Novel risk assessment of reactive metabolites from discovery to clinical stage

Abstract

This study was aimed to predict drug-induced liver injury caused by reactive metabolites. Reactive metabolites covalently bind to proteins and could result in severe outcomes in patients. However, the relation between the extent of covalent binding and clinical hepatotoxicity is still unclear. From a perspective of body burden (human in vivo exposure to reactive metabolites), we developed a risk assessment method in which reactive metabolite burden (RM burden), an index that could reflect the body burden associated with reactive metabolite exposure, is calculated using the extent of covalent binding, clinical dose, and human in vivo clearance. The relationship between RM burden and hepatotoxicity in humans was then investigated. The results indicated that this RM burden assessment exhibited good predictability for sensitivity and specificity, and drugs with over 10 mg/day RM burden have high-risk for hepatotoxicity. Furthermore, a quantitative trapping assay using radiolabeled trapping agents ([³⁵S]cysteine and [¹⁴C]KCN) was also developed, to detect reactive metabolite formation in the early drug discovery stage. RM burden calculated using this assay showed as good predictability as RM burden calculated using conventional time- and cost-consuming covalent binding assays. These results indicated that the combination of RM burden and our trapping assay would be a good risk assessment method for reactive metabolites from the drug discovery stage.

INTRODUCTION

Metabolism is a detoxification system, and most drugs are metabolized and excreted as non-toxic compounds (Meyer, 1996). Some drugs are metabolized to chemically reactive species such as quinone, α,β-unsaturated carbonyl, iminium, etc. (Edwards and Sturino, 2011). These chemically reactive species covalently bind to proteins and this sometimes results in drug-induced liver toxicity in humans (Leung et al., 2012). These metabolite-induced undesirable adverse events are difficult to predict in non-clinical animal studies and limited scale clinical studies, because they are influenced by multiple factors of patients of different populations (e.g., age, sex, disease states, genetic predisposition). Thus, the majority of drug-induced liver injury (DILI) is idiosyncratic and it rarely occurs (only 1 in 10,000 to 100,000 treated patients) (Bell and Chalasani, 2009). Despite the difficulty of prediction, it is very important to avoid or mitigate the risk induced by reactive metabolites, since post-marketing idiosyncratic drug toxicity would seriously impact the pharmaceutical company as well as the patients, and therefore many pharmaceutical companies are making a huge effort to screen out compounds that form reactive metabolites (Park et al., 2011). To detect reactive metabolite formation, covalent binding assays using radiolabeled compounds (³H, ¹⁴C) are often performed in drug development. However, since the synthesis of radiolabeled compounds includes long and costly processes, they are rarely available in the drug discovery stage. (Isin et al., 2012) . From the aspect of time and cost, radiolabeled drugs are not obtained in the drug discovery stage. Therefore, trapping assays, instead of covalent binding assays, are often performed as a high-throughput screening (Brink et al., 2017). When reactive metabolites were produced in vitro, some agents can trap these chemically unstable intermediates and the complex of reactive metabolites and trapping agents can be detected using LC/MS. However, since this method is not quantitative due to the variability in the MS response intensity of compounds, the use of radiolabeled trapping agents is considered necessary to obtain quantitative data.

Based on the Hard and Soft Acid and Base theory, reactive metabolites are classified as “soft” and “hard” electrophiles. Quinone, α,β-unsaturated carbonyl (Michael acceptor), and epoxide are classified as “soft” electrophiles, and they react with the sulfhydryl group (-SH) of glutathione, acetyl cysteine, and cysteine, which are “soft” nucleophiles. On the other hand, iminium is classified as a “hard” nucleophile, and it reacts with “hard” nucleophiles such as cyanide (CN^-) and lysine (-NH₂) (Kalgutkar and Soglia, 2005). In addition to these, some trapping assays such as those using dansyl GSH (Gan et al., 2005), stable isotope GSH (Rousu et al., 2009), and quaternary amine GSH (Soglia et al., 2006) were developed as high-throughput assays. However, based on these assays, it is very difficult to interpret the extent of reactive metabolites in vitro, because there is no clear correlation between in vitro reactivity alone and clinical adverse events. It was reported that the combination of the extent of covalent binding and daily doses showed good predictability (Obach et al., 2008). There have also been some reports that indicated daily doses are very important for prediction (Reese et al., 2011; Nakayama et al., 2009). However, to characterize ADME properties (absorption, distribution, metabolism, and excretion) of a drug, in addition to the daily dose and the extent of covalent binding, the fraction absorbed and the fraction metabolized also need to be considered. For understanding of human in vivo risk, the body burden of reactive metabolites (human in vivo exposure to reactive metabolites) would be most important.

In this study, we developed an evaluation method of reactive metabolite formation in humans in vivo. To improve the predictability of idiosyncratic drug toxicity caused by reactive metabolites, we tried to estimate the “reactive metabolite burden (RM burden)”; that is, in vivo human exposure to reactive metabolites. We took human oral clearance of drugs into consideration and investigated their relationship to drug safety information. These parameters are essential to consider human in vivo ADME profile. In addition to the risk assessment method, we also developed quantitative trapping assays using radiolabeled trapping agents. Since conventional covalent binding assays require significant cost and time as described above, we developed an alternative assay for determining RM burden in the drug discovery stage. Two types of trapping agents ([¹⁴C]KCN, [³⁵S]cysteine ([³⁵S]Cys)) are used to assess two different types of reactive metabolites: “soft” electrophiles and “hard” electrophiles (Inoue et al., 2009). These radiolabeled trapping agents are more easily available than each radiolabeled drug. Using these assays, the extent of reactive metabolite formation can be measured quantitatively without each radiolabeled drug. The combination of RM burden and this quantitative trapping assay enables us to assess the risk of reactive metabolites from the early discovery stage.

MATERIALS AND METHODS

Materials

Pooled human microsomes (n = 50, mixed gender) were purchased from Sekisui XenoTech, LLC (Kansas City, KS, USA). NADPH-generating system was purchased from BD Bioscience (Franklin Lakes, NJ, USA). Troglitazone, clozapine, indinavir and acetonitrile (MeCN) were purchased from Wako Pure Chemicals (Osaka, Japan). Nefazodone hydrochloride, buspirone hydrochloride, ciprofloxacin, diclofenac sodium, ketoconazole and carbamazepine were purchased from Sigma-Aldrich (St Louis, MO, USA). Quetiapine fumarate was purchased from Toronto Research Chemicals (North York, Canada). Rimonabant was purchased from Santa Cruz Biotechnology (Dallas, TX, USA). [³⁵S]Cys, [¹⁴C]KCN, Ultima FLO-M, and Deepwell LumaPlate-96 were purchased from PerkinElmer Life and Analytical Science (Boston, MA, USA). XBridge BEH C18 Column, 130Å, 3.5 µm, 4.6 mm × 150 mm and Sep-Pak tC18 96-well µElution Plates were purchased from Waters Corporation (Milford, MA, USA). Formic acid (FA) and dimethyl sulfide (DMSO) were purchased from Nacalai Tesque (Kyoto, Japan).

In vitro trapping study using [³⁵S]Cys and [¹⁴C]KCN

The experimental procedure was based on a conventional covalent binding study (Evans et al., 2004). The stock solutions of test compounds were prepared in DMSO at 10 mM and the stock solutions were diluted with MeCN. The final incubation mixture consisted of the following: 10 µM test compound, 1 mg/mL human liver microsomes, 100 mM potassium phosphate buffer (pH 7.4), 1.3 mM NADP+, 3.3 mM glucose 6-phosphate, 3.3 mM MgCl₂, 0.45 units/mL glucose-6-phosphate dehydrogenase, and 1 mM [¹⁴C]KCN or [³⁵S]Cys. The final incubation volume was 0.5 mL. The incubation mixtures without test compound were prepared as the control samples. After incubation of the mixtures for 1 hr in a 37°C water bath, the reaction was terminated by loading the incubated sample fractions onto the wells of SPE plates (Sep-pak tC18). The trapping assay using [¹⁴C]KCN and the assay using [³⁵S]Cys were performed independently (n = 1 for [³⁵S]Cys, n = 2 for [¹⁴C]KCN).

Analytical conditions

Radio-HPLC was conducted using a Shimadzu Nexera X2 system equipped with a binary pump, autosampler, thermostat column compartment, and PDA detector coupled with an online radiometric detector with a 500-µL cell (Radiomatic 625TR, PerkinElmer Life and Analytical Science, Boston, MA, USA).

Analytical methods: Elution was conducted using an Xbridge C18 BEH column, 3.5 µm, 4.6 mm × 150 mm (Waters Corporation, Milford, MA, USA) at 40°C with a gradient B conc. 5% (0 to 5 min), 95% (5 to 35 min), 95% (35 to 40 min) (Mobile phase A: 0.1% FA aq. B: 0.1% FA MeCN) at a flow rate of 1.0 mL/min over 40 min. Initial 5 min eluent from the HPLC was wasted through a diverter valve to eliminate unreacted trapping agents. Liquid scintillation cocktail (Ultima Flo-M) was mixed with the eluent from the HPLC at a flow rate of 3.0 mL/min. The radioactivity of LC eluent was measured by monitoring ¹⁴C scintillation period (almost the same as ³⁵S). The obtained chromatograms of each sample were compared with that of the control. To identify the peak, the mass spectrometric analysis was also conducted on LTQ Orbitrap XL (Thermo Fisher Scientific, Waltham, MA, USA).

Calculation of adduct formation rates

The peaks observed on radio-chromatograms were integrated and the obtained peak areas were used for further calculation as follows:

In this calculation, the efficiency and the recovery from SPE was regarded as 1.0 for simplicity.

Data collection

The values of the covalent binding (CVB) of marketed drugs were obtained from literature references (Bergström et al., 2011; Inoue et al., 2009; Nakayama et al., 2009; Obach et al., 2008; Usui et al., 2009). The values of human oral clearance (CL/F) and maximum dose (mg/day) of 57 drugs were collected from literature references, drug labels, and the drug approval documents from FDA and EMA (Barbhaiya et al., 1996; Boik and Newman, 2008; Durand et al., 1992; Gasser et al., 1987; Kerremans et al., 1982; Li et al., 2012; Mather et al., 1981; Ward and Smith, 2004; Wiseman and Chiaini, 1972). When different values were obtained for one drug, the mean was used for further calculation. The safety profiles of the drugs were obtained from their drug labels (FDA or PMDA labels).

Classification of tested drugs

The safety information of tested drugs was obtained from US labels and Japan labels in reference to Nakayama et al. (2009).

Calculation formulas for reactive metabolite burden (RM burden)

The body burden of reactive metabolites (RM burden) was calculated as follows:

RM burden = Dose × Fa × CL_RM/CL_total

where Dose is the maximum clinical dose (mg/day), CL_RM is the formation clearance of reactive metabolites, CL_total is the total human clearance of each prescription drug, Fa is “fraction absorbed,” and F is bioavailability:

F = Fa × Fg × Fh ↔ Fa = F/(Fg × Fh)

where Fg is “the fraction that escapes gut metabolism,” Fh is “the fraction that escapes hepatic first-pass metabolism.” For metabolically stable drugs, Fg × Fh is regarded as 1.0. An approximate formula of RM burden is then described as follows:

RM burden = CL_RM × Dose /(CL_total/F) = CL_RM × AUC_{oral =} X_RM

The CL_RM was calculated as follows:

CL_RM = CVB (pmol/mg/hr)/10 (µM) × 45 (mg protein/g liver) × 20 (g liver/kg)

For simplicity, the metabolite-forming rate was calculated under the assumption that the concentration of test compounds was constantly 10 µM under the covalent binding assay or trapping assay conditions, and in vitro reactive metabolite-forming rate was scaled up to intrinsic CL_RM according to well-stirred model without f_u,inc or f_u,p (Obach, 1999).

Calculation of CL_RM using trapping assay data

CL_RM from the radiolabeled trapping assays were calculated as follows:

CL_RM = Adduct formation rate (pmol/mg/hr)/10 (µM) × 45 (mg protein/g liver) × 20 (g liver/kg)

where adduct formation rate was calculated using the equation described above.

RESULTS

Calculation of RM burden

The data for 57 drugs were collected and RM burden of 48 drugs was calculated. All data are listed in Table 1. Safe drugs and warning drugs were classified as low-risk drugs. Black-boxed warning and withdrawn drugs were classified as high-risk drugs.

Table 1. The collected data and the calculated RM burden.

Compound	Dose [mg/day]	CL/F [L/hr/kg]	Drug Label	CVB (Ms) [pmol/mg/hr]	RM burden [mg/day]	Reference for CL/F	Reference for CVB
Acetaminophen	4000	0.243	Boxed warning	113	167.9	Approval document	Nakayama et al., 2009, Usui et al., 2009, Obach et al., 2008
Alpidem	150	0.390	Withdrawn	440	15.2	Durand et al., 1992	Usui et al., 2009
Aminopyrine	3000	0.177	Withdrawn	31	47.1	Boik and Newman, 2008	Nakayama et al., 2009
Amlodipine	10	0.398	Safe	7	0.0	Approval document	Nakayama et al., 2009
Atorvastatin	80	0.517	Warning	352	4.9	Approval document	Nakayama et al., 2009
Benzbromarone	150	0.090	Boxed warning	390	58.6	Japanese drug label	Nakayama et al., 2009
Buspirone	60	44.800	Safe	139	0.0	Boik and Newman, 2008	Obach et al., 2008
Caffeine	900	0.084	Safe	6	6.2	Approval document	Nakayama et al., 2009, Usui et al., 2009
Carbamazepine	1200	0.069	Boxed warning	49	77.3	Approval document	Usui et al., 2009, Obach et al., 2008
Celecoxib	400	0.467	Warning	13	1.0	Approval document	Nakayama et al., 2009
Clopidogrel	75	--	Warning	417	--	--	Nakayama et al., 2009
Clozapine	900	0.313	Boxed warning	195	50.4	Li et al., 2012	Nakayama et al., 2009, Inoue et al., 2009
Ciprofloxacin	400	1.005	Warning	--	--	Gasser et al., 1987	--
Dexamethasone	10	0.268	Warning	445	3.0	Japanese drug label	Usui et al., 2009
Diclofenac	200	0.498	Warning	57	2.1	Approval document	Nakayama et al., 2009, Usui et al., 2009, Obach et al., 2008, Inoue et al., 2009
Diphenhydramine	150	--	Safe	76	--	--	Obach et al., 2008
Donepezil	10	0.115	Safe	30	0.2	Approval document	Nakayama et al., 2009
Erythromycin	1000	0.840	Warning	57	6.1	Mather et al., 1981	Nakayama et al., 2009
Ethinylestradiol	0.035	0.300	Safe	938	0.0	Ward and Smith, 2004	Nakayama et al., 2009
Fluoxetine	80	0.108	Warning	15	1.0	Approval document	Nakayama et al., 2009
Furosemide	80	0.092	Warning	79	6.2	Kerremans et al., 1982	Nakayama et al., 2009
Imipramine	300	2.411	Warning	202	2.3	Boik and Newman, 2008	Nakayama et al., 2009, Usui et al., 2009
Indinavir	1600	1.135714	Warning	--	--	Boik and Newman, 2008	--
Indomethacin	200	--	Warning	25	--	--	Nakayama et al., 2009, Obach et al., 2008
Ketoconazole	400	0.285	Boxed warning	--	--	Approval document	--
Levofloxacin	750	0.108	Safe	0	0.1	Approval document	Nakayama et al., 2009
Losartan	100	1.528	Warning	105	0.6	Approval document	Usui et al., 2009
Meloxicam	15	0.008	Safe	60	9.5	Approval document	Obach et al., 2008
Nefazodone	600	2.153	Boxed warning	1848	46.3	Barbhaiya et al., 1996	Usui et al., 2009, Obach et al., 2008
Nevirapine	400	0.023	Boxed warning	19	30.6	Approval document	Nakayama et al., 2009
Olanzapine	20	0.393	Safe	139	0.6	Approval document	Nakayama et al., 2009
Olmesartan	40	0.066	Safe	3	0.2	Approval document	Nakayama et al., 2009
Paroxetine	50	0.516	Warning	2669	23.3	Boik and Newman, 2008	Usui et al., 2009, Obach et al., 2008
Phenytoin	600	0.036	Warning	4	6.5	Ward and Smith, 2004	Nakayama et al., 2009
Pioglitazone	45	0.051	Warning	295	23.4	Approval document	Nakayama et al., 2009, Usui et al., 2009
Pravastatin	80	0.432	Safe	4	0.1	Approval document	Nakayama et al., 2009
Procainamide	4000	3.800	Warning	5	0.5	Approval document	Nakayama et al., 2009
Propranolol	480	2.586	Warning	221	3.7	Boik and Newman, 2008	Nakayama et al., 2009, Obach et al., 2008
Quetiapine	400	1.159	Warning	--	--	Boik and Newman, 2008	--
Rimonabant	20	0.087	Safe	920	19.0	Approval document	Bergström et al, 2011
Ritonavir	1200	0.126	Boxed warning	253	217.6	Approval document	Nakayama et al., 2009
Rosiglitazone	8	0.081	Safe	853	7.6	Approval document	Nakayama et al., 2009, Usui et al., 2009
Sertraline	200	2.364	Warning	49	0.4	Boik and Newman, 2008	Usui et al., 2009
Sudoxicam	50	0.006	Withdrawn	35	26.5	Wiseman and Chiaini, 1972	Obach et al., 2008
Sulfamethoxazole	1600	--	Warning	3	--	--	Nakayama et al., 2009
Tacrine	160	11.683	Warning	201	0.2	Approval document	Nakayama et al., 2009, Usui et al., 2009
Tamoxifen	40	0.102	Warning	12	0.4	Approval document	Nakayama et al., 2009
Ticlopidine	600	2.678	Boxed warning	1072	21.6	Approval document	Nakayama et al., 2009, Usui et al., 2009
Tienilic Acid	500	0.056	Withdrawn	935	754.8	Kerremans et al., 1982	Nakayama et al., 2009, Usui et al., 2009, Obach et al., 2008
Troglitazone	600	0.571	Withdrawn	1170	110.7	Approval document	Usui et al., 2009
Valproic Acid	4200	0.012	Boxed warning	6	200.8	Japanese drug label	Nakayama et al., 2009
Valsartan	320	0.136	Safe	1	0.3	Boik and Newman, 2008	Nakayama et al., 2009
Venlafaxine	225	1.510	Safe	5	0.1	Approval document	Usui et al., 2009
Verapamil	480	--	Warning	66	--	--	Nakayama et al., 2009
Warfarin	10	0.004	Safe	16	3.7	Approval document	Nakayama et al., 2009
Zafirlukast	40	0.480	Warning	36	0.3	Approval document	Nakayama et al., 2009
Zolpidem	10	0.244	Safe	23	0.1	Approval document	Usui et al., 2009

--: No data CVB: the extent of covalent binding Ms: microsomes RM burden: reactive metabolite burden Approval document obtained from EMA or FDA

Relation between RM burden and drug label

The safety information was plotted against the extent of CVB, CVB × Dose, and RM burden, respectively. CVB extent alone did not have clear relation with the drug label (Fig. 1A). Consideration of the dose improved the relation (Fig. 1B). Furthermore, in the RM burden analysis, all of the 14 high-risk drugs showed above 10 mg/day RM burden, and 31 of the 34 low-risk drugs showed below 10 mg/day RM burden (Fig. 1C). The specificity and selectivity were 100% and 90%, respectively. Additionally, our in-house compounds, which had been discontinued during clinical development due to several reasons including the increase of hepatic parameters, showed over 10 mg/day RM burden. The analysis based on RM burden showed good predictability. In this analysis, the extent of covalent binding (pmol/mg/hr) from the literature was used for the calculation.

Fig. 1

The safety information of drugs against the extent of covalent binding (A), CVB × Dose (B), RM burden (C). The red line in (C) shows 10 mg/day, which is reported as a safety threshold of a daily dose (Uetrecht, 1999).The solid circles indicate reference compounds, while the open circles indicate our in-house compounds.

Sample cleanup procedure development ([¹⁴C] KCN trapping assay)

After the reaction, the incubation samples were mixed with twice the volume of MeCN and centrifuged (10,000 rpm, 4°C, 5 min), and the supernatants were collected. The supernatant was analyzed using radio-HPLC. High background levels were shown on the radio-chromatogram of the control sample (Fig. 2AB). To reduce the background due to the unreacted cyanide, the samples were cleaned up using SPE. The reaction was terminated by loading the incubation sample fractions onto SPE and the wells were washed with 2% MeCN/H₂O four times. To prevent the production of toxic HCN gas, washed solutions were received in plates containing NaOH solution. After washing, the samples were eluted with MeCN (200 µL). The SPE procedures were performed in duplicate. The eluted samples were analyzed using radio-HPLC and LC/MS. The background in the chromatogram of the control sample was reduced effectively (Fig. 2C). The same clean-up procedure was performed for [³⁵S]Cys trapping assay samples. The detection limit of each assay was approximately 100 pmol/hr/mg protein. (Fig. 3).

Fig. 2

The radio-chromatograms of the control sample of [¹⁴C]KCN trapping assay. From 0 to 15 min, high background levels were detected due to the unreacted [¹⁴C]cyanide. (A: Full scale, B: Magnified, C: after SPE cleanup procedure).

Fig. 3

The radio-chromatograms of the trapping assay samples using radiolabeled trapping agents. The chromatograms show that the detection limit of each trapping assay is approximately 100 pmol/hr/mg protein.

Comparison of the trapping assay data and the CVB extent

For 11 drugs (buspirone, carbamazepine, ciprofloxacin, clozapine, diclofenac, indinavir, ketoconazole, nefazodone, rimonabant, troglitazone and quetiapine), the obtained values of the extent of covalent binding from the trapping assay were compared with those in literature references. These drugs exhibited a variety of extent of covalent binding, and from their structures, their reactive metabolites were supposed to be trapped by Cys and/or KCN. The values obtained by the trapping assay were not largely different from CVB values in the references (within 5 fold). Therefore, this assay was thought to be a useful tool to surrogate conventional covalent binding assays considering the hurdle of time and cost. (Table 2).

Table 2. The extent of covalent binding and the trapping assay data.

Compound	CVB(Ms) [pmol/mg/hr]	Cys [pmol/mg/hr]	KCN [pmol/mg/hr]	Reference for CVB
Buspirone	139	98	382	Obach et al., 2008
Carbamazepine	49	<100	--	Usui et al., 2009, Obach et al., 2008
Clozapine	195	429	<100	Nakayama et al., 2009, Inoue et al., 2009
Diclofenac	57	257	--	Nakayama et al., 2009, Usui et al., 2009, Obach et al., 2008, Inoue et al., 2009
Nefazodone	1848	2333	934	Usui et al., 2009, Obach et al., 2008
Rimonabant	920	<100	939	Bergström et al, 2011
Troglitazone	1170	265	<100	Usui et al., 2009
Ciprofloxacin	--	--	138	--
Indinavir	--	--	193	--
Ketoconazole	--	<100	88	--
Quetiapine	--	93	243	--

--: No data CVB: the extent of covalent binding Ms: microsomes

In the clozapine, diclofenac and troglitazone samples, cysteine adducts were detected. In the ciprofloxacin, indinavir, rimonabant, and ketoconazole samples, cyanide adducts were detected. In the buspirone and nefazodone and quetiapine samples, both cysteine and cyanide adducts were detected. In the carbamazepine sample, no adducts were detected. In this study, cysteine was used to trap soft electrophiles, because cysteine has a reacting sulfhydryl group as does glutathione. Diclofenac-cysteine adduct was detected in the form of “Cys+O-2H”, which indicated reactive quinone imine was formed from diclofenac. In addition, cysteine can also trap aldehydes because of its suitable structure (Inoue et al., 2015). Quetiapine-cysteine adduct was detected in the form of “Cys-4H-C₂H₄O.” Moreover, when Cys-Gly was used as a trapping agent, “CysGly-4H-C₂H₄O” was also detected. However, when GSH was used as a trapping agent, no GSH-adducts were detected (data not shown). These results indicated that reactive aldehyde was formed from quetiapine. On the other hand, KCN was used for trapping of hard electrophiles, e.g., iminium, and rimonabant-cyanide adduct was detected in the form of “+CN-H”, which indicated that reactive iminium was formed from rimonabant. The postulated structures of the adducts are shown in Fig. 4.

Fig. 4

The postulated structures of reactive metabolites and their adducts with trapping agents. (diclofenac, quetiapine and rimonabant)

For the compounds that produce hard electrophiles, the trapping assay data of rimonabant was the only example that could be compared with covalent binding data. The trapping assay data (939 pmol/mg/hr) was very close to the extent of CVB (920 pmol/mg/hr).

Relation between RM burden using the trapping assay data and drug label

To predict the risk of reactive metabolite formation in the discovery stage, risk assessments without radiolabeled test compounds are preferable. Thus, the RM burden calculated based on the trapping assay was plotted against safety information using the extent of covalent binding. When the drugs were trapped by both of Cys and KCN, the greater adduct rate was used for further calculation. The results are shown in Fig. 5.

Fig. 5

Reactive metabolite (RM) burden calculated using the trapping assay data. The solid stars indicate the plots for which RM burden was calculated using greater one from two kinds of trapping assay data. The solid circles indicate the plots for which RM burden was calculated using the extent of covalent binding (CVB) data. The red line shows 10 mg/day, a boundary that would separate low risk drugs and high risk drugs.

Ciprofloxacin, diclofenac, quetiapine, and buspirone are low-risk drugs (safe or warning) and their RM burden is below 10 mg/day. Clozapine, nefazodone, troglitazone, and ketoconazole are high-risk drugs (boxed warning or withdrawn) and their RM burden is above 10 mg/day. The predictability of RM burden using radiolabeled trapping agents was high.

DISCUSSION

In this study, we developed a risk assessment method based on RM burden, which is calculated using the extent of in vitro covalent binding, human in vivo oral clearance, and the daily dose. In the RM burden calculation, we tried to estimate human in vivo exposure to reactive metabolites. The results of this analysis indicated that over 10 mg/day RM burden increases hepatotoxicity risk. Due to the difficulty in risk assessment of reactive metabolites, pharmaceutical companies are making huge efforts to eliminate drug candidates that form reactive metabolites. From the empirical analysis, the daily dose has been considered as an important factor. However, daily dose in humans cannot be obtained in the early stage. When a drug candidate is selected, the clinical daily dose is generally predicted based on the predicted clearance in several ways (Hosea et al., 2009; Nakayama et al., 2018). In cases in which the dose is over/under-estimated, the risk of reactive metabolites will be judged incorrectly. However, in our RM burden method, the key factors, the daily dose and CL/F in animals, can be converted to human AUC, and RM burden can be calculated using the AUC. This would be more useful for DMPK scientists, because first of all, DMPK scientists predict human clearance, the volume of distribution, and then dose is predicted based on pharmacologically required concentrations and/or AUC. When the human clearance was not predicted properly, the clinical dose would be changed to fit the required AUC level. On the basis of AUC, the risk of underestimation or overestimation of the risk of reactive metabolites due to human prediction would be reduced as compared with the dose prediction. Since AUC is dependent on dose, using AUC is compatible with reports that show close relation between daily dose and hepatotoxicity risk (Nakayama et al., 2009; Thompson et al., 2012). Moreover, the result that 10 mg/day RM burden is a threshold is in a good agreement with a well-known report that “drugs given at a daily dose of 10 mg or less are rarely if ever associated with a high incidence of idiosyncratic drug reactions.” (Uetrecht, 1999). We believe that our analysis will help risk assessment in drug development and would also help to reduce drug-induced toxicity in patients.

Although the RM burden analysis showed good predictability in our collected data set, there were some exceptions (such as indinavir and rimonabant). Indinavir exhibited over 10 mg/day RM burden, and it was classified as a warning drug in the FDA label. However, indinavir is reported to cause acute hepatic hepatitis, and the use is not recommended. Rimonabant is classified as a safe drug, in spite of the > 10 mg/day RM burden. Although rimonabant covalently binds to microsomal proteins to a high extent, no adverse events related to reactive metabolites have been reported. This may be because of the relatively short period of usage. Rimonabant was approved in Europe in 2006 for treatment of obesity and withdrawn in 2007 with the concern of pharmacologically associated CNS adverse efects. The time period might have been insufcient to detect low frequency idiosyncratic adverse drug reactions (Senior, 2007).

If a test compound is metabolically labile, RM burden would be underestimated because CL_RM cannot be calculated properly in this assay and Fg × Fh could not be regarded as 1.0. We have to remind the possibility of underestimation of the risk, although the recently launched drugs and drug candidates would show high metabolic stability owing to the effort of CYP metabolic stability screening (Cerny, 2016).

For the assay development, there are some topics to be noticed. With effective cleanup by SPE, the concentration of KCN was set at 1 mM and a similar value (939 pmol/mg/hr (trapping assay) to 920 pmol/mg/hr (CVB)) was obtained. It is reported that the amount of cyanide-adducts increased in a dose-dependent manner of KCN (from 80 to 1000 µM), while another study was conducted at lower KCN concentrations because of high background. (Inoue et al., 2009; Meneses-Lorente et al., 2006). We considered that SPE cleanup method successfully reduced background noise caused by highly-polar substances. Therefore, we used radiolabeled cysteine, which is more polar and less costly than glutathione (GSH).

On the basis of the RM burden assessment, we developed a risk assessment strategy for reactive metabolites (Fig. 6). The parameters used could be updated as drug development proceeds. Generally, definitive data in humans are obtained in the late development stage. Therefore, in the discovery, the trapping assay data and the pharmacologically required AUC would be used. After the first-in-human study, oral clearance and the covalent binding data would be obtained using radiolabeled compounds. The risk assessment could then be refined step-by-step.

Fig. 6

Step-wise refinement of the risk assessment using RM burden. In the lead generation stage, formation of reactive metabolites is determined in high throughput screening. In the lead optimization stage, a quantitative trapping assay is performed, and using the AUC value required for the pharmacological action, RM burden is determined for quantitative risk assessment. In the preclinical stage, if the radiolabeled drug is obtained, CL_RM, trapping is updated to a more definitive value, CL_RM,CVB. During the clinical stage, observed human oral clearance data is available. RM burden would be helpful to reduce hepatotoxicity risk at higher doses in further clinical studies.

Although we focused on covalent binding to human microsomes via oxidative metabolism, there are also metabolites activated by non-CYP activation such as acyl glucuronide and acyl CoA metabolites. Regarding non-CYP metabolic activation, our trapping assay cannot detect the risk of hepatotoxicity because microsomes cannot catalyze phase II metabolism in the absence of appropriate co-factors. Non-CYP metabolism and phase II metabolism are often characteristic of substance structures (Gan et al., 2016; Stepan et al., 2013). Therefore, if a test compound has a characteristic structure, covalent binding assay using human hepatocytes and radiolabeled drugs would be suitable. With more insight into in vitro - in vivo extrapolation, the accuracy of CL_RM and the practical usage of this method would be improved. On the other hand, the risk of reactive metabolites is also studied in the aspect of immune responses (Shenton et al., 2005). Finally, both the production of reactive metabolites and responses of living body should be investigated.

In this study, we have developed the risk assessment of reactive metabolites in the aspect of the body burden of reactive metabolites in human body. Moreover, we developed the quantitative trapping assay using [³⁵S]Cys and [¹⁴C]KCN. In the combination of these, the risk assessment of reactive metabolites will be improved.

ACKNOWLEDGMENTS

We thank Saki Yamauchi and Michihide Maekawa for their technical assistance, and Dr. Kota Asahina for scientific advice. We would also like to thank ASCA Corporation for English language editing.

Conflict of interest

The authors declare that there is no conflict of interest.

REFERENCES

• Barbhaiya, R.H., Dandekar, K.A. and Greene, D.S. (1996): Pharmacokinetics, absolute bioavailability, and disposition of [14C]nefazodone in humans. Drug Metab. Dispos., 24, 91-95.
• Bell, L.N. and Chalasani, N. (2009): Epidemiology of idiosyncratic drug-induced liver injury. Semin. Liver Dis., 29, 337-347.
• Bergström, M.A., Isin, E.M., Castagnoli, N. Jr. and Milne, C.E. (2011): Bioactivation pathways of the cannabinoid receptor 1 antagonist rimonabant. Drug Metab. Dispos., 39, 1823-1832.
• Boik, J.C. and Newman, R.A. (2008): Structure-activity models of oral clearance, cytotoxicity, and LD50: a screen for promising anticancer compounds. BMC Pharmacol., 8, 12.
• Brink, A., Pähler, A., Funk, C., Schuler, F. and Schadt, S. (2017): Minimizing the risk of chemically reactive metabolite formation of new drug candidates: implications for preclinical drug design. Drug Discov. Today, 22, 751-756.
• Cerny, M.A. (2016): Prevalence of Non-Cytochrome P450-Mediated Metabolism in Food and Drug Administration-Approved Oral and Intravenous Drugs: 2006-2015. Drug Metab. Dispos., 44, 1246-1252.
• Durand, A., Thénot, J.P., Bianchetti, G. and Morselli, P.L. (1992): Comparative pharmacokinetic profile of two imidazopyridine drugs: zolpidem and alpidem. Drug Metab. Rev., 24, 239-266.
• Edwards, P.J. and Sturino, C. (2011): Managing the liabilities arising from structural alerts: a safe philosophy for medicinal chemists. Curr. Med. Chem., 18, 3116-3135.
• Evans, D.C., Watt, A.P., Nicoll-Griffith, D.A. and Baillie, T.A. (2004): Drug-protein adducts: an industry perspective on minimizing the potential for drug bioactivation in drug discovery and development. Chem. Res. Toxicol., 17, 3-16.
• Gan, J., Harper, T.W., Hsueh, M.M., Qu, Q. and Humphreys, W.G. (2005): Dansyl glutathione as a trapping agent for the quantitative estimation and identification of reactive metabolites. Chem. Res. Toxicol., 18, 896-903.
• Gan, J., Ma, S. and Zhang, D. (2016): Non-cytochrome P450-mediated bioactivation and its toxicological relevance. Drug Metab. Rev., 48, 473-501.
• Gasser, T.C., Ebert, S.C., Graversen, P.H. and Madsen, P.O. (1987): Ciprofloxacin pharmacokinetics in patients with normal and impaired renal function. Antimicrob. Agents Chemother., 31, 709-712.
• Hosea, N.A., Collard, W.T., Cole, S., Maurer, T.S., Fang, R.X., Jones, H., Kakar, S.M., Nakai, Y., Smith, B.J., Webster, R. and Beaumont, K. (2009): Prediction of human pharmacokinetics from preclinical information: comparative accuracy of quantitative prediction approaches. J. Clin. Pharmacol., 49, 513-533.
• Inoue, K., Shibata, Y., Takahashi, H., Ohe, T., Chiba, M. and Ishii, Y. (2009): A trapping method for semi-quantitative assessment of reactive metabolite formation using [³⁵S]cysteine and [¹⁴C]cyanide. Drug Metab. Pharmacokinet., 24, 245-254.
• Inoue, K., Fukuda, K., Yoshimura, T. and Kusano, K. (2015): Comparison of the Reactivity of Trapping Reagents toward Electrophiles: Cysteine Derivatives Can Be Bifunctional Trapping Reagents. Chem. Res. Toxicol., 28, 1546-1555.
• Isin, E.M., Elmore, C.S., Nilsson, G.N., Thompson, R.A. and Weidolf, L. (2012): Use of radiolabeled compounds in drug metabolism and pharmacokinetic studies. Chem. Res. Toxicol., 25, 532-542.
• Kalgutkar, A.S. and Soglia, J.R. (2005): Minimising the potential for metabolic activation in drug discovery. Expert Opin. Drug Metab. Toxicol., 1, 91-142.
• Kerremans, A.L., Gribnau, F.W., Tan, Y. and van Ginneken, C.A. (1982): Pharmacokinetic and pharmacodynamic studies of tienilic acid in healthy volunteers. Eur. J. Clin. Pharmacol., 22, 515-521.
• Leung, L., Kalgutkar, A.S. and Obach, R.S. (2012): Metabolic activation in drug-induced liver injury. Drug Metab. Rev., 44, 18-33.
• Li, L.J., Shang, D.W., Li, W.B., Guo, W., Wang, X.P., Ren, Y.P., Li, A.N., Fu, P.X., Ji, S.M., Lu, W. and Wang, C.Y. (2012): Population pharmacokinetics of clozapine and its primary metabolite norclozapine in Chinese patients with schizophrenia. Acta Pharmacol. Sin., 33, 1409-1416.
• Mather, L.E., Austin, K.L., Philpot, C.R. and McDonald, P.J. (1981): Absorption and bioavailability of oral erythromycin. Br. J. Clin. Pharmacol., 12, 131-140.
• Meneses-Lorente, G., Sakatis, M.Z., Schulz-Utermoehl, T., De Nardi, C. and Watt, A.P. (2006): A quantitative high-throughput trapping assay as a measurement of potential for bioactivation. Anal. Biochem., 351, 266-272.
• Meyer, U.A. (1996): Overview of enzymes of drug metabolism. J. Pharmacokinet. Biopharm., 24, 449-459.
• Nakayama, S., Atsumi, R., Takakusa, H., Kobayashi, Y., Kurihara, A., Nagai, Y., Nakai, D. and Okazaki, O. (2009): A zone classification system for risk assessment of idiosyncratic drug toxicity using daily dose and covalent binding. Drug Metab. Dispos., 37, 1970-1977.
• Nakayama, K., Ito, S., Suzuki, M., Takubo, H., Yamazaki, H. and Nomura, Y. (2018): Prediction of human pharmacokinetics of typical compounds by a physiologically based method using chimeric mice with humanized liver. Xenobiotica, 19, 1-11.
• Obach, R.S. (1999): Prediction of human clearance of twenty-nine drugs from hepatic microsomal intrinsic clearance data: an examination of in vitro half-life approach and nonspecific binding to microsomes. Drug Metab. Dispos., 27, 1350-1359.
• Obach, R.S., Kalgutkar, A.S., Soglia, J.R. and Zhao, S.X. (2008): Can in vitro metabolism-dependent covalent binding data in liver microsomes distinguish hepatotoxic from nonhepatotoxic drugs? An analysis of 18 drugs with consideration of intrinsic clearance and daily dose. Chem. Res. Toxicol., 21, 1814-1822.
• Park, B.K., Boobis, A., Clarke, S., Goldring, C.E., Jones, D., Kenna, J.G., Lambert, C., Laverty, H.G., Naisbitt, D.J., Nelson, S., Nicoll-Griffith, D.A., Obach, R.S., Routledge, P., Smith, D.A., Tweedie, D.J., Vermeulen, N., Williams, D.P., Wilson, I.D. and Baillie, T.A. (2011): Managing the challenge of chemically reactive metabolites in drug development. Nat. Rev. Drug Discov., 10, 292-306.
• Reese, M., Sakatis, M., Ambroso, J., Harrell, A., Yang, E., Chen, L., Taylor, M., Baines, I., Zhu, L., Ayrton, A. and Clarke, S. (2011): An integrated reactive metabolite evaluation approach to assess and reduce safety risk during drug discovery and development. Chem. Biol. Interact., 192, 60-64.
• Rousu, T., Pelkonen, O. and Tolonen, A. (2009): Rapid detection and characterization of reactive drug metabolites in vitro using several isotope-labeled trapping agents and ultra-performance liquid chromatography/time-of-flight mass spectrometry. Rapid Commun. Mass Spectrom., 23, 843-855.
• Senior, J.R. (2007): Drug hepatotoxicity from a regulatory perspective. Clin. Liver Dis., 11, 507-524.
• Shenton, J.M., Popovic, M., Chen, J., Masson, M.J. and Uetrecht, J.P. (2005): Evidence of an immune-mediated mechanism for an idiosyncratic nevirapine-induced reaction in the female Brown Norway rat. Chem. Res. Toxicol., 18, 1799-1813.
• Stepan, A.F., Mascitti, V., Beaumont, K. and Kalgutkar, A.S. (2013): Metabolism-guided drug design. MedChemComm, 4, 631-652.
• Soglia, J.R., Contillo, L.G., Kalgutkar, A.S., Zhao, S., Hop, C.E., Boyd, J.G. and Cole, M.J. (2006): A semiquantitative method for the determination of reactive metabolite conjugate levels in vitro utilizing liquid chromatography-tandem mass spectrometry and novel quaternary ammonium glutathione analogues. Chem. Res. Toxicol., 19, 480-490.
• Thompson, R.A., Isin, E.M., Li, Y., Weidolf, L., Page, K., Wilson, I., Swallow, S., Middleton, B., Stahl, S., Foster, A.J., Dolgos, H., Weaver, R. and Kenna, J.G. (2012): In vitro approach to assess the potential for risk of idiosyncratic adverse reactions caused by candidate drugs. Chem. Res. Toxicol., 25, 1616-1632.
• Uetrecht, J.P. (1999): New concepts in immunology relevant to idiosyncratic drug reactions: the “danger hypothesis” and innate immune system. Chem. Res. Toxicol., 12, 387-395.
• Usui, T., Mise, M., Hashizume, T., Yabuki, M. and Komuro, S. (2009): Evaluation of the potential for drug-induced liver injury based on in vitro covalent binding to human liver proteins. Drug Metab. Dispos., 37, 2383-2392.
• Ward, K.W. and Smith, B.R. (2004): A comprehensive quantitative and qualitative evaluation of extrapolation of intravenous pharmacokinetic parameters from rat, dog, and monkey to humans. I. Clearance. Drug Metab. Dispos., 32, 603-611.
• Wiseman, E.H. and Chiaini, J. (1972): Anti-inflammatory and pharmacokinetic properties of sudoxicam N-(2-thiazolyl)-4-hydroxy-2-methyl-2H-1,2-benzothiazine-3-carboxamide 1,1-dioxide. Biochem. Pharmacol., 21, 2323-2334.