 were quantified in human liver, intestine, and kidney primarily by LC/MS/MS (Fallon et al., 2013a,b; Milne et al., 2011; Harbourt et al., 2012; Sato et al., 2012; Schaefer et al., 2012; Sridar et al., 2013). Because of differences in the protein abundance values across different studies, caution should be taken when these values are used for the generation of ISEF values. It is advisable that either the median values or a range of values be used.

Chemical Inhibitors of UGT

Incubations in microsomes in the presence of selective UGT inhibitors may also be useful as a complementary method to assess the fractional metabolism by UGT. Tables 2–5 summarize examples of proposed selective substrates and inhibitors of five major hepatic UGT isoforms and recommended inhibitor concentrations relative to their inhibition potency (IC₅₀ values). Examples of selective inhibitors are atazanavir for UGT1A1/1A3 (Zhang et al., 2005); hecogenin for UGT1A4 (Uchaipichat et al., 2006a,b; Walsky et al., 2012); troglitzone for UGT1A6 (Ito et al., 2001), digoxin, transilast, and niflumic acid for UGT1A9 (Vietri et al., 2002; Lapham et al., 2012); and β-phenyllongifolol-2 for 2B7 (Bichlmaier et al., 2007). Careful optimization of experimental conditions (Court, 2014; Zhou and Miners, 2014) (also more in Supplemental Materials) should be done for the selection of suitable inhibitor concentrations, e.g., when considering the presence of BSA, in particular for UGT1A9 and 2B7 reactions. These methods have not been well established yet for several other UGT isoforms. As with any chemical inhibitor (described in Determining Fractional Metabolism by P450: P450 Reaction Phenotyping), chemical inhibitors of UGT are rarely absolutely specific at higher concentrations used for phenotyping studies. It is therefore warranted to carefully assess and account for cross-reactivity when using chemical inhibitors. Additional tools appropriate for in vitro reaction phenotyping for an extended number of UGT enzymes considered to be important in drug metabolism need to be developed.

The use of selective inhibitors in liver microsomal incubations together with commercially available recombinant human UGT isoforms, selective probe substrates, and protein abundance information in intestine, liver, and kidney for some of the major UGT isoforms has led to the emergence of a more semiquantitative approach toward determination of the contribution of these UGT isoforms to overall drug clearance. However, current tools only allow quantitative scaling for limited UGT isoforms (UGT1A1, 1A6, 1A9, and 2B7), because this area remains an area of active growth. For drugs metabolized by both P450 and UGT, incubations in human liver, kidney, and intestinal microsomes together with P450 and UGT cofactors in the presence of BSA to improve f_m,CYP and f_m,UGT prediction have been proposed and is also an emerging area (Cubitt et al., 2011) (Supplemental Materials). IVIVC analyses (Miners et al., 2006) (Supplemental Materials) of limited dataset of UGT-mediated drugs suggest mixed success in the prediction of total UGT-mediated whole body clearance of drugs. Under-predictions were observed in cases like zidovudine and morphine, whereas reasonable success was demonstrated for drugs like laropiprant (Gibson et al., 2013), where inclusion of BSA was suggested to improve the predictions. Low incidence of UGT-mediated DDI in the clinic has also made it challenging to extrapolate in vitro f_m,UGT values to those in the clinic. Analysis of limited cases of UGT-mediated DDIs as in lamotrigine-valproic acid (Rowland et al., 2006), zidovudine-fluconazole (Uchaipichat et al., 2006b), zidovudine-vaproic acid interactions also suggest improvement in predictions of DDI magnitude with addition of BSA. Addition of BSA to improve predictions of UGT-mediated clearance and for determining f_m,UGT has not been standardized across companies and continues to be an area of further research.

Relative Contribution of Glutathione-S-Transferases

Fractional metabolism by GST may be assessed by intrinsic clearance in hepatocytes or human liver S9 fractions fortified with reduced glutathione (GSH). GST-mediated metabolism may also be qualitatively assessed by determining intrinsic clearance in the presence of GSH versus intrinsic clearance in cytosol in the presence of other cofactors that may be needed for other known or suspected pathways of metabolism (e.g., NADPH for P450-mediated oxidation or uridine diphosphoglucuronic acid for UGT). Intrinsic nonenzymatic reactivity of NME with GSH in buffer needs to be determined and factored in while establishing GST-mediated metabolism. Inhibitors such as ethacrynic acid may be used to confirm the involvement of GST isoforms (Ahokas et al., 1985; Ploemen et al., 1990), although sufficiently selective inhibitors for the different GST enzymes are needed to conduct effectively reaction phenotyping in subcellular or cellular systems.

Human recombinant GST isoforms such as GSTA1, A2, M1, M2, P1, and T1 are available commercially and may be used to assess the qualitative contribution of the different GST isoforms in GSH conjugation. However, tissue abundance of various GST isoforms (including different GST genotypes) and GST isoform-selective substrates (Ginsberg et al., 2009) need to be further evaluated to develop scaling factors for reaction phenotyping using human recombinant GST isoforms.

Because of the expression of many GST isoforms in extrahepatic tissues (e.g., GSTP1 in erythrocytes) (Awasthi et al., 1994; Whalen and Boyer, 1998), in vitro-in vivo extrapolation of human clearance needs to take into consideration the contribution from extrahepatic organs (including blood). In vitro-in vivo extrapolation of GST-mediated metabolic clearance remains to be established, and hence data from in vitro experiments should be interpreted with caution when assessing overall metabolic clearance by GST. There are no reported metabolism-based DDIs via GST of which this working group is aware.

Estimating Relative Contribution of Sulfotransferases

Sulfotransferases (SULTs) are cytosolic enzymes that transfer a sulfonate group from 3′-phosphoadenosine-5′-phosphosulfate to a drug molecule, thereby making it more polar and readily excreted mostly in urine (Parkinson and Ogilvie, 2007). In addition to biotransformation of many endogenous substrates such as cholesterol, dehydroepoandro-strerone, or estradiol and other estrogens, several hepatic SULTs are responsible for biotransformation of xenobiotics. These SULTs include SULT1A1, SULT1A2, and SULT2A1. Phenols and aliphatic alcohols are the major sites of sulfation on drug molecules. Other sites such as aromatic amines and primary amines can also undergo sulfation. However, unlike glucuronidation, sulfation usually does not occur on carboxylic acids. SULT and UGT often share similar substrate properties, such as conjugation on phenol groups. However, SULT is an easily saturable, high-affinity low-capacity enzyme class, whereas UGT is a low-affinity high-capacity enzyme class (Parkinson and Ogilvie 2007), due to which UGTs are often the predominant enzymes if phase 2 conjugation is the major metabolic pathway. In vitro studies need to be designed carefully to elucidate the sulfation pathway, because sulfation is easily saturated at low concentrations (Li et al., 1999; James, 2014).

Relative contribution by SULT for a given compound can be done using recombinant SULT isoforms to determine the intrinsic clearance value of a NME. The abundance of SULT isoforms in human liver, kidney, intestine, and lung has been reported using Western blot quantification (Riches et al., 2009) (Table 5), although the current literature is limited. The intrinsic clearance values in theory can be normalized by the reported relative abundance of SULT in human tissues (e.g., ng SULT/g cytosol) to obtain the relative contributions of each SULT toward the total clearance (assumption is that relative abundance of various SULT isoforms determined by Western blot is reflective of relative activity). 2,6-Dichloro-4-nitrophenol has been reported to inhibit various SULTs with varying potency (Wang and James, 2006). By using a combination of rhSULT and inhibition by 2,6-dichloro-4-nitrophenol (Tables 2–5) one can study the relative contribution of SULT isoforms in liver S9, cytosol, or hepatocytes. Improved predictions of f_m,SULT using hepatic and intestinal scaling factors are also emerging (Gertz et al., 2011) and will be an area of continued research to see whether additional incorporation of f_m,SULT will also refine predictions of f_m,CYP and f_m,UGT for NMEs that are metabolized by all the three enzymes (e.g., troglitazone). This working group is not aware of any reported cases of metabolic DDIs via SULT.

Identifying Role of N-acetyltransferases NAT1 and NAT2

The N-acetyltransferases (NATs) are an important family of metabolizing enzymes that catalyze direct conjugation of aryl and alkyl amine compounds such as hydrazine with the acetyl group of acetyl-CoA (Sim et al., 2008b). These cytosolic enzymes have been shown to be important in arylamine-containing drug detoxification and carcinogen activation and are present in liver, esophagus, stomach, small intestine, colon, ureter, bladder, and lung (Windmill et al., 2000). In humans, there are two functional NAT genes encoding two isoenzymes, NAT1 and NAT2. A genetic polymorphism at the NAT2 gene locus encoding for polymorphic NAT2 results in individuals with rapid, intermediate, or slow acetylator phenotypes (Sim et al., 2008a). NAT1, which also metabolizes a variety of arylamines including p-amino benzoic acid but not isoniazid, is also polymorphic, although the clinical effect of NAT1 polymorphism is less well defined compared with NAT2 polymorphisms. NAT polymorphism leads to different rates of inactivation of drugs, such as isoniazid, hydralazine, and sulfonamides, which undergo NAT-mediated metabolism.

NAT1 and NAT2 have distinct substrate specificities (Wu et al., 2007): NAT2 acetylates hydralazine, isoniazid, and sulfamethazine, whereas human NAT1 acetylates sulfamethoxazole, p-aminosalicylate, and p-aminobenzoylglutamate (folate catabolite). Human NAT2 is mainly in the liver and gut, whereas human NAT1 is expressed in many tissues (Windmill et al., 2000). Caffeic acid (Tables 2–5), ferulic acid, and gallic acid have been shown to selectively inhibit NAT1, whereas scopuletin and curcumin (Tables 2–5) have been shown to selectively inhibit NAT2 (Kukongviriyapan et al., 2006).

Once involvement of NAT1 or NAT2 is confirmed using recombinant NAT1 or NAT2, the relative contribution of NAT1 or NAT2 may be assessed in human hepatocytes, using selective chemical inhibitors, as outlined in Fig. 7. A RAF-type approach, using NAT1 or NAT2 probe substrates in commercially available NAT1 and NAT2 (fortified with cofactors acetyl-CoA and an acetyl-CoA regenerating system composed of acetyl-dl-carnitine and carnitine acetyltransferase) system can theoretically be scaled to total metabolism in human hepatocytes. However, such scaling approaches are not well established and should be interpreted with caution.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 7.

Confirming role of NAT1- and NAT2-mediated metabolism of an NME.

Because only a few drugs have N-acetylation as the major metabolic clearance pathway, documented incidences of NAT-mediated DDI are rare. However, potential inhibitions of NAT activity were reported by gypenosides, cisplatin, and nitrosoarene metabolites of carcinogenic arylamines (Chiu et al., 2004; Ragunathan et al., 2008; Liu et al., 2008). When NAT is identified to be the primary enzyme responsible for the metabolic clearance of a NME, it is advisable to assess the contributions of the NAT and isoforms using currently available tools.

Determining Relative Role of Carboxylesterase CES1 and CES2

Carboxyesterase (CES) enzymes are serine esterases responsible for the hydrolysis of esters, amides, thioesters, and carbamates (Laizure et al., 2013). In humans, the two major carboxyesterases known are human carboxyesterase-1 and human carboxyesterase-2. CES1 is primarily expressed in the liver, where it plays an important role in the metabolism of many prescribed medications including clopidogrel (Kazui et al., 2010; Hagihara et al., 2009) and methylphenidate (Nemoda et al., 2009). CES2 is present predominately in the intestine, where it has been shown to hydrolyze anticancer prodrugs gemcitabine (Pratt et al., 2013), capecitabine (Ribelles et al., 2008), and irinotecan. (Humerickhouse et al., 2000).

During in vitro assessment if NADPH-independent metabolism is observed with a NME exhibiting an ester or amide linkage, metabolism using rhCES1 and rhCES2 (commercially available) should be investigated. If no metabolism is observed in the recombinant CES, the probability of CES involvement is low and no further evaluation is warranted. If NME metabolism is observed in the recombinant CES system, further assessment can be conducted using subcellular fractions, e.g., human liver (CES1 enriched, low CES2 levels) and intestinal (CES2 only) microsomal and/or S9 systems, in combination with specific substrates and inhibitors listed in Tables 2–5, to evaluate the relative contribution of CES to overall clearance (Zhu et al., 2009; Ross et al., 2012; Nishimuta et al., 2014; Shimizu et al., 2014). There are no known CES-mediated metabolic DDIs reported, making this enzyme class as a potentially low risk for causing metabolic DDIs.

Special Considerations: Low Turnover, Extrahepatic Metabolism, Inhibition of Parallel Metabolic Pathways

Extrahepatic Extraction

Several P450 and non-P450 enzymes discussed above are expressed in intestine and kidney in addition to liver, which is recognized to be the major metabolizing organ. So for refined assessments of victim DDI interactions, primary site of metabolism needs a case by case consideration once the major DMEs involved have been identified. Intestinal CYP3A4-mediated metabolism has been implicated to be major contributor of the overall first-pass metabolism of a large number of CYP3A4 substrates, such as cyclosporine, tacrolimus, midazolam, felodipine, simvastatin, atorvastatin, verapamil, and nifedipine, where inhibition of intestinal CYP3A4 may significantly contribute to the overall magnitude of a DDI. Analysis and comparison of observed and estimated ratio change of fraction extracted in gut (F_G) F_G,inhibited/F_G for CYP3A4 victim drugs have highlighted the need to accurately estimate and incorporate F_G,inhibited/F_G ratio for prediction of a DDI magnitude of a victim drug (Galetin et al., 2007, 2008), in particular for highly extracted victim drugs whose F_G ≤ 0.5. For such drugs that undergo high first-pass intestinal extraction (≥50%), magnitude of DDI will be very sensitive to inaccuracies in estimating the F_G,inhibited/F_G ratio. Because F_G is not a commonly measured parameter in clinic it is strongly recommended to understand the impact of uncertainty of F_G,inhibited/F_G ratio via sensitivity analysis modeling (more in modeling and simulation section). Metabolism-based DDIs in kidneys have not been reported.

Impact of Simultaneous Inhibition of Parallel Metabolic Pathways/Enzymes

It is generally desirable for a NME to have multiple metabolic pathways of clearance so when one pathway is inhibited the impact on its exposure change is not drastic. However, investigators should appreciate cases where parallel metabolic pathways of a victim NME are simultaneously inhibited (Ogilvie and Parkinson, 2014). This can potentially occur because of simultaneous loss of activity of multiple DMEs involved in the parallel metabolic pathways of the victim DME or DME and transporters, via any combination of factors such as genetic polymorphism, chemical inhibition, and disease state. The result is a dramatic change in exposure of a victim NME. Several victim DDI examples are well documented, including those of propranol (genetic polymorphism); omeprazole, lansoprazole (genetic polymorphism + P450 inhibitor); telithromycin (disease state + P450 inhibitor); and ramelteon (multi-P450 inhibitor) (Obach and Ryder, 2010). Investigators are encouraged to read some excellent publications for detailed understanding of the complex nature of such victim DDIs (Ito et al., 2005; Isoherranen et al., 2012; Ogilvie and Parkinson, 2014 ) and consider this on a case by case basis during NME development, as also recommended in regulatory guidances (http://www.ema.europa.eu/docs/en_GB/document_libraryScientificguideline/2012/07/WC500129606.pdf; http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm292362.pdf).

Current Status for Phenotyping of P450 and Non-P450 Enzymes

Substantial research over the past several years to mitigate P450-mediated victim DDI in the clinic has enhanced our ability to predict f_m,CYP of NMEs with high confidence, using commonly used methods of RAF/ISEF and use of P450-specific chemical inhibitors. Several of these examples have been summarized in literature publications (Youdim et al., 2008; Lu et al., 2008) and in recent regulatory submissions as in the case of ibrutinib (http://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm432240.html). Substantial higher incidence of P450-mediated compared with non-P450-mediated DDI cases in the clinic has allowed scientists to assess the IVIVC of the observed P450-mediated victim DDIs and therefore to gain better confidence in the currently available in vitro tools, such as scaling methods, recombinant enzymes, probe substrates, and selective chemical inhibitors of P450s. The knowledge gained from the P450s has been applied to other non-P450 enzymes, which has led to tremendous growth in the understanding of the non-P450 enzymes and identification of selective inhibitors and probe substrates for several of the non-P450 enzymes, as discussed in this section and summarized in Tables 2–5. This has remarkably increased the confidence among researchers to predict overall P450 versus non-P450 contribution toward the hepatic metabolism of a NME (Table 6). However, extrahepatic tissue abundance and activity information for majority of the non-P450s (including several UGT isoforms) is still largely absent. This poses a challenge in estimating f_m,non-CYP toward NME clearance if extrahepatic enzymes are substantially involved as primary metabolic pathways. This is demonstrated from lack of predictive power to accurately assess contribution of non-P450 enzymes AO and UGT toward overall drug clearance, observed in the analysis of known, limited set of drugs metabolized by AO (Zientek et al., 2010) and UGT (Miners et al., 2006). Scaling factors for human liver, kidney, and intestinal matrices are emerging and will continue to be an area of active research to refine predictions of f_m,CYP and f_m,non-CYP enzymes. Because of scarce incidence of reported non-P450-mediated clinical DDIs (such as those of XO-mediated allopurinol- azathioprine, UGT-mediated lamotrigine-valproic acid, zidovudine-fluconazole and, zidovudine-vaproic acid), extrapolation of in vitro f_m,non-CYP to a clinical setting continues to be a challenge and will be monitored closely by companies as more NMEs, with involvement of non-P450 enzymes emerge in coming years.

Knowledge of f_{CL,metabolism} as it grows with the progression of a potential drug candidate through various development stages is used in concert with the f_m data and allows better informed estimates of the overall victim DDI potential of a NME. The next section will discuss common approach how information gathered at various stages of drug development is used to estimate the contribution of metabolism to overall NME clearance.

Prediction from Preclinical Species and In vitro Systems—Learning from an In Vitro:In Vivo Correlation Approach (Pre-FIH)

Early qualitative assessment of clearance routes in human starts in preclinical stage via assessment of IVIVC in preclinical species. A typical IVIVC approach is outlined in Fig. 8. Assessments of the contribution of metabolism, renal, and biliary elimination (f_{CL, metabolism/renal/biliary}) toward total NME clearance are commonly made in at least two preclinical species. Preclinical PK studies, conducted with unlabeled or radiolabeled NME using intact or bile duct-cannulated (BDC) animals can provide information as to the route(s) of clearance based on quantification of unchanged drug in urine, feces, and bile (in case of BDC). Assessment of f_CL quantitatively can be best done with the use of a radiolabeled NME. However, useful information can also be obtained from studies using unlabeled NMEs under careful considerations such as the potential lack of mass balance and the assumption that all unrecovered NME was metabolized (i.e., major route of clearance was via metabolism), which may not be correct.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 8.

Estimating fractional metabolic clearance of an NME using information from preclinical studies and in vitro information using human liver matrices-IVIVC approach.

The route of clearance derived from in vivo preclinical animal studies may be used to provide some qualitative insight into a NME’s route of clearance in humans. Experience among different pharmaceutical companies suggests that if the NME is predominantly metabolized in animals and there is a good IVIVC in multiple preclinical species, metabolism is likely the major clearance pathway in humans. Similarly, if a NME is primarily cleared unchanged renally in animals, a substantial proportion of the NME may be expected to be cleared unchanged renally in humans. Thus insight into the qualitative involvement of metabolic or renal clearance in humans may be obtained based on results from multiple preclinical species.

Prediction of f_CL,renal in humans from preclinical species has been done with reasonable success when the NME is cleared via passive filtration (Paine et al., 2011), but knowledge is still emerging for NMEs cleared via active secretion and reabsorption. Because these latter processes usually involve transporters, interspecies differences can be expected and extrapolation from animals to humans may not be straightforward. Based on analysis of a large dataset of renally excreted compounds (Varma et al., 2009), physicochemical properties such as ionization state, lipophilicity, and polar descriptors have been proposed as important determinants of human renal clearance, taking into consideration both net reabsorption as well as net secretion. Such early assessments may be valuable in identifying compounds with potentially high f_{CL, renal} and aid to refine overall contribution of metabolism toward the clearance of the compound. The BDC rat is a routinely relied upon model for the prediction of biliary excretion in humans and to provide qualitative insight into hepatobiliary disposition in humans. However, because of significant species differences in the function, substrate specificity, and regulation of transporter proteins, it is difficult to directly extrapolate animal hepatobiliary data quantitatively to humans (Ishizuka et al., 1999; Wang and LeCluyse, 2003; Ghibellini et al., 2006; Swift et al., 2010). In vitro models have begun to emerge to characterize hepatobiliary elimination (Pfeifer et al., 2013) and predict the extent of biliary excretion of drugs in humans. Establishing IVIVC of biliary excretion in humans remains a challenge in part because of the lack of high-quality and quantitative biliary excretion data in humans. Hence, the predictive power of in vitro models for estimation of f_CL,biliary in humans remains debatable.

During preclinical PK studies, metabolite profiles from in vivo animal studies are usually compared with those from in vitro animal studies in subcellular fraction or hepatocytes. A good IVIVC of metabolic pathways in multiple preclinical species can offer some confidence in predicting major metabolic pathways in humans from in vitro human data. An early estimate of in vitro contribution of phase 1 (P450 versus non-P450 oxidation) and phase 2 (conjugation) metabolic pathways toward metabolic clearance of a NME also gives a preliminary estimate of any potential victim DDI risk associated with a pathway (such as those involving P450). Availability of current robust in vitro human models including human-derived subcellular fractions, recombinant enzymes, or hepatocytes, have allowed researchers to gain better confidence from successful prediction of in vivo human profiles before Phase I studies.

This IVIVC exercise allows early qualitative assessment of the role of metabolism and potential metabolic pathways in humans and raises any alert of potential victim DDI risks. Researchers should recognize that the metabolic pathways in vitro may not reflect primary metabolic pathways in vivo. Also, reliance on animal data to predict human metabolic pathways is not recommended, because the extent of metabolism and metabolic pathways in humans can be very distinct from that in animals. This is especially true when non-P450 metabolizing enzymes and transporters are involved in the disposition of a NME.

Learning from First-in-Human Studies

In early clinical development, information of metabolic clearance pathways is gained from metabolite profiling studies using plasma and urine samples routinely collected from the FIH single and/or multiple dose (SAD and/or MAD) studies. Innovative advances in high-efficiency sample preparation and response normalization-based bioanalytical techniques (¹⁴C-calibrant approach, use of quantitative NMR; Yu et al., 2007; Vishwanathan et al., 2009; Ramanathan et al., 2010; Dear et al., 2011) have made it possible for investigators to gain reliable quantitative estimates of metabolic pathways in plasma and urine in the absence of authentic metabolite reference standards. Quantitation of unchanged drug in urine enables preliminary estimation of f_CL,renal.

Assessment of metabolic clearance pathways in feces in cold FIH studies is not a common practice because of the lack of established quantitative methods to assess drug/metabolite recovery in cold fecal samples. It is therefore recommended to exercise caution for victim DDI risk assessments with estimates of metabolic clearance pathways in plasma and urine only, because the metabolism information based on these two matrices will be incomplete. A common and often erroneous assumption at this stage is that metabolites detected in plasma and urine truly represents all the major metabolic pathways of a NME. This assumption may hold in cases where a NME and its metabolites are predominantly cleared renally, where f_{CL,metabolism} and f_CL,renal may be obtained with reasonable confidence. However, elimination of unchanged NME and/or metabolites via biliary excretion into feces may also contribute substantially toward overall clearance, in which case clearance pathway information derived from fecal samples will be an important determinant of f_CL and f_m. Recent advances in clinical devices as Entero-Test (Guiney et al., 2011; Bloomer et al., 2013) have enabled direct assessment of drugs and metabolites excreted into the bile of humans. This method can be applied in early clinical development to provide qualitative information on the risk of interactions for drugs that are metabolized and eliminated in bile. Increased conduct of ¹⁴C-microtracer studies (trace ¹⁴C-NME administered as part of SAD and MAD studies), coupled with accelerator mass spectrometry (AMS) detection (Lappin and Stevens, 2008) have also enabled investigators to gain quantitative information of metabolic clearance pathways in circulation and in feces early in clinical development.

Information gathered from early metabolite profiling in FIH studies is valuable to get a first look into the major metabolic clearance pathways in plasma and urine and in certain cases bile (quantitative information when ¹⁴C-microtracer studies have been incorporated into FIH study designs or qualitative when techniques such as Entero-Test have been used). In addition, metabolite profiling in FIH studies can prove valuable in identifying primary metabolic pathway in vivo that may not have been previously identified in in vitro human models pre-FIH, although such cases are of low incidence. With certain caveats as discussed above, the overall learning from FIH study is helpful in refining victim DDI risk assessments made at pre-FIH stages.

Route of Clearance Determination in Human ¹⁴C-ADME Study

Human ¹⁴C-ADME study is deemed the gold standard in vivo study to quantitatively determine routes of clearance and metabolic pathways of a NME (ensuring good mass balance is observed), where unchanged drug and drug-derived metabolites are measured in plasma, urine, and feces. Data from the radiolabeled metabolite profiling in human are informative in the planning and design of definitive in vitro reaction phenotyping and clinical DDI studies required in support of a NME’s registration. When formation of secondary or tertiary metabolites is observed, it is assumed that the enzyme involved in the formation of the primary metabolite is of the most interest for DDIs (assuming a worst-case scenario).

A caveat to consider in these studies is the difficulty sometimes in interpretation of clearance pathways in feces based on metabolite profiles observed in fecal samples. Gut microflora can potentially convert certain metabolites back to their parent (e.g., known for glucuronides and N-oxides). After an oral dose (dosing route for majority of drugs), unchanged NME detected in feces could be due to: 1) either unabsorbed drug or hepatically formed metabolites excreted in bile and further metabolized to parent drug by gut microflora, or 2) absorbed drug excreted unchanged in bile via liver, or 3) intestinal secretion of unchanged drug. Metabolites in feces, especially if observed at substantially higher levels in feces compared with lower/nonexisting levels in circulation, may also indicate substantial metabolism via intestinal enzymes. Potential contribution of intestinal and hepatic metabolizing enzymes (and transporters) toward NME metabolism and excretion into feces versus unabsorbed drug component is important to understand for victim DDI risk evaluation.

Ideally the best estimate of f_CL,biliary can be made after intravenous dosing of a NME (radiolabeled or radiotracer) followed by determination of unchanged NME in feces but intravenous ¹⁴C-ADME studies are not routinely done (approximately 20% of ¹⁴C-ADME studies done across IQ companies were via intravenous route). Enhanced appreciation of information gathered from such studies has resulted in increased conduct of microtracer intravenous ¹⁴C-ADME in recent years with primary goals of providing better understanding absolute bioavailability/PK, mass balance, and routes of clearance. It should be noted that intestinal secretion cannot be confidently determined even after an intravenous dose because of the fact that quantitative, uninterrupted bile collection from healthy humans is not possible.

Investigators need to comprehensively understand the contribution of metabolism toward the overall routes of clearance of a NME via careful quantitative evaluation of metabolite profiles in plasma, urine, and feces. Understanding the site of formation of primary metabolites (e.g., intestine, liver, or kidney) is crucial in evaluating metabolizing enzymes involved in the primary metabolic pathways and responsible for precipitation of metabolic victim DDI.

Incorporating robust in vitro f_m data and f_{CL,metabolism} information with appropriate modeling and simulation may also provide better confidence in prediction of victim DDI liability in the clinic. Modeling and simulation allows investigators to better assess dependence/sensitivity (i.e., the uncertainties in measurement of these parameters) of these key victim DDI parameters and their impact on clinical DDI timing and study design at different stages of drug development. The following section will briefly discuss various commonly used modeling approaches along with an explanation of their suitability in specific development phases based on available data and specific questions of interest.

Simple Static Models

Static models are based on a single static perpetrator concentration. In the most simplified versions, f_mCYP for victim drug is assumed to be 100% (worst case scenario). Numerous reference concentrations for perpetrator (total or unbound) have been proposed previously (Ito et al., 2004; Obach et al., 2007). Generally for the perpetrator, the total average concentration at steady state, maximum concentration at steady state, or hepatic inlet concentration are recommended and used to predict victim drug interactions in the liver using the static model (http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm292362.pdf). Therefore, this approach often leads to overpredictions of victim DDIs (Ito et al., 2003). The static model can be a useful tool for early DDI risk assessment when supporting data are limited (e.g., during early discovery phase).

Mechanistic Static Models

Mechanistic static models incorporate more victim drug information, such as estimated f_mCYP and impact of the DDI on both hepatic and gut metabolism (Obach et al., 2006) or the integrated impact of competitive or mechanism-based inhibition and induction (Fahmi et al., 2008). The mechanistic static model is often applied once in vivo victim drug data are available, although this is not required (Table 7, example 6).

Mechanistic Dynamic Models

Mechanistic dynamic models including physiologically-based pharmacokinetic (PBPK) simulations incorporate the time-dependent change of both perpetrator and victim drug concentrations over the time course of the DDI. Therefore, this approach permits investigation of nonlinear kinetics and time-dependent phenomena, which include drug accumulation, additional impact due to the formation of inhibitory metabolite(s), and effect of dosing regimen. Commercial software packages are available that allow incorporation of literature data on human physiology, changes in special populations and disease states, and drug effects on physiologic rates [e.g., Simcyp (Certara), Phoenix (Certara), GastroPlus (Simulations Plus), PKSim (Bayer)]. PBPK modeling requires the greatest degree of parameterization based on both in vitro and in vivo data. Key victim drug parameters required for PBPK model development include volume of distribution (V_dss) and clearance (CL) (Supplemental Table 4S). Although PBPK approaches may be applied at the stage of discovery or development, it is most powerful in later stages when more data become available.

Model Application: “When and What”

Model selection must always be issue driven and based on particular development stage and available data. During early clinical development (i.e., clinical candidate selection), predictions are primarily used to improve mechanistic understanding and support internal decision making. At this stage the understanding of victim drug disposition is mainly based on preclinical and/or in vitro data, which are generally limited. Applications of modeling at this stage include support of prediction of clinical DDI risk, DDI study design, and victim drug dose selection (especially critical for a victim drug that has narrow safety window). In these situations a simplified model may be more suitable. In situations where more definitive in vitro ADME data and Phase I clinical PK data are available, PBPK simulation is applied. Retrospective model simulation of in vivo data can be used to estimate f_m,_CYP and the impact of alternative trial design on outcomes. In later stages of development when data from human mass-balance and DDI studies with a specific inhibitors (e.g., itraconazole for the inhibition of CYP3A4 for drugs metabolized by this enzymatic pathway) become available, the model is further refined and used for simulation of victim DDIs with other inhibitor(s) or inducer(s) of the same pathway, or inhibitors of a minor metabolic pathway, victim drug PK in special populations (e.g., hepatic or renal impaired), dose selection, influence labeling, or justifying of waivers for additional DDI studies.

Building Confidence in Model Predictions

Several approaches have been developed to build confidence in and ensure mechanistic relevance of the simulation results, which will lead to increased impact of modeling and simulation efforts. These approaches include verification of assumptions, sensitivity analysis, and assessment of variability. The level of model verification or validation will depend upon the stage of development and its impact on critical decision making.

Recommended Strategy in Modeling and Simulation

Modeling and simulation is a valuable tool in predicting victim DDI throughout every stage of drug development. In each stage, fit-for-purpose strategies should be based on the issue to be addressed. During early development stages, modeling and simulation will support internal decision making by leveraging mechanistic understanding when only preclinical and in vitro data are available. When data are limited, a more simplified modeling approach is advised. At later stages, modeling aids study design, dose selection, labeling, and justification of delay/waiver of clinical victim DDI studies. As more data become available more complex models may be developed. For initial metabolic victim DDI risk assessments, a qualitative understanding of contribution of metabolism toward overall clearance of a NME (f_{Cl,metabolism}) and fractional contribution of a DME toward overall metabolism (f_m) is necessary. Typically some human PK data (after single dose with reasonable expectation of linear PK or after multiple dose if not) will be necessary for more quantitative prediction of in vivo victim DDI.

PBPK modeling requires the greatest amount of data. However, a PBPK approach allows the greatest potential for “what-if” scenario analysis and simulations leading to improved mechanistic understanding. Specifically, PBPK-based predictions allow analyses of how f_m,CYP characterization may impact victim DDI prediction. This information may then be used to inform exclusion criteria (i.e., if a polymorphic enzyme involved), victim drug dose selection, sampling schedule, and optimal chemical inhibitor.

Sensitivity analysis is strongly recommended in all situations but especially when data are incomplete or uncertain (Zhao et al., 2012; Jones et al., 2015). This assessment can help identify to what extent the model can tolerate uncertainty and the range of uncertainties, as well as indicate additional experimental data to be generated and/or confirmed. For victim drug DDI prediction, the fraction metabolized (f_m) and fraction escaping first-pass metabolism in the gut (F_G) are common parameters that need to be considered for sensitivity analysis. The worst case scenario from sensitivity analysis may be included in the risk assessment. It is essential that confidence and acceptance/cut-off criteria be well defined at the beginning of model development to ensure a well-established metric for comparing model performance.