Clobetasol propionate is a heme-mediated selective inhibitor of human cytochrome P450 3A5

Abstract

The human cytochrome P450 (CYP) enzymes CYP3A4 and CYP3A5 metabolize most drugs and have high similarities in their structure and substrate preference. Whereas CYP3A4 is predominantly expressed in the liver, CYP3A5 is upregulated in cancer, contributing to drug resistance. Selective inhibitors of CYP3A5 are, therefore, critical to validating it as a therapeutic target. Here we report clobetasol propionate (clobetasol) as a potent and selective CYP3A5 inhibitor identified by high-throughput screening using enzymatic and cell-based assays. Molecular dynamics simulations suggest a close proximity of clobetasol to the heme in CYP3A5 but not in CYP3A4. UV–visible spectroscopy and electron paramagnetic resonance analyses confirmed the formation of an inhibitory type I heme–clobetasol complex in CYP3A5 but not in CYP3A4, thus explaining the CYP3A5 selectivity of clobetasol. Our results provide a structural basis for selective CYP3A5 inhibition, along with mechanistic insights, and highlight clobetasol as an important chemical tool for target validation.

Graphical abstract

Introduction

The cytochrome P450 (CYP) superfamily of heme-containing enzymes is responsible for catalyzing a wide range of biological processes. The heme group is critical for the catalytic activity of CYPs, with a highly reactive heme iron-oxo species (compound I) being responsible for the oxidation of CYP substrates bound close to the heme iron^1,2. The 3A subfamily of CYPs (CYP3A) is critical for xenobiotic clearance in humans and is reported to metabolize more than half of all currently prescribed drugs^3,4. Although the members of this family include CYP3A4, CYP3A5, CYP3A7, and CYP3A43, it is CYP3A4 and CYP3A5 that are the primary CYP enzymes expressed in adults^5,6. These enzymes are promiscuous and have exceptionally broad substrate specificity. They metabolize various pharmaceuticals, natural products, and endogenous small molecules. This is possible, in part, because of their ability to bind diverse, structurally unrelated compounds within their large active sites (as we recently reviewed⁷). Such indiscriminate binding is the basis for many compounds being able to modulate CYP3A, thereby decreasing drug efficacy or drug–drug interactions^8–11. Ligand promiscuity of CYP3A4 and CYP3A5 homologs is one reason why these enzymes are considered functionally redundant¹². Compared to CYP3A5, CYP3A4 is more predominantly expressed in the normal liver, the primary site of drug metabolism, which explains why CYP3A4 is regarded as the representative member of the CYP3A family⁵. Furthermore, the US Food and Drug Administration (FDA) recommends testing potential pharmaceuticals for CYP3A inhibition and does not differentiate between CYP3A4 and CYP3A5^13,14. For these reasons, the enzymes are ubiquitously grouped together as “CYP3A4/5” in most expression- and metabolism-related studies.^15–19 However, emerging evidence suggests that these isozymes are not as redundant as previously thought.

CYP3A5 is reportedly overexpressed in pancreatic ductal adenocarcinoma (PDAC) and mediates chemoresistance in different subtypes of this cancer²⁰. Importantly, RNA interference (RNAi)-mediated knockdown of CYP3A5 re-sensitizes the drug-resistant PDAC cells, confirming the role of elevated CYP3A5 levels in drug resistance²⁰. Although this is the first evidence that selectively inhibiting CYP3A5 might be clinically significant, previous reports have discussed the need for a CYP3A5-selective inhibitor as a tool compound with which to differentiate CYP3A5 from CYP3A4²¹. The challenge of finding such a tool compound is intensified by CYP3A4 being generally more catalytically active, although this property aided the development of CYP3A4-selective inhibitors^{22, 23}. Interestingly, among the numerous overlapping substrates, a few compounds, including the anticancer drug vincristine and the immunosuppressant tacrolimus have been reported which characterize the selective catalytic activity of CYP3A5^24–27. These compounds are metabolized to a greater extent by CYP3A5 than by CYP3A4. Moreover, the N-oxide metabolite of the phosphodiesterase inhibitor T-1032 has been reported to be catalyzed almost exclusively by CYP3A5, further demonstrating the distinct function of the enzyme²⁸. Unsurprisingly, each of these compounds has a different chemical structure, upending the notion that this selective activity is chemotype-dependent. The demonstration that CYP3A5 selectively metabolizes certain compounds, however few, suggests that CYP3A5 plays a discrete role which might potentially be selectively targeted. A compound capable of selectively inhibiting CYP3A5 will be important for circumventing the inaccurate assignment of cumulative enzymatic activity to CYP3A4 alone, as well as for delineating between the two homologs. Such an inhibitor could prove especially beneficial in cases in which CYP3A5 is the target (e.g., when extrahepatic CYP3A5 is elevated and causes drug resistance) but CYP3A4 inhibition should be avoided (e.g., so as not to alter normal drug metabolism in the liver).

Here, we exploited high-throughput screening techniques to identify a selective inhibitor of CYP3A5. The compound we identified, clobetasol propionate (abbreviated hereafter as clobetasol), can potently block CYP3A5 catalytic activity without inhibiting CYP3A4 or other major CYPs. We further identified pancreatic cancer cell lines as suitable models for studying the selective modulation of CYP3A5 in vitro. Clobetasol proved to be a nontoxic, potent, and selective inhibitor of CYP3A5 in these cells. We leveraged molecular dynamics (MD) simulations to predict the potential mechanism of CYP3A5-selective inhibition. Our simulations suggested that clobetasol resides closely to the heme group in CYP3A5 but binds to CYP3A4 too distally from the heme in that protein for effective clobetasol–heme interaction to occur. This computational prediction is consistent with the essential role of heme in the catalytic activity of CYPs, and it was further supported experimentally with multiple biophysical techniques, including UV–visible spectroscopy and electron paramagnetic resonance (EPR) analyses, which can detect the heme-dependent interactions with compounds. Although it has been suggested that identifying a CYP3A5-selective inhibitor would be challenging because of the high structural similarity between CYP3A4 and CYP3A5^{20, 21}, our work demonstrates that it is indeed feasible. We anticipate that the structural basis and mechanistic insights revealed by this work will aid in the further development of CYP3A5-selective inhibitors, facilitate further investigations of the structural and functional regulation of CYP3A4 and CYP3A5, and enable exploration of the therapeutic potential of targeting CYP3A5.

Results

High-throughput screening identifies clobetasol as a selective CYP3A5 inhibitor.

The high degree of structural similarity between CYP3A4 and CYP3A5 makes it challenging to use a structure-based approach to design a CYP3A5-selective inhibitor. Therefore, to search for potential selective CYP3A5 inhibitors, we first used a luminescence-based enzymatic assay to screen and identify inhibitors of the catalytic function of CYP3A5. We then eliminated those compounds which also inhibited CYP3A4. The assay we used leverages the ability of CYP3A5 to metabolize a “pro-luciferin” substrate into D-luciferin, which is then converted into luminescence when luciferase is added to the system. For the primary screen, we used the St. Jude bioactive compound library of 11,200 total compounds, which contains FDA-approved drugs, drug candidates, and other compounds with known activity^{29, 30}. The propensity of a given small molecule to inhibit CYP3A5 was clear; hundreds of compounds showed greater than 50% inhibition at the final tested concentration of 5 μM (Fig. 1a, Supplementary Data 1). To confirm the CYP3A5-inhibitory activity and determine which of these compounds were selective for CYP3A5, we chose compounds from the primary screen that conferred at least 60% CYP3A5 inhibition and evaluated them in dose-response analyses. The 252 chosen compounds were then screened against CYP3A4 and CYP3A5 in parallel, using the exact same conditions for each (described in the Methods section). As a demonstration of nonselective inhibition, we tested the antiretroviral compound ritonavir—a known inhibitor of the CYP3A family³¹ and the only compound to have been co-crystalized with both CYP3A4³² and CYP3A5³³. Ritonavir-mediated inhibition aligned with values reported in the literature^34,35, and the inhibitory curves demonstrated high potency for each enzyme (Fig. 1b). A systematic comparison of all the dose-response curves led to the identification of clobetasol propionate (clobetasol) as a potent and selective CYP3A5 inhibitor. Clobetasol showed the greatest selective inhibition, having IC₅₀ values of 0.206 μM for CYP3A5 and 15.6 μM for CYP3A4 (Fig. 1b). While a few other compounds from the screen demonstrated comparatively marginal selective inhibition, none could match the approximately 76-fold CYP3A4/CYP3A5 IC₅₀ ratio of clobetasol. Interestingly, clobetasol contains a four-ring steroid scaffold that is analogous to several known substrates of CYP3A5 within the same drug class³⁶ (Fig. 1c).

Figure. 1. Clobetasol propionate is a potent and selective inhibitor of CYP3A5.

(a) CYP3A5 inhibition profile of the St. Jude bioactive library when screened at a concentration of 5 μM. Compounds selected for downstream dose-response testing (those having ≥60% CYP3A5 inhibition) are shown between the gray dotted lines. The green circle indicates clobetasol. (b) Dose-response curves for CYP3A4 and CYP3A5 with ritonavir (Rit) or clobetasol (Clob). IC₅₀ values for each curve are shown in parenthesis. The IC₉₀ for clobetasol–CYP3A5 are shown in red text. (c) Chemical structure of clobetasol. (d) Inhibition of CYP3A4 or CYP3A5 by using 1.89 μM ketoconazole (Keto) or clobetasol (Clob). The DMSO concentration is 0.1%. *** P ≤ 0.0001, ns = not significant (P ≥ 0.05); one-way analysis of variance (ANOVA). (e) Inhibition profile of 1.85 μM clobetasol against a panel of major human CYPs.

To further demonstrate the CYP3A5-selective inhibition of clobetasol, we compared the effect to CYP3A4 using clobetasol at a concentration of 1.8 μM (the IC₉₀ concentration for CYP3A5) and used the same concentration of ketoconazole (the gold-standard pan-CYP3A inhibitor³⁷ that served as a normalization control in our biochemical assays). At 1.8 μM, clobetasol inhibited CYP3A5 by 90%, but no inhibition of CYP3A4 was observed (Fig. 1d). As expected, the same concentration of ketoconazole completely inhibited both enzymes (Fig. 1d). Although clobetasol demonstrated remarkable selectivity for CYP3A5 as compared to CYP3A4, we wanted to test whether it inhibited the catalytic activity of other major human CYPs. We screened clobetasol against CYP3A4 and a panel of six other human CYPs (CYP1A2, 2B6, 2C8, 2C9, 2C19, and 2D6) recommended by the FDA for drug–drug interaction testing for candidate pharmaceuticals^13,14. Each of these enzymes was tested for catalytic inhibition by directly measuring the formation of product derived from its own substrate. In the case of CYP3A4, we tested the two structurally unrelated substrates midazolam and testosterone, as recommended by the FDA^13,14. Not only was CYP3A4-mediated catalysis of these substrates uninhibited by 1.8 μM clobetasol but other major CYPs also showed little or no inhibition (Fig. 1e). Taken together, these data suggest that clobetasol is a potent and selective inhibitor of CYP3A5 in biochemical systems.

Pancreatic cancer is an appropriate and clinically relevant model for studying selective modulation of CYP3A5.

After validating clobetasol as a CYP3A5-selective inhibitor in biochemical assays, we next sought to determine whether this selectivity was maintained in cell-based systems. Hepatocellular carcinoma serves as a widely used model for studying CYP-related biology, including that of the CYP3A family^38–42. However, although CYP3A5 is expressed appreciably in hepatocellular carcinoma, the (often higher) expression of CYP3A4 makes delineation between these homologs complex and arduous. Accordingly, we needed a model system that expressed CYP3A5 but lacked CYP3A4 expression. To this end, we turned to the PanCancer analysis project from The Cancer Genome Atlas (TCGA), which hosts RNA-seq datasets derived from more than 10,000 tumor samples spanning 33 different cancer types⁴³. We reanalyzed all data from this project by using our in-house pipeline (our stepwise protocol is made available in Protocol Exchange⁴⁴) and ranked each cancer type according to the level of CYP3A5 expression. As expected, the liver and hepatocellular carcinoma (LIHC) cohort exhibited the highest CYP3A5 expression, followed by the bile duct (CHOL) and pancreatic (PAAD) cancer cohorts (Fig. 2a). Conversely, CYP3A4 expression was primarily contained within the LIHC and CHOL cancers (Supplementary Fig. 1a). In view of the apparent selective overexpression of CYP3A5 in pancreatic cancer, we decided to pursue our studies using cell models of this type. Moreover, CYP3A5 has been reported to be overexpressed in pancreatic adenocarcinoma and to mediate its chemoresistance²⁰, adding a clinical significance to studying CYP3A5-mediated drug metabolism in this cancer type.

Figure. 2. CYP3A5 is over-expressed in pancreatic cancer models; AsPC-1 cells are ideal for studying CYP3A5 modulation.

(a) Expression of CYP3A5 across all cancers within the PanCancer dataset from The Cancer Genome Atlas (TCGA). Expression is derived from 10,534 tumor samples across 33 cancers and is ranked by median values. The pancreatic adenocarcinoma cohort (PAAD) is highlighted in red. (Sample IDs, the expansion of the cohort abbreviations, and the full CYP3A5 expression data are located in Supplementary Data 2; the stepwise normalization protocol is available in Protocol Exchange⁴⁴). (b) CYP3A5 expression in pancreatic adenocarcinoma cell lines compared to that in the noncancerous pancreatic cell line HPNE. *** P ≤ 0.0001, ns = not significant (P ≥ 0.05); t statistic–derived P values adjusted by the Benjamini-Hochberg false discovery rate. (c) Expression of the xenobiotic-metabolizing CYPs from all pancreatic adenocarcinoma and control cell lines. Units of expression for all panels are log2 (normalized CPM+1).

To set about assessing CYP3A5 expression in pancreatic cancer models, we obtained eight commercially available pancreatic ductal adenocarcinoma (PDAC) cell lines (AsPC-1, HPAF-II, PANC-1, MIA PaCa-2, SU.86.86, Capan-2, CFPAC-1, and Panc 02.13) and one noncancerous immortalized pancreatic cell line (hTERT-HPNE) from the American Type Culture Collection (ATCC). We performed RNA-seq on each sample and compared the CYP3A5 expression levels. Relative to the noncancerous control cell line, all but one of the eight cancer cell lines highly overexpressed CYP3A5 and demonstrated clear statistical significance (Fig. 2b). The AsPC-1 cell line displayed the highest CYP3A5 levels among all cell lines, having greater than 40-fold higher expression compared to the control HPNE cell line (Fig. 2b). The need for a cell model in which clobetasol-mediated CYP3A5 modulation could be studied meant that high CYP3A5 expression by itself was insufficient; it was imperative to examine the abundance of CYP3A4 and other relevant CYPs. Therefore, we surveyed the expression of known xenobiotic-metabolizing CYPs, as previously categorized by Guengerich⁴. Most of the cell lines tested had low (or no) expression of these enzymes and, critically, CYP3A4 was not expressed (Fig. 2c). Collectively, our sequencing-based data suggested that CYP3A5 was indeed overexpressed in pancreatic cancer. Moreover, this clinically relevant cell model demonstrated good potential for use as a cellular system for studying CYP3A5 selectivity because of the lack of CYP3A4 and other relevant CYPs. We found AsPC-1 cells to have particularly high and selective CYP3A5 expression, warranting further investigation into how clobetasol works in these cells.

Clobetasol selectively inhibits CYP3A5 in cells.

To determine whether clobetasol inhibited CYP3A5 in a cell-based context, we modulated the expression of CYP3A5 and/or CYP3A4 in parental AsPC-1 cells (referred to as wild-type, or WT), which express high levels of endogenous CYP3A5 but no detectable CYP3A4. To further elevate CYP3A5 levels, we produced doxycycline (Dox)-inducible CYP3A5 overexpression in AsPC-1 parental cells (hereafter referred to as “WT + 3A5^OE” cells). We also used CRISPR/Cas9 technology to knock out CYP3A5 from the parental AsPC-1 cells (a total CYP3A5 genetic deletion). In these AsPC-1^{CYP3A5−/−} cells (hereafter abbreviated as 3A5^−/− cells), we also produced doxycycline inducible CYP3A4- and CYP3A5-overexpression systems (referred to as “3A5^−/− + 3A4^OE” and “3A5^−/− + 3A5^OE” cells, respectively). We validated each cell line by measuring and quantifying the relevant protein expression. As expected, the results (Fig. 3a and Supplementary Fig. 2) demonstrated (1) that the 3A5^−/− cells had no detectable CYP3A5 protein, (2) that “WT + 3A5^OE” cells could be induced to express a higher level of CYP3A5, (3) that “3A5^−/− + 3A5^OE” cells could be induced to express CYP3A5 at a level comparable to that in “WT + 3A5^OE” cells, and (4) that CYP3A4 could be successfully overexpressed in the 3A5^−/− cells upon induction (“3A5^−/− + 3A4^OE” cells). To probe for CYP3A4 or CYP3A5 catalytic activity, we treated these cell lines with midazolam and measured the product formation via liquid chromatography with tandem mass spectrometry (LC-MS/MS). Midazolam is a well-established substrate of CYP3A4 and CYP3A5¹³, and the CYP3A-catalyzed 1-hydroxymidazolam product (1OH-MDZ) is readily amenable to detection by LC-MS/MS^45–47. Furthermore, because no other enzyme has been reported to catalyze the formation of 1OH-MDZ, the presence of this product is generally attributed directly to CYP3A activity. In experiments using parental AsPC-1 (WT) cells, which have high endogenous CYP3A5 expression but no CYP3A4 expression, clobetasol produced a sigmoidal dose-response inhibition curve (Fig. 3b, upper left). Surprisingly, clobetasol was even slightly more potent at inhibiting CYP3A5 activity than was the control compound, ketoconazole (a gold-standard, nonselective pan-CYP3A inhibitor). When we probed for activity in the 3A5^−/− cells, we found CYP3A5 activity to be totally abolished, as indicated by the absence of detectable 1OH-MDZ formation (Fig. 3b, upper right), thus providing evidence that CYP3A5 was completely knocked out, and indeed no other enzyme is able to catalyze the formation of 1OH-MDZ in the 3A5^−/− cells. When CYP3A5 was induced (+Dox) to overexpress in either WT or 3A5^−/− cells, midazolam metabolism occurred, and both clobetasol and ketoconazole inhibited the CYP3A5 activity (Fig. 3b, middle panels). Having demonstrated clobetasol to be a potent inhibitor of CYP3A5 catalytic activity in cells, we sought to test its selectivity by repeating the experiment using 3A5^−/− cells overexpressing CYP3A4. As expected, ketoconazole proved to be a potent and nonselective inhibitor, whereas clobetasol displayed remarkable selectivity and completely avoided CYP3A4 inhibition (Fig. 3b, bottom right). Interestingly, at higher tested concentrations, clobetasol appeared to increase the enzymatic activity of CYP3A4, producing higher levels of 1OH-MDZ than were present at baseline (Fig. 3b, bottom right). Although this phenomenon occurred only with higher concentrations of clobetasol, we found the result interesting and followed it up by examining whether clobetasol increased CYP3A4 protein levels. As shown in Fig. 3c and Supplementary Fig. 2, clobetasol did not increase the protein level of CYP3A4 (note that the anti-CYP3A4 antibody detects both CYP3A4 [marked with a start] and CYP3A5). Interestingly, ketoconazole at a concentration of 1 μM increased CYP3A4 protein expression, but this had no effect on our activity assay because the tested concentration of 1 μM completely abolished CYP3A4 activity. The compounds used did not affect cell growth in any cell line that we tested (Supplementary Fig. 3). Moreover, we tested the catalytic activity of each DMSO-treated cell line by monitoring the generation of 1OH-MDZ in the absence of compound treatment. Our results showed that the actual catalytic activity of CYP3A4 in the “3A5^−/− + 3A4^OE” cells was less than 10% of that of CYP3A5 in the “3A5^−/− + 3A5^OE” cells (Fig. 3d). Together, these data demonstrate that clobetasol is a potent and selective inhibitor of CYP3A5 in the context of relevant cell models.

Figure. 3. Clobetasol selectively inhibits CYP3A5 in vivo.

(a) Representative Western blots of CYP3A5 (top 3 panels) or CYP3A4 (bottom panel). Corresponding quantifications are shown to the right. Quantification was derived from triplicate experiments. *** P ≤ 0.0001; unpaired two-tailed t-test. The relative intensity of WT, WT + 3A5^OE (+Dox), 3A5^−/− + 3A5^OE (+Dox) and 3A5^−/− + 3A4^OE (+Dox) was set as 1.0, as indicated in each panel. Dox, doxycycline (100 ng/mL). (b) CYP3A4 or CYP3A5 activity measured by 1-hydroxymidazolam formation. All panels were normalized to the DMSO control as 100% activity. (c) Western blot of CYP3A4 or CYP3A5 expression after 24 h treatment with DMSO (left), 1 μM ketoconazole (Keto, middle), or 1 μM clobetasol (Clob, right). CYP3A4 is indicated by the slightly higher bands (star icon); the antibody is known to also detect CYP3A5. Quantification is shown below the gel images, with DMSO treatment set as 1.0. *** P ≤ 0.0001, ** P ≤ 0.005, * P ≤ 0.05, and ns = not significant (P ≥ 0.05); One way ANOVA with Tukey adjustment. (d) CYP3A-mediated activity across all cell lines, showing the relative catalytic contribution; all samples were treated with the same concentration of DMSO for 24 h.

Molecular dynamics simulations predict the mechanism of selective inhibition by clobetasol.

We next asked how clobetasol achieved selectivity between such highly homologous enzymes as CYP3A4 and CYP3A5. We hypothesized that the selectivity of clobetasol was in part due to the subtle differences in their active site shapes which were recently reported⁴⁸. Furthermore, we reasoned that clobetasol probably interacted in some way with the heme moiety of CYP3A5, but not with that of CYP3A4. Catalytic activation of CYPs occurs through single electron reduction of the CYP heme iron to its ferrous state, enabling binding of dioxygen. A further reduction and two subsequent protonation reactions facilitate the production of the reactive iron-oxo species compound I, which oxidatively attacks the bound substrate to form the product⁴⁹. CYPs can also be inhibited by the binding of inhibitors that coordinate to the CYP heme iron⁵⁰. Prototypical inhibitors such as ketoconazole function by forming tight, direct interactions with the heme iron and by blocking access to potential substrates^32,51–53. To obtain insight into whether this might be the mechanism underlying the CYP3A5 selectivity of clobetasol, we used molecular dynamics (MD) simulations. We started by docking clobetasol into CYP3A4 and CYP3A5. Fortunately, both enzymes have been crystalized with the same ligand (ritonavir^32,33), which provided us with a solid starting point for our in silico studies. As expected, superimposing the two structures revealed no significant differences in their secondary structure (Supplementary Fig. 4, upper panels). Interestingly, docking results produced only one pose of clobetasol for both CYP3A4 and CYP3A5; however, the ligand adopted a different orientation within their active sites (Supplementary Fig. 4, lower panels). To circumvent any bias in binding estimation as a result of starting structures being forced into low-energy, non-physiologically relevant conformations, we ran MD simulations independently for the CYP3A4–clobetasol and CYP3A5–clobetasol structures and ensured that each system had properly equilibrated by monitoring the root mean squared deviation (RMSD) (Supplementary Fig. 5). Additionally, we solvated the systems in the same water-based solvent model, ran 200-nanosecond (ns) simulations for each, and kept all other parameters the same between systems to enable direct comparisons (as detailed in the Methods section).

Our simulations showed clear differences between the interactions of clobetasol with CYP3A4 and its interactions with CYP3A5. Clobetasol could bind to CYP3A4, but it preferred an area of the active site too distant from the heme group to be considered a classical inhibitor (Fig. 4a, upper panel). Remarkably, clobetasol was stabilized at a sufficient distance from the heme in CYP3A4 for water to enter the cavity from the solvent, and we observed a water molecule become stabilized on the heme iron in the distal position (Fig. 4a, upper panel, Supplementary Video 1). A water molecule coordinated in this way is representative of the resting state of the enzyme^54–56, and our MD-derived CYP3A4–clobetasol structure closely matched the published crystal structure of water-bound CYP3A4⁵⁷ (Supplementary Fig. 6). Conversely, clobetasol interacted with the active site in closer proximity to the heme in CYP3A5 than to that in CYP3A4 (Fig. 4a, lower panel). Moreover, we observed this heme-ligand interaction as stable and able to be maintained over the course of the simulation (Supplementary Video 2).

Figure. 4. Molecular dynamics simulations suggest that clobetasol forms a heme-ligand coordination in CYP3A5 but not in CYP3A4.

(a) Snapshots from stable segments of molecular dynamics simulations showing clobetasol bound to CYP3A4 (top) or CYP3A5 (bottom). (b) Percentage residue interaction diagram demonstrating the residues interacting with clobetasol in CYP3A4 (left) or CYP3A5 (right). Only residues with a combined CYP3A4+CYP3A5 interaction of ≥1% over the simulation course are shown. (c) The root mean squared fluctuation (RMSF) of clobetasol for CYP3A4 (blue line) or CYP3A5 (red line) simulations. The atom numbers of clobetasol (left) correspond to the RMSF plot x-axis (right). Star (*) represents the carbonyl oxygen closest to the heme. (d) Ångstrom distance measurement of the closest non-hydrogen atom of clobetasol to the heme iron of CYP3A4 (blue) or CYP3A5 (red) over simulations. (e) Docked poses of clobetasol in the CYP3A4 (left) or CYP3A5 (middle) active sites, with 5-Å binding sites shown in gray mesh. The superimposed structures are rotated to show the active site differences (right), which are shown in slate gray (for CYP3A4) or raspberry red (for CYP3A5) mesh. Oxygen atoms are shown in red, nitrogen atoms in blue, sulfur atoms in yellow, fluorine atoms in teal, chlorine atoms in green, and carbon atoms in black (except for in panels a and e, in which carbon is shown in slate gray for CYP3A4 or raspberry red for CYP3A5).

To predict the residues with which clobetasol might be interacting inside the active site of CYP3A4 or CYP3A5, we quantified the percentage interaction across both simulations and looked at the interaction types (Fig. 4b). We also measured the binding stability of clobetasol itself as a function of relative movement by using the root mean squared fluctuation (RMSF) metric. Quantifying the RMSF showed that most atoms of clobetasol fluctuated less in CYP3A5 than in CYP3A4; these included the carbonyl oxygen atoms, closest to the heme iron in CYP3A5 (Fig. 4c). This finding can be attributed to clobetasol binding more stably (or moving less) in CYP3A5 than in CYP3A4. Additionally, we wanted to measure the clobetasol–heme distance in both simulations, since proximity to the heme group is such a hallmark of effective substrates and inhibitors^58–60. We measured the clobetasol–heme distance for every frame of the simulation and observed that clobetasol was closer to the heme in CYP3A5 than in CYP3A4 (Fig. 4d). The ability of clobetasol to preferentially and stably coordinate with the heme iron of CYP3A5 but not that of CYP3A4, thereby blocking solvent access and presumably serving as the mechanism of selective inhibition, was puzzling. We investigated why this might occur by examining the active site shapes of each enzyme. As reported when the crystal structure of CYP3A5 was first solved, its active site is slightly taller and narrower than that of CYP3A4^{33, 48}. Indeed, this difference was sufficient to permit clobetasol to adopt a vertical binding orientation in CYP3A5, as compared to its more horizontal orientation in CYP3A4 (Fig. 4e, left and middle). Furthermore, neither active site could reciprocate its homolog’s conformation to clobetasol, since the ligand would clash with the ceiling of the cavity of CYP3A4 or with the wall of the cavity of CYP3A5 (Fig. 4e, right). Collectively, our in silico data have presented a model providing predictive insights into how and why clobetasol selectively inhibits CYP3A5.

Differential interaction of clobetasol with the heme in CYP3A4 and CYP3A5.

To determine experimentally the direct interaction of clobetasol with CYP3A4 and CYP3A5, we performed UV-visible spectroscopic analysis of clobetasol with each of these CYPs and compared the results to those obtained with their bona fide inhibitor ketoconazole³⁷ and substrate midazolam⁶¹ using the recombinant purified enzymes. CYPs possess a prosthetic heme b group that is essential for their catalytic activity. Ligand interactions involving the heme in CYPs can be monitored by UV–visible spectroscopy, based on the ability of the ligands to induce either a type I Soret shift (blue shift), reflecting the conversion of the low-spin (LS) heme iron to the high-spin (HS) state through binding of a substrate, resulting in the displacement of the axial water ligand; or a type II Soret shift (red shift), typically occurring through the binding of an inhibitor that displaces the axial water ligand and coordinates to the heme iron (Fig. 5a, 1–3). The titration of clobetasol and midazolam with CYP3A5 in both cases produced a typical type I CYP spectral shift with a Soret band shift from 417 nm (LS) to approximately 395 nm (HS) (Fig. 5c and Supplementary Fig. 7). As expected, titration of ketoconazole with CYP3A5 produced a typical type II (LS) spectrum with a Soret band shift from 417 nm to 424 nm (Fig. 5b). To determine the dissociation constant, the changes in the absorbance spectra of the heme-bound complex for both the clobetasol and ketoconazole titrations were determined by absorbance difference spectral titrations (Supplementary Fig. 7). Spectral data were obtained by successive additions of clobetasol or ketoconazole until no further heme absorbance change was observed. Thereafter, in each case, absorbance difference spectra were generated by subtracting the initial (ligand-free) spectrum from each successive spectrum, and then identifying the absorbance maximum (peak) and minimum (trough) values. The peak minus trough values were determined from each difference spectrum, and these values were plotted against the relative ligand concentration, enabling the production of ligand-binding curves, leading to the generation of the respective dissociation constants (K_d) of 0.1 ± 0.3 μM for clobetasol binding to CYP3A5 and a K_d of 3.1 ± 0.5 μM for ketoconazole binding to CYP3A5. In addition, a similar degree of conversion to the ligand bound form was achieved in titrations with both clobetasol and ketoconazole where both showed a near complete conversion to the high-spin or inhibitor bound forms respectively albeit at different concentrations of ligand (Fig. 5d). In contrast, titrating clobetasol with CYP3A4 produced only a small type I spectral shift to the substrate-bound form, with a shoulder on the Soret band at 386 nm forming concomitant with a small decrease in the low-spin Soret maximum at 416 nm. Similar to CYP3A5, ketoconazole induced a type II spectral shift in CYP3A4 with a K_d value of 0.9 ± 0.2 μM (Fig. 5e–g and Supplementary Fig. 7). These data clearly suggest a selective heme iron interaction by clobetasol, particularly for CYP3A5.

Figure. 5. UV-Vis and EPR spectroscopy of clobetasol binding to CYP3A4 and CYP3A5.

(a1–3) Schematic representation of ligand-induced CYP heme ferric iron (Fe^III) transition spin states from resting (green) to either high spin (blue) or low spin (red) states. (b,c,e,f) UV-Vis titrations of ketoconazole (Keto) (b,e) and clobetasol (Clob) (c,f) on CYP3A5 and CYP3A4, respectively. All titrations were normalized (normalized Abs) for easy comparison (see Methods). Green, blue or red line and numbering corresponds to the heme state in a1–3. (d) Plot of clobetasol (Clob) (ΔAbs _{386 nm – 418 nm}) and ketoconazole (Keto) (ΔAbs _{432 nm – 410 nm}) induced heme absorption change (ΔAbs of Heme Bound) in CYP3A5 versus ligand concentration and data fitted with equation 2 (see Methods) to give a K_d of 0.1 ± 0.3 μM and 3.1 ± 0.5 for clobetasol and ketoconazole, respectively. (g) Plot of the maximal clobetasol (ΔAbs _{388 nm – 418 nm}) and ketoconazole (ΔAbs _{432 nm – 410 nm}) induced heme absorption change (ΔAbs of Heme Bound) in CYP3A4 versus ligand concentration and data fitted with equation 2 (see Methods). A K_d of 0.9 ± 0.2 μM was calculated for ketoconazole. (h,i) EPR spectra for CYP3A5 and CYP3A4 in their ligand-free forms and for their complexes with ketoconazole, midazolam, and clobetasol with colored number labels corresponding to the heme ferric iron (Fe^III) state as in a1–3.

To further confirm the selective heme interaction of clobetasol with CYP3A5 versus CYP3A4, continuous wave (CW) X-band electron paramagnetic resonance (EPR) spectra were generated with ligand-free CYP3A4 and CYP3A5 and with these CYPs in complex with clobetasol, midazolam, and ketoconazole. Ligand-free CYP3A5 produced a rhombic LS EPR spectrum with g-values of 2.41 (g_z), 2.24 (g_y), and 1.91 (g_x), respectively, indicative of a single dominant ferric heme species coordinated by a proximal cysteine thiolate ligand and a weakly coordinated axial water ligand (Figs. 5a1 and 5h). In ligand-free CYP3A4, a small signal was seen for an HS species with g-values apparent at 8.17 and 3.56 (the g-values were not labeled in the spectrum), while in CYP3A5 no such small HS signal was seen in the resting form of the enzyme. The EPR spectra for CYP3A5 in complex with clobetasol or midazolam revealed spectral signals consistent with the formation of a five-coordinate HS ferric heme iron state (which usually indicates the ligand-dependent displacement of the CYP3A5 distal water ligand) with HS g-values of 8.04/3.65 (with clobetasol) and 8.10/3.56 (with midazolam) (Figs. 5a2 and 5h). A proportion of the LS ferric signals was retained, indicative of the retention of a water coordinated state in the absence of substrate, with g-values of 2.4½.24/1.92 for both clobetasol and midazolam (Figs. 5a1 and 5h). In contrast, CYP3A4 revealed no new HS formation with clobetasol; the spectra were consistent with the corresponding LS ligand-free form with g-values of 2.40/2.24/1.91, indicating the retention of the axial water ligand in the presence of clobetasol in CYP3A4 (Figs. 5a1 and 5i). However, an HS signal was induced by midazolam when the compound was bound to the enzyme with HS g-values of 8.17/3.55 and LS g-values of 2.40/2.24/1.91. As expected, both CYP3A4 and CYP3A5 produced new LS species when bound to ketoconazole, with g-values of 2.55, 2.49/2.24/1.88 for CYP3A4 and 2.47, 2.42/2.24/1.88 for CYP3A5, which are typical signals seen for a CYP enzyme with an azole inhibitor bound (Fig. 5h and 5i). These data provide further evidence of the selectivity of clobetasol for CYP3A5.

Discussion and Conclusions

Until now, a CYP3A5 inhibitor that does not also inhibit CYP3A4 has not been identified. In addition to the difficulty arising from the fact that CYP3A5 and CYP3A4 have a high degree of structural homology, there was a lack of commercially available tools with which to screen for such a selective inhibitor by using requisite throughput. Many assays for screening CYP3A4 inhibitors are available with various readouts, such as luminescence and fluorescence, and are compatible with high-throughput screening⁶², but there are no such assays for CYP3A5. We hypothesized that we could exploit the overlapping substrate specificity of the enzymes and use an assay originally intended for CYP3A4 to screen for CYP3A5 inhibitors. The luminescence-based assay from Promega that we used (as detailed in the Methods section) is advertised as having no cross-reactivity with CYPs outside the CYP3A family, with the kit being optimized for CYP3A4. Nevertheless, we hypothesized that, with proper controls and normalization, the assay would work for CYP3A5. Our hypothesis was validated when we began testing the assay. Although the higher catalytic activity of CYP3A4 was clear, the properly controlled assay produced exceptional signal windows that were suitable for parallel screening and cross-enzyme comparisons. Our screen yielded results showing both zero CYP3A5 inhibition and total CYP3A5 inhibition (Fig. 1a). Furthermore, many known inhibitors were successfully identified from the screen, and a few cross-plate duplicates that served as internal controls lined up very well and demonstrated assay reproducibility (interactive results are available in Supplementary Data 1). Interestingly, our screen also identified a few CYP3A5 enzymatic activators, a phenomenon already reported in the literature for CYP3A enzymes ^{45, 63}. Our dose-response analysis further confirmed the high performance of the assay, as indicated by the tight replicates and sigmoidal inhibitory curves (Fig. 1b). After demonstrating the potential of clobetasol to selectively inhibit CYP3A5 while avoiding CYP3A4, we used a different assay to test the effect of clobetasol on six other major human CYPs. The IC₉₀ concentration of clobetasol for CYP3A5 (1.8 μM) showed no significant inhibition of these enzymes. This validated the selectivity of clobetasol not just in the context of CYP3A4 (Fig. 1e). The lack of CYP3A4 inhibition observed with three different substrates and two different readouts gave us confidence that clobetasol was potent and selective for CYP3A5 in cell-free systems.

We identified a cell model that was suited to studying the in vitro effect of clobetasol by leveraging publicly available RNA-seq data and confirming the results with in-house sequencing experiments using a panel of cell models that we assembled. We simultaneously normalized the expression results of the entire TCGA PanCan analysis dataset (described in detail in our stepwise protocol available in the Protocol Exchange⁴⁴). This benefited our study twofold: 1) it enabled proper and confident interpretation of the results when comparing CYP3A4 and CYP3A5 expression levels, and 2) it used the same processing pipeline that we used for our own samples from our cell panel, enabling much more direct comparisons. Our transcriptomic studies culminated in the understanding that CYP3A5 was highly expressed in pancreatic cancer (Fig. 2a) but CYP3A4 was not (Supplementary Fig. 1a). Notably, CYP3A4 was predominantly expressed in two cancer types: liver hepatocellular carcinoma and bile duct cancer (Supplementary Fig. 1a), whereas CYP3A5 was highly expressed not only in these two cancers but also in several others, including PDAC (Fig. 2a). These findings informed our decision to procure various pancreatic cancer cell lines. We performed RNA-seq on each of these cell lines and clearly demonstrated that CYP3A5 was overexpressed in PDAC (Fig. 2b). Moreover, we sought to examine the expression of other CYPs in these models. The CYP superfamily is broad, with humans expressing 57 CYP enzymes⁶⁴. We focused on xenobiotic-metabolizing CYPs because of their relevance to our project, and we showed that most of them, including CYP3A4, are not expressed in any of these cells (Fig. 2c). Although it is entirely plausible that CYPs outside this category are present, they are unlikely to be expressed at levels as high as those of CYP3A5 or to interfere with our selectivity studies. There is a dearth of information on CYP3A5 versus CYP3A4 expression in cancer, but these data helped us to conclude that pancreatic cancer cells were appropriate models for our cell-based selectivity studies.

We went on to demonstrate the selectivity of clobetasol in vitro by modulating the expression levels of CYP3A5 or CYP3A4 (Fig. 3a). In WT cells that endogenously expressing CYP3A5 but not CYP3A4, we were surprised to see clobetasol behaving as a more potent inhibitor than ketoconazole, however slightly. Importantly, in CYP3A4-null cells with a genetic deletion of CYP3A5, the metabolic activity for midazolam was completely abolished, thus demonstrating that no other enzymes were catalyzing the midazolam hydroxylation (Fig. 3b). The midazolam-metabolizing activity was rescued by overexpression of CYP3A5 but was again abolished by clobetasol, confirming that the effect of clobetasol was indeed CYP3A5 dependent. Overexpression of CYP3A4 in the CYP3A5-null cells rescued the midazolam-metabolizing activity but could not be inhibited by clobetasol, thus confirming that clobetasol inhibits CYP3A5 but not CYP3A4 (Fig. 3b, bottom). We were intrigued by the apparent clobetasol-induced enzymatic activation of CYP3A4 observed at high concentrations of the compound. Although our Western blot analysis indicated that clobetasol did not increase the protein levels of CYP3A4 (Fig. 3c), we cannot exclude the possibility that there is an endogenous CYP3A4 inhibitor present in our cell system that was somehow replaced by clobetasol. This intriguing phenomenon warrants further investigation. We also noticed that ketoconazole increased CYP3A4 protein levels, as reported previously⁶⁵, but that this increase in protein levels did not hinder our ability to enzymatically inactivate CYP3A4. Additionally, the relative CYP3A4-specific contribution to midazolam catalysis was drastically lower than that of CYP3A5 in DMSO-treated samples (Fig. 3d), which appears to be consistent with the lower protein levels of CYP3A4, as compared to CYP3A5, in DMSO-treated samples. (Fig. 3c).

In silico approaches have been used to study differential ligand interactions with CYP3A4 and CYP3A5. These reports use techniques such as ligand docking and molecular dynamics simulations to understand the differential behavior of ligands with the two homologs, usually focusing on heme–ligand interactions^66–68. We applied the same approaches to investigate how clobetasol selectively inhibited CYP3A5 but not its nearly identical homolog CYP3A4. Clobetasol formed a tight heme–ligand interaction throughout the simulation for CYP3A5 but preferred a site further from the heme in the CYP3A4 simulation (Fig. 4a). To our surprise, the simulations revealed that solvent enters the binding pocket and becomes stabilized on the heme only in CYP3A4, which is indicative of the resting (non-inhibited) enzyme state^54–56 (Supplementary Fig. 6, Supplementary Video 1). Conversely, clobetasol blocked solvent access throughout the simulation for CYP3A5, coordinating with the anchoring cysteine residue through the heme iron (Figs. 4b–d, Supplementary Video 2). We determined that the subtle differences in the shape of the active sites of the two enzymes were sufficient to cause discrete binding orientations for clobetasol, and this ultimately helped us to propose a mechanism for the selective inhibition (Fig. 4e). Furthermore, the MD simulations may be consistent with the enzymatic CYP3A4 activation that we observed at high concentrations of clobetasol. It is possible that clobetasol stabilizes CYP3A4 in an orientation that allows solvent (and, therefore, substrate) access to the heme more frequently, but this is of no significant consequence as only high concentrations produced this effect (and no concentration produced CYP3A4 inhibition).

We proceeded to experimentally test the model predicted by our MD simulations by using classical biophysical techniques for studying CYPs, namely UV–visible and EPR spectroscopy to interrogate the spin states and the overall electronic environment of the ferric heme iron in both CYPs, using recombinantly expressed proteins. With CYP3A5, UV–visible titrations distinctly showed a typical type I ligand-binding mode (indicating the displacement of the axial water ligand) for clobetasol, resulting in the formation of a five-coordinate ferric heme iron, whereas a type II spectrum shift was produced for ketoconazole, indicating the replacement of the axial water ligand with the nitrogen atom from the imidazole moiety of the azole compound to retain the six-coordinate ferric heme iron state. Both phenomena are consistent with the results of previous studies on human and bacterial CYPs^{33, 69–71}. To our surprise, clobetasol also showed a profoundly higher binding affinity for CYP3A5 when compared to ketoconazole, the pan-CYP3A inhibitor (Figs. 5b-d). In contrast, UV–visible titrations could not detect a significant spectral shift by which to assign a binding mode for clobetasol in CYP3A4, although minor spectral perturbations were observed with higher clobetasol concentrations. Moreover, the spectral binding data generated showed clobetasol to have a significantly weaker affinity for CYP3A4 when compared to ketoconazole. However, ketoconazole also produced a type II binding mode in CYP3A4, with binding affinities similar to those noted with CYP3A5 (Figs. 5e–g). These findings support the prediction from the MD simulations that clobetasol occupies a site closer to the heme in CYP3A5 compared to that in CYP3A4. We further validated this heme-dependent selectivity by using EPR to monitor the effect of clobetasol on the ferric heme iron in the resting state of both enzymes. Interestingly, clobetasol induced an LS to an HS ferric heme iron spin state in only CYP3A5, maintaining an LS ferric heme iron spin state similar to the ligand-free water-ligated resting form of the protein in CYP3A4 (Figs. 5h–i). In addition, as expected, midazolam and ketoconazole induced HS and LS spin states corresponding to typical type I and type II ligand-bound CYP species in both enzymes. A small proportion of HS was seen in the ligand-free form of CYP3A4 corresponding to a small portion of the enzyme existing without the axial water ligand to keep it in a complete LS spin state. However, this ligand-free HS does not correspond to a typical HS produced in the presence of a true type I ligand-bound form that usually presents concurrently with a significant size reduction in the g_z and g_y signals as noticed in the midazolam-bound forms of both CYPs (Figs. 5h-i). This feature was also observed in other CYPs from bacteria^{72, 73}. Our data confirm beyond reasonable doubt that the selective type I binding mode induced by clobetasol in the heme of CYP3A5, but not in that of CYP3A4, represents a true CYP3A5-inhibitor complex. Indeed, inhibitors that mimic substrates with a type I binding mode are not novel in the CYP field. Previous studies have shown that bromocryptine displays similar atypical type I–like binding features in CYP3A4 but is arguably classified as an inhibitor for the enzyme^{74, 75}.In addition, previous studies on bacterial CYPs involving CYP126A1 in Mycobacterium tuberculosis also identified potent inhibitors that displayed similar type I–like features⁷⁶. Our proposed mechanism is that clobetasol initially presents as a substrate on approaching the heme in CYP3A5, but the enzyme fails to oxidize the compound, leading to the formation of a pseudosubstrate–CYP complex that in turn inhibits the enzyme.

Identifying a CYP3A5-selective inhibitor has been challenging for many reasons, not least of which is the 83% sequence homology between CYP3A4 and CYP3A5⁷⁷. The potential applications for such a compound range from delineating substrates of CYP3A4 to studying the role of CYP3A5 as a mediator of drug resistance²⁰ or its involvement in oncogenic signaling⁷⁸. We have identified and characterized clobetasol as being capable of potently and selectively inhibiting CYP3A5. Our work lays the foundation for developing CYP3A5-selective inhibitors and can be expanded in various ways to uncover the roles of CYP3A5 both in catalytic mechanisms and in disease-relevant contexts.

Experimental Section

Compounds.

All compounds were obtained from commercial sources and the purity was 95% or higher. The compound name, vendor, catalog number, and analytical method used to determine purity are: Ritonavir, Toronto Research Chemicals, R535000, ¹H NMR; 1-hydroxymidazolam, Cayman Chemical, 10385, ¹H NMR; Clobetasol propionate, AK Scientific, F535, HPLC; Ketoconazole, Abovchem, AC513267, ¹H NMR; Midazolam, U.S. Pharmacopeia, 1443599, UHPLC and LC-MS/MS.

CYP3A4 and CYP3A5 biochemical inhibition.

To measure the inhibition of CYP3A4 or CYP3A5, the P450-Glo luminescence assay using the substrate Luciferin-IPA was purchased from Promega (Madison, WI; cat. no. V9002). Supersomes from Corning Life Sciences (Tewksbury, MA) were used as the source of purified recombinant human enzymes; they contained P450 oxidoreductase (POR), cytochrome b₅, and either CYP3A4 (Corning, cat. no. 456202) or CYP3A5 (Corning, cat. no. 456256), each at 1000 pmol/mL. Insect cell control supersomes were used as a negative assay control (Corning, cat. no. 456200). The assay protocol was performed in accordance with the P450-Glo manual, using final concentrations of 0.1 pmol of enzyme, 100 mM KPO₄ (pH 7.4) buffer, and 3 μM Luciferin-IPA as substrate. Upon the reactions being initiated by the addition of NADPH, the plates were incubated for 10 min at 37 °C. Reactions were then quenched for 20 min at room temperature, using the luciferin detection reagent. Luminescence was recorded with an EnVision 2102 Multilabel Plate Reader from PerkinElmer Life Sciences (Hopkinton, MA). All assays were performed using 384-well, polystyrene, white, opaque, non-treated plates (Corning, cat. no. 8850BC). The percentage inhibition was normalized to the average of six replicates containing either 30 μM ketoconazole (100% inhibition) or DMSO (0% inhibition) within each plate. The final concentration of DMSO for all compound and control wells of the assay plates was 0.1%.

For primary screening, an in-house library of bioactive compounds (n = 11,200, but some compounds are redundant) having diverse bioactivity²⁹ was screened at a single concentration of 5 μM against CYP3A5. Compounds conferring at least 60% inhibition (n = 423; 252 unique compounds) were then screened in a dose-response format against CYP3A4 and CYP3A5 in parallel. All compounds were screened in technical triplicate as 1:2 dilutions with concentrations ranging from 0.0073 μM to 15 μM. Scale transformation, normalization, curve fitting, and inhibitory concentration calculations were performed using GraphPad Prism 8.2.0 (GraphPad Software, La Jolla, CA). Specifically, the “log(inhibitor) vs. normalized response” equation was used to fit a standard slope curve based on normalized data and to subsequently extrapolate IC₅₀ and IC₉₀ values. Statistical comparisons were also performed in GraphPad Prism, using one-way ANOVA assuming equal standard deviations and comparing mean values.

Clobetasol inhibition profiling against a panel of major human CYPs.

Clobetasol propionate was prepared as a 25 mM stock solution in DMSO and stored at −20 °C. The compound was screened against a panel of major human CYPs by using the cytochrome P450 inhibition service from Cyprotex US, LLC (Watertown, MA). Serial dilutions of clobetasol were made in acetonitrile:DMSO (9:1) to yield final concentrations at 1:3 intervals ranging from 0.068 μM to 50 μM. The final DMSO concentration across all reactions was 0.2%. Clobetasol was incubated with pooled human liver microsomes (Bioreclamation-IVT, Baltimore, MD) in the presence of 2 mM NADPH in 100 mM KPO₄ buffer, pH 7.4, containing 5 mM MgCl₂ and the respective CYP probe substrate. The final volume of all reactions was 200 μL, and the assay was performed in technical triplicate. The probe substrate, probe concentration, microsomal protein concentration, incubation time, and positive control compound for each tested CYP were as follows: for CYP1A2: tacrine, 5 μM, 0.2 mg/mL, 10 min, α-naphthoflavone; for CYP2B6: bupropion, 100 μM, 0.25 mg/mL, 10 min, ticlopidine; for CYP2C8: amodiaquine, 5 μM, 0.25 mg/mL, 10 min, quercetin; for CYP2C9, tolbutamide, 100 μM, 0.5 mg/mL, 15 min, sulfaphenazole; for CYP2C19: mephenytoin, 100 μM, 0.25 mg/mL, 60 min, ticlopidine; for CYP2D6: dextromethorphan, 5 μM, 0.5 mg/mL, 10 min, quinidine; and for CYP3A4: midazolam, 2.5 μM, or testosterone, 50 μM, 0.25 mg/mL, 10 min, ketoconazole. Each reaction mixture was incubated at 37 °C. Reactions were terminated by adding methanol containing an internal standard for analytical quantification. Quenched samples were then incubated at 4 °C for 10 min and centrifuged at 6,102 × RCF. Supernatants were analyzed by LC-MS/MS for the following metabolites: hydroxytacrine for CYP1A2; hydroxybupropion for CYP2B6; desethylamodiaquine for CYP2C8; α-hydroxytolbutamide for CYP2C9; 4-hydroxymephenytoin for CYP2C19; dextromethorphan for CYP2D6; and 1-hydroxymidazolam or 6β-hydroxytestosterone for CYP3A4.

Samples were analyzed using a 5500 QTrap mass spectrometer (AB Sciex, Framingham, MA) in positive ionization mode, a 1290 Infinity Series autosampler and solvent delivery system (Agilent, Santa Clara, CA) at 10 °C, and an Acquity UPLC HSS T3 1.8 μm, 2.1 mm × 50 mm column (Waters, Milford, MA) at 50 °C. For the UPLC gradient, samples were collected at 0.00, 0.05, 1.00, 1.80, 1.81, and 2.80 min. At the collection times, the respective flow rates (in mL/min) were 0.6 for all times; the mobile phase A percentages were 98, 98, 5.0, 5.0, 98, and 98, respectively; and the mobile phase B percentages were 2.0, 2.0, 95, 95, 2.0, and 2.0. Mobile phase A consisted of water containing 0.1% formic acid, and mobile phase B consisted of acetonitrile containing 0.1% formic acid. Injection volumes were 10 μL for all samples.

Data mining to profile CYP3A5 expression across cancer types.

To determine the CYP3A5 expression levels in publicly available cancer samples, data were obtained from the UCSC TOIL recompute⁷⁹ of samples originating from the Pan-Cancer analysis project conducted by The Cancer Genome Atlas (TCGA)⁴³. Expression data from 33 cancer types (10,534 samples) were loaded into the R statistical environment [www.r-project.org] in the form of gene-level counts produced as the output of RSEM⁸⁰. The edgeR package⁸¹, with Limma⁸² and Voom⁸³, was used for linear modeling, empirical Bayes smoothing, and TMM normalization to produce normalized expression in units of log2(normalized CPM+1). CYP3A5 expression levels were plotted using the ggplot2 package⁸⁴ as ranked levels in descending order according to their median values. Data on CYP3A5 expression by sample, cohort abbreviation, and cancer type are available in Supplementary Data 2. The stepwise data-mining protocol is available in the Protocol Exchange⁴⁴.

RNA extraction and sequencing.

Total RNA was extracted from each cell line by using the Maxwell 16 LEV automation system and a Maxwell simplyRNA Purification Kit (Promega, cat. no. 1280) in accordance with the manufacturer’s protocol. The extracted RNA was quantified by spectrophotometry, using a NanoDrop 8000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA) to ensure a reading for OD260/OD280 between 1.8 and 2.0.

The RNA quality was further checked by TapeStation 4200 high-sensitivity RNA screen tape (Agilent, Santa Clara, CA) before library generation. Only high-quality samples with an RNA integrity number (RIN) of 8 or higher were used to construct the sequencing library. The RNA was fragmented using fragmentation reagent. For cDNA synthesis, the first-strand cDNA was generated using random hexamer-primed reverse transcription, after which the second-strand cDNA was synthesized. Libraries were prepared from total RNA with the TruSeq Stranded mRNA Library Preparation Kit (Illumina, San Diego, CA). Libraries were quantified using the Quant-iT PicoGreen dsDNA Assay (Life Technologies, Carlsbad, CA). One hundred–cycle paired-end sequencing was performed using an Illumina NovaSeq 6000 System to produce 100 bp paired-end reads.

RNA-seq data analysis.

Raw sequence files were merged across lanes according to sample and subjected to a first round of quality control by using the FastQC tool (www.bioinformatics.babraham.ac.uk/projects/fastqc/). Illumina universal adapters were trimmed from all samples using Trim Galore! ([www.bioinformatics.babraham.ac.uk/projects/trim_galore/), and samples were then subjected to a second round of quality control. Reads were mapped to the human Hg38 reference genome by using Bowtie2⁸⁵ in sensitive mapping mode. Gene-level quantification was obtained using RSEM⁸⁰ to produce raw read counts. The edgeR package⁸¹, with Limma⁸² and Voom⁸³, was used for linear modeling, empirical Bayes smoothing, TMM normalization, and all statistical comparisons. All P values were calculated based on t statistics then adjusted with the Benjamini–Hochberg false discovery rate. RNA-seq data from all included cell lines are available at the Gene Expression Omnibus (GEO) repository (accession no. GSE138437).

Cell culture.

AsPC-1 wild-type (WT), AsPC-1^{CYP3A5−/−} (3A5^−/−), and SU.86.86 cells were grown in culture in RPMI 1640 medium containing phenol red (Gibco, cat. no. 11875–093) with added 10% fetal bovine serum (HyClone, cat. no. SH30071.03), 1% Glutamax (Gibco, cat. no. 35050–061), and 1% penicillin streptomycin (Gibco, cat. no. 15140–122). AsPC-1 cells with CYP3A5 overexpression (WT + 3A5^OE cells), AsPC-1^{CYP3A5−/−} cells with CYP3A5 overexpression (“3A5^−/− + 3A5^OE” cells), and AsPC-1^{CYP3A5−/−}cells with CYP3A4 overexpression (“3A5^−/− + 3A4^OE” cells) were grown in culture in RPMI 1640 medium containing phenol red (Gibco) with added 10% tetracycline-free fetal bovine serum (Takara, cat. no. 631101), 2 μg/mL puromycin (Sigma, cat. no. P9620), 1 mg/mL G418 (Gibco, cat. no. 10131–027), and 1% Glutamax (Gibco). MIA PaCa-2 cells were grown in culture in DMEM containing phenol red (Gibco, cat. no. 11965–092) with added 10% fetal bovine serum (HyClone), 1% penicillin streptomycin (Gibco), and 2.5% horse serum (ATCC, cat. no. 30–2040). PANC-1 cells were grown in culture DMEM (Gibco) with added 10% fetal bovine serum (HyClone) and 1% penicillin streptomycin (Gibco). hTERT-HPNE cells were grown in culture in DMEM containing phenol red (Gibco) with added 5% fetal bovine serum (HyClone), 150 ng/mL puromycin (Sigma), 10 ng/mL human epidermal growth factor (Fisher Scientific, cat. no. 50400346), and 5.5 mM D-glucose (Sigma, cat. no. G8769). CFPAC-1 cells were grown in culture in IMDM containing phenol red (Life Technologies, cat. no. 12440–053) with added 10% fetal bovine serum (HyClone) and 1% penicillin streptomycin (Gibco). Capan-2 cells were grown in culture in McCoy’s 5A modified medium containing phenol red (ATCC, cat. no. 30–2007) with added 10% fetal bovine serum (HyClone) and 1% penicillin streptomycin (Gibco). HPAF-II cells were grown in culture in EMEM containing phenol red (ATCC, cat. no. 30–2003) with added 10% fetal bovine serum (HyClone) and 1% penicillin streptomycin (Gibco). Panc 02.13 cells were grown in culture in RPMI containing phenol red (Gibco) with 15% fetal bovine serum (Gibco), 1% penicillin streptomycin (Gibco), and 10 units/mL insulin (Sigma, cat. no. 91077C). All cell lines were obtained from ATCC (Manassas, VA), and all were maintained at 37 °C in 5% CO2. All cell lines have been authenticated by short tandem repeat (STR) DNA profiling and were routinely verified to be free of mycoplasma contamination.

Generation of stable AsPC-1 cells with inducible overexpression of CYP3A4 and CYP3A5.

pLVX-TRE3G-ZsGreen1 (cat. no. 631361) and pLVX-EF1a-TRE3G (cat. no. 631359) were obtained from Clontech Laboratories, Inc. (Mountain View, CA). CYP3A5 cDNA (OriGene RC207432) and CYP3A4 cDNA (OriGene SC125488) were obtained from OriGene Technologies Inc. (Rockville, MD). The following PCR primers were used to amplify the CYP3A5 MluI/EcoRI fragment and the CYP3A4 MluI/NdeI fragment for subcloning into the corresponding sites of the pLVX-TRE3G-ZsGreen1 vector, after gel extraction and purification using the QIAquick Gel Extraction Kit (cat. no. 28407) from Qiagen Science Inc. (Germantown, MD):

Tet-ZsGreen-3A5-MluI-F: 5′-GCCCCCGGGACGCGTGATGGACCTCATCCCAAATTTGG-3′

Tet-ZsGreen-3A5-EcoRI-R: 5′-CTACCCGGTAGAATTCTCATTCTCCACTTAGGGTTCCA-3′

Tet-ZsGreen-3A4-MluI-F: 5′-GAAAACGCGTATGGCTCTCATCCCAGACTTGGCCA-3′

Tet-ZsGreen-3A4-NdeI-R: 5′-AGCATATGTCAGGCTCCACTTACGGTGCCATC-3′

The resulting constructs, pLVX-TRE3G-ZsGreen1-CYP3A4 and pLVX-TRE3G-ZsGreen1-CYP3A5, were confirmed by sequencing. Lentiviruses for pLVX-TRE3G-ZsGreen1-CYP3A5 or pLVX-TRE3G-ZsGreen1-CYP3A4 and pLVX-EF1a-TRE3G transactivator were packaged and generated in 293T cells (ATCC), using medium with 10% Tet System Approved FBS (Takara Bio, Mountain View, CA; cat. no. 631101). AsPC-1 cells were transduced with lentiviral pLVX-TRE3G-ZsGreen1-CYP3A5 or pLVX-TRE3G-ZsGreen1-CYP3A4 and pLVX-EF1a-TRE3G transactivator. Treatment with 100 ng/mL doxycycline for 16 hours was sufficient to induce CYP3A4 or CYP3A5. Stable cells were selected by using 5 μg/mL puromycin and 1.5 mg/mL G418 for 2 weeks.

Western blot analysis.

Cells were rinsed once with cold PBS (Gibco 14190–144) then lysed in RIPA lysis buffer (Thermo Fisher Scientific) containing protease inhibitor (Roche). Lysates were loaded onto NuPAGE 4%–12% Bis-Tris gels (Invitrogen) with NuPAGE MES SDS running buffer (Invitrogen). The proteins were transferred from the gel to a nitrocellulose membrane by using the iBlot gel-transfer system (Invitrogen). The membrane was then blocked for 1 h with Odyssey Blocking Buffer (LI-COR Biosciences, Lincoln, NE) and probed with mouse monoclonal antibodies against CYP3A4 (K03; stock concentration: 2.5 mg/mL, diluted 1:5000) ⁸⁶, or CYP3A5 (Abcam, ab108624, 0.24 μg/mL), followed by probing with antibodies against β-actin (Sigma, cat. no. A5441, lot no. 043M4840V, diluted 1:5000). The membranes were then incubated with 1:5000 dilutions of secondary antibodies conjugated with species-specific infrared dyes (goat anti-rabbit IRDye® 800CW and goat anti-mouse IRDye® 680RD; LI-COR Biosciences). An Odyssey infrared imager (LI-COR Biosciences) was used to visualize the protein bands. At least three independent experiments were performed, and a representative gel is shown. For quantification, gel images were imported into Image Studio Lite. Equal size quantification rectangles were placed around each band to be measured. Measurements were visually inspected to ensure the total protein band was captured. After quantification of triplicate gels, the relative intensity of each protein band was then determined by normalizing the intensity of each protein band to that of actin. The relative intensity of control was set as 1.0.

CRISPR/Cas9-mediated deletion of CYP3A5.

AsPC-1 cells with full deletion of CYP3A5 were generated using CRISPR/Cas9 technology at the St. Jude Center for Advanced Genome Engineering (CAGE). Briefly, sgRNAs were designed with at least 2 bp of mismatch to any other site in the human genome to mitigate the risk of off-target editing. Two sets of sgRNAs were used sequentially to delete the entire ORF of both alleles (Set 1—g8: 5′-UGGCUGAAGACUGCUGUGCA-3′ and g4: 5′-UAAUGUACUGCAUGAGUAGU −3′; Set 2—g12: 5′-AACAGCAGCACUCAGCUAAA-3′ and g7: 5′-AGUUGAAAUCUCUGGUGUUC-3′). To generate the AsPC-1^{CYP3A5−/−} line, 400,000 cells were transiently co-transfected with 100 μmol of each sgRNA (Synthego) in Set 1, 35 μmol SpCas9 protein (from the St. Jude Protein Production Core), and 200 ng of pMaxGFP via nucleofection (with a 4D-Nucleofector™ X-unit; Lonza) in a small cuvette, using solution P3 and program EN-158, in accordance with the manufacturer’s recommended protocol. Five days post nucleofection, cells were single-cell sorted into 96-well plates by FACS for transfected cells based on pMaxGFP expression. After sorting, cells were clonally expanded and screened for the desired deletion by PCR-based assays and confirmed with targeted deep sequencing. Specifically, the deletion was detected using primers hCYP3A5.Del.F (5′-ACCCTTGGACTCCCCGATAACACTGA-3′) and hCYP3A5.Del.R (5′-TCTGATGAGAGCTCAGGAGGAGTTGA-3′). The nucleofection, cloning, and screening process was repeated with sgRNA Set 2 as described above, and the deletions were sequence confirmed by targeted deep sequencing. Additionally, the internal primers hCYP3A5.inner.F (5′-AGTCACAATCCCTGTGACCTG −3′) and hCYP3A5.inner.R (5′-GAAACCTCAGAACTCCCTCCC-3′) were used to verify the loss of the intervening sequence (by the absence of the band). CYP3A5 deletion was further verified by Western blot analysis, qRT-PCR, and functional testing.

LC-MS/MS detection of midazolam and 1-hydroxymidazolam in cells.

Cells were plated at a concentration of 10,000 cells per well in 384-well white, opaque, tissue culture–treated microplates (Corning, cat. no. 8804BC). Overexpression cell lines were induced with 100 ng/mL doxycycline. Twenty-four hours later, the cells were treated with clobetasol, ketoconazole, or DMSO. Both clobetasol and ketoconazole were tested in a dose-response format, using 16 concentrations in 1:2 dilutions ranging from 0.0003 μM to 10 μM. Additionally all groups were treated with 5 μM midazolam. The final DMSO concentration for all wells of the assay plates, including the controls, was 0.66%. Twenty-four hours after drug treatment, the reaction was quenched by adding acetonitrile (containing 4 μg/mL warfarin as an internal standard) to the wells in a 2:1 volume ratio. The plates were centrifuged for 20 min at 4,000 × g, and the reaction supernatants were diluted into an equal volume of dH₂O. Reference standards of midazolam and 1-hydroxymidazolam were also prepared in the same fashion, using culture medium and concentration ranges of 1.19 × 10⁻⁶ μM to 10 μM (for midazolam) or 1.19 × 10⁻⁷ μM to 1 μM (for 1-hydroxymidazolam). Samples were then frozen at −20 °C until analyzed by mass spectrometry.

LC-MS/MS analysis was performed using a SCIEX Triple Quad 6500 triple-quadrupole mass spectrometer (SCIEX, Forster City, CA) coupled to an ACQUITY UPLC system (Waters Corporation, Milford, MA). For each unknown sample and calibration sample, 10 μL was injected onto an Acquity UPLC HSS C18 2.1 mm × 50 mm column (particle size: 1.8 μm) (Waters Corporation). Chromatographic separation was performed by gradient elution at a constant flow rate of 1 mL/min for 2 min. The mobile phase consisted of 0.1% formic acid–water (solvent A) and 0.1% formic acid–acetonitrile (solvent B). The gradient applied was 0.0 min, 90% A–10% B; 0.3 min, 80% A–20% B; 1.35 min, 80% A–20% B; 1.65 min, 5% A–95% B; and 1.95 min, 10% A–90% B. The first 0.5 min of eluate was desalted to waste by an integrated Valco valve. The remaining eluates were directed to the triple quadrupole mass spectrometer, which was equipped with an electrospray ionization source. LC-MS/MS was performed in positive polarity (at 3000 V), and the source temperature was 650 °C. Gas 1 and gas 2 settings for nitrogen were set to 60. The curtain gas and collision gas were also nitrogen and were set to 20 and 10, respectively. Multiple reaction monitoring transitions were m/z 326 to m/z 291 for MDZ, m/z 342 to m/z 324 for 1-OH MDZ, and m/z 309 to m/z 163 for the IS (warfarin). The declustering potentials, entrance potential, collision energy, and collision cell exit potential were as follows: 120 V, 12 V, 35V, and 27 V, respectively, for MDZ; 70 V, 12 V, 30V, and 40 V for 1-OH MDZ; and 57 V, 12V, 44V, and 20 V for the IS. Data acquisition was conducted with Analyst 1.6.3 (SCIEX) and the data processes were operated with MultiQuant 2.1.1 software (SCIEX). The stepwise protocol for cell-based LC-MS/MS detection of CYP3A4 or CYP3A5 activity in cells will be available in the Protocol Exchange.

Confluence imaging.

Cells were plated at a concentration of 10,000 cells per well in 384-well, black, tissue culture–treated, clear-bottom polystyrene microplates (Corning, cat. no. 3712BC). All samples were processed in parallel with samples used for LC-MS/MS (and using the same methods) until just before the reaction-quenching stage. Cells were then imaged using a Lionheart FX Automated Live Cell Imager (BioTek, Winooski, VT). Single images of each well were acquired using phase-contrast microscopy with a 4× objective. An LED intensity of 10 was used, with a 100-millisecond integration time and a camera gain of 6.8. The laser was set to autofocus, and the wells were scanned at a distance of 600 μm in increments of 50 μm with a 0-μm well offset. The vibration CV threshold was set to 0.01. The instrument was set to add a 30-millisecond delay after plate movement. Lionheart FX software version 3.05.11 was used to obtain the cell confluence and export the well-by-well results. Confluence data were normalized in GraphPad Prism 8.2.0, using the average of six replicates of 0.66% DMSO (set to 100% confluence) for each cell line. Nonlinear regression curves were fitted using the “log(inhibitor) vs. response (three parameters)” equation applied to normalized data and were plotted to show normalized cell confluence over the tested concentration ranges.

Ligand docking and molecular dynamics simulations.

Published crystal structures of ritonavir-bound CYP3A4 (PDB: 3NXU, Chain A) or CYP3A5 (PDB: 5VEU, Chain A) were loaded into Maestro software (Schrödinger Release 2019–3). To prepare the protein for docking and simulations, the protein preparation wizard was used to assign bond orders, add hydrogens, create zero-order bonds to metals, create disulfide bonds, and fill in missing side chains and loops. Default parameters were used for the optimization of hydrogen-bond assignment (sampling of water orientations and use of pH 7.0). Waters beyond 5 Å of het groups or with fewer than three hydrogen bonds to non-waters were removed. Restrained energy minimization was applied using the OPLS3e⁸⁷ forcefield. Prepared protein systems were further checked by Ramachandran plots, ensuring there were no steric clashes. To generate receptor grids, ritonavir was selected as the grid-defining ligand for both the CYP3A4 and CYP3A5 systems. Default Van der Waals radius scaling parameters were used (scaling factor of 1, partial charge cutoff of 0.25).

For docking clobetasol into CYP3A4 and CYP3A5, the 3D structure of clobetasol was first obtained from the PubChem database (www.pubchem.ncbi.nlm.nih.gov). The virtual screening workflow panel was used to prepare clobetasol (by generating possible states at pH 7.0 ± 2.0 and retaining the specified stereochemical properties) and dock it into CYP3A4 and CYP3A5 in parallel. The most stringent docking mode (extra precision, “XP”) of Glide⁸⁸ was used, with the following parameters: dock flexibly, perform post-docking minimization, and keep 100% of scoring compounds. Of note, only one pose of clobetasol was returned for each system. For molecular dynamics simulations, systems were built for clobetasol-docked CYP3A4 and CYP3A5 by using the system builder panel of Desmond (Schrödinger Release 2019–3). The SPC solvent model was used, and the forcefield was set to OPLS3e. Solvated systems were loaded into the workspace by using the molecular dynamics panel. The total simulation time for each system was set to 200 nanoseconds, with 200-picosecond trajectory recording intervals. The system energy was set to 1.2, and the ensemble class used was NPT. Simulations were set to run at 300.0 K and at 1.01325 bar. The option to relax model systems before simulations was selected.

Cloning of CYP3A4 and CYP3A5 for bacterial protein expression.

The cloning procedure closely followed that previously described^{33, 89}. Cloning was performed in pCW ori⁺ plasmids harboring either CYP2C8 or CYP2C9 from Addgene: CYP2C8 and CYP2C9 in pCW ori⁺ were gifts from Joyce Goldstein (Addgene plasmid # 69604; http://n2t.net/addgene:69604; RRID:Addgene_69604 and plasmid # 69554; http://n2t.net/addgene:69554; RRID:Addgene_69554). The pCW3A4His plasmid was generated by replacing the CYP2C8 sequence of the pCW ori+-CYP2C8 plasmid with a codon-optimized CYP3A4 coding sequence at the NdeI/XbaI sites by GenScript (Piscataway, NJ), which expresses an N-terminal truncated version of CYP3A4 (with amino acids 3–24 truncated) with the His-4 tag at the C-terminal as previously described⁸⁹. The pCW3A5His expression plasmid was generated by PCR from a cDNA template coding for full-length CYP3A5 (OriGene, Rockville, MD) by using the forward and reverse primers 5′-ggattcGGCATATGGACTATCTATATGGGACCCGTACACATGGAC-3’ and 5′-cggaattccgAAGCTTTTAGTGGTGGTGGTGTCCATCTCTTGAATCCACCTTTAGAAC-3′ (Invitrogen, Carlsbad, CA). The underlined letters in the primers indicate engineered restriction sites (NdeI/HindIII). The bold letters indicate engineered ATG start and stop codons. The PCR reaction was accomplished using a ProFlex PCR system (Life Technologies, Carlsbad, CA) and Fusion High-Fidelity Master Mix (New England Biolabs, Ipswitch, MA). The amplification conditions were 95 °C for 2 min followed by 30 cycles of 95 °C for 30 s, 60 °C for 30 s, and 68 °C for 5 min, with a final polymerization step of 68 °C for 5 min. The PCR product was then digested using NdeI/HindIII and inserted into the restriction sites in a pCW ori⁺ plasmid (previously digested with the corresponding enzymes to remove CYP2C9) by using a Quick Ligation Kit (New England Biolabs). The resulting pCW3A5His construct encoded an N-terminal 3–24 trans-membrane helix amino acid deletion and a C-terminal 498–501 amino acid deletion. The C-terminal deletion in CYP3A5 produces a more soluble protein, as observed previously³³. In addition, a His-4 tag to aid nickel affinity chromatography was engineered at the C-terminal. No other mutations were made in the enzymes. All constructs were confirmed by DNA sequencing.

Expression and purification of CYP3A4 and CYP3A5.

The expression and purification methods closely followed those previous described, albeit with a few modifications^{33, 89}. The cloned pCW3A4His and pCW3A5His constructs were co-transformed with a pGro7 plasmid (Takara Bio, Japan) harboring the E. coli chaperone proteins groES and groEL (henceforth, groESL) into E. coli DH5-α cells, and transformed cells were co-selected using ampicillin and chloramphenicol (Gold Biotechnology, Olivette, MO). The chaperone proteins were co-transformed to aid solubility and improve protein yield. Expression of the individual (CYP3A4 or CYP3A5) gene constructs was achieved by using an isopropyl β-D-thiogalactopyranoside (IPTG) (Gold Biotechnology)–inducible tac promoter system in the pCW ori⁺ plasmid.

Protein production was typically done in 12-L cultures of Terrific Broth (TB) growth medium (Fisher Scientific, Hampton, NH) distributed between six 4-L conical flasks. Each flask contained 2000 mL of growth medium supplemented with ampicillin (50 μg/mL) and chloramphenicol (35 μg/mL). The medium was then inoculated with 20 mL of transformant cells from an overnight culture in the same medium and under the same conditions. The cells were then grown at 37 °C with 200 rpm agitation to an exponential phase with an OD₆₀₀ of 0.5, then the temperature was dropped to 28 °C and the agitation to 170 rpm. At an OD₆₀₀ of 0.7–0.8, IPTG (1 mM) and delta-aminolevulinic acid (1 mM) (Gold Biotechnology) were added to induce P450 production and to promote heme synthesis, respectively. Approximately 3 g/L of L-(+)-arabinose (Sigma-Aldrich, St. Louis, MO) was also added to induce expression of the E. coli chaperone proteins groESL. The transformant cells were grown for a further 48–50 h, then the cells were harvested by centrifugation at 6000 × g for 10 min at 4 °C by using an F9–6X1000 LEX rotor in a Sorvall Lynx6000 centrifuge (Thermo Fisher Scientific). The supernatant was discarded, and the cell pellets were resuspended in approximately 500 mL of 0.5 M potassium phosphate (KPi, pH 7.4) containing 20% glycerol and 1 mM phenylmethanesulfonyl fluoride (PMSF) protease inhibitor (Sigma-Aldrich). A few granules of lysozyme (Sigma-Aldrich) were added to facilitate cell lysis. The cells were lysed by passing them once through a microfluidizer (Microfluidics, Newton, MA); thereafter, the resulting lysate was supplemented with 10 mM CHAPS and incubated for 2 h (for CYP3A5) or 6 h (for CYP3A4) to release the proteins from the spheroplasts. The lysates were then centrifuged at 40,000 × g for 60 min at 4 °C and the supernatants were collected.

Each supernatant was mixed in batch with Ni-NTA resin (Qiagen, Hilden, Germany) pre-equilibrated with buffer A (0.5 M potassium phosphate, pH 7.4, 1 mM PMSF, 10 mM CHAPS, 20% glycerol) and incubated overnight at 4 °C with stirring. Next day, the supernatant–Ni-NTA resin was loaded onto a column and washed with 2 bed volumes (BV) of buffer A. The resin was then washed/buffer exchanged with 10 BV of buffer B (25 mM potassium phosphate, pH 6.8, 10 mM 2-mercaptoethanol, 20 mM NaCl, 2% Tween-20, 20% glycerol), followed by another 10 BV of buffer C (buffer B supplemented with 50 mM MgCl₂ and 5 mM ATP) to remove the groESL chaperone proteins. The resin was next washed with 5 BV of buffer B to remove excess ATP. The CYPs were then eluted from the resin by using buffer B supplemented with 200 mM imidazole and were analyzed spectrally (at 250–800 nm) and by SDS-PAGE.

The CYPs were then subjected to a second round of purification using a cation exchange chromatography column packed with carboxymethyl (CM) sepharose fast-flow resin (GE Healthcare) pre-equilibrated with buffer B. The column was mounted on an automated AKTA Avant purification system (GE Healthcare) and eluted via a linear gradient of NaCl (20–500 mM) with buffer B and buffer B2 (buffer B supplemented with 500 mM NaCl), and each fraction was analyzed by both spectroscopy and SDS-PAGE as before. Eluted fractions with high A₄₂₀/A₂₈₀ (Reinheitszahl, Rz) ratios (≥ 1) were pooled and concentrated to approximately 3–5 mL by ultrafiltration with a Vivaspin 20 concentrator at 4 °C. The CYPs were then subjected to a final purification step using a Sephacryl S-200 size-exclusion chromatography column and an AKTA Avant purification system with buffer C (100 mM HEPES, pH 7.4, 10 mM 2-mercaptoethanol, 200 mM NaCl, 20% glycerol). Fractions with Rz values of 1.3 or higher were pooled, concentrated, and stored at −80 °C until use.

UV–visible absorbance titrations with CYP3A4 and CYP3A5.

UV–visible experiments were carried out in accordance with previously described methods^{76, 90–92}. Optical titrations were performed to determine the dissociation constant (K_d) values for interactions of clobetasol, ketoconazole, and midazolam with CYP3A4 or CYP3A5. Titrations were performed using 1-cm path-length cuvettes in a SPECTRAmax PLUS³⁸⁴ UV–Visible spectrophotometer (Molecular Devices, San Jose, CA). Stock solutions (10 mM) of all compounds were made in DMSO-d₆ and titrated (in 1.0 μL aliquots) in cuvettes containing either ligand-free CYP3A4/CYP3A5 (4–6 μM) or buffer alone (as a negative control). The total DMSO-d₆ concentrations after saturation of enzyme were kept below 2.2%, and the absorbance spectra of CYP3A4 and CYP3A5 were not affected by DMSO-d₆ within this range. Continuous absorbance spectra (250–800 nm) were recorded at 25 °C. The negative control spectra generated from the buffer alone were subtracted from the protein absorbance spectra to eliminate the absorbance due to optical interference from small molecules. In addition, difference spectra were generated by subtracting the initial ligand-free protein absorbance spectrum from the ligand-bound spectrum, and the maximum change in absorbance calculated from each difference spectrum was then plotted against the corresponding ligand concentration. Data for midazolam binding were then fitted using the equation 1:

$$\mathrm{\Delta }A=\frac{\mathrm{\Delta }{A}_{max}{[L]}_T}{{[L]}_T+K_D}$$ (1)

where ΔA and ΔA_max are absorbance and maximum absorbance changes, respectively, [L]_T is the total ligand concentration, and K_d is the equilibrium dissociation constant. Data for ketoconazole and clobetasol binding, in which the assumption that [L]_total = [L]_free is not valid, were fitted using the equation 2:

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

where [P]_T is the total protein concentration⁹³. All data analysis was performed using GraphPad Prism (GraphPad Software, San Diego, CA).

EPR spectroscopic analysis of CYP3A4 and CYP3A5.

A continuous-wave X-band EPR spectrum was collected for both CYP3A4 and CYP3A5. Spectra were obtained at 10 K by using a Bruker ELEXSYS E500 EPR spectrometer equipped with an ER4122SHQ Super High Q cavity. An Oxford Instruments ESR900 cryostat connected to an ITC503 was used to control the temperature. The microwave power was set to 0.5 mW, with the frequency and modulation amplitude set to 10 GHz and 5 G, respectively. Spectra were collected for both CYPs in the ligand-free state (200 μM) and with the addition of exogenous compounds (400 μM).

Statistics.

For CYP inhibition (Fig. 1d), we performed statistical calculations by using one-way analysis of variance (ANOVA) assuming Gaussian distribution of residuals and equal standard deviation. Tuckey correction for multiple comparisons was applied. For CYP3A5 expression in RNA-seq experiments (Fig. 2b), the edgeR package with Limma and Voom was first used for linear modeling, empirical Bayes smoothing, and TMM normalization. Subsequently, all P values were derived from t statistics, then adjusted with the Benjamini-Hochberg false discovery rate (FDR) adjustment. For protein expression we either performed unpaired, two-tailed t tests, assuming Gaussian distribution and all populations having equal standard deviation (Fig. 3a), or used a one-way ANOVA, assuming Gaussian distribution of residuals and equal standard deviation (Fig. 3c). The Tuckey method of adjustment for multiple comparisons was applied. Information on sample size selection is indicated in the figures and corresponding Methods section, and no data were excluded from the analyses. For all statistical analyses, *** P ≤ 0.0001, ** P ≤ 0.005, * P ≤ 0.05, and ns = not significant (P ≥ 0.05).

Abbreviations Used

3A4^OE

CYP3A4 overexpression

3A5^OE

CYP3A5 overexpression

3A5^−/−

CYP3A5 genetic deletion

CHOL

cholangiocarcinoma

Clob/Clobetasol

clobetasol propionate

CW

continuous wave

CYP3A

cytochrome P450 3A family

Dox

doxycycline

HS

high-spin

Keto

ketoconazole

LIHC

liver and hepatocellular carcinoma

LS

low-spin

MDZ

midazolam

1OH-MDZ

1’-hydroxymidazolam

PDAC

pancreatic ductal adenocarcinoma

RMSF

root-mean-square fluctuation

RNAi

RNA interference

RNA-Seq

RNA sequencing

TCGA

The Cancer Genome Atlas