Fluoxetine hydrochloride, norfluoxetine (NFLX) and formic acid were purchased from Sigma‒Aldrich (St. Louis, MO, United States). Twenty-seven natural products were obtained from Shanghai Chuangsai Technology Co., Ltd. (Shanghai, China). Paroxetine hydrochloride, as an internal standard (IS), was purchased from Toronto Research Chemicals, Inc. (Toronto, ON, Canada). sf21 insect microsomes expressing wild-type and variant CYP2D6, cytochrome b5 and reductase were obtained from Beijing Hospital (Beijing, China). NADPH was obtained from Roche Pharmaceuticals Ltd. (Basel, Switzerland), and the NADPH regeneration system was obtained from Promega (Madison, WI, United States). Rat liver microsomes (RLMs) were obtained from Corning Life Sciences Co., Ltd. The organic solvents used were purchased from Merck (Darmstadt, Germany).

To research the possible DDIs of fluoxetine, 27 natural products were screened with the incubation assay of RLMs. The reactions were set up as follows. RLMs (0.3 mg/mL), NADPH (1 mM), fluoxetine (10 μM) and one of 27 natural products (100 μM) were mixed in Tris-HCl (100 μM) in a final volume of 200 μL. The concentration of fluoxetine was selected based on the K_m value reported previously (Ring et al., 2001). After preincubation for 5 min at 37 °C, the reaction was started by the addition of the NADPH regeneration system (3.3 mmol/L glucose 6-phosphate, 3.3 mmol/L MgCl₂, 1.3 mmol/L NADP⁺ and 0.4 units/mL glucose-6-phosphate dehydrogenase) and stopped after 30 min with the addition of precooled acetonitrile (400 μL) containing IS (0.5 μg/mL). After vortexing for 2 min, the mixture was centrifuged at 13,000 rpm for 10 min. The supernatant (100 μL) was mixed with 200 μL of ddH₂O, and an aliquot of the mixture (3 μL) was analyzed by ultraperformance liquid chromatography–tandem mass spectrometry (UPLC‒MS/MS). The relative enzymatic activity was measured after the formation of norfluoxetine.

According to the screening results, the effect of miltirone on the metabolic parameters of fluoxetine in RLMs was studied. Different concentrations of miltirone (0, 0.01, 0.1, 1, 10, 25, 50 and 100 μM) were mixed into the reaction system as mentioned above. Then, these reactions were started and stopped following the abovementioned steps, and the products were analyzed by UPLC‒MS/MS.

To investigate whether miltirone has the potential for causing time-dependent CYP inhibition, IC₅₀ shift experiments were further carried out (Bao et al., 2018; Dong et al., 2021; Wang et al., 2020). RLMs (0.3 mg/mL) and different concentrations of miltirone were mixed and preincubated with or without 1 mM NADPH in Tris-HCl (100 μM) in a final volume of 200 μL at 37 °C for 30 min to identify weak inactivation. Then, fluoxetine (10 µM) was added to the mixture for another 30 min of incubation, and the supernatant was analyzed by UPLC‒MS/MS.

Molecular docking analysis was performed to confirm the drug binding conformations of miltirone to human CYP2C9 (PDB code 1R9O), CYP2C19 (PDB code 4KFO) and CYP2D6 (PDB code 4WNW). The ligands and proteins needed for molecular docking were prepared and docked with the software programs AutoDock Vina and PyRx software programs. Discovery Studio 2020 Client was used for visual analysis.

Sprague‒Dawley rats (male, 200 ± 10 g) were obtained from the Experimental Animal Research Center of Wenzhou Medical University (Wenzhou, China) and housed at 25 °C under a natural light-dark cycle. The animal experiments were approved by the Animal Ethics Committee of Wenzhou Medical University (wydw 2023-0461). All rats were randomly divided into 2 groups (6 in each group): the control group and the miltirone (coadministration) group. Miltirone and fluoxetine were dissolved in 5% Tween and vegetable oil, respectively. The rats in the miltirone group were orally administered miltirone (40 mg/kg), while those in the control group were given 5% Tween in the same manner for comparison. After 0.5 h, fluoxetine (8 mg/kg) was administered to all rats by gavage. Blood samples (approximately 300 μL) were collected via the caudal vein and placed in heparinized 1.5 mL tubes at 0.5, 1, 2, 3, 4, 6, 8, 12, 24 and 48 h after fluoxetine administration. Adequate amounts of urine were obtained by directly catching rats to collect urine at 12, 24 and 48 h. These samples were centrifuged (10,000 × g) immediately for 10 min after collection. Then, 100 μL of each plasma or urine sample was stored at −80 °C before analysis.

Wild-type CYP2D6 and 24 CYP2D6 variants (2 common variants and 22 defective allelic variants) were used for the enzymatic studies. The incubation procedures (60 min) were as follows. First, equal amounts of the P450 from insect microsomes (5–20 pmol), cytochrome b5 (5–20 pmol) and fluoxetine (10–300 μM) were mixed in potassium phosphate buffer in a final volume of 200 μL (100 mM, pH 7.4). These reactions were then started and stopped following the abovementioned steps. The supernatants were analyzed by UPLC‒MS/MS. The experiments were performed in triplicate.

The levels of fluoxetine, norfluoxetine and paroxetine were determined by a Waters ACQUITY UPLC‒MS/MS system (Waters, Milford, MA, United States) according to our previous articles (Ji et al., 2015; Fang et al., 2017).

The Michaelis‒Menten equation or the substrate inhibition equation was used to analyze and obtain the kinetic parameters (K_m and V_max) after incubation, with calculations performed in GraphPad Prism 5 software (San Diego, CA, United States). One-way analysis was used to analyze the in vitro variances. Noncompartmental analysis was used for the analyte pharmacokinetic parameters of analytes, including the area under the time-concentration curve (AUC), half-life (T_1/2), the maximum of blood concentration (C_max), blood clearance rate (CL), apparent volume of distribution (V) and mean retention time (MRT), in vivo with DAS 3.0 software. The mean values in the miltirone and control groups were analyzed by t tests.

The inhibitory effects of 27 natural products on RLMs are summarized in Figures 1A, B. These products exhibited different inhibitory effects on CYP activity in vitro. Among these compounds, miltirone had the greatest inhibitory effect. Moreover, our results showed that the CYP inhibitions by miltirone was greater than that of tanshinone IIA, which was consistent with the results of a previous study (Zhou et al., 2013). The miltirone inhibition of RLMs data are summarized in Figure 1C, which gave an IC₅₀ value of 2.9 μM. The results of the IC₅₀ shift assays in Figure 1D show that the IC₅₀ values were 26.12 and 14.24 μM after preincubation with and without NADPH, respectively.

To confirm the ligand binding conformations of miltirone in the active sites of certain human CYP450 isoforms, a molecular docking study was performed. The free binding energies indicated that miltirone had stronger binding affinity than did fluoxetine (Table 1). As shown in Figure 2A, molecular docking analysis revealed that three key residues (ILE50, PHE69 and PHE212) were critical for the binding of miltirone and fluoxetine. Rings B and C of miltirone have strong pi-alkyl interactions with ILE50, PHE 69 and PHE 212. PHE 69 and PHE 212 also interact with fluoxetine through pi-pi T-shaped interactions. Moreover, ILE50 participated in a sigma–pi interaction with fluoxetine (Figure 3A). These interactions might be responsible for the inhibitory effect of miltirone on CYP2C9-mediated fluoxetine N-demethylation. As shown in Figure 2B, miltirone interacts with the side chains of ALA264 and CYS400 via alkyl interactions. Furthermore, miltirone also binds to CYS400 via pi–donor hydrogen bonding or pi–sulfur interactions. However, fluoxetine participates in an alkyl–pi interaction and/or conventional hydrogen bond with ALA264 and CYS400 (Figure 3B). These results suggested that miltirone could bind to the fluoxetine binding site on human CYP2C9. As shown in Figure 2C and Figure 3C, miltirone and fluoxetine do not share the same binding site in the crystal structure of CYP2D6. However, a previous article reported that the type of interaction between miltirone and CYP2D6 is a pi–pi interaction with PHE120 in another crystal structure of CYP2D6 (PDB code 3QM4), which is located in the fluoxetine binding site in our docking study (Zhou et al., 2013). Thus, PHE120 may be a key residue leading to the inhibition of CYP2D6-mediated fluoxetine N-demethylation by miltirone.

As shown in Figure 4, the plasma concentration of fluoxetine decreased with miltirone exposure after comedication. The AUC_(0–t), T_1/2z and CLz of fluoxetine, which are summarized in Table 2, were significantly different between the control and miltirone groups. The CLz of fluoxetine in the coadministration group decreased by 69.6% (p < 0.05) compared with that in the control group. Moreover, the miltirone group exhibited significantly greater AUC_(0–t) (2.6-fold) and T_1/2z (2.4- fold) values. In addition, the C_max of fluoxetine increased by 96.1% (p < 0.05) in the miltirone group. As shown in Table 3, there were no significant differences in the pharmacokinetic parameters of norfluoxetine between the two groups (p > 0.05). However, the levels of norfluoxetine in the urine at different times after miltirone administration were lower than those in the control group, which was different from the findings for fluoxetine and indicated the potential for excretion-related DDIs when fluoxetine was coadministered with miltirone (Figures 4C, D).

The Michaelis‒Menten curves are shown in Figure 5, and the corresponding K_m and V_max values are summarized in Table 4. As illustrated in Table 4, all the variants exhibited significantly reduced V_max values, and almost all exhibited significantly changed K_m values compared with those of wild-type protein. Therefore, the intrinsic clearance (V_max/K_m) values for fluoxetine demethylation were significantly altered in more than half of the tested allelic variants. CYP2D6*92 and *96 were inactive in terms of fluoxetine metabolism in this study and were thus classified as null alleles. The other 22 variants were classified into three types according to their relative clearance values compared to wild-type: the first type included 10 variants (CYP2D6*88/V104A, *90/K147R, *91/C161S, *94/D337G, *95/R388H, *98/H463D, *V327M, *F219S, *F164L and *D336N) that showed similar intrinsic clearance; 7 variants (CYP2D6*2, *93/T249P, *10, *87/A5V, *E215K, *R25Q and *R440C) exhibited significantly reduced V_max/K_m values (6.62%–66.79%, p < 0.05) and were classified into the second type; and the remaining variants (CYP2D6*89/L142S, *R497, *97/F457L, *V342M and *R344Q) exhibited increased V_max/K_m values of 578.83%, 212.57%, 201.62%, 188.50% and 125.07%, respectively, compared with wild type (p < 0.05).

As shown in Table 5, to understand the differences in the metabolism of diverse antidepressants by reported CYP2D6 variants, we compared the relative clearance values of venlafaxine, citalopram, olanzapine, and fluvoxamine reported with those of fluoxetine in this study (Hu et al., 2016; Ye et al., 2022; Zhan et al., 2016; Zhou et al., 2016).