Interaction of CYP3A4 with the inhibitor cobicistat: Structural and mechanistic insights and comparison with ritonavir

Abstract

Cobicistat is a derivative of ritonavir marketed as a pharmacoenhancer for anti-HIV therapy. This study investigated the interaction of cobicistat with the target protein, drug-metabolizing cytochrome P450 3A4 (CYP3A4), at the molecular level using spectral, kinetic, functional, and structural approaches. It was found that, similar to ritonavir, cobicistat directly coordinates to the heme via the thiazole nitrogen but its affinity and the binding rate are 2-fold lower: 0.030 μM and 0.72 s⁻¹, respectively. The newly determined 2.5 Å crystal structure of cobicistat-bound CYP3A4 suggests that these changes arise from the inability of cobicistat to H-bond to the active site S119 and establish multiple stabilizing contacts with the F–F’ connecting fragment, which becomes disordered upon steric clashing with the bulky morpholine moiety. Nonetheless, cobicistat inhibits recombinant CYP3A4 as potently as ritonavir (IC₅₀ of 0.24 μM vs 0.22 μM, respectively) due to strong ligation to the heme and formation of extensive hydrophobic/aromatic interactions via the phenyl side-groups. To get insights into the inhibitory mechanism, the K257 residue, known to be solely and irreversibly modified by the reactive ritonavir metabolite, was substituted with alanine. Neither this nor control K266A mutation changed the extent of time-dependent inhibition of CYP3A4 by cobicistat and ritonavir, suggesting the existence of alternative inactivation mechanism(s). More importantly, K257 was found to be functionally important and contributed to CYP3A4 allosterism, possibly by modulating protein-ligand interactions through conformational dynamics.

1. Introduction

Cobicistat and ritonavir (Fig. 1) are potent inhibitors of cytochrome P450 3A (CYP3A)¹ enzymes involved in drug metabolism. Ritonavir was originally developed as an HIV-1 protease inhibitor [1] but later found to exert a pharmacoenhancing (boosting) effect on co-administered antiretroviral drugs predominantly metabolized by CYP3A [2]. The ritonavir-containing combination therapy enabled more effective dosing regiments of antiviral medications and led to advances in HIV treatment [3]. However, ritonavir has poor bioavailability and multiple off-target activities that could lead to adverse side effects and a wide range of clinically significant drug-drug interactions [4].

Fig. 1.

Chemical structures of cobicistat and ritonavir.

To address these limitations, a new pharmacoenhancer was developed, cobicistat [5,6]. This drug is a derivative of ritonavir that lacks the backbone hydroxyl group critical for the binding to HIV protease and, as a result, is devoid of anti-HIV activity. Another distinction is the presence of a bulky and more polar morpholine side-group instead of valine. This structural change decreases cross-reactivity with other CYPs and increases aqueous solubility, while the inhibitory potency for CYP3A remains sufficiently high. Therefore, cobicistat is considered as a more specific CYP3A inhibitor with improved physicochemical properties and safety profile [7]. Recently, the usage of CYP3A-based pharmacoenhancers has been expanded. Ritonavir boosting is currently used to improve the efficacy of antiviral therapy for HCV and COVID-19 patients [8–10], whereas cobicistat is being tested as a booster for anticancer and antimalarial drugs [11,12].

CYP3A4 is the major and most clinically relevant CYP3A isoform, whose interaction with ritonavir has been extensively investigated (reviewed in Ref. [13]). The current notion is that ritonavir inhibits CYP3A4 via multiple mechanisms that include direct ligation to the heme and time-dependent (mechanism-based) inactivation. In vitro studies on isolated recombinant CYP3A4 showed that ritonavir can easily displace substrates from the active site and, upon coordination to the heme via the thiazole nitrogen, decreases the heme redox potential and precludes catalytic turnover [14]. The predominant inactivation pathway is thought to proceed via bioactivation of ritonavir and formation of reactive metabolites that covalently modify the heme or apoprotein [15,16]. The chemical nature of the reactive metabolites remains unknown, but the isopropyl thiazole moiety, strictly required for potent inhibition, is the likely site of oxygen attack [1,16,17].

Our main interest is elucidation of structure-activity relationship (SAR) of ritonavir-like compounds for rational inhibitor design. Toward this goal, we determined crystal structures of ritonavir-bound CYP3A4 in two different crystal forms [14,18] and defined the contribution of its structural determinants to the binding affinity and inhibitory potency using close analogues [19–22]. In contrast to well-studied ritonavir, interaction of CYP3A4 with cobicistat, widely used in anti-HIV therapy for over a decade, has not been studied at the molecular level. To fill this knowledge gap and gain further insights into SAR of ritonavir-like inhibitors, we investigated the complex formation between isolated recombinant CYP3A4 and cobicistat using spectral, biochemical, and structural approaches. The crystal structure of the inhibitory complex provided the first direct insights into the binding mode of cobicistat, helped interpret experimental results, and highlighted differences with the ritonavir-bound complex. Moreover, we identified K257 as a functionally important residue contributing to CYP3A4 allosterism but could not confirm its role in time-dependent inactivation under our experimental conditions.

2. Materials and methods

Protein Expression and Purification -

Codon-optimized full-length and Δ3–22 human CYP3A4 were produced as reported previously [18] and used for assays and crystallization, respectively. The K257A and K266A mutations were introduced using QuikChange Site-directed mutagenesis kit (Agilent) and verified by sequencing. Both variants were expressed and isolated using protocols developed for purification of the WT enzyme [18]. Expression plasmid for the full-length rat cytochrome P450 reductase (CPR) was received from Dr. Charles Kasper. CPR was expressed in E. coli C41 (DE3) strain and purified according to the previously reported procedure [23].

Spectral Binding Titrations –

Equilibrium ligand binding to CYP3A4 was monitored in a Cary 300 spectrophotometer at ambient temperature in 0.1 M phosphate pH 7.4, containing 20 % glycerol and 1 mM dithiothreitol. Cobicistat (Cayman Chemical), ritonavir (Toronto Research Chemicals) and ketoconazole (Tocris Bioscience) were dissolved in DMSO to prepare 50 mM stock solutions, which were further diluted with DMSO to 0.25–5.0 mM working solutions. The inhibitors were added to the 1.5–2 μM protein sample in small aliquots, with the final solvent concentration <2 %. Spectral dissociation constant (K_s) was determined from quadratic fits to titration curves using KaleidaGraph software, where stoichiometry was used as a fitting parameter.

Kinetics of Cobicistat Binding –

Kinetics of cobicistat binding to CYP3A4 were monitored in a SX.18 MV stopped flow apparatus (Applied Photophysics, UK) at ambient temperature in 0.1 M phosphate, pH 7.4. CYP3A4 (2 μM) was mixed with an equal volume of 1–50 μM solution of cobicistat, and conversion of the heme iron to the low-spin form was followed at 425 nm. For each condition, five measurements were conducted, and the average was used for fitting. Kinetic data were analyzed with manufacturer’s PROKIN software using double exponential equation: $\mathrm{Y}={\mathrm{A}}_1\bullet {\mathrm{e}}_1^{\bullet \mathrm{k}\mathrm{X}}+{\mathrm{A}}_2\bullet {\mathrm{e}}_2^{\bullet \mathrm{k}\mathrm{X}}+\mathrm{C}$, where $\mathrm{Y}$ is absorbance, $\mathrm{X}$ – time, ${\mathrm{k}}_1$ and ${\mathrm{k}}_2$ -observed rate constants for the fast and slow phases, ${\mathrm{A}}_1$ and ${\mathrm{A}}_2$– total absorbance change during each phase, and $\mathrm{C}$ – initial absorbance.

Thermal Denaturation –

Melting curves were recorded in 0.1 M phosphate, pH 7.4, in a Cary 300 spectrophotometer. Before measurements, CYP3A4 (1 μM) was incubated with 10 μM inhibitor or DMSO (2 % final concentration) for 10 min at room temperature. Thermal denaturation was monitored at 260 nm using a 0.2 °C measurement step, 0.9 °C/min ramp rate, and 50–73 °C temperature range. A denaturation midpoint (melting temperature; ${\mathrm{T}}_{\mathrm{m}}$) was determined from non-linear fitting to a Boltzmann sigmoidal curve using the equation: $\mathrm{Y}={\mathrm{A}}_{\mathrm{m}\mathrm{i}\mathrm{n}}+\left({\mathrm{A}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{A}}_{\mathrm{m}\mathrm{i}\mathrm{n}}\right)/\left(1+\mathrm{e}\mathrm{x}\mathrm{p}\left({\mathrm{T}}_{\mathrm{m}}-\mathrm{X}/\right.\right.\mathrm{S}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}\left.)\right)$, where $\mathrm{Y}$ - absorbance; $\mathrm{X}$ - temperature; ${\mathrm{A}}_{\text{min}}$ and ${\mathrm{A}}_{\text{max}}$ - absorbance at the lowest and highest temperature, respectively; and Slope - steepness of the curve.

H₂O₂-dependent Heme Depletion –

Heme bleaching in ligand-free and inhibitor-bound CYP3A4 was monitored in a Cary 300 spectrophotometer at ambient temperature in 0.1 M phosphate buffer, pH 7.4. CYP3A4 (1.6 μM) was incubated with 16 μM inhibitors for 10 min before addition of 10 mM H₂O₂ (final concentration). Heme decay kinetics were followed at 420 nm for 2 h. The percentage of heme destroyed at the end of the reaction was calculated relative to that in ligand-free CYP3A4 (100 % decay). The initial rates were calculated from a linear portion of kinetic traces using Igor Pro software (WaveMetrics).

CYP3A4 Activity Assay -

Activity of WT and mutant CYP3A4 toward 7-benzyloxy-4-(trifluoromethyl)coumarin (BFC) was measured fluorometrically at 37 °C in a soluble reconstituted system containing 0.1 M potassium phosphate, catalase and superoxide dismutase (2 Units/ml each), 0.0025 % CHAPS (3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonate), 0.1 μM CYP3A4 and 0.15 μM CPR. Protein solution was incubated with 1–60 μM BFC for 1 min prior to addition of 100 μM NADPH. Formation of the fluorescent product, 7-hydroxy-4-trifluoromethyl coumarin (HFC), was monitored for 2 min in a Hitachi F400 fluorimeter with the excitation and emission wavelengths of 404 nm and 500 nm, respectively. Within this time interval, fluorescence change was linear. The amount of product formed was calculated from linear fittings and a calibration curve built using commercial HFC (Santa Cruz Biotechnology). Measurements were done in triplicate. Kinetic parameters were derived from sigmoidal fittings to the reaction velocity vs [BFC] plots. For data analysis, the Hill equation was used: $\mathrm{Y}={\mathrm{V}}_{\mathrm{m}\mathrm{a}\mathrm{x}}\mathrm{*}{\mathrm{X}}^{\mathrm{n}}/\left({\mathrm{S}}_{50}^{\mathrm{n}}+{\mathrm{X}}^{\mathrm{n}}\right.)$, where Y is enzyme activity, $\mathrm{X}$-substrate concentration, ${\mathrm{V}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ –maximum velocity, S₅₀ - concentration of substrate giving half maximal velocity, and n – cooperativity factor (Hill coefficient).

Inhibitory Potency Assays –

Inhibitory potency of the investigated compounds was assessed for the BFC debenzylase activity of CYP3A4 as described above. However, NADPH was replaced with a NADPH-regenerating system consisting of 1 mM glucose-6-hosphate, 100 μM NADPH, and 2 Units/ml glucose-6-phosphate dehydrogenase (final concentration). The protein solution was mixed with the inhibitor or DMSO and incubated for 2 min at room temperature, followed by addition of 40 μM BFC. One min later, the NADPH regeneration system was added and, after an additional 1 min incubation at 37 °C, formation of HFC was monitored in a Hitachi F400 fluorimeter as described above. For time-dependent inactivation, the NADPH-regenerating system was added to the protein mixture prior to BFC, and the sample was incubated at room temperature for 20 min. All measurements were done in triplicate. Reaction rates were calculated from linear fittings, and the remaining activity was calculated relative to the DMSO-containing control (100 % activity). The IC₅₀ values were derived from [% remaining activity] vs [inhibitor] plots by fitting the data to a four-parameter logistic nonlinear regression equation: $\mathrm{y}=\mathrm{D}+((\mathrm{A}-\mathrm{D})/(1+{10}^{(\mathrm{X}-\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{C})\bullet \mathrm{B}}))$, where A and D are the maximal and minimal activity, respectively, $\mathrm{X}$ - inhibitor concentration, $\mathrm{B}$ - a slope factor, and $\mathrm{C}$ is IC₅₀.

Determination of the X-ray Structure –

Δ3–22 CYP3A4 was co-crystallized with cobicistat at ambient temperature by a sitting drop vapor diffusion method. Protein (70 mg/ml) in 0.1 M potassium phosphate was incubated with a 3-fold molar excess of cobicistat for 10 min and centrifuged to remove the precipitate. Cobicistat-bound CYP3A4 (0.5 μl) was mixed with an equal volume of 50 % solution #18 from Molecular Dimensions Morpheus II kit: 0.5 mM MgCl₂·4H₂O, 0.5 mM CoCl₂·6H₂O, 0.5 mM NiCl₂·6H₂O, 0.5 mM Zn(CH₃CO₂)₂·2H₂O, 20 % hexanetriol, and 12.5 % PEG 4000 (final concentration). The mixture was equilibrated at room temperature against the same solution in a CrystalQuick 96 well plate (Greiner). Crystals appeared the next day and, after harvesting, were cryoprotected with Paratone-N oil and frozen in liquid nitrogen. X-ray diffraction data were collected at the Advanced Light Source beamline 8.2.2. Crystal structure was solved by molecular replacement with PHASER [24] and the 3NXU structure as a search model. The ligand was built with eLBOW [25] and manually fit into the density with COOT [26]. The initial model was rebuilt and refined with COOT and PHENIX [25]. Ligand occupancy was refined with PHENIX after its value was set below unity. Polder omit electron density map and correlation coefficients were calculated with PHENIX using default settings and verified according to Liebschner et al. (Table S1) [27]. Data collection and refinement statistics are summarized in Table 1. The atomic coordinates and structure factors were deposited to the Protein Data Bank (PDB) with the ID code 9BBB.

Table 1.

Data collection and refinement statistics.

Data statistics
Space group	I222
Unit cell parameters	a = 77 Å, b = 102 Å, c = 126 Å; α, β, γ = 90°
Molecules per asymmetric unit	1
Resolution range	(Å) 79.31–2.50 (2.64–2.50) a
Total reflections	81,298 (12,119)
Unique reflections	17,357 (2500)
Redundancy	4.7 (4.8)
Completeness	99.3 (100.0)
Average	I/σI 9.9 (1.6)
Rmerge	0.066 (0.704)
Rpim	0.034 (0.348)
CC ½	0.997 (0.808)
Refinement statistics
R/R free b	22.7/25.9
Number of atoms:
Protein	3518
Solvent	31
R.m.s. deviations:
Bond lengths, Å	0.003
Bond angles, °	0.753
Wilson B-factor, Å2	76
Average B-factor, Å2:
Protein	80
Ligand	111
Ramachandran plotc (residues; %)
Preferred	408 (94.7 %)
Allowed	23 (5.3 %)
Outliers	0

^aValues in brackets are for the highest resolution shell.

^bR_free was calculated from a subset of 5 % of the data that were excluded during refinement.

^cAnalyzed with PROCHECK.

3. Results

Cobicistat is a High-affinity Type II Ligand –

Spectral measurements showed that cobicistat induces a red shift in the Soret band of CYP3A4 (type II spectral change), indicative of heme iron coordination to the nitrogen atom. Similar to ritonavir-bound CYP3A4 (Fig. S1A), the cobicistat-bound protein has absorption maximum at 421 nm (Fig. 2A). The distinctive features were incomplete reduction by sodium dithionite and the equal amplitude of the α- and β-bands of the ferrous form (green spectra). In contrast, ritonavir-bound CYP3A4 can be fully reduced by sodium dithionite and its ferrous species has a more pronounced β-band (Fig. S1A). Spectral dissociation constant (K_s) of cobicistat, estimated based on titration data (Fig. 2B and C), was nearly 2-fold higher compared to that of ritonavir: 0.030 μM vs 0.017 μM, respectively (Table 2). Nonetheless, similar to ritonavir, cobicistat remained bound to CYP3A4 during gel filtration chromatography and after multiple dilution/concentration cycles. This demonstrates that cobicistat is a high-affinity type II ligand that binds tightly to CYP3A4.

Fig. 2.

Spectral changes induced by cobicistat in CYP3A4. A – Equilibrium titration of CYP3A4 with cobicistat. Ferric ligand-free CYP3A4 (2 μM; solid black spectrum) was titrated with increasing concentrations of cobicistat (dotted black spectra). Upon saturation, the cobicistat-bound form (red spectrum) was reduced by sodium dithionite (green spectrum) and reacted with carbon monoxide (blue spectrum). Inset is a magnified view at the long-wavelength region showing absorption maxima and the relative amplitude of the characteristic α- and β-bands in the spectrum of ferrous cobicistat-bound CYP3A4. B and C-Difference absorbance spectra and titration plot with quadratic fitting, respectively. The derived K_s is 0.030 μM.

Table 2.

Binding affinity and inhibitory potency of the investigated compounds.

Ks (μM)		IC50 (μM)
		-preincubation	+ preincubation	change
Cobicistat
WT	0.030 ± 0.004	0.24 ± 0.04	0.12 ± 0.03	↓ 50 %
K257A	0.033 ± 0.003	0.30 ± 0.03	0.11 ± 0.02	↓ 64 %
K266A	0.028 ± 0.002	0.36 ± 0.05	0.13 ± 0.02	↓ 64 %
Ritonavir
WT	0.017 ± 0.003	0.22 ± 0.03	0.12 ± 0.03	↓ 45 %
K257A	0.018 ± 0.002	0.27 ± 0.02	0.11 ± 0.02	↓ 59 %
K266A	0.017 ± 0.002	0.30 ± 0.03	0.11 ± 0.01	↓ 63 %
Ketoconazole
WT	0.047 ± 0.003	0.35 ± 0.06	0.55 ± 0.04	↑ 57 %
K257A	0.052 ± 0.005	0.28 ± 0.03	0.37 ± 0.07	↑ 32 %
K266A	0.050 ± 0.004	0.25 ± 0.04	0.36 ± 0.05	↑ 44 %

Kinetics of Cobicistat Binding –

Kinetics of cobicistat ligation to CYP3A4 was followed at 425 nm in a stopped flow spectrophotometer. Kinetic traces recorded at increasing cobicistat concentrations are shown in Fig. 3A. All reactions were biphasic, with an equal percentage of the absorbance change occurring during the fast and slow phases. The limiting rate constants were derived from the plots of the observed rate constants (k_obs) vs cobicistat concentration (Fig. 3B) and were equal to 0.72 ± 0.01 s⁻¹ and 0.078 ± 0.005 s⁻¹ for the fast and slow phases, respectively. Both values were 2-to-3-fold lower than those measured for ritonavir under similar conditions [14], meaning that cobicistat ligates to CYP3A4 considerably slower. The dissociation constants (K_d) for the fast and slow phases estimated from kinetic data were 0.93 ± 0.08 μM and 2.9 ± 0.6 μM, respectively. Biphasicity of the ligand binding kinetics might be due to conformational heterogeneity of CYP3A4. As shown earlier, in aqueous solutions ligand-free CYP3A4 exists as a mixture of monomers, dimers and higher molecular weight oligomers [28], among which some conformers could interact with the ligand faster than the other (conformational selection mechanism). Another factor that could affect the ligand binding kinetics is the ability of cobicistat and ritonavir to enter the active site with the isopropyl thiazole end-group. This is evidenced by the fact that the isopropyl thiazole moiety in both inhibitors is the primary site of CYP3A4-dependent metabolism [17,29]. Moreover, as metabolic profiles suggest [29,30], cobicistat and ritonavir are flexible and have some degree of freedom inside the CYP3A4 active site. The necessity for ligand re-entry/reorientation to allow heme ligation could further complicate the binding kinetics and contribute to the disparity between K_s and K_d.

Fig. 3.

Formation and properties of the CYP3A4-cobicistat complex and comparison with the ritonavir-bound form. A and B - Kinetics of cobicistat binding. A – Kinetic traces recorded at 425 nm in a stopped flow spectrophotometer upon mixing of 1 μM CYP3A4 with 0.5–25 μM cobicistat (final concentration). Each trace represents an average of five individual measurements. Biexponential fits are shown as solid lines. The dashed horizontal line corresponds to the maximal absorbance at the end of the cobicistat binding reaction (determined in a separate experiment). B – Plots of the observed rate constants (k_obs) for the fast and slow phases vs cobicistat concentration. The k_lim and K_d values derived from hyperbolic fittings were 0.72 s⁻¹ and 0.93 μM, respectively, for the fast phase and 0.078 s⁻¹ and 2.9 μM for the slow phase. C – Melting curves of CYP3A4 in the presence of inhibitors and solvent DMSO, used as a control. Melting temperatures (${\mathrm{T}}_{\mathrm{m}}$) are indicated. D – Kinetcs of H₂O₂-dependent heme depletion in ligand-free and inhibitor-bound CYP3A4. The initial rates were 5.8 × 10⁻³ min⁻¹ for ligand-free CYP3A4 and 0.43 × 10⁻³ and 0.39 × 10⁻³ min⁻¹ for cobicistat- and ritonavir-bound forms, respectively. The percentage of heme destroyed within 2 h is indicated. E – Inhibitory plots for the BFC debenzylation activity of CYP3A4 measured without and with preincubation with NADPH to assess the contribution of time-dependent inactivation. The derived IC₅₀ values are listed in Table 2.

Thermostability and Heme Accessibility of Cobicistat-bound CYP3A4 –

Tighter ligands tend to increase protein stability to a higher degree. Thermal denaturation experiments showed that, relative to the DMSO-containing control, cobicistat increased the melting temperature (${\mathrm{T}}_{\mathrm{m}}$) of CYP3A4 nearly as much as ritonavir: by 4.8C° vs 5.4C°, respectively (Fig. 3C).

Bulky ritonavir-like molecules can inhibit the activity of CYP3A4 not only by coordinating to the heme and decreasing the redox potential but also by sterically blocking the active site and preventing the substrate access to the catalytic center. To compare heme accessibility in cobicistat- and ritonavir-bound CYP3A4, kinetics of heme bleaching was measured in the presence of hydrogen peroxide, a small oxidizing agent that can easily reach and destroy the heme without altering CYP structure [31]. As seen from Fig. 3D, association of both inhibitors drastically slowed down heme decay. Relative to ligand-free CYP3A4, over 70 % of the cofactor in the inhibitor-bound protein remained intact. Due to complexity and multiphasicity of the heme decay process, only the initial rates were compared. In the presence of cobicistat and ritonavir, the rates were similar and an order of magnitude lower than for ligand-free CYP3A4: ~0.4 × 10⁻³min⁻¹ vs 5.8 × 10⁻³ min⁻¹, respectively.

Inhibitory Potency of Cobicistat –

Inhibitory potency of cobicistat was measured for the BFC debenzylase activity of CYP3A4 and was close to that of ritonavir: IC₅₀ of 0.24 μM and 0.22 μM, respectively. If substrate was added after 20 min preincubation of CYP3A4 with the inhibitor and NADPH, the IC₅₀ value decreased by 2-fold (Table 2), indicative of time-dependent inactivation. The same trend was observed for ritonavir. Thus, under both experimental settings, cobicistat displays similar inhibitory potential despite the lower affinity and binding rate. Considering the inhibitory, spectral and kinetic results, it can be concluded that cobicistat binds to CYP3A4 similarly but less optimal than ritonavir. To confirm this and to get direct insights into the binding mode of cobicistat, the X-ray structure of the CYP3A4-cobicistat complex was determined.

Crystal Structure of the CYP3A4-cobicistat Complex –

Although cobicistat was identified as a high-affinity ligand, it was challenging to co-crystallize it with CYP3A4. The first trials were conducted using a microbatch under oil crystallization setup, previously employed for co-crystallization of CYP3A4 with ritonavir [14]. This approach was unsuccessful because cobicistat dissociated from CYP3A4 and, possibly, partitioned into paraffine oil covering the protein solution, thereby promoting formation of ligand-free CYP3A4 crystals. The I222 crystals of cobicistat-bound CYP3A4 were eventually obtained using the vapor diffusion method and the X-ray structure was solved to 2.5 Å resolution. Data collection and refinement statistics are shown in Table 1.

In the crystal structure, cobicistat binds to the active site and, as spectral data predicted, coordinates to the heme iron via the thiazole nitrogen, with the Fe–N distance of 2.15 Å (Fig. 4A and B). Despite the high occupancy of cobicistat (0.92), only its lower part closest to the heme is well defined, whereas the tail portion is partially disordered due to high thermal motion. The phenyl group adjacent to the thiazole moiety (Phe-1) inserts into a grove above the I-helix to form hydrophobic and aromatic interactions with F108, M114, V240, F241, I301, and F304, distorting the I-helix by ~2.0 Å. The second phenyl group (Phe-2) lies near the heme and thiazole ring (3.3–3.6 Å distance) and close to the guanidinium group of R105 (4.5 Å away). This promotes π-π and cation-π aromatic interactions, further strengthening the complex. As seen from the overlay with ligand-free CYP3A4 (5VCE; Fig. 4B), there is no steric hindrance between Phe-2 and the nearby 369–371 fragment, but residues 211–217 from the F–F’ connecting fragment become fully disordered due to clashing with the morpholine ring. As a result, the morpholine moiety forms only two van der Waals contacts with F108 and F220. The isopropyl thiazole end-group, on the other hand, is engaged in multiple interactions with the surrounding residues, forming T-shaped π-π stacking and S-aromatic interactions with F57 and van der Waals contacts with Y53, L221, E374 and G481. However, these interactions are not strong enough to minimize thermal motion of the end-group. Finally, there is one repulsive long-range polar interaction between the carbonyl oxygens of cobicistat (O11 atom) and P106 (4.0 Å away).

Fig. 4.

Crystal structure of cobicistat-bound CYP3A4 (PDB ID 9BBB) and comparison with the ligand-free (5VCE) and ritonavir-bound models (5VC0). A – Crystallographic binding mode of cobicistat. Blue and green mesh are 2F_o-F_c and polder omit electron density maps contoured at 1σ and 3σ level, respectively. B – Positioning of cobicistat in the active site. The ligand-free structure is shown to demonstrate clashing of the morpholine ring with the F–F′ fragment, part of which becomes fully disordered in cobicistat-bound CYP3A4. C - Comparison of CYP3A4 bound to cobicistat (in gray/cyan) and ritonavir (in beige/orange). D – Another view at the ligands showing that ritonavir but not cobicistat can form stabilizing H-bonds with the S119 hydroxyl group. E – Ligands colored according to the B-factor. Atoms with high and low B-factors are in red and blue, with the respective thresholds of 70 Ų and 140 Ų, respectively. In cobicistat, over half of the atoms have high thermal motion but only the isopropyl thiazole end-group in ritonavir.

To better understand why cobicistat has a lower binding affinity and tends to dissociate during crystallization, we compared cobicistat- and ritonavir-bound structures. Because crystal packing can perturb the ligand binding mode [18], the isomorphous I222 structure of ritonavir-bound CYP3A4 (5VC0) was used for comparative analysis. Structural superposition shows (Fig. 4C) that the inhibitors’ orientation is generally similar but with a few notable distinctions. In particular, (i) the ritonavir backbone curves and bends to a higher degree, leading to a more compact fold stabilized through H-bonds with S119 (Fig. 4D); (ii) ritonavir’s Phe-1 distorts the I-helix to a larger extent (by 3.0 Å) and comes closer to and orients more favorably toward F304 to allow π-π stacking interactions lacking in the cobicistat-bound structure; (iii) Phe-2 clashes with and displaces A370, and lies near parallel to and with a larger overlap with the heme but, unlike Phe-2 of cobicistat, cannot form cation-π interactions with the R105 guanidine; (iv) due to a lower positioning in the active site, ritonavir does not clash with and, instead, forms multiple hydrophobic and van der Waals contacts with the F–F′ fragment; (v) the end-group of ritonavir mediates a larger number of hydrophobic and van der Waals contacts, including aromatic interactions with F57, F213 and F215; and (vi) the carbonyl of R372 rather than P106 forms the repulsive long-range polar contact with ritonavir.

Collectively, these structural differences suggest that replacement of the valine group with morpholine rather than elimination of the backbone hydroxyl weakens the complex and leads to a 2-fold drop in the binding affinity of cobicistat. As studies on desoxyritonavir analogues showed [22], removal of the backbone hydroxyl is beneficial because it increases the backbone flexibility and allows a better fit into the CYP3A4 active site. In contrast, by clashing with the F–F′ fragment, the bulky morpholine disrupts/prevents formation of stabilizing ligand-protein contacts and increases the ligand’s thermal motion (Fig. 4E). Insignificant differences in the IC₅₀, ${\mathrm{T}}_{\mathrm{m}}$ and heme accessibility (Fig. 3C–E; Table 2), in turn, provide further evidence that strong heme ligation and extensive hydrophobic/aromatic interactions mediated by the phenyl side-groups are the key factors that define the inhibitory strength of ritonavir-like compounds and their ability to stabilize the CYP3A4 structure [22].

Structural Role of K257 –

Rock et al. reported that, upon ritonavir bioactivation, its reactive metabolite specifically reacts with K257 [16]. The fact that only one out of 38 lysines present in CYP3A4 undergoes covalent modification by the oxidation product was proposed to be due to a strategic location of K257 in the egress channel. To better understand the role of K257 in structure/function of CYP3A4, the protein architecture near this residue was compared in ligand-free (water--coordinated), ritonavir- and cobicistat-bound protein (Fig. 5). In ligand-free CYP3A4 (column A), K257 is H-bonded to Q279 and D292, forming a triad that brings the G-, H- and I-helices closer and stabilizes the fold. In this protein form, there is no passage from the active site cavity to the solvent channel formed by the F’-, G- and I-helices. Upon binding of ritonavir (column B), the H-bond between K257 and Q272 becomes broken and the passage to the solvent channel opens up. The association of cobicistat (column C) not only widens the passage but also leads to partial disorder of the G-helix, possibly due to a loss of the K257-centered H-bonding triad. It is unclear whether the newly formed exit channel could accommodate molecules as large as ritonavir and allow product egress. Nonetheless, the ligand-dependent structural changes suggest that K257 could act as a gating residue and regulate the channel opening via inter-helical H-bonding interactions. To test whether elimination of the polar side chain of K257 affects catalytic activity of CYP3A4 and its inhibition by ritonavir and cobicistat, this residue was replaced with alanine. The nearby K266 from the G-H connecting loop was mutated to alanine as a control.

Fig. 5.

Ligand-dependent changes in the vicinity of K257. Columns A - C – The X-ray models of ligand-free (5VCE), ritonavir-(5VC0) and cobicistat bound CYP3A4 (9BBB), respectively. Selected helices are color-coded and labeled. Cavities with semitransparent surface (in pale blue) were displayed with PYMOL, with the cavity detection radius and cavity detection cutoff set at 5 solvent radii. Panels in the top and middle rows are the front and side views of the CYP3A4 molecule, respectively. The lower row panels are magnified views at the K257 site. Residues interacting with K257 as well as the nearby K266 are shown in stick representation; blue dotted lines are H-bonds. In the middle row, arrows point at the junction of the active site cavity with the solvent channel formed between the G-, H- and I-helices. This junction is closed in ligand-free CYP3A4, partially opens upon ritonavir association, and widens in cobicistat-bound CYP3A4. In the lower row, arrows point at the channel exit.

Mutational Effect on the Functional Activity of CYP3A4 –

The K257A and K266A variants were spectrally indistinguishable from the wild type (WT) but had 28 % and 44 % higher V_max, respectively, for the BFC debenzylase activity (Fig. 6; Table 3). Sigmoidal dependence of the velocity on substrate concentration was observed for WT and K266A CYP3A4 (Hill coefficient (n_H) of 1.8 and 1.9, respectively), manifesting cooperativity in substrate binding. In contrast, the respective plot for the K257A mutant was hyperbolic (n_H of 1.0). The R² coefficients (goodness of fit) for hyperbolic and sigmoidal fittings are compared in Fig. S2. A switch from sigmoidal to non-sigmoidal saturation kinetics caused by elimination of K257 implies that this remote surface residue is functionally important and contributes to the allostericity of CYP3A4. Because BFC does not cause spin transition in CYP3A4, it was not possible to spectrally determine using this substrate whether the K257A mutation changes the multi-site to single-site ligand binding.

Fig. 6.

Mutational effect on the BFC debenzylase activity of CYP3A4. The activity was measured in a soluble reconstituted system with CPR by following the formation of a fluorescent product, HFC. Parameters derived from sigmoidal fittings are listed in Table 3.

Table 3.

Mutational effect on the BFC debenzylase activity of CYP3A4.

	S50 (μM)	Vmax (min−1)	nH
WT	9.3 ± 0.4	0.025 ± 0.005	1.8 ± 0.1
K257A	8.4 ± 0.5	0.036 ± 0.008	1.0 ± 0.1
K266A	6.1 ± 0.1	0.032 ± 0.003	1.9 ± 0.1

Mutational Effect on the Inhibitory Potency of Cobicistat and Ritonavir -

Cobicistat and ritonavir also could not be used for assessing changes in the allostericity because they have only one binding site in the WT and mutant CYP3A4 (stoichiometry of 1.1 and 1.0, respectively). Moreover, their affinity is virtually unaffected by the K257 and K266 elimination (Fig. 7 and S1A, B; Table 2). However, compared to WT, the inhibitory potency of both compounds for K257A and K266A CYP3A4 was 25–50 % lower under conditions without preincubation with NADPH (Fig. 8; Table 2). With preincubation, the mutants were inhibited as effectively as WT. Consequently, when the IC₅₀ values for each protein are compared in a pairwise fashion, it is evident that preincubation with NADPH had a more pronounced effect on the mutants: 59–64 % decrease in IC₅₀ relative to 45–50 % for WT. Thus, the K257A and K266A variants behave similarly toward the inhibitors and neither mutation protects from time-dependent inactivation, which suggests the existence of alternative mechanisms in addition to the covalent K257 modification.

Fig. 7.

Spectral changes induced by cobicistat in K257A (A) and K266A (B) CYP3A4. Main panels show spectral changes observed during equilibrium titrations of 2 μM ferric CYP3A4 (solid black spectra) with increasing concentrations of cobicistat (dotted black spectra). The ligand-bound forms (red spectra) were reduced by sodium dithionite (green spectra) and reacted with carbon monoxide (blue spectra). Left and right insets show the difference absorbance spectra and titration plots with quadratic fittings, respectively. Spectral changes induced by ritonavir in K257A and K266A CYP3A4 are shown in Figs. S1A and B. The derived K_s values are listed in Table 2.

Fig. 8.

Inhibition of the BFC debenzylase activity of K257 and K266A CYP3A4 by cobicistat and ritonavir. The activity was measured under conditions without and with preincubation with NADPH to assess the contribution of time-dependent inactivation. The derived IC₅₀ values are listed in Table 2.

To validate our in-vitro inhibitory assay, the antifungal drug ketoconazole was used as a negative control. Ketoconazole acts as a type II ligand for CYP3A4 and displays the binding affinity and IC₅₀ close to those for cobicistat and ritonavir (Figs. S3 and S4; Table 2). Regardless of whether mutations were present or not, preincubation of CYP3A4 with ketoconazole and NADPH prior to substrate addition led to a ~30–60 % increase rather than decrease in IC₅₀. This result supports the notion that ketoconazole is a reversible inhibitor that does not cause time-dependent inactivation of CYP3A4 [32] and strengthens conclusions on ritonavir and cobicistat.

4. Discussion

This study investigated the interaction of CYP3A4 with the pharmacoenhancer cobicistat at the molecular level. Along with the experimental results (Figs. 2 and 3), we report the first crystal structure of cobicistat-bound CYP3A4 that revealed the inhibitor binding mode and distinctions with the ritonavir-bound complex (Fig. 4). In accord with our previous work on ritonavir analogues [14,19–22], this study suggests that interactions mediated by the end-moiety mostly contribute to the binding affinity, whereas the strength of heme ligation and contacts established by the phenyl side-groups define the inhibitory strength. Also, we conclude that steric clashing between the morpholine moiety and the F–F’ fragment, disallowing formation of stabilizing protein-ligand contacts, and the lack of H-bonding interactions with the active site S119 lead to a 2-fold decrease in the affinity and binding rate of cobicistat, with no feasible effect on its inhibitory potency.

Chemical similarity (Fig. 1), resemblance of the binding modes (Fig. 4C) and equal inhibitory potency of cobicistat and ritonavir (Fig. 3E; Table 2) point to a similar mechanism of time-dependent inactivation. While the reactive metabolites are yet to be identified, one oxidation product of ritonavir was found to selectively attack and form a covalent adduct with K257, supposedly located in the product egress channel [16]. Structural analysis showed (Fig. 5) that, indeed, a passage from the active site to the solvent channel near K257 becomes open in the inhibitor-bound CYP3A4 and could be regulated through K257-mediated H-bonding interactions. To disrupt the inter-helical contacts and prevent an adduct formation, we replaced K257 with alanine. Both K257A CYP3A4 and the control K266A mutant displayed higher activity toward BFC, but only elimination of K257 changed the sigmoid dependency of the reaction velocity on substrate concentration to hyperbolic (Fig. 6; Table 3). To our knowledge, this is the first case when a single mutation of the surface residue remote from both the active site and allosteric peripheral ligand-binding site (~30 Å away) could abolish sigmoidicity of saturation kinetics. That the K257A mutation alters the manner of ligand binding would need to be confirmed using compounds with multiple binding sites that either change fluorescence or cause a feasible spin shift upon binding to CYP3A4. Molecular dynamics simulations, on the other hand, could clarify whether protein dynamics and product egress are affected in K257A CYP3A4 and how conformational changes at the mutational site are transmitted to the ligand binding areas.

The lack of protective effect of the K257A mutation on time-dependent inactivation of CYP3A4 is another important finding (Fig. 8; Table 2). Given that K257 is solely and irreversibly modified by the ritonavir metabolite [16], one would expect a decreased or no contribution of the time-dependent component in the inactivation of K257A CYP3A4. On the contrary, the K257A and K266A variants were equally prone to time-dependent inhibition by cobicistat and ritonavir. This discrepancy could arise from differences in the assay system: lipid-free in our study vs cytochrome b₅-containing membranous Supersomes used by Rock et al. [16]. The existence of alternative modifiable residues and/or inactivation mechanisms is another explanation. One such mechanism could be heme destruction and formation of a heme-protein adduct, which was detected in the b₅/lipid-free assay system similar to ours [15] but not in Supersomes [16]. Owing to the complexity and dependence of the inhibitory action of ritonavir on experimental settings, further mechanistic studies are required to fully understand the underlying mechanism and reconcile the accumulated data.

5. Conclusions

This study utilized spectral, kinetic, functional, and structural approaches to investigate the interaction of CYP3A4 with cobicistat. The newly solved 2.5 Å crystal structure of the inhibitory complex provided the first direct insights into the binding mode of cobicistat, improved understanding of SAR for ritonavir-like compounds, and helped interpret experimental results. Collectively, our data suggest that the bulky morpholine moiety rather than elimination of the backbone hydroxyl is responsible for the lower binding affinity of cobicistat and its slower association to CYP3A4. The inhibitory potency was unaffected due to the ability of cobicistat to strongly coordinate to the heme and form extensive interactions via the phenyl side-groups, blocking access to the catalytic center as effectively as ritonavir. Based on structural comparison and mutagenesis data, K257 was identified as a functionally important residue that contributes to CYP3A4 allosterism, possibly through conformational dynamics, as its replacement with alanine eliminates sigmoidicity of the BFC oxidation kinetics. However, the K257A mutation had no effect on time-dependent inactivation of CYP3A4 by cobicistat and ritonavir, which implies that their inhibitory action is more complex and, in addition to the covalent K257 modification, may involve alternative mechanism(s).

Abbreviations:

BFC

7-benzyloxy-4-(trifluoromethyl)coumarin

CPR

cytochrome P450 reductase

CYP3A4

cytochrome P450 3A4

HFC

7-hydroxy-4-trifluoromethyl coumarin

SAR

structure-activity relation

WT

wild type

Data availability statement

All experimental data generated during this study are included in this article and Supplementary Material. Coordinates and structure factors for the X-ray model of cobicistat-bound CYP3A4 are freely available at the Protein Data Bank (https://www.rcsb.org/).