2.3.1. Crystal Structure of the CYP3A4-DHB Complex

We attempted to co-crystallize CYP3A4 with both bergamottin and DHB but succeeded in obtaining only the DHB-bound crystals. The latter structure was solved to 2.2 Å resolution (Table S1) and contains one DHB molecule in the active site (Figure 4C–E). The psoralen ring of DHB is placed 3.6–5.2 Å above the heme (~15◦ tilt) and parallel to the I-helix, with the carbonyl oxygen H-bonded to the S119 hydroxyl group. As the spectral data predicted, the heme iron remains hexa-coordinate but the water ligand shifts toward the I-helix, altering the Fe-O bond perpendicularity by ~7◦ (Figure 4C). This subtle perturbation in the heme ligation explains why only small spectral changes could be detected during the equilibrium titrations.

The dihydroxygeranyl group curls along the substrate channel and anchors to its wall through the water-mediated H-bond, linking 7- -OH to the R372 and E374 side chains. F57, F215, A370 and M371 create a hydrophobic environment for the aliphatic portion of the geranyl moiety, whereas R212 and F304 flank the psoralen ring. The F304 side group adopts two alternative conformers: one pointing away and another toward the active site, as in ligand-free CYP3A4 (Figure 4E). Since the DHB site is fully occupied, this conformational heterogeneity is likely caused by steric hindrance with the psoralen ring. Considering the largely different affinities for the two binding sites (Table 1), it is reasonable to conclude that DHB occupies the high affinity site. However, the crystallographic binding mode is non-productive, because the nearest to the iron C22 atom of psoralen's phenyl ring (indicated in Figure 4E) is too far for oxidation (>5Å).

#### 2.3.2. Possible Mechanism for DHB Bioactivation

Bergamottin and DHB mediate food-drug interactions primarily through reversible and mechanism-based inhibition of CYP3A4 (IC50 of 2–23 μM) [14,19,20]. The structure-activity studies on natural furanocoumarins showed that a plain tricyclic ring containing the furan moiety is strictly required for inhibiting CYP3A4 activity [21]. This led to a suggestion that bioactivation of bergamottin and DHB proceeds via epoxidation and opening of the furan ring [22,23]. In the crystal structure, however, the furan double bond (indicated in Figure 4E) is too remote and cannot approach the catalytic center without rotation of the psoralen ring, which might be difficult to achieve due to close proximity of the heme and I-helix.

To reconcile our and the previous results [22,23], we propose the following mechanism (Figure 5). Having a more hydrophobic aliphatic chain, bergamottin preferably enters the active site with the geranyl being the closest to the heme iron (step 1), as observed in the 6DWM structure of CYP1A1 [24]. This binding mode is productive and leads to formation of DHB and singly hydroxylated products (step 2). Being more hydrophilic, the oxidation products would dissociate and reenter the active site with the psoralen group approaching the heme (steps 3–4). This orientation could enable association of the second DHB molecule, which may alter the conformation of DHB bound to the high-affinity site (step 5). Occupation of both binding sites would enhance the inhibitory potential, as furanocoumarin dimers are known to inhibit CYP3A4 stronger than DHB [21,25,26]. Nonetheless, the higher-affinity ligands could fully displace DHB and, thus, its metabolic activation would be a prerequisite for the potent inhibition. Among multiple DHB orientations, only some would be suitable for the enzymatic bioactivation of the furan ring (steps 6–7). Alternatively, since CYP3A4 catalysis is highly uncoupled [27], the furan double bond could be oxidized by a by-product, hydrogen peroxide, even when DHB adopts the crystallographic binding mode (step 7a). Upon epoxidation and opening of the furan ring, the radical intermediate could attack the heme (step 8). The epoxide product, in turn, could diffuse out of the active site and modify the apoprotein [22]. This could lead to alterations in conformational dynamics and structural integrity of CYP3A4 and its accelerated degradation [28].

**Figure 5.** Possible DHB binding and bioactivation mechanism. The parent compound, bergamottin, preferably binds to CYP3A4 with the hydrophobic geranyl group entering the substrate channel (step 1). This binding mode is productive and leads to formation of DHB, which dissociates and re-enters the active site with the psoralen group approaching the heme iron (steps 2–3). DHB could have multiple orientations and two binding sites (steps 4–6). In some binding modes, psoralen's furan ring will be close enough to the iron to allow enzymatic bioactivation (step 7). Alternatively, the furan double bond can be oxidized by a by-product, hydrogen peroxide (step 7a). The reactive metabolite(s) could attack and destroy the heme or escape the active site and modify the apoprotein (step 8), leading to alterations in conformational dynamics and structural integrity of CYP3A4. The bioactivation site is indicated by an asterisk.

In summary, this study provided the first direct insights into the interaction of human drug-metabolizing CYP3A4 with three structurally diverse suicide substrates: mibefradil, azamulin and DHB. Only minimal conformational adjustments were needed to accommodate these compounds which, instead, were molded or stretched for a better fit and optimization of protein-ligand contacts. In addition to S119 and R212, frequently involved in the ligand binding, three more polar residues, E308, R372 and E372, were identified as important for the formation/stabilization of the substrate-bound complexes. The CYP3A4-DHB structure identified the high-affinity area where furanocoumarins and other small hydrophobic molecules bearing few polar groups could preferably dock. The binding manner of mibefradil and azamulin, on the other hand, suggested that these compounds could exert their inhibitory action not only upon bioactivation of the newly identified or other oxidation sites, but also by forming slowly dissociable complexes that disallow other substrates to access the catalytic site. Together, the spectral and structural data help better understand the suicide substrate binding and inhibitory mechanism, enable more accurate mapping of the CYP3A4 active site, and can be used to improve computational tools for the prediction of the binding ability, metabolic sites and MBI/DDI potential of newly developed medicines and other chemicals relevant to human health.

#### **3. Materials and Methods**

Mibefradil was obtained from Tocris Bioscience (Minneapolis, MN, USA) azamulin and DHB from Cayman Chemical (Ann Arbor, MI, USA), and bergamottin from Sigma-Aldrich (St. Louis, MO, USA).
