Decoding the selective chemical modulation of CYP3A4

Abstract

Drug-drug interactions associate with concurrent uses of multiple medications. Cytochrome P450 (CYP) 3A4 metabolizes a large portion of marketed drugs. To maintain the efficacy of drugs metabolized by CYP3A4, pan-CYP3A inhibitors such as ritonavir are often co-administered. Although selective CYP3A4 inhibitors have greater therapeutic benefits as they avoid inhibiting unintended CYPs and undesirable clinical consequences, the high homology between CYP3A4 and CYP3A5 has hampered the development of such selective inhibitors. Here, we report a series of selective CYP3A4 inhibitors with scaffolds identified by high-throughput screening. Structural, functional, and computational analyses reveal that the differential C-terminal loop conformations and two distinct ligand binding surfaces disfavor the binding of selective CYP3A4 inhibitors to CYP3A5. Structure-guided design of compounds validates the model and yields analogs that are selective for CYP3A4 versus other major CYPs. These findings demonstrate the feasibility to selectively inhibit CYP3A4 and provide guidance for designing better CYP3A4 selective inhibitors.

To maintain the efficacy of drugs metabolized by CYP3A4, pan-CYP3A inhibitor is often co-administered, but the high homology between CYP3A4 and CYP3A5 has hampered the development of selective CYP3A4 inhibitors. Here, the authors report a series of selective CYP3A4 inhibitors and show that differential C-terminal loop conformations and two distinct ligand binding surfaces disfavour the binding of selective CYP3A4 inhibitors to CYP3A5.

Introduction

An ever-increasing number of people are subjected to polypharmacy¹, whereby multiple medications are taken concurrently to treat diverse ailments and health conditions. As a result, drug-drug interactions raise major clinical concerns. Cytochrome P450 (CYP) enzymes, particularly CYP3A4 and CYP3A5 are notably responsible for the metabolism of most marketed drugs, resulting in reduced or loss of drug efficacy^2,3. In addition, metabolites generated by CYPs may be toxic⁴. The variability in CYP expression among different populations makes it hard to control drug dosage, and some drugs, such as tacrolimus, may even require personalized dosage adjustment based on the patient’s genotype⁵. Accordingly, the U.S. Food and Drug Administration (FDA) has issued guidelines to help drug developers better evaluate interactions between drug candidates and CYPs^6,7.

Because of the critical role of CYP3A4 and CYP3A5 in drug metabolism and disposition, their inactivation can enhance the therapeutic efficacy of drugs by slowing down their metabolism. Some CYP3A-metabolized drugs, such as darunavir and lopinavir, are often co-administered with a non-selective pan-CYP3A inhibitor such as ritonavir to maintain their effective plasma concentrations^8–10. A recent example of this is the anti-COVID drug Paxlovid^TM, which uses ritonavir to decrease the metabolism of its active ingredient, nirmatrelvir, mainly by CYP3A4 in the liver^11,12. However, CYP3A4 and CYP3A5 are not functionally identical and have their own metabolic preferences for drug clearance. A study of 32 drugs known to be largely metabolized by CYP3A showed that at least 17 drugs are predominantly metabolized by CYP3A4, and 5 drugs are metabolized more by CYP3A5¹³. Therefore, pan-CYP3A inhibitors may lead to unnecessary inhibition of all CYP3A family members, resulting in dangerously elevated plasma levels of drugs highly dependent on those unintended CYP3A enzymes for clearance and life-threatening drug reactions. For instance, the co-administration of ritonavir, which aims to slow the metabolism of the active ingredient nirmatrelvir, can increase the risk associated with the immunosuppressive drug tacrolimus in kidney transplant recipients, as tacrolimus metabolism is largely influenced by CYP3A5 activity^14,15. Similarly, co-administration of ritonavir and vincristine can increase the neurotoxicity of vincristine because the latter is largely metabolized by CYP3A5^16–18. In addition, a significant portion of the global population expresses active CYP3A5¹⁹. Consequently, co-administering a selective CYP3A4 inhibitor rather than a pan-CYP3A inhibitor would provide much more needed benefits in maintaining a safe concentration range for drugs largely metabolized by CYP3A5 such as tacrolimus and vincristine.

The development of selective CYP3A4 inhibitors has been hampered by three main challenges: 1) The high homology between CYP3A4 and CYP3A5 (83% sequence identity and high structural similarity)²⁰ makes it difficult to identify structural features unique to CYP3A4 that can be exploited for selectivity, particularly in the absence of a molecular mechanism of differential selectivity between CYP3A4 and CYP3A5; 2) The prominent ligand promiscuity characteristic of the CYPs as a result of their large and flexible ligand binding pockets that can accommodate a variety of ligands in multiple orientations²¹; 3) Certain ligands are activated by CYPs to form reactive metabolites that permanently and irreversibly inactivate CYPs (suicide inhibitors) in a process known as mechanism-based inhibition, which can result in toxicity due to unintended lower drug clearance and prolonged drug exposure²².

In the current work, we identified non-suicide selective CYP3A4 inhibitors with distinct chemical scaffolds from an unbiased high-throughput screening (HTS) campaign and subsequent chemical analog design. Importantly, we uncovered the basis of such selectivity based on crystallographic studies of 9 CYP3A4–ligand complexes, one CYP3A5-ligand complex, functional studies of a series of chemical derivatives and computational docking analysis. The differential positioning of the C-terminal loop between CYP3A4 and CYP3A5 would prevent these compounds from binding to CYP3A5 but not to CYP3A4 in a favorable conformation, thus leading to selective inhibition. A unique surface within the ligand binding pocket of CYP3A4 was identified as essential for the preferential binding of these inhibitors, while analogs with reduced selectivity can occupy a secondary surface with a less favorable chemical environment that is present in both orthologs. These findings open additional venues in the design of selective CYP3A4 modulators with vast clinical implications.

Results

Identification of selective CYP3A4 inhibitors

To identify selective CYP3A4 inhibitors of different chemotype, we conducted an unbiased HTS campaign for CYP3A4 inhibitors using the P450-Glo luminescence-based biochemical assay at a fixed compound concentration of 10 µM (Fig. 1a and Supplementary Table 1). The diverse collection of 9299 compounds consisted of chemicals with drug-like physicochemical properties and β-turn peptidomimetics. The screening assay displayed a clear separation between positive and negative controls with adequate robustness, as reflected by the average Z′ factor of 0.85 (Supplementary Fig. 1a)²³. Based on a cutoff of ≥80% inhibition, 956 compounds were selected for dose–response analysis against CYP3A4 and CYP3A5, using the same P450-Glo assay as described above. A total of 41 commercially available compounds with high potency against CYP3A4 (IC₅₀ < 0.1 µM and maximum inhibition > 50%) and high selectivity (CYP3A5/CYP3A4 IC₅₀ ratio > 100) in the dose–response analysis were selected and further validated using powders re-purchased from vendors (Supplementary Table 2). These results showed that most compounds remained selective for CYP3A4, although they displayed lower CYP3A4 selectivity when compared with the screening results, probably due to the degradation or oxidation of the screening compounds during long-term storage.

Fig. 1. Primary screen against CYP3A4.

a Scatter plot of percent inhibition determined using the biochemical P450-Glo assay. Data were normalized to 15 µM ritonavir as positive controls (100% inhibition, green) and DMSO as negative controls (0% inhibition, orange). Screened compounds are in blue. b Clustering by structural similarities of the 956 compounds selected for dose–response studies against CYP3A4 and CYP3A5, with the three lead selective CYP3A4 inhibitors indicated as blue, green and pink. The positional spread of the three compounds in the circular clustering tree indicates variety in chemical scaffold. c Chemical structures of the selective CYP3A4 inhibitors SCM-01, SCM-02, and SCM-03, respectively. The colors represent their labeled locations in (b). Source data for (a) are provided as a Source Data file.

The hierarchical clustering of the screened compounds, presented as a circular cladogram in Fig. 1b, shows the compounds grouped by structural similarity. Three CYP3A4-selective compounds, SCM-01, SCM-02, and SCM-03 (Fig. 1b, c, shown in blue, green, and pink, respectively) featured in this work are structurally different from each other and provide unique chemical scaffolds based on their positional separation within the cladogram. Most of the compounds screened and the hits obtained from the primary screen have physicochemical properties (AlogP and molecular weight) within the ranges characteristic of drug-like compounds (Supplementary Fig. 1b)²⁴. No common chemical scaffold that correlated with high CYP3A4 potency or selectivity was observed. This is not surprising, because CYP3A4/5 have large and promiscuous binding pockets that can accommodate structurally diverse compounds^21,25. The lack of a common scaffold for CYP3A4 selectivity also justifies the use of an unbiased HTS approach, as it is not feasible to perform structure-based design without a fixed scaffold.

The results of P450-Glo inhibition assays confirmed that SCM-01, SCM-02, and SCM-03 selectively inhibited CYP3A4 in the biochemical setting (Fig. 2a–c and Supplementary Table 3) and in the cellular context (Supplementary Fig. 2). Binding of SCM-01, SCM-02, and SCM-03 to CYP3A4 resulted in spectral shifts toward higher wavelengths, indicating that these compounds bind and inhibit CYP3A4 as type II inhibitors that coordinated with the heme iron (Supplementary Fig. 3)²⁶. SCM-01 and SCM-02 also induced type II spectral shifts when bound to CYP3A5, but their binding to CYP3A5 was weaker than that to CYP3A4, as indicated by their spectral binding constants (K_s) (Supplementary Fig. 3 and Supplementary Table 4). As SCM-01 and SCM-02 have only one heme coordinating group, the UV-vis results suggest that both compounds bind to CYP3A5 in a pose similar to that of CYP3A4 with less favorability. By comparison, binding of SCM-03 to CYP3A5 resulted in a type I spectral shift toward lower wavelengths, indicating the absence of heme coordination and a CYP3A5 binding conformation very different from that in CYP3A4 (Supplementary Fig. 3). Finally, all three compounds showed higher binding affinity to CYP3A4 than to CYP3A5. Although the K_s and IC₅₀ values are not strictly proportional, which may be caused by the atypical kinetics of CYP3A4²⁷ and allosteric communication with the cytochrome P450 oxidoreductase²⁸, these results indicate that selective CYP3A4 inhibition is a result of preferential inhibitor binding to CYP3A4. Together, these data demonstrated that SCM-01, SCM-02, and SCM-03 are selective inhibitors of CYP3A4.

Fig. 2. The C-terminal loop plays an important role in selective CYP3A4 inhibition.

a–c Concentration-dependent inhibition of CYP3A4 and CYP3A5 for (a) SCM-01, (b) SCM-02, and (c) SCM-03, respectively. Percentage inhibition data were normalized against their DMSO controls (as 0% inhibition) and against 30 µM ketoconazole controls (as 100% inhibition). The reported numerical values are the means and standard deviations of the IC₅₀ values derived from triplicate experiments (n = 3). d–f Crystal structures of CYP3A4 in complex with (d) SCM-01, (e) SCM-02, and (f) SCM-03 highlighting the interactions between ligands and adjacent residues. g Superimposition of the three selective inhibitor–bound CYP3A4 crystal structures (green, blue, and pink) and all five published CYP3A5 structures (yellow), showing that the selective inhibitors would clash with the CYP3A5 C-terminal loop (circled region, purple) if bound to CYP3A5 in the same conformation as CYP3A4. h Representative CYP3A5-apo crystal structure showing H-bonds associated with the C-terminal loop residues. i Representative CYP3A4–SCM-01 crystal structure showing that the D214-G481 is the only H-bond associated with the C-terminal loop in CYP3A4. The inset shows the sequence alignment of C-terminal loop residues in CYP3A4 and CYP3A5. The heme groups in all panels are colored in wheat. Source data for (a–c) are provided as a Source Data file.

SCM-01 is a non-suicide selective CYP3A4 inhibitor

A major disadvantage of CYP3A4 irreversible inhibitors is that they permanently incapacitate CYP3A4’s ability to dispose of drugs that can be toxic due to prolonged exposure²². In contrast, those that are not CYP3A4 irreversible inhibitors are desirable because they can be more easily employed within a particular timeframe to increase plasma concentrations of drugs that can be metabolized quickly by CYP3A4. Thus, these types of inhibitors are more amenable in the design of controlled dosage regimens.

An irreversible inhibitor is usually generated from the CYP-mediated chemical modification of a drug into a reactive metabolite, which permanently inactivates the enzyme. This class of inhibitors is known as suicide or mechanism-based inhibitors²⁹. An assay commonly used for determining mechanism-based inhibitors is the time-dependent inhibition assay, in which the test compound is pre-incubated with the enzyme in the presence or absence of the cofactor NADPH before adding the substrate^30,31. In this assay, mechanism-based inhibitors, such as CYP3cide, are catalyzed by CYPs and display a different IC₅₀ value upon pre-incubation, whereas non–mechanism-based inhibitors, such as ketoconazole, are not metabolized and maintain the same IC₅₀ value^32,33.

The results of the time-dependent inhibition assay indicated that SCM-01, similar to ketoconazole, is not a mechanism-based inhibitor because it did not undergo IC₅₀ changes upon pre-incubation with CYP3A4/5 and NADPH (Supplementary Fig. 4 and Supplementary Table 5). Pre-incubating SCM-02 with CYP3A4 and SCM-03 with CYP3A5 also had little effect on the respective inhibitory potencies of these compounds. However, SCM-02 and SCM-03 exhibited IC₅₀ shifts when pre-incubated with CYP3A5 and CYP3A4, respectively, indicating that the two compounds were metabolized (Supplementary Fig. 4 and Supplementary Table 5). Consequently, SCM-01 was highlighted as the ideal candidate for further development and to investigate the mechanism of selective CYP3A4 inhibition because it acted as a non-suicide inhibitor of both CYP3A4/5. Nevertheless, it should be noted that the parent compounds, SCM-02 and SCM-03, are more selective for CYP3A4 than are their metabolites; therefore, their crystal structures are still valuable in elucidating the mechanism of selective CYP3A4 inhibition.

CYP3A4/5 C-terminal loop differences dictate selective inhibition

To investigate the mechanism of selective CYP3A4 inhibition, we solved the crystal structures of CYP3A4 in complex with each of the three compounds, SCM-01, SCM-02, and SCM-03, at resolutions of 2.7 Å, 3.1 Å, and 3.2 Å, respectively (Fig. 2d–f, Supplementary Table 6). The three crystal structures have the same overall folding, and the conformations of compounds are well defined by electron density maps (Supplementary Fig. 5). All three compounds formed coordination with the heme iron, which is consistent with the results obtained from UV-vis measurements.

In addition to heme coordination, structural analysis indicates that the binding of SCM-01, SCM-02, and SCM-03 to CYP3A4 is primarily stabilized by hydrophobic interactions. In the CYP3A4–SCM-01 structure, the trifluoromethyl moiety forms multiple hydrophobic interactions with a cluster of phenylalanine residues consisting of F108, F213, F215, F241, and F304 (Fig. 2d). These residues form the ceiling of the ligand-binding pocket and stabilize the ligand binding in the observed conformation. The hydrophobic interactions associated with residue I369 and I120 may have also contributed to compound binding in this conformation (Fig. 2d). The binding of SCM-02 to CYP3A4 is also mostly driven by hydrophobic interactions. The methyl thiazole moiety of the compound is embedded in a hydrophobic cavity formed by residues F108, V111, M114, I120, F241, and I301 (Fig. 2e). Additionally, SCM-02 binding is stabilized by the H-bond between the S119 side chain and the methyl thiazole moiety, as well as hydrophobic interaction between residue F304 and the backbone of SCM-02. SCM-03 is a mixture of R/S stereoisomers; however, only the S-configuration fits well according to the electron density map, suggesting that this isomer is primarily responsible for CYP3A4 binding and inhibition. The crystal structure shows that SCM-03 binds CYP3A4 in a compact, donut-shaped conformation, probably caused by the flexibility of the compound and the π–π interaction between the chlorophenyl moiety and the pyridine (Fig. 2f). The hydrophobic cluster formed by F108, F213, F215, and F304 may also contribute to ligand binding in the observed conformation.

Superimposition of the three selective inhibitor–bound CYP3A4 structures on all five published CYP3A5 crystal structures (PDB: 7LAD, 6MJM, 8SG5, 5VEU, and 7SV2) allowed us to explore the structural basis of preferential inhibitor binding to CYP3A4^{20,25,34–36}. Comparison of the CYP3A4/5 ligand binding pockets revealed structural differences in two regions, the F–F′ loop and the C-terminal loop, that may contribute to selective CYP3A4 inhibition (Supplementary Fig. 6a). Our previous study indicated that the F–F′ loop plays an important role in the selective inhibition of CYP3A5, because the selective inhibitor induces the loop to adopt a conformation that is favorable for ligand binding³⁴. However, the binding of SCM-01 and SCM-03 to CYP3A4 does not induce any structural changes to the F–F′ loop that favors ligand binding. On the contrary, SCM-02 pushes the F–F′ loop further away from the ligand-binding pocket and leads to partial disorder of the loop, thus preventing interactions between SCM-02 and the F–F′ loop (Supplementary Fig. 6b). Therefore, the F–F′ loop is unlikely to play a major role in selective CYP3A4 inhibition.

Importantly, the major CYP3A4/5 structural difference that may contribute to differential ligand binding is the orientation of the C-terminal loop consisting of residues 476–486. This loop is structurally conserved within selective inhibitor–bound CYP3A4 structures and within all CYP3A5 structures (Supplementary Fig. 6c). Structural alignment indicated that the conformation of the C-terminal loop in CYP3A5 is 2–3 Å closer to the heme group than that in CYP3A4, which creates a narrower ligand-binding pocket for CYP3A5. This would prevent selective CYP3A4 inhibitors from binding to CYP3A5 in the CYP3A4 conformation, as all three compounds would clash with residue L481 on the C-terminal loop (Fig. 2g). Consequently, these compounds may have either bound to CYP3A5 in alternative and less favorable poses (such as in the case of SCM-03) or altered the conformation of the C-terminal loop to create a larger ligand binding pocket, both of which generate energetic penalty for ligand binding.

Further structural analysis revealed that the C-terminal loop in CYP3A5 is highly rigid and resistant to ligand-induced conformational change. Analysis of the CYP3A5 crystal structures revealed that the conformation of the C-terminal loop is fixed by several H-bonds, including a backbone H-bond, V369–L482, and three side-chain H-bonds, W58–D477, D477–Q483, and Q483–E308 (Fig. 2h). These bonds are present in other CYP3A5 structures and result in a conserved and rigid C-terminal loop conformation (Supplementary Fig. 6d). This observation suggests that the steric clash mentioned above cannot be mitigated by altering the conformation of the C-terminal loop. In contrast, none of these H-bonds are present in CYP3A4-apo or in selective inhibitor–bound structures due to sequence variations (Fig. 2i). Instead, the conformation of the C-terminal loop in CYP3A4 is stabilized by a single H-bond between D214 and G481, which pulls the loop away from the heme group and creates a wider ligand binding space for CYP3A4. Taken together, these studies suggest that sequence differences between CYP3A4 and CYP3A5 lead to different conformations of the C-terminal loop and ultimately to selective chemical inhibition of CYP3A4.

SCM-01 analogs support the critical role of the C-terminal loop

To understand the selective CYP3A4 inhibition by SCM-01, we tested a series of chemical analogs selected by substituting the tert-butyl (region A) group with moieties of different sizes, ranging from small isobutyl to large biphenyl (Fig. 3a, b and Supplementary Fig. 7). Since the region A group is responsible for the clash with the CYP3A5 C-terminal loop, we hypothesize that altering it would have a large impact on CYP3A5 inhibition and consequently CYP3A4 selectivity.

Fig. 3. Structure-activity relationship analysis of SCM-01 analogs indicates that the region A group is important for selective CYP3A4 inhibition.

a The chemical scaffold of SCM-01 analogs. The R group represents the region A group to be modified. b Dose–response curves showing concentration-dependent inhibition of CYP3A4 and CYP3A5 by SCM-01 and its analogs. The structure of the region A group of each compound is shown in the inset. Percentage inhibition data were normalized against their DMSO controls (as 0% inhibition) and 30 µM ketoconazole controls (as 100% inhibition). The reported numerical values are the means and standard deviations of the IC₅₀ values derived from triplicate experiments (n = 3). c Superimposition of CYP3A4–SCM-01 (gray) and CYP3A4–SCM-08 (green) structures showing that SCM-08 binds to CYP3A4 the same as SCM-01, with the region A cyclobutane group residing on surface 1. Source data for (b) are provided as a Source Data file.

Consistent with our hypothesis, SCM-01 analog–mediated CYP3A5 inhibition is sensitive to the size of the region A group. Substituting the region A tert-butyl group with similarly sized moieties, such as the isobutyl group (SCM-04), had no effect on both CYP3A4 and CYP3A5 inhibition (Supplementary Fig. 7). By contrast, larger hydrophobic region A groups, such as cyclopropane (SCM-07), cyclobutane (SCM-08), thiophane (SCM-09, log K_ow = 1.81³⁷), cyclohexane (SCM-10), and chlorophenyl (SCM-11), impaired CYP3A5 but not CYP3A4 inhibition, thereby increasing the selectivity for CYP3A4 (Fig. 3b). Notably, SCM-10 almost completely failed to inhibit CYP3A5, making it one of the most selective CYP3A4 inhibitors we have synthesized.

The crystallized SCM-08 in complex with CYP3A4 shows that the inhibitor binds to CYP3A4 the same as SCM-01 (Fig. 3c, Supplementary Fig. 8), suggesting that modifications in region A do not affect the overall ligand binding conformation. The cyclobutane group in region A of SCM-08 resides in a hydrophobic concave surface formed by residues I369, A370, M371, L482, and L483 (termed surface 1) in the same manner as the tert-butyl group of SCM-01 (Fig. 3c). However, surface 1 can only accommodate substitutions of limited size, because extension of the region A group, as observed with SCM-12, SCM-13, SCM-14, SCM-15, reduces the inhibition of CYP3A4 to different degrees (Fig. 3b). As a result, linear extension of the region A group must be carried out with caution to avoid unwanted loss of CYP3A4 inhibition, although some of these large substitutions resulted in weaker and sometimes complete loss of CYP3A5 inhibition. Together, these results highlight the critical role of region A in selective CYP3A4 inhibition and indicate that relatively larger region A groups tend to be more favorable for selective CYP3A4 inhibition.

We also noticed that SCM-16 and SCM-17, which have relatively large region A groups, retained the inhibitory effect of SCM-01 on CYP3A5 while showing weaker CYP3A4 inhibition (Fig. 3b). Compared to other SCM-01 analogs, SCM-16 and SCM-17 have hydrophilic region A groups that may not tolerate the mostly hydrophobic surface 1. Therefore, these results indicate that hydrophobicity of the region A group is as important as the substitution size for selective CYP3A4 inhibition. Relatively large and hydrophobic region A groups are favorable for CYP3A4 but not CYP3A5 inhibition, whereas hydrophilic region A groups would offset the effect of large region A groups and lead to the loss of CYP3A4 selectivity.

Most of SCM-01 analogs are non-suicide CYP3A4/5 inhibitors as they did not demonstrate time-dependent IC₅₀ shift upon pre-incubation (Supplementary Fig. 9 and Supplementary Table 5). However, SCM-09, SCM-10, and SCM-14 are mechanism-based CYP3A4 and/or CYP3A5 inhibitors. The highly selective CYP3A4 analog SCM-10 gained potent CYP3A5 inhibition after pre-incubation with CYP3A5 and NADPH, and the inhibitory potencies of SCM-09 and SCM-14 were also enhanced when they were pre-incubated with either of the enzymes. It is likely that the cyclohexane moiety of SCM-10 was metabolized to hydrophilic cyclohexanol that favors CYP3A5 binding and inhibition³⁸, whereas enhanced CYP inhibition by SCM-09 may be caused by the metabolism of the thiophene moiety and subsequent covalent modification of the enzyme³⁹. The mechanism-based inhibition by these compounds, unfortunately, reduced their selectivity for CYP3A4 (Supplementary Fig. 9 and Supplementary Table 5). This also makes these compounds not ideal for mechanistic studies and further drug development.

Selective inhibitors interact with a less favorable surface in CYP3A5

Comparative analysis of our crystal structures of CYP3A4 in complex with SCM-01, SCM-02, and SCM-03 in conjunction with reported CYP3A5 structures provided us with a general explanation for CYP3A4 selectivity: SCM-01 analogs that can potently inhibit CYP3A4 contain substitutions in region A that can interact with surface 1 in CYP3A4 (Fig. 4a), but they clash with a wall formed by the C-terminal loop position in CYP3A5 (which makes surface 1 almost non-existent in CYP3A5) (Fig. 4b). To further understand the lingering CYP3A5 inhibition by these analogs with the expectation to improve selectivity towards CYP3A4, we conducted molecular docking studies. Docking of SCM-01 to CYP3A4 showed that the inhibitor binds to CYP3A4 in two heme coordinating poses (Fig. 4a, c). A primary ligand pose (pose 1) is similar to that observed in the crystal structure and validates the docking result (Supplementary Fig. 10a). The CYP3A4–SCM-01 docking pose 1 shows that the region A group points towards the mostly hydrophobic surface 1 in CYP3A4 (Supplementary Fig. 10b) but would clash with CYP3A5 if bound in pose 1 (Supplementary Fig. 10c). A second pose (pose 2) was identified, with the region A group rotating about 60° relative to that of pose 1 and interacting with an alternative surface formed by residues R105, A370, R372, L373, E374, and R440 (surface 2) whose environment and dimensions are different from that of surface 1 (Fig. 4c and Supplementary Fig. 10d). This surface has also been reported to play an important role in CYP3A4 binding by ritonavir and its analogs^40–42. In the case of SCM-01, the tert-butyl group interacts more favorably with the mostly hydrophobic surface 1 than with the highly polar surface 2, which is predominantly formed by charged residues and a carboxylic acid group of the heme (Supplementary Fig. 10b, d). Docking of SCM-01 to CYP3A5 also identified pose 2, with the region A group interacting with surface 2 (Fig. 4d and Supplementary Fig. 10e), suggesting that it is the most likely conformation for SCM-01 to bind to CYP3A5. Notably, this surface is relatively conserved between CYP3A4 and CYP3A5 (Supplementary Fig. 10d, e), suggesting surface 2 in both CYP3A4/5 is unfavorable for binding hydrophobic compounds, especially analogs with relatively large hydrophobic region A groups.

Fig. 4. The identification of two distinct ligand binding surfaces of CYP3A4/5 by docking studies reveals the mechanism of selective CYP3A4 inhibition.

a Illustration of the favorable CYP3A4–SCM-01 docking pose 1 and its corresponding surface 1. b Illustration of the clash between SCM-1 and CYP3A5 (circled region) if SCM-01 were to bind to CYP3A5 the same way as CYP3A4. c Illustration of the less favorable CYP3A4–SCM-01 docking pose 2 and its corresponding surface 2. d Illustration of the CYP3A5–SCM-01 docking pose 2 and its corresponding surface 2. For stick representations of SCM-01 in all panels, carbon atoms are colored in yellow, fluoride atoms colored in cyan, nitrogen atoms colored in blue, and oxygen atoms colored in red. The heme group is colored in wheat.

The fact that an inhibitor can occupy surface 2 in CYP3A5 as an alternative region to compensate for the absence of surface 1 that is present in CYP3A4 enhances our understanding of why the size and hydrophobicity of the region A group are both critical for selective inhibition (Fig. 4 and Supplementary Fig. 10). First, SCM-01 analogs with hydrophobic region A groups bind to CYP3A4 more favorably to the mostly hydrophobic surface 1, whereas their binding to CYP3A5 is limited to the less favorable pose 2 due to the clash with the C-terminal loop. As a result, these compounds bind to CYP3A4 with higher affinity and demonstrate selectivity for CYP3A4. Second, SCM-01 analogs with larger hydrophobic region A groups, such as cyclohexane in SCM-10 and chlorophenyl in SCM-11, bind to CYP3A5 even less favorably than SCM-01 due to the proximity of their region A groups to the highly polar surface 2. In contrast, analogs that contain region A groups with increasing polarity, such as SCM-16 and SCM-17, can tolerate the polar environment of surface 2. Finally, while surface 1 in CYP3A4 can accommodate various region A groups, excessively large groups may still clash with the C-terminal loop in CYP3A4, resulting in weakening of CYP3A4 inhibition. In summary, the combined structural, functional, and docking studies revealed that the clash with the CYP3A5 C-terminal loop, together with the difference in hydrophobicity and dimensions between the two surfaces in the binding pocket, are mainly responsible for the selective inhibition of CYP3A4 by SCM-01 and its analogs.

Identification of a highly selective, non-suicide CYP3A4 inhibitor

Our previous structure-activity relationship (SAR) analyses have shown that SCM-10, SCM-14, and SCM-15 do not inhibit CYP3A5 at concentrations up to 60 µM (Fig. 3b). However, SCM-10 and SCM-14 are mechanism-based inhibitors that lost CYP3A4 selectivity upon pre-incubation, whereas SCM-15 has much weaker CYP3A4 inhibition and therefore worse selectivity (Supplementary Fig. 9 and Supplementary Table 5). To obtain a non-suicide selective CYP3A4 inhibitor with selectivity profile similar to or better than that of SCM-10, we based on the potent and non-mechanism based selective CYP3A4 inhibitor SCM-11 for further chemical modifications and identified SCM-18 (Fig. 5a). Compared to SCM-11, SCM-18 has a urea linker instead of the amide of region A, which would allow the chlorophenyl group to extend closer to surface 2 when bound to CYP3A5. Consequently, SCM-18 would have weaker CYP3A5 inhibition and better CYP3A4 selectivity due to unfavorable interactions with surface 2 residues. Consistent with our premise, the biochemical inhibition assay showed that SCM-18 completely lost its inhibition of CYP3A5 without perturbing the inhibitory potency of CYP3A4 (Fig. 5b). The time-dependent inhibition assay showed that SCM-18 is not a mechanism-based inhibitor (Fig. 5c), and UV-vis measurements confirmed that SCM-18 is a type II inhibitor that binds to CYP3A5 much weaker than CYP3A4 (Fig. 5d–f).

Fig. 5. SCM-18 is a highly selective and non-suicide CYP3A4 inhibitor.

a The chemical structure of SCM-18. The region A group is colored red. b Dose–response curves showing concentration-dependent inhibition of CYP3A4/5 by SCM-18. c Dose–response curves showing time-dependent inhibition of CYP3A4/5 indicate that SCM-18 is not a suicide (irreversible or mechanism-based) inhibitor. For panels b and c, percentage inhibition data were normalized against DMSO controls (as 0% inhibition) and 30 µM ketoconazole controls (as 100% inhibition). The reported numerical values are the means and standard deviations of the IC₅₀ values calculated from triplicate experiments (n = 3). d, e Normalized UV-vis spectra showing spectral changes upon SCM-18 binding to CYP3A4/5. Insets are difference spectra calculated by subtracting the spectrum of ligand-free CYP3A4/5 from each of the ligand-bound state. Spectra of the ligand-free state are shown as the black solid lines, and spectra at the highest ligand concentration are shown as red solid lines. f ΔA_peak – ΔA_trough vs compound concentration plots showing the effect of selective inhibitor binding on CYP3A4/5 spectral signals. The reported numerical values are the means and standard deviations of K_s values derived from triplicate experiments (n = 3). g Overlay of CYP3A4–SCM-01 (gray), CYP3A4–SCM-18 (pink), and CYP3A5–apo (yellow) crystal structures showing that SCM-18, like SCM-01, would clash with the CYP3A5 surface 1 formed by C-terminal loop residues. h CYP3A4–SCM-18 docking results showing that SCM-18 binds CYP3A4 in 2 poses facing two distinct surfaces (green and blue). i Overlay of the CYP3A4–SCM-18 pose 2 and CYP3A5-apo crystal structure showing that ligand binding to CYP3A5 in pose 2 is possible but not favorable due to polar residues. For stick presentations in (h, i), fluoride atoms are colored in cyan, chloride atoms are colored in green, nitrogen atoms are colored in blue, and oxygen atoms are colored in red. The heme group is colored in wheat. Source data for (b–f) are provided as a Source Data file.

The crystal structure of CYP3A4 in complex with SCM-18 was solved at the resolution of 2.95 Å (Supplementary Fig. 8, Supplementary Table 7). Structural analysis indicated that CYP3A4–SCM-18 and CYP3A4–SCM-01 have almost identical crystal structures with an average RMSD of 0.375 Å, and the conformation of the SCM-18 is also highly similar to that of SCM-01 (Fig. 5g). The region A chlorophenyl group interacts with surface 1, which would clash with CYP3A5 if bound in the same conformation due to a wall formed by the C-terminal loop. The similarity in crystal structures also explains why SCM-01 and SCM-18 have the same inhibitory potencies for CYP3A4.

Subsequent docking analysis showed that the selective CYP3A4 inhibition by SCM-18 follows the same mechanism as we have proposed. The docking of SCM-18 to CYP3A4 resulted in two binding poses like SCM-01 (Fig. 5h). Pose 1 is similar to that observed in the crystal structure and validates docking parameters. Pose 2 is consistent with our hypothesis that a longer urea linker would push the hydrophobic region A group closer to the highly polar surface 2. This resulted in more unfavorable interactions and reduced ligand binding to surface 2. Docking experiments also provided a putative binding pose of SCM-18 to CYP3A5, showing that the hydrophobic chlorophenyl is near an unfavorable polar environment on surface 2 formed by residues E374 and R372, resulting in poor binding to CYP3A5 (Fig. 5i and Supplementary Fig. 11). In summary, SCM-18 was identified as the most selective CYP3A4 inhibitor, and the combined functional, structural, and computational studies supported our mechanistic model for selective CYP3A4 inhibition.

Trade-off between CYP3A4 selectivity and potency

To further understand why SCM-18 can display such robust preference in inhibiting CYP3A4 over CYP3A5 and to investigate the possibilities of increasing the potency of SCM-18, we explored the effect of various region A substitutions in inhibiting CYP3A4/5. The analogs tested, SCM-19, SCM-20, SCM-21, SCM-22, SCM-23, have meta- or para-substituted benzene groups instead of ortho-substituted chlorophenyl (Supplementary Fig. 12a). Although all these analogs showed no or minimal CYP3A5 inhibition consistent with the model, they also revealed the loss of CYP3A4 inhibitory potency (Supplementary Fig. 12a). This can be explained by the steric clash between the analogs and residues R212 and I369 in CYP3A4, as their distances to meta- and para-substituents in the CYP3A4–SCM-18 structure are only 4.4 and 3.6 Å, respectively (Supplementary Fig. 12b). This view is supported by analogs with the amide linker, SCM-14 and SCM-15, in which excessively long region A groups also weakened CYP3A4 inhibition (Fig. 3b). Therefore, only ortho-substituted benzene derivatives are favorable for CYP3A4 inhibition and selectivity, and further extension at para- or meta-positions will lead to the loss of CYP3A4 inhibition and selectivity.

Given the importance of the ortho-substituted chloride in SCM-18, we designed SCM-24 to have the longer ethyl chloride substitution group, which may allow the compound to establish additional hydrophobic interactions through the chloride in CYP3A4 (Fig. 6a). Indeed, the results of the inhibition assay showed that SCM-24 is a more potent inhibitor of CYP3A4 than SCM-18, and time-dependent inhibition assay showed that SCM-24, like SCM-18, is a non-mechanism-based inhibitor (Supplementary Fig. 13a). Consistent with our hypothesis, the CYP3A4 crystal structure in complex with SCM-24 showed that the compound forms an additional hydrophobic interaction with F215, thereby stabilizing compound binding and inhibition of CYP3A4 (Fig. 6d). However, the modification also resulted in stronger CYP3A5 inhibition and reduced CYP3A4 selectivity (Fig. 6a). Docking studies give a rationale for the observed CYP3A5 inhibition by showing that the ethyl chloride group forces the hydrophobic benzene ring of SCM-24 to be positioned farther away from the unfavorable polar environment formed by E374 and R372 as compared with that of SCM-18 (Supplementary Fig. 14a). Additionally, the ethyl group is seen favorably interacting with a hydrophobic patch, although the chloride is seen close to a polar spot formed by S119 and R105. Therefore, SCM-24 has a more favorable surface 2 binding environment than SCM-18 does, resulting in enhanced CYP3A5 inhibition.

Fig. 6. Trade-off between potency and selectivity of SCM-18 analogs.

a–c Chemical structures of (a) SCM-24, (b) SCM-25 and (c) SCM-26 and dose–response curves showing their concentration-dependent inhibition of CYP3A4/5. Percentage inhibition data were normalized against DMSO controls (as 0% inhibition) and 30 µM ketoconazole controls (as 100% inhibition). The reported numerical values are the means and standard deviations of the IC₅₀ values calculated from triplicate experiments (n = 3). d–f Overlay of CYP3A4–SCM-18 (pink) with (d) CYP3A4–SCM-24 (yellow), (e) CYP3A4–SCM-25 (yellow), (f) CYP3A4–SCM-26 (yellow) crystal structures highlighting unique interactions associated with these analogs. For stick representations in (d–f), fluoride atoms are colored in cyan, chloride atoms are colored in green, nitrogen atoms are colored in blue, and oxygen atoms are colored in red. Source data for (a–c) are provided as a Source Data file.

Because the urea linkage in SCM-18 has such a dramatic effect on CYP3A4 selectivity compared to the amide linker in SCM-11 due to spacing differences, SCM-25 was designed to have a methyl amide which has a similar linker length as urea (Fig. 6b). Compared to SCM-18, SCM-25 is a more potent inhibitor of both CYP3A4 and CYP3A5. Time-dependent inhibition assay confirmed that SCM-25 acts as a non-mechanism-based inhibitor (Supplementary Fig. 13b). Analysis of the CYP3A4–SCM-25 crystal structure indicated that the improved potency in inhibiting CYP3A4 could be due to increased hydrophobic interaction of the methyl amide with A370 compared to that of the urea linker in SCM-18 (Fig. 6e and Supplementary Fig. 14b, c). Additionally, the potential H-bond between the ligand and residue R105 may contribute to enhanced CYP3A4 inhibition because the amide carbonyl in SCM-25 has more electronegativity than the urea carbonyl in SCM-18 (Fig. 6e). In the case of enhanced CYP3A5 inhibition, docking studies suggested that the hydrophobic region A group of SCM-25 increased favorability of surface 2 binding similar to SCM-24, in which the chlorophenyl is positioned farther away from the unfavorable polar environment formed by E374 and R372 as compared to that of SCM-18 (Supplementary Fig. 14d). Therefore, altering the urea linker to a more flexible methyl amide would enhance inhibitory potency for both CYP3A5 and CYP3A4.

The analogs SCM-26 and SCM-27 were synthesized with branched phenol groups on the urea linker to investigate whether occupying surface 1 and surface 2 simultaneously would increase potency in CYP3A4 (Fig. 6c and Supplementary Fig. 13d, e). SCM-26 and SCM-27 showed a significant increase in CYP3A4 inhibition potency compared to SCM-18 (Fig. 6c, Supplementary Fig. 13e, and Supplementary Table 3). The crystal structure of CYP3A4–SCM-26 revealed that the hydrophobic chlorophenyl in SCM-26 is placed on surface 1 the same as SCM-18, and the hydrophilic phenol group is oriented towards surface 2 and forms an H-bond with the backbone residue of E374, thereby enhancing ligand binding and inhibition of CYP3A4 (Fig. 6f). In the case of SCM-26 binding to CYP3A5, the time-dependent inhibition assay indicated that the compound is a mechanism-based inhibitor (Supplementary Fig. 13c). Therefore, more than one species could contribute to SCM-26—mediated CYP3A5 inhibition. Nevertheless, docking of the parent compound SCM-26 to CYP3A5 showed that the phenol hydroxyl of SCM-26 formed a H-bond with E374 as predicted, but the chlorophenyl group is flipped 180° and points against the C-terminal loop to avoid the clash with the C-terminal loop (Supplementary Fig. 14e).

Among our SCM analogs, SCM-27 has the highest potency in inhibiting both CYP3A4 and CYP3A5 (Supplementary Fig. 13d, e). Since the high potency of SCM-27 in both CYP3A paralogs may allow us to understand its binding modes to CYP3A4 and CYP3A5, we co-crystallized SCM-27 with CYP3A4 and CYP3A5 to a resolution of 3.40 Å and 2.25 Å, respectively (Supplementary Fig. 8f, g, Supplementary Fig. 14f, g, Supplementary Table 8). Analysis of these structures revealed that the orientation of the main scaffold (trifluoromethylphenyl imidazole) is consistent with our mechanistic model of CYP3A4 selective inhibition, with the two branched aromatic groups facing towards surfaces 1 and 2 in the CYP3A4–SCM-27 structure (Supplementary Fig. 14f). In the CYP3A5–SCM-27 structure, the phenol group is oriented towards surface 2 and fixed by H-bonds with R105 and S119 (Supplementary Fig. 14g), and the chlorophenyl moiety is positioned at an opening surface to avoid the clash with the C-terminal loop and interacts with the rearranged F–F’ loop (Supplementary Fig. 14h), thereby explaining the strong CYP3A5 inhibition by SCM-27.

In summary, combined functional, structural, and computational studies indicate there is a trade-off between potency and selectivity. The large, flexible, and promiscuous ligand-binding pocket of CYP3A4/5 can accommodate the same ligand in multiple orientations. It is therefore not surprising that the addition of a functional group can enhance inhibitory potency for both CYP3A4 and CYP3A5. In addition, the metabolism of compounds into more potent inhibitors adds an extra layer of complexity in designing potent and selective CYP3A4 inhibitors.

Selective CYP3A4 inhibitor SCM-08 does not target other major CYPs

While CYP3A4/5 are responsible for a large portion of CYP-mediated drug metabolism, other major human CYPs also contribute substantially to drug clearance^2,3. To test whether the most selective CYP3A4 inhibitor SCM-18 targets other major CYPs, we measured its inhibition of a panel of CYPs recommended by FDA guidelines^6,7. Unfortunately, SCM-18 demonstrated potent inhibition of CYP1A2 and CYP2C19, making it a less ideal selective CYP3A4 inhibitor for possible clinical use (Fig. 7).

Fig. 7. SCM-08 is a highly selective CYP3A4 inhibitor that does not target other CYPs.

Dose–response curves showing concentration-dependent inhibition of a panel of CYPs as determined using the P450-Glo and the Vivid P450 assays. Percentage inhibition data were normalized against DMSO controls (as 0% inhibition) and 30 µM control inhibitors (as 100% inhibition). The reported numerical values are the means and standard deviations of the IC₅₀ values calculated from triplicate experiments (n = 3). Source data are provided as a Source Data file.

To identify selective CYP3A4 inhibitors that do not inhibit other major CYPs, we investigated the SAR of CYP1A2 and CYP2C19 inhibition using SCM-01 analogs. The results indicated that compounds with larger region A groups, such as SCM-10, tend to have more potent inhibition of CYP1A2 and CYP2C19 (Supplementary Fig. 15). The observed SAR of CYP1A2 and CYP2C19 inhibition created a dilemma, because large hydrophobic region A groups are also favorable for selective CYP3A4 inhibition. Nevertheless, SCM-08 stands out as a highly selective CYP3A4 inhibitor that avoids inhibiting both CYP1A2 and CYP2C19 (Supplementary Fig. 15). Subsequent tests confirmed that SCM-08 does not inhibit any of the major human CYPs, including CYP3A7 in the CYP3A subfamily (Fig. 7 and Supplementary Table 9). In addition, the inhibition by SCM-08 is not mechanism based, as illustrated by the results of time-dependent IC₅₀ shift assays (Supplementary Fig. 16a, b). The compound also remains selective for CYP3A4 in cell-based assays and is not cytotoxic up to 60 µM (Supplementary Fig. 16c, d). In contrast to ritonavir, the pan-CYP3A inhibitor in current clinical use, SCM-08 not only exhibits a 46-fold difference in the inhibition of CYP3A4 and CYP3A5 but also avoids undesired inhibition of CYP2C9, CYP2C19, and CYP3A7 compared to ritonavir (Fig. 7 and Supplementary Table 9). Additionally, the metabolic stability of SCM-08 in human liver microsomes is similar to that of ritonavir, with half-lives of 0.99 ± 0.05 h at 1 µM and >2 h at 5 µM (Supplementary Table 10). Therefore, SCM-08 may serve as a starting point for developing more potent and selective CYP3A4 inhibitors.

Discussion

In this work, we embarked on the challenging undertaking to develop selective CYP3A4 inhibitors and uncovered the structural basis of such selectivity. Our findings provide much needed understanding in the design of chemicals that can be used in lessening drug-drug interactions. A growing number of patients take multiple medications concurrently to treat various and unrelated illnesses¹. The higher prevalence of polypharmacy can increase health risks due to drug-drug interactions, leading to unintended and potentially fatal consequences. CYP3A4 and CYP3A5 are main players in drug metabolism, which are responsible for the chemical modification and degradation of the majority of FDA approved drugs^2,3. It is estimated that more than 90% of common chemicals⁴³, 66% of carcinogens⁴⁴, and more than half of currently marketed drugs are metabolized by CYPs⁴⁵. Hence, CYP3A4/5 can affect the plasma concentration of drugs or alter their chemical structure to generate toxic metabolites.

The inactivation of CYP3A4 can enhance the therapeutic efficacy of drugs by slowing down their metabolism. For instance, drugs used to treat HIV are often co-administered with the broad spectrum CYP inhibitor ritonavir to prevent their rapid metabolism and disposition^8–10. However, pan-CYP3A inhibitors can lead to undesirable inhibition of CYP3A5 that affect the pharmacokinetics of drugs with narrow therapeutic index, raising their levels to toxic concentrations. A representative example is the immunosuppressant tacrolimus, which is mainly metabolized by CYP3A5 and can cause nephrotoxicity and neurotoxicity when its blood level is elevated due to reduced CYP3A5 activity^46,47. Co-administration of ritonavir in Paxlovid^TM has been shown to cause elevated tacrolimus toxicity among kidney transplantation patients¹⁵. Moreover, CYP3A4 and CYP3A5 are not functionally identical or redundant, and they have their own metabolic preferences for drug clearance¹³. Clinical studies have revealed a number of important drugs that are metabolized selectively by CYP3A4 or CYP3A5. For example, CYP3A4 selectively metabolizes everolimus⁴⁸, quetiapine⁴⁹, and testosterone⁵⁰, and CYP3A5 selectively metabolizes vincristine¹⁶, and atazanavir⁵¹. To better understand the frequency of prescription drugs being selectively metabolized by CYP3A4/5, we evaluated 1491 FDA-approved drugs for their metabolism by CYP3A4 and CYP3A5 using mass-spectrometry. Further evaluation of 161 putative substrates revealed that 13 are selective for CYP3A4 and 11 are selective for CYP3A5 (Supplementary Fig. 17, Supplementary Table 11), indicating that selective CYP3A4 or CYP3A5 substrates are not rare. The differences between CYP3A4 and CYP3A5 with respect to substrate specificity indicate that the two enzymes should not be inhibited non-selectively. Therefore, co-administration with a selective CYP3A4 inhibitor rather than a pan CYP3A inhibitor provides a more valuable alternative for clinical applications in polypharmacy.

The development of selective CYP3A4 inhibitors has been difficult because CYP3A4 and CYP3A5 are highly homologous proteins with 83% sequence identity and high structural similarity⁵². However, by solving crystal structures of CYP3A4 in complex with several selective inhibitors, together with docking studies and functional assays, we managed to identify structural differences between CYP3A4 and CYP3A5 that enable selective inhibition of CYP3A4. We discovered that the clash between the ligand and the C-terminal loop of CYP3A5, but not that of CYP3A4, is responsible for such selectivity. The structural difference between CYP3A4 and CYP3A5 C-terminal loop has been noticed previously^20,53. Our work provides experimental evidence that such differences can be exploited in the selective inhibition of CYP3A4. The differential positioning of the C-terminal loop is an inherent property of the two enzymes, and the C-terminal loop in CYP3A5 is highly rigid as it is stabilized by four H-bonds. This suggests that the mechanism is not unique to SCM-01 analogs and is applicable to inhibitors with other scaffolds. Indeed, the three selective inhibitors identified by HTS (SCM-01, SCM-02 and SCM-03) have different scaffolds and binding conformations, yet they have the same potential to clash with the CYP3A5 C-terminal loop.

Another significant component in the mechanism of selective CYP3A4 inhibition involves two distinct areas in the ligand binding pockets, termed surface 1 and 2 (Fig. 8). The lead compound SCM-01 and its potent analogs interact favorably with the mostly hydrophobic surface 1 in CYP3A4, particularly those that possess a hydrophobic substitution of limited size in region A group, such as SCM-08 and SCM-18. Surface 1 found in CYP3A4 is replaced by a wall in CYP3A5 due to the position of the C-terminal loop, which forces the region A group of the analogs to interact with an alternative space (surface 2) that has been shown to be important for ligand binding in CYP3A4^40–42. However, the highly polar environment of surface 2 discourages strong interactions with those analogs with hydrophobic region A substitutions, resulting in attenuated CYP3A5 inhibition.

Fig. 8. Molecular basis for the selective CYP3A4 inhibition by SCM-1 analogs.

a SCM-1 analogs bind preferentially to the mostly hydrophobic surface 1 of CYP3A4 when region A groups (indicated as R in red) are hydrophobic of limited size. b Due to a wall created by the C-terminal loop in CYP3A5, SCM-1 analogs are forced to interact with the alternative surface 2, but the highly polar environment discourages interactions with analogs containing hydrophobic region A groups.

Because of their large and flexible ligand binding pockets, CYPs have a broad substrate profile and are responsible for removing many toxins and xenobiotics^52,54,55. This ligand promiscuity complicates the design of more potent selective CYP3A4 inhibitors. As revealed by the hierarchical clustering analysis based on structural similarity using crystal structures of CYP3A4–ligand complexes, our SCM-01 analogs represent a distinct class of compounds among all ligands crystallized to date (Supplementary Fig. 18). To increase the binding affinity of our analogs, we designed compounds that were aimed to concomitantly occupy surface 1 and surface 2 in CYP3A4, while trying to maintain selectivity towards CYP3A4 because of the absence of surface 1 in CYP3A5. Surprisingly, analogs such as SCM-26 and SCM-27 were more potent for both enzymes. Based on the CYP3A5–SCM-27 crystal structure, we conclude that the spacious interior and plasticity of the CYP3A5 ligand binding pocket are responsible for the loss of selectivity of SCM-27: (1) The large space allowed for the reorientation of bulky inhibitors so that the added moiety can form favorable H-bonds with charged residues (Supplementary Fig. 14g); (2) The flexible ligand binding pocket allowed for the rearrangement of the F–F’ loop and the formation of hydrophobic interactions with the chlorophenyl moiety (Supplementary Fig. 14h). Hence, there can be a trade-off between selectivity and potency when it comes to the expansive ligand binding site characteristic of CYPs. This highlights the critical role of the region A group in CYP3A4 selectivity and demonstrates that the attempt to achieve CYP3A4 potency and selectivity simultaneously remains a challenging task. Future work to increase potency while retaining selectivity may include chemical modifications of the trifluoromethyl phenyl moiety opposite of region A in SCM-01 analogs, because region A was shown to be highly impactful to CYP3A4 selectivity.

Ideally, CYP-metabolized drugs are co-administered with a CYP inhibitor to reach a desired therapeutic concentration for a required amount of time, after which drug clearance removes excess drug that could otherwise be toxic. The wide-ranging substrate profile of CYP3A4 adds an extra layer of complexity in obtaining selective inhibitors. Many compounds with differing physicochemical properties are metabolized by CYP3A4 to reactive intermediates that permanently and irreversibly inactivate the enzyme; those inhibitors are known as suicide or mechanism-based inhibitors, such as CYP3cide⁵⁶. Therefore, suicide inhibitors can lead to drug overexposure and toxicity, especially for drugs with a very narrow therapeutic window. We paid particular attention to our efforts in developing selective CYP3A4 inhibitors that are not suicide inhibitors as indicated by the time-dependent inhibition assays. Most SCM-01 analogs were not shown to increase potency over time, indicative of not being suicide inhibitors, including SCM-08 and SCM-18. On the other hand, SCM-10 was omitted for further development even though it showed marked selectivity towards CYP3A4 as opposed to CYP3A5.

There are a total of 57 known human CYPs classified into 18 families and 43 subfamilies⁵⁷. Our study focuses on two most important enzymes, CYP3A4 and CYP3A5, as they alone are responsible for 30%–50% of CYP-mediated drug clearance^2,3. We also tested the inhibitory potency of SCM-01 and its analogs against a panel of CYPs recommended by FDA guidelines^6,7. Interestingly, the inhibition of CYP1A2 and CYP2C19 was negatively correlated with the inhibition of CYP3A5 (Supplementary Fig. 15). As the ligand-binding pockets of CYP1A2 and CYP2C19 are very different from that of CYP3A5^58,59, their mechanisms of inhibition by our compounds remain unknown, but it is reasonable to assume that these compounds adopt different conformations when binding to CYP1A2 and CYP2C19. Nevertheless, SCM-08 maintains a high selectivity for CYP3A4 relative to CYP3A5 and other CYPs, showing that it is possible to develop selective inhibitors targeting a specific enzyme among its homologs. In our case, the C-terminal loop may be exploited for the design of other selective inhibitor because the sequence alignment shows that the C-terminal loop region in major human CYPs (CYP1A2, CYP2B6, CYP2C8, CYP2C9, CYP2C19, CYP2D6, CYP3A4, CYP3A5, and CYP3A7) displays significant variability (Supplementary Fig. 19).

In summary, we obtained a series of non-suicide and highly selective CYP3A4 inhibitors with chemical scaffolds different from previously published based on an unbiased HTS campaign. Guided by structural analysis, functional data and docking studies, we uncovered the basis for the differential inhibition of CYP3A4. The 10 X-ray crystal structures of CYP3A4 and CYP3A5 in complex with the investigated inhibitors revealed that the differential positioning of the C-terminal loop between CYP3A4 and CYP3A5 would prevent these compounds from binding to CYP3A5 but not to CYP3A4 in a favorable conformation, thus leading to the observed selective inhibition. Additionally, two distinct surfaces in the ligand binding site were identified that explain the weak or nonexistent inhibition of CYP3A5. The proposed mechanistic model provides a roadmap for the development of improved selective CYP3A4 inhibitors with valuable clinical ramifications in reducing drug-drug interactions.

Methods

Compounds and chemical synthesis

The identities of compounds purchased from vendors were verified using mass spectrometry, and their purities were determined to be >95% by high-performance liquid chromatography, using a system equipped with UV-vis and evaporative light-scattering detectors. The names of commercially available compounds, which are shown in the form of “vendor name + catalog number”, are presented in Supplementary Tables 2–5 and 9 under “alternative name”.

Chemical structures of compounds studied in this manuscript, i.e., SCM-01 to SCM-27, are displayed in Supplementary Fig. 20. The details of the chemical synthesis of SCM-01 analogs are available in the Supplementary Information (Supplementary Figs. 21-50).

High-throughput screening campaign

The chemical library was composed of compounds acquired from several commercial sources (mostly ChemBridge, Enamine, ChemDiv, Life Chemicals, Otava, and InterBio) with molecular descriptors that conformed to drug-like physicochemical characteristics while minimizing pan-assay interference compounds (PAINs)²⁴. The P450-Glo luminescence assays were conducted in 384-well white plates (PerkinElmer, Waltham, MA, catalog no. 6007290), as described below, at a single-compound concentration of 10 µM. In total, 956 compounds with ≥80% inhibition were selected for dose–response confirmatory screens against both CYP3A4 and CYP3A5, using the same assay. HTS data were analyzed using in-house software written in Pipeline Pilot (BIOVIA, San Diego, CA). The analysis included fitting data points from dose–response experiments to sigmoidal models to obtain IC₅₀ values and determining Z′ factors for assay quality assessment⁶⁰.

Tanimoto coefficient–based hierarchical clustering analysis of compounds was performed using ChemMine Tools⁶¹, and the results were displayed as a circular cladogram constructed using Interactive Tree of Life (iTOL)⁶².

P450-Glo biochemical inhibition assay

The inhibitory effects of compounds on CYP3A4, CYP3A5, and CYP3A7 were measured using the P450-Glo luminescence assay (Promega, Madison, WI, catalog no. V9002 and V8912)^34,63. The inhibitory profiles of CYP3A4 and CYP3A5 were determined using the substrate luciferin-IPA, and the inhibition of CYP3A7 was quantified using the substrate luciferin-PPXE. The CYP3A4, CYP3A5, and CYP3A7 Supersome™ human recombinant enzymes (catalog nos. 456202, 456256, and 456237) were purchased from Discovery Life Sciences (Huntsville, AL). These enzymes contain 1000 pmol/mL of the full-length CYP protein mixed with NADPH–cytochrome P450 reductase, cytochrome b5, and membrane components.

The P450-Glo assays were carried out in 384-well white assay plates (PerkinElmer, Waltham, MA, catalog no. 6007290) with a 25 µL reaction volume. To each well was added 25 nL (in the primary HTS) or 37.5 nL (in all other assays) of the test compound in DMSO, using an Echo 650 Acoustic Liquid Handler (Beckman Coulter, Brea, CA), followed by 12.5 µL of 2× enzyme–substrate mixture. The reaction was initiated by adding 12.5 µL of 2 × NADPH Regeneration System (Promega, catalog no. V9510), and the plate was incubated for 25 minutes at room temperature. The final concentrations of reaction components were as follows: 100 mM potassium phosphate pH 7.4, 4 nM (with luciferin-IPA as the substrate) or 10 nM (with luciferin-PPXE as the substrate) CYP3A4/5/7, 3 µM luciferin-IPA or 25 µM luciferin-PPXE substrate, and 0.00017–60 µM of the test compound. The final DMSO concentration was 0.25%. The reaction was quenched by adding 25 µL of luciferin detection reagent as supplied in the P450-Glo kit, and the mixture was incubated at room temperature for an additional 20 minutes. Luminescence signals were determined with an EnVision Multimode Plate Reader (PerkinElmer, Waltham, MA). The experiment was repeated three times for each compound concentration, except for the primary HTS, which was carried out only once at a final compound concentration of 10 µM.

The luminescence signals were normalized to DMSO (set as 0% inhibition) and 30 µM ketoconazole or 15 μM ritonavir (set as 100% inhibition) within each plate. The resulting % inhibition values were analyzed by plotting the percentage inhibition against the ${\log }_{10}\left[compound\right]$ and were fitted to Eq. 1 in GraphPad Prism (GraphPad, San Diego, CA):

$$\%inhibition=bottom+\frac{\left(top-bottom\right)}{1+{10}^{\left({\log }_{10}{IC}_{50}-{\log }_{10}\left[compound\right]\right)\times hillslope}}$$ 1

where the $top$ and $bottom$ are the minimum and maximum percentage inhibition values, respectively, and the $hillslope$ is the factor for possible non–single-site binding events.

P450-Glo cell-based inhibition assay

The P450-Glo cell-based inhibition assay was performed as previously reported³⁴. The 293 T/17 cell line (catalog no. CRL-11268) used for this assay was obtained from the American Type Culture Collection (Manassas, VA), and the cell line was authenticated by short tandem repeat DNA profiling. The cells were maintained in growth medium (Dulbecco Modified Eagle Medium with 10% fetal bovine serum) in a humidified atmosphere at 37 °C with 5% CO₂ and were verified to be mycoplasma free with the MycoProbe Mycoplasma Detection Kit (R&D Systems, Minneapolis, MN). Cell counts were obtained with a Countess II Automated Cell Counter (Thermo Fisher Scientific) and trypan blue exclusion staining.

The transfection mixture was prepared by mixing 90 µL of FuGENE 6 transfection reagent (Promega, catalog no. E2692) and 30 µg of pcDNA3 plasmids encoding ZsGreen1-IRES-CYP3A4 or ZsGreen1-IRES-CYP3A5 in 2 mL of Opti-MEM I Reduced Serum Medium (Thermo Fisher Scientific) for 15 minutes at room temperature³⁴. The 2 mL mixture was subsequently added to a 10-cm tissue culture–treated dish containing 10 × 10⁶ 293 T/17 cells in 8 mL of growth medium. The medium was replaced with 10 mL of fresh growth medium at 24 hours post transfection, and the cells were trypsinized and diluted in fresh growth medium to prepare them for the P450-Glo cell-based inhibition assay at 48 hours post transfection.

The inhibition assay was carried out using the same P450-Glo luminescence assay kit (Promega, catalog no. V9002) in white 384-well tissue culture–treated plates. To each well was added 5 × 10⁴ cells in 25 µL of growth medium. This was followed by 37.5 nL of the test compound and 25 nL of 3 mM luciferin-IPA in DMSO, added using an Echo 650 Acoustic Liquid Handler (Labcyte). The final concentrations of the reaction components were 3 µM luciferin-IPA substrate, 0.00017–60 µM of the test compound. The final DMSO concentration was 0.25%. Plates were shaken for 3 minutes at room temperature to mix the reagents, centrifuged at 500 × g for 1 minute, and then incubated at 37 °C in 5% CO₂ for 1 hour. After the incubation, 25 µL of luciferin detection reagent was added to each well and the plates were incubated for an additional 20 minutes at room temperature to allow the reaction to go to completion. Luminescence signals were measured and analyzed as described for the biochemical inhibition assay. The experiment was repeated three times for each compound concentration.

Time-dependent IC₅₀ shift assay

The time-dependent IC₅₀ shift assay was performed using the same P450-Glo CYP3A4 Assay kit (Promega, catalog no. V9002) as described above. The experiment was carried out in 384-well white assay plates (PerkinElmer, catalog no. 6007290) with a 25 µL reaction volume. To each well was added 37.5 nL of the test compound in DMSO, using an Echo 650 Acoustic Liquid Handler (Beckman Coulter), followed by 12.5 µL of 2× enzyme mixture in the absence or presence of 2× NADPH Regeneration System (Promega, catalog no. V9510). The plate was incubated for 30 minutes at room temperature, then 12.5 µL of substrate solution, with or without 2× NADPH Regeneration System, was added to each well. The final concentrations of the reaction components were as follows: 100 mM potassium phosphate pH 7.4, 4 nM CYP3A4/5, 3 µM luciferin-IPA, and 0.00017–60 µM of the test compound. The final DMSO concentration was 0.25%. The reaction was allowed to proceed for 25 minutes at room temperature then quenched by adding 25 µL of luciferin detection reagent as supplied in the P450-Glo kit. The mixture was then incubated at room temperature for an additional 20 minutes. Luminescence signals were determined with an EnVision Multimode Plate Reader (PerkinElmer, Waltham, MA). The experiment was repeated three times for each compound concentration.

Cytotoxicity assay

The cytotoxicity assay was carried out using the CellTiter-Glo Luminescent Cell Viability Assay kit (Promega, catalog no. G7572). The experiment was performed in 384-well white tissue culture-treated plates (Revvity, catalog no. 6007688). Each well was added with 2500 293 T/17 cells in 25 μL growth medium (DMEM supplemented with 10% FBS) and 62.5 nL of test compound at final concentrations of 0.00017–60 µM. Wells with 62.5 nL of DMSO instead of test compound were negative controls (0% toxicity), while wells without cells (25 µL growth medium only) were positive controls (100% toxicity). Cells were incubated with compounds for 72 hours at 37 °C and mixed with 25 µL of CellTiter-Glo reagent. The cell viability was determined by measuring the luminescence signal in an EnVision Multimode Plate Reader (PerkinElmer). The final DMSO concentration was 0.25% for all wells.

Metabolic stability assay

The metabolic stability assay of SCM-08 and ritonavir was performed by measuring their degradation in human liver microsomes using methods described in a previous publication with minor modifications⁶⁴. The reaction was conducted in a solution containing 0.5 mg/mL protein in human liver microsomes (XenoTech, pooled mixed gender, catalog no. H0630), 1.29 mM of NADPH solution A, 0.4 U/mL of NADPH solution B (Fisher Scientific, catalog no. NC1096251 and NC1095811), 91 mM potassium phosphate (pH 7.4), and 1 mM EDTA (Fluka, catalog 03690). 50 nL of a 5 mM DMSO stock solution of the test compound (final concentration 5 µM, 0.1% DMSO) was dispensed individually into a 96-well assay plate (Corning, catalog no. 3363) using an Echo 655 liquid handler (Beckman Coulter). Next, 50 µL of the reaction solution was added to the plate, and the plate was incubated at 37 °C. Samples were collected at 0, 0.25, 0.5, 1, and 2 hours, followed by quenching with 150 µL of acetonitrile containing 100 ng/mL warfarin as the internal standard. The quenched samples were centrifuged at 4000 rpm for 20 minutes. The supernatant was then mixed with MilliQ water in a 1:1 ratio before analysis by UPLC-MS/MS, as described in the sections below. The metabolic stability was determined by calculating the half-life (T_1/2) from the linear fit of Ln percentage remaining vs incubation time⁶⁵. The intrinsic clearance (Cl_int) was calculated using the previously reported formula⁶⁶. All assays were performed in triplicates.

Expression and purification of CYP3A4 and CYP3A5

The CYP3A4^WT and CYP3A5^WT proteins used for UV-vis measurements and protein crystallization were expressed and purified using methods adapted from a previous publication³⁴. CYP3A4^WT was modified to have a deletion of N-terminal amino acids 3–23 and a C-terminal 4× His-tag⁶⁷, whereas CYP3A5^WT was constructed to have N-terminal and C-terminal deletions of amino acids 3–23 and 498–502, respectively, as well as a C-terminal 4× His-tag²⁰. Both CYP3A4 and CYP3A5 expression constructs were verified to have the correct coding sequence by DNA sequencing, and each construct was co-transformed with a pGro7 plasmid (Takara Bio USA, Mountain View, CA) that expresses the chaperonin protein GroES−GroEL into E. coli DH5α_F^’_I^q cells (New England Biolabs, Ipswich, MA, catalog no. C2992H). The transformed cells were selected by both ampicillin and chloramphenicol resistance.

In a typical 12 L protein production process, the cell culture was grown at 37 °C in terrific broth until the OD₆₀₀ reached 0.4-0.5. The growth temperature was then lowered to 28 °C, and cells were induced 30 minutes later at the same temperature with 1 mM isopropyl β-D-1-thiogalactopyranoside, 0.5 mM (CYP3A4) or 5 mM (CYP3A5) 5-aminolevulinic acid hydrochloride, and 3 g/L L-arabinose. Cells were harvested 44-48 hours after induction by centrifugation, and cell pellets were resuspended in cold lysis buffer containing 500 mM potassium phosphate pH 7.4, 20% glycerol, 10 mM β-mercaptoethanol (BME), and 1 mM phenylmethylsulfonyl fluoride. Cells were lysed by two passes through an LM-10 microfluidizer (Microfluidics, Westwood, MA) at 18000 psi, and the lysate was incubated with 8 mM CHAPS for 1 hour at 4 °C to extract CYP proteins from spheroplasts. After incubation, the lysate was centrifuged at 40000 × g for 2 hours, and the supernatant was collected for subsequent protein purification. All purification steps were carried out at 4 °C.

CYP3A4 was purified using Ni affinity and size-exclusion chromatography columns. Lysate supernatant from 12 L culture were loaded on a 20 mL HisPrep FF 16/10 Ni-NTA column (Cytiva, Marlborough, MA, catalog no. 28936551) that was pre-equilibrated in Buffer A (500 mM potassium phosphate pH 7.4, 10 mM BME, 8 mM CHAPS, and 20% glycerol). The column was washed with Buffer A containing 5 mM ATP, 10 mM MgCl₂, and 100 mM KCl to remove the chaperone protein, followed by another wash with Buffer A containing 20 mM imidazole to remove loosely bound contaminants. The CYP3A4 was eluted with a 20 − 250 mM imidazole gradient in Buffer A. Fractions containing partially purified CYP3A4 were concentrated using 30 kDa molecular-weight cutoff concentrators, diluted in Buffer A to lower imidazole concentration, and further purified using 2 × 5 mL HisTrap FF Ni-NTA columns (Cytiva, catalog no. 17525501) using the same method mentioned above. The eluted CYP3A4 was concentrated again and passed through a HiLoad 26/600 Superdex 200 pg size-exclusion column (Cytiva, catalog no. 28989336) pre-equilibrated in Buffer B (50 mM potassium phosphate pH 7.4, 500 mM NaCl, 10 mM BME, 20% glycerol). Purified CYP3A4 was concentrated and stored at −80 °C.

CYP3A5 was purified using Ni affinity, ion exchange, and size-exclusion chromatography columns. Lysate supernatant from 12 L culture was incubated with 30 mL Ni-NTA Superflow resin (Qiagen, Germantown, MD, catalog no. 30430) overnight with stirring. The resin was packed in a gravity column, washed with Buffer A containing 5 mM ATP, 10 mM MgCl₂, and 100 mM KCl to remove the chaperone protein, followed by another wash with Buffer A containing 3 mM histidine to remove loosely bound contaminants. The CYP3A5 was eluted with Buffer A containing 40 mM histidine. CYP3A5-containing fractions were concentrated using 30 kDa molecular-weight cutoff concentrators, diluted 20-fold in Buffer C (100 mM HEPES pH 7.4, 20 mM potassium acetate, 8 mM CHAPS, 10 mM BME, and 20% glycerol) to reduce salt concentrations, and incubated with 20 mL CM Sepharose FF ion exchange resin (Cytiva, catalog no. 17071910) overnight with stirring. The resin was packed in a gravity column, washed with Buffer C containing 60 mM potassium phosphate, and eluted with Buffer C containing 250 mM potassium phosphate. Purified CYP3A5 fractions were concentrated again and passed through a Superdex 200 Increase 10/300 GL column (Cytiva, catalog no. 28990944) pre-equilibrated in Buffer D (100 mM HEPES pH 7.4, 50 mM potassium acetate, 10 mM BME, and 20% glycerol). Purified CYP3A5 was concentrated and stored at −80 °C. CYP3A4 and CYP3A5 protein concentrations were determined using the CO difference spectra method described by Guengerich et al. ⁶⁸.

Crystal structure determination of CYP3A4/5 complexed with ligands

CYP3A4 complexes were screened for crystallization using commercial sparse matrix (Qiagen) in 96-well format using a Mosquito robot (SPT Labtech) via the hanging drop or sitting drop method. All crystallization trays were set up using CYP3A4 and ligand solutions in DMSO at 20 °C after pre-incubation of protein and ligand for 1 hour at 4 °C.

Crystals of CYP3A4 with SCM-01 or SCM-03 were optimized using the microbatch under-oil method. Crystallization drops were prepared by mixing 1 µL of protein–ligand solution and 1 µL of precipitant in a 96-well MRC Under Oil Plate (Hampton Research, Aliso Viejo, CA) and were covered with 20 µL of paraffin oil. For the CYP3A4–SCM-01 complex, the protein–ligand solution contained 20 mg/mL protein and 4 mM ligand in storage buffer (with a final DMSO concentration of 10%) and the precipitant solution was 0.1 M sodium citrate, pH 5.0, 14% PEG 3350. For the CYP3A4–SCM-03 complex, the protein–ligand solution contained 15 mg/mL protein and 4 mM ligand in storage buffer (with a final DMSO concentration of 10%) and the precipitant solution was 0.1 M MIB, pH 6.0, 0.1 M NaCl, 15% PEG 3350.

Crystals of CYP3A4 with SCM-02 were obtained using the hanging-drop vapor diffusion method. Crystallization drops were prepared by mixing 1 µL of protein–ligand solution and 1 µL of precipitant on a coverslip, and the drops were equilibrated against 1 mL of precipitant solution in a 24-well VDX plate (Hampton Research). The protein–ligand solution contained 30 mg/mL protein and 4 mM ligand in storage buffer (with a final DMSO concentration of 10%). The precipitant solution was 0.1 M sodium malonate, pH 7.0, 22% PEG 2000.

CYP3A4 with SCM-08, SCM-18, SCM-24, SCM-25, SCM-26 and SCM-27 were crystallized using the sitting-drop vapor diffusion method in a 24-well Cryschem M plate (Hampton Research). The protein was dialyzed against the glycerol-free storage buffer and concentrated to 40 mg/mL. Crystallization drops were prepared by mixing 1 µL of protein–ligand solution with a final DMSO concentration of 2.5% and 1 µL of precipitant. Co-crystals of complexes were obtained in 0.1 M HEPES, pH 7.0, 10–12% PEG 3350 and grew within 48 hours.

CYP3A5 with SCM-27 was crystallized using the sitting-drop vapor diffusion method in a 24-well Cryschem M plate (Hampton Research). The protein was dialyzed against the storage buffer with 5% glycerol and concentrated to 12 mg/mL. Crystallization drops were prepared by mixing 1 µL of protein–ligand solution with a final DMSO concentration of 2.5% and 1 µL of precipitant. Co-crystals of CYP3A5 with SCM-27 were obtained in 0.2 M Li₂SO₄, 20% PEG 3350 and grew within 48 hours.

Crystals were cryoprotected in reservoir solution supplemented with 15-20% ethylene glycol. All the X-ray diffraction data were collected from single crystals at SER-CAT Beamline 22-ID at the Advanced Photon Source at the Argonne National Laboratory or AMX and FMX Beamlines (17-ID-1 and 17-ID-2) at the National Synchrotron Light Source II at the Brookhaven National Laboratory. Data were processed using XDS⁶⁹ or XDSGUI⁷⁰, scaled using XSCALE or aimless⁷¹, and all CYP3A4 crystals belonged to space group I222 with one molecule in the asymmetric unit. The CYP3A5 structure contained two molecules per asymmetric unit in space group C222₁. The structures were solved by molecular replacement in Phaser⁷² with the Phenix package using PDB ID 1TQN as the search model⁶⁷. Subsequent refinements were performed in Coot⁷³ and Phenix⁷⁴. The 2F_o-F_c maps of each ligand in the asymmetric unit are shown in Supplementary Figs. 5 and 8. All crystallographic figures were generated in PyMOL (Schrödinger). Data processing and refinement statistics are shown in Supplementary Tables 6 and 7.

UV-vis spectroscopy assay

The ligand binding–induced spectral change was monitored using UV-vis spectroscopy. The assay was carried out in Corning 384-well clear-bottom plates (catalog no. 8807BC) as previously reported³⁴. To each well was added 50 µL of protein solution that contained 5 µM CYP3A4 or CYP3A5 and 0–100 µM test compound in 100 mM potassium phosphate buffer, pH 7.4. The final concentration of DMSO was 0.25% for all wells. Absorbance spectra between 350 nm and 700 nm, with a step size of 2 nm, were recorded for each well by using a SpectraMax Plus 384 spectrophotometer (Molecular Devices, San Jose, CA). Each experiment was repeated at least three times.

Data analysis was carried out by subtracting the ligand-free (DMSO) spectrum from the raw data within the same set of measurements. The peak and trough of each resulting difference spectrum were identified, and their difference was defined as the compound-induced spectral change (∆A). The ∆A values determined at different compound concentrations were subsequently plotted against the compound concentrations, and the data were fitted to Eq. 2 with GraphPad Prism (GraphPad):

$$\mathrm{\Delta }A=\mathrm{\Delta }{A}_{\max }\left(\frac{{{\left[P\right]}_T+\left[L\right]}_T+K_s-\sqrt{{\left({\left[P\right]}_T+{\left[L\right]}_T+K_s\right)}^2-{4{\left[P\right]}_T\left[L\right]}_T}}{2{\left[P\right]}_T}\right)$$ 2

where spectral dissociation constant (K_s) and maximum absorbance change $\left(\mathrm{\Delta }{A}_{\max }\right)$ are the values to be fitted, and ${\left[P\right]}_T$ and ${\left[L\right]}_T$ are the known total protein and ligand concentrations, respectively.

Docking studies

Docking of inhibitors to CYP3A4/5 was conducted in a manner similar to that described previously⁷⁵ and was performed using AutoDock Vina version 1.1.1⁷⁶. All water and ligand molecules were removed from the crystal structure by using the Pymol molecular graphics system⁷⁷ before the protein and ligand PDBQT files needed for docking were generated using AutoDockTools version 1.5.6 (http://mgltools.scripps.edu/).

Vivid P450 biochemical inhibition assay

The inhibitory potency of CYP1A2, 2B6, 2C8, 2C9, 2C19, and 2D6 were determined using the corresponding Vivid CYP450 Screening Kits (Thermo Fisher Scientific, catalog nos. P2863, P3019, PV6141, P2860, P2864, and P2972, respectively). Similar assay kits for CYP3A4 and CYP3A5 (Thermo Fisher Scientific, catalog nos. P2858 and P2970, respectively) were also used as controls. All the reagents, including the CYP enzymes, NADPH Regeneration System, and substrates, were supplied with the assay kits.

The assays were carried out in 384-well black plates (Corning, catalog no. 3571BC) with a 25 µL reaction volume. To each well was added 37.5 nL of test compound in DMSO, using an Echo 650 Acoustic Liquid Handler (Beckman Coulter), followed by 12.5 µL of 2× enzyme–substrate mixture. The reaction was initiated by adding 12.5 µL of 2× NADPH Regeneration System. The final concentrations of the reaction components were set up according to the vendor recommendations, with 0.00017–60 µM test compound and 0.25% DMSO. The reaction mixture was incubated for 25 minutes at room temperature before being quenched with 50 µL of 0.5 M Tris base solution. Fluorescence signals were measured using a CLARIOstar Plus Microplate Reader (BMG Labtech, Ortenberg, Germany), using excitation/emission wavelengths of 415 nm/460 nm or 485 nm/520 nm. The data were analyzed using the same method as described for the P450-Glo assay.

Acoustic Ejection Mass Spectrometry (Echo-MS) based metabolic assay

Screening of FDA-approved drugs (APExBio, Huston, TX) for their metabolism by CYP3A4 and CYP3A5 was performed using an approach similar to that published previously⁷⁸. The reaction was carried out in 96-well plates (Corning, catalog no. 3363) with a 25 µL reaction volume and setup the same as the P450-Glo biochemical inhibition assay, with the exceptions that no substrate was added, the CYP3A4/5 protein concentration was 50 nM, and the test compound concentration was 2 µM with a final DMSO concentration of 0.2%. The reaction was quenched with 75 µL of acetonitrile containing 800 ng/mL warfarin at time points 0 and 120 minutes, followed by a centrifugation at 4000 rpm for 20 minutes to remove precipitated protein. 25 µL of supernatant and 25 µL of water were then transferred and mixed in an Echo-qualified 384-well polypropylene plate (Beckman Coulter, catalog no. C74290) for Echo-MS detection. All assays were performed in triplicate, and samples were acquired using the Echo-MS system (SCIEX, Concord, Ontario, Canada) with a modified method as described previously^79,80. The MRM transitions were determined using Discovery Quant software on a SCIEX 6500 + UPLC-Qtrap (SCIEX) system. All data were analyzed with SCIEX OS Analytics Software, and the half-life (T_1/2) was calculated from the linear fit of Ln percentage remaining vs incubation time⁶⁵.

Echo-MS setup included an externalized transducer assembly from an Echo acoustic liquid handler, an open port interface (OPI) connected to a carrier solvent pump, and a transfer capillary leading to a standard IonDrive Turbo V ESI source on an AB Sciex Triple Quad 6500+ system. The carrier liquid consisted of methanol supplemented with 1 mM ammonium fluoride, delivered at a flow rate of 400 µL/min. 10 nL of sample was ejected directly from the microtiter plate wells into the carrier liquid vortex of the OPI, with a sampling frequency of 2 seconds per sample. The ESI source of the triple quadrupole MS instrument operated in positive ionization mode with the following parameters: nebulizer gas (GS1) at 90 psi, heater gas (GS2) at 70 psi, curtain gas at 20 psi, and CAD gas at 9 units. The ion source temperature was set to 500 °C, and the spray voltage was 5500 V. The dwell time was 15 milliseconds, with a pause time of 5 milliseconds. The multiple reaction monitoring (MRM) transitions were monitored in each sample for quantification. These MRM transitions were determined using Discovery Quant software on a SCIEX 6500 + UPLC Qtrap (AB SCIEX) system and are provided in the Source Data file for Supplementary Fig. 17. Q1 and Q3 were operated at unit resolution. Data processing was performed with the following integration parameters: minimum signal-to-noise (S/N) threshold of 2, minimum peak height of 100, the expected retention time was set to a minimum of 0.02 min. All sample acquisition and data analysis were performed using AB Sciex OS Analytics Software (v2.1.6), and the half-life (T_1/2) was calculated from the linear fit of Ln percentage remaining vs incubation time⁶⁵.

Liquid Chromatography Mass Spectrometry (LC-MS) based metabolic assay

The measurement of CYP3A substrates for their time-dependent metabolism by CYP3A4 and CYP3A5 was performed using UPLC-MSMS similar to that published previously⁷⁸. The reaction was carried out identically as described above for Echo-MS detection, with the exception that the reaction was quenched at time points 0, 10, 20, 40, 60, 90, and 120 minutes. All assays were performed in triplicate, and samples were acquired using an Acquity UPLC coupled (Waters, Milford, MA) with SCIEX TQD 6500 system (SCIEX). All data were analyzed with SCIEX OS Analytics Software, and the half-life (T_1/2) was calculated from the linear fit of Ln percentage remaining vs incubation time⁶⁵.

For sample acquisition, appropriate volume of the extract was separated on an Acquity UPLC BEH C18 column (1.7 µm, 2.1 × 50 mm) maintained at 60 °C. The mobile phases consisted of 0.1% formic acid in MilliQ water (solvent A) and 0.1% formic acid in acetonitrile (solvent B). The gradient method utilized a flow rate of 0.9 mL/min with the following profile: 0–0.2 minutes at 99% solvent A and 1% solvent B; 0.2–0.5 minutes with a linear gradient from 99% to 60% solvent A and 1% to 40% solvent B; 0.5–1.2 minutes with a linear gradient from 60% to 5% solvent A and 40% to 95% solvent B; 1.2–1.95 min holding at 5% solvent A and 95% solvent B; 1.95–2 minutes with a linear gradient back to 99% solvent A and 1% solvent B. The first 0.4 minutes of eluate were diverted to waste using an integrated Valco valve to remove salts. The mass spectrometer operated in positive ion mode with electrospray ionization, with the following parameters: ion source gas 1 and 2 pressures at 60 psi, curtain gas at 30 psi, collision gas at 10 psi, ion spray voltage at 5.5 kV, and source temperature at 600 °C. The multiple reaction monitoring (MRM) transitions were monitored in each sample for quantification. All data were analyzed with AB Sciex OS Analytics Software (v2.1.6), and the half-life (T_1/2) was calculated from the linear fit of Ln percentage remaining vs incubation time⁶⁵.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

J.W., S.C.C., and T.C. conceived and organized the project. S.C.C. and J.W. designed and performed HTS experiments, and S.C.C. analyzed the data. J.W. performed biochemical and cell-based inhibition assays, purified proteins, and performed the UV-vis assays. S.N. and J.W. designed and performed crystallography studies and analyzed the data. J.W., S.N., D.M. collected X-ray data. S.C.C. performed docking studies. Y.-H.J. and H.W.O. synthesized all custom-designed compounds. Y.Lei. performed the metabolic stability assay. Y.Lei., Y.Li., J.W., and Y.Z. performed Echo-MS and LC-MS measurements. J.W., S.C.C., S.N., Y.-H.J., and T.C. wrote the manuscript. All authors reviewed the final manuscript.

Peer review

Peer review information

Nature Communications thanks Michael Cameron, Guangbo Ge and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Source data for relevant figures are provided with this paper and are available from the corresponding author upon request. PDB files and structure factors have been deposited in the RCSB PDB under the codes: 9BV5; Crystal structure of human CYP3A4 in complex with SCM-01 (SJ000362065) [10.2210/pdb9BV5/pdb]. 9BV6; Crystal structure of human CYP3A4 in complex with SCM-02 (SJ000388260) [10.2210/pdb9BV6/pdb]. 9BV7; Crystal structure of human CYP3A4 in complex with SCM-03 (SJ000310315) [10.2210/pdb9BV7/pdb]. 9BV8; Crystal structure of human CYP3A4 in complex with SCM-08 (SJYHJ-114) [10.2210/pdb9BV8/pdb]. 9BV9; Crystal structure of human CYP3A4 in complex with SCM-18 (Enamine Z56791366) [10.2210/pdb9BV9/pdb]. 9BVA; Crystal structure of human CYP3A4 in complex with SCM-24 (SJYHJ-106) [10.2210/pdb9BVA/pdb]. 9BVB; Crystal structure of human CYP3A4 in complex with SCM-25 (SJYHJ-075) [10.2210/pdb9BVB/pdb]. 9BVC; Crystal structure of human CYP3A4 in complex with SCM-26 (SJYHJ-110) [10.2210/pdb9BVC/pdb]. 9MS1; Crystal structure of human CYP3A4 in complex with SCM-27 (SJYHJ-111) [10.2210/pdb9MS1/pdb]. 9MS2; Crystal structure of human CYP3A5 in complex with SCM-27 (SJYHJ-111) [10.2210/pdb9MS2/pdb]. Previously published crystal structure data are available in the RCSB PDB database with the following codes: 6MJM; Crystal structure of substrate-free human CYP3A5 [10.2210/pdb6MJM/pdb]. 7LAD; Crystal structure of human CYP3A5 in complex with clobetasol propionate [10.2210/pdb7LAD/pdb]. 8SG5; Crystal structure of human CYP3A5 in complex with clotrimazole [10.2210/pdb8SG5/pdb]. 5VEU; Crystal structure of human CYP3A5 in complex with ritonavir [10.2210/pdb5VEU/pdb]. 7SV2; Crystals structure of human CYP3A5 in complex with azamulin [10.2210/pdb7SV2/pdb]. 1TQN; Crystal structure of substrate-free human CYP3A4 [10.2210/pdb1TQN/pdb] Source data are provided with this paper.

Competing interests

Authors T.C., J.W., Y.-H.J., S.N., and S.C.C. declared the following competing financial interest(s): The authors have the following pending patent related to this manuscript: Chen T, Wang J, Jung Y-H, Nithianantham S. Chai SC. Selective chemical modulation of human cytochrome P450 3A4. US Provisional Application No. 63/664,080. Filing date: June 25th, 2024. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-58749-8.