Discussion

Detection and identification of major and disproportionate human circulating metabolites in a timely manner are very important to properly assess safety liability of these metabolites without delaying the development timeline of a new chemical entity (NCE). The exposure level of human circulating metabolites accounting for more than 10% of total drug-related materials needs to be covered in preclinical safety species, or additional nonclinical study of the metabolites is warranted (https://www.fda.gov/media/72279/download). Despite the importance of major and/or disproportionate human circulating metabolites, their identification is not available until samples from early clinical trials have been analyzed for metabolite profiling and identification. If the major or disproportionate metabolites for a given NCE were identified later in the clinical development, the follow-up studies on those metabolites may take significant time and can potentially delay the development program to registration. It is, therefore, essential that potential disproportionate human metabolites are identified in early stages and would be extremely valuable if they were identified before initiation of clinical studies. Additionally, understanding of the clearance mechanism of a drug is often obscured by lack of biliary excretion metabolic profiles in humans. The fecal metabolic profiles often do not provide information on the extent of metabolism because of the presence of unabsorbed drug, instability of certain metabolites, and deconjugation of conjugative metabolites by gut microflora in the gastrointestinal (GI) tract (Pellock and Redinbo, 2017). Furthermore, the fecal metabolism data are available only from the radiolabeled human ADME study, which is performed quite late in the clinical development of NCEs. An early assessment or understanding of a clearance mechanism can inform clinical investigators of the potential for DDI liability when combined with in vitro clearance mechanism and reaction phenotyping data.

DL is a tricyclic H1 antagonist that is used to treat allergies (Agrawal, 2001). DL was found to be extensively metabolized in five of six healthy volunteers in a clinical trial (Ramanathan et al., 2007). The major biotransformation pathway was via the hydroxylation at the three position (M40) and the subsequent O-glucuronidation (M13) (Fig. 1A) in those five subjects. DL was poorly metabolized in the sixth subject, presumably because of the polymorphism of CYP2C8 and/or UGT2B10, which were the key enzymes that contributed to the formation of M40 (Kazmi et al., 2015). For the five extensive metabolizers, M13 accounted for 51.0%, 70.2%, and 79.0% of the total circulating radioactivity at 3, 6, and 12 hours, respectively (Table 5). M13 was the predominant peak in urine, accounting for 22.2% of dose, and M40 was the predominant peak in feces, accounting for 28.4% of dose (Ramanathan et al., 2007). On the other hand, M13 and M40 were either not observed or detected at a trace level in plasma, urine, and feces from monkey, rat, and mouse, as shown in Supplemental Fig. 1 (Ramanathan et al., 2006). M13 was thus a disproportionate metabolite in human plasma and urine. M40, the aglycone of M13, was a disproportionate metabolite in human feces. Therefore, the significance of M13 and M40 in humans might not be realized for this drug until the clinical samples were analyzed for metabolites. However, the data from a PXB-mouse experiment could have predicted it before the drug was introduced in the clinic.

A comparison of the plasma metabolic profiles between PXB-mice and human subjects after administration of [¹⁴C]DL is shown in Table 5. M13, the predominant and disproportionate metabolite in human plasma at 3, 6, and 12 hours, was detected at a similar level in 4-hour plasma of PXB-mice. There was a minor difference in metabolic profile between PXB-mouse and human plasma (Fig. 2; Table 5; Ramanathan et al., 2007). DL was more abundant in PXB-mice than in human plasma. Nevertheless, this was not surprising, as the PXB-mouse liver contained a small fraction of mouse hepatocytes, whereas DL in SCID mice contributed to almost twice the circulating radioactivity in PXB-mice. Additionally, the metabolic profiles in urine and feces were similar between human and PXB-mice (Fig. 2; Supplemental Fig. 1). The results suggested that PXB-mice could have accurately predicted, at least semiquantitatively, human in vivo metabolism of DL (Table 5). The PXB-mouse was a much better model for predicting human metabolism than any other preclinical animals evaluated for this drug. Uehara et al. (2020) recently investigated the metabolism of DL by using another chimeric model, TK-NOG mice. A similar conclusion from this mouse model was drawn to that from PXB-mice in comparison with human metabolic profiles in plasma, urine, and feces. Nevertheless, nonradiolabeled DL was dosed in that study. Therefore, only metabolites with authentic standards were quantified, and the level was unknown for unmeasured metabolites. In our study, [¹⁴C]DL was dosed to PXB-mice. All drug-related materials could be captured in radiochromatographic profiles. Additionally, unlike the analysis of biliary excretion of metabolites investigated in our study, a bile sample was not collected in the previous study with TK-NOG mice. In the current study, M13 and M40 were detected as major metabolites in the PXB-mouse bile (Fig. 2E), indicating that 3-hydroxylation and subsequent glucuronidation are the primary metabolism-related clearance pathways in the bile. As M13 was missing in the fecal sample (Fig. 2G) of PXB-mice, it must be hydrolyzed to M40 after bile entered the GI tract (Fig. 1A), i.e., a fraction of the M40 observed in the fecal sample was derived from deconjugation of M13. Similar to the data from PXB-mice, M13 was the predominant peak in human plasma and urine, although it was not detected in human feces (Ramanathan et al., 2007). Considering the similarity of metabolic profiles in plasma, urine, and feces between human and PXB-mice, it is reasonable to believe that biliary metabolic profiles, at least semiquantitatively, were also similar between the two species. Therefore, M13 and M40 represented the biliary excretion metabolic pathway of DL. The proposed biliary excretion of M40 based on the human fecal profile might be overestimated in the clinical trial (Ramanathan et al., 2007). This observation is particularly informative for compounds undergoing significant direct glucuronidation. On the contrary, neither M13 nor M40 was observed in SCID mice, whose metabolic profiles resembled those in CD-1 mice (Supplemental Fig. 1).

Mianserin is a tetracyclic piperazinoazepine antidepressant that is used primarily in the treatment of depression (de Jongh et al., 1981). After oral administration of [¹⁴C]mianserin to healthy volunteers, M4, a quaternary N-glucuronide, was found in human urine, but it was not found in urine and/or feces from rat, mouse, rabbit, and guinea pig (Delbressine et al., 1992). This observation was not surprising, as species difference was observed previously in N-glucuronidation. For example, quaternary glucuronides were preferentially formed in humans and higher primates such as chimpanzees (Chiu and Huskey, 1998). Therefore, M4 was a human disproportionate metabolite at least in urine compared with the preclinical safety species evaluated. After administration of [³H]mianserin to PXB-mice, two quaternary N-glucuronides, M4a and M4b, were detected in plasma, urine, bile, and feces (Table 4). The two metabolites had very similar retention times in the radiochromatogram (Fig. 3), which suggests that they might be diastereomers, as mianserin had a chiral carbon, and the formation of the quaternary N-glucuronide generated another chiral center (Fig. 1B). This fact was consistent with the previous study showing that two diastereomers were isolated for M4 (Delbressine et al., 1992), which suggests that M4 was the combined M4a and M4b. The metabolic profile of mianserin has not been reported in human plasma. Considering M4a and M4b combined was one of the most abundant metabolites in the plasma of PXB-mice, M4 was possibly one of the major circulating metabolites in humans. Interestingly, M28, a quaternary N-glucoside, was detected in urine, bile, and feces of PXB-mice (Fig. 1B; Table 4). This metabolite has never been reported in human ADME studies after administration of radiolabeled mianserin (de Jongh et al., 1981; Delbressine et al., 1992), but the presence of M28 in humans cannot be ruled out. The two human ADME studies reported by de Jongh et al. (1981) and Delbressine et al. (1992) were conducted at a time when the LC-MS was not available or just emerging. Because of the limitation of analytical technology, minor metabolites might be missed during the study. A recent study indicated that M28 was generated after incubation of mianserin in human liver microsomes in the presence of uridine diphosphate glucose (Lu et al., 2018). The formation of M28 was catalyzed by UGT2B10. The data in the PXB-mice from this study and in vitro data from human liver microsomes suggest that M28 might be generated in humans. In PXB-mice, M28 was the predominant component in the bile, but it was a minor metabolite in the feces. Additionally, mianserin was not observed in the bile, but it was one of the most abundant components in the feces (Fig. 3). Therefore, a fraction of M28 was probably hydrolyzed to mianserin by glucuronidase and/or glucosidase in the GI tract (Fig. 1B) (Samara et al., 1990; Flores et al., 2012). This conclusion was corroborated by the fact that the drug was well absorbed in mice, as minimal mianserin was recovered in the feces of SCID mice. In comparison with PXB-mice, none of M4a, M4b, and M28 was observed in SCID mice.

Cyproheptadine is a drug with antihistamine, anticholinergic, and antiserotonergic properties (Porter et al., 1975). After oral administration of [¹⁴C]cyproheptadine to human subjects, M7, a quaternary N-glucuronide, was found to be the major metabolite in human urine (Fig. 1C) (Porter et al., 1975). M7 was also detected at a comparable level in urine from chimpanzees and with a significantly smaller amount in urine from various monkeys (Fischer et al., 1980). In contrast, M7 was not detected in urine from rat, dog, and cat (Hucker et al., 1974). Therefore, M7 was a disproportionate human metabolite relative to rat and dog, which are typical preclinical safety species. In our study, M7 was detected in plasma, urine (data not shown), bile, and feces of PXB-mice after administration of cyproheptadine (Fig. 4). Despite the difference in MS sensitivity for different metabolites, M7 might be the major urinary and biliary metabolite in PXB-mice, as it was the most abundant MS peak in urine and bile. M7 was the second most abundant MS peak of the metabolites in plasma, indicating that it might be one of the major circulating metabolites in humans. Cyproheptadine itself was barely detected in bile and was the most abundant MS peak in feces. Since this drug was well absorbed in animals (Hucker et al., 1974), a fraction of cyproheptadine might be produced from deconjugation of M7 by microbiota in feces (Fig. 1C). In this study, M7 was not detected in SCID mice. The formation of the quaternary N-glucuronide in both human and PXB-mice for both mianserin and cyproheptadine suggests that PXB-mice may be able to accurately predict the quaternary N-glucuronidation in humans and could be a better model than the other preclinical animals in this regard. Oxidative metabolite M19 was eluted after cyproheptadine on the reversed phase LC. It was potentially the N-oxide of cyproheptadine (Fig. 1C), which has been detected in dog and cat urine (Hucker et al., 1974). LC-MS data indicated that M19 was a major metabolite in the plasma of PXB-mice (Fig. 4A), whereas it was a minor metabolite in the same matrix of SCID mice (Fig. 4B), suggesting that it might be a human circulating metabolite. However, the metabolic profile of cyproheptadine has never been reported for human plasma.

Carbazeran was developed as a positive inotropic agent for the treatment of congestive heart failure (Taylor et al., 1981; Kaye et al., 1984). After oral administration of [¹⁴C]carbazeran to healthy volunteers, M21 was detected as the only drug-related component in the plasma (Kaye et al., 1984). In contrast, in the dog plasma, M21 was not observed, whereas unchanged carbazeran accounted for 74% and 61% at 0.5 and 8 hours, respectively. Further investigation revealed that M21 was generated by AO (Kaye et al., 1985). Species variability was observed in AO with high activity in humans, low activity in rats and mice, and no activity in dogs (Kitamura et al., 2006). In our study, as ³H was labeled at the fourth position of the phthalazine moiety, the ³H label was lost during the formation of M21 (Fig. 1D). Therefore, the metabolite was not detected in the radioprofile, and the radiochromatogram is not shown for this compound. However, the LC-MS data show that, in PXB-mice, M21 was the predominant circulating metabolite, whereas carbazeran was a minor component (Fig. 5), which was consistent with the human plasma profile (Kaye et al., 1984). In comparison, M21 was a minor metabolite, whereas carbazeran was the predominant component in the plasma of SCID mice. This result suggests that AO activity in PXB-mice was much higher than SCID mice. Therefore, PXB-mice may be a good animal model for prediction of AO-derived metabolites in humans.

In summary, PXB-mice successfully predicted various human disproportionate metabolites for DL, mianserin, cyproheptadine, and carbazeran compared with preclinical safety species. These metabolites include those formed by cytochrome P450, UGT, and AO. Those metabolites were not detected or observed at trace levels in the control SCID mice. The main difference between PXB- and SCID mice is that the liver of the former is replaced with more than 70% of human hepatocytes. Therefore, the accurate prediction of human metabolites by PXB-mice is primarily due to the humanized liver. In this study, we have not investigated the true quantitative nature of this prediction. However, the data presented here, although limited to only four compounds, together with other published reports (Inoue et al., 2009; De Serres et al., 2011; Sanoh et al., 2012a; Nakada, 2017) indicate that at least a semiquantitative prediction of human metabolism can be obtained in the PXB-mouse model. PXB-mice, therefore, can be a useful model to predict human major and disproportionate circulating and biliary excreted metabolites prior to clinical investigation of an NCE and thus can potentially avoid delay in clinical development of a drug should disproportionate metabolites be identified. One caveat, however, is that the PXB-mouse model may not be able to predict human circulating metabolites if the metabolites are generated from organs other than liver, such as gut or kidney.