CYP26 Homology Modeling.
Characterization of the three-dimensional structure of a protein and, perhaps more importantly, its ligand binding regions can prove crucial to understanding the biological functions of a given drug target or metabolic enzyme. Though a definitive assessment of a protein’s structure generally relies on experimentally-determined structural data, the prediction of the protein’s structure through homology modeling can also prove valuable in understanding the structural features which contribute to its ligand binding characteristics and biological mechanism of action (Hillisch et al., 2004; Cavasotto and Phatak, 2009). A homology modeling approach aims to develop a computationally derived three-dimensional model for a protein with an unknown structure based on its similarity to other proteins with experimentally determined structures (Lesk and Chothia, 1986; Murzin, 2001; John and Sali, 2003; Zhang, 2008). The success of such an approach relies heavily on the degree of similarity between the two proteins as well as the generation of a correct sequence alignment between the two protein sequences (Khan et al., 2016). The data generated through the use of a homology model can be used to identify endogenous substrates or rationally design new xenobiotic ligands for the protein of interest, whose observed binding properties in the active site of the protein can then be used to further refine the model (Hillisch et al., 2004).
The first step in developing a homology model is the selection of a template protein with a known three dimensional structure and high degree of sequence similarity to the target protein with unknown structure using either comparative sequence assessment, multiple sequence assessments such as hidden Markov models and intermediate sequence searches for secondary structure prediction and fold recognition or a threaded template matching approach (Berman et al., 2000; Westbrook et al., 2002; Saxena et al., 2013). In general, a template structure with 30% or greater structural similarity to the target protein can provide the basis for a reliable homology model (Khan et al., 2016). After identification of an appropriate template, the two protein sequences must be aligned in such a manner to provide the greatest degree of sequence alignment, using either dynamic programming algorithms designed to take a relatively insensitive approach to aligning two sequences or a more complex approach where the target sequence is aligned to the protein sequences of multiple related proteins or where position specific information is incorporated into the sequence alignment (Sanchez and Sali, 1997a; Sanchez and Sali, 1997b). The latter approach becomes necessary for two proteins with less than optimal degrees of structural similarity (Xu et al., 1996; John and Sali, 2003).
Upon achievement of an acceptable degree of sequence alignment, a number of computational approaches are available with which to design and optimize the resulting homology model. Perhaps the most widely used approach is model assembly using a subset of 24 rigid protein sequences that are derived from the overall target sequence (Browne et al., 1969; Blundell et al., 1987; Greer, 1990; Blundell et al., 2006). A common approach to sub-dividing the protein sequence involves individually modeling the core backbone regions of the protein, followed by connecting variable loops and ultimately optimizing the individual amino acid side chains (Saxena et al., 2013). Additional approaches include coordinate reconstruction, where the segmentation of the target protein into hexapeptide segments leads to the ultimate structural assignments, spatial restraint modelling, where geometrical deviations from the template structure using a set of pre-defined spatial restraints are minimized, or loop modeling, used to define flexible regions in a protein structure (generally less than eight amino acid residues) based on known libraries and conformational searching or energy-based approaches (Krieger et al., 2003; Saxena et al., 2013). Model optimization follows the initial model design. Owing to the dynamic interplay between the predicted structure of the backbone core regions and the geometry of the individual side chains, an iterative approach is often utilized where the effects of energy minimization of the side chains on the backbone and vice-versa are taken into account over multiple cycles until the entire model eventually converges into a global energy minimum (Krieger et al., 2003). In general, the energy minimization step is achieved by using either classical molecular dynamics or quantum force fields, which incorporate the positions of each atomic nucleus as well as the inherent charge distribution and treat the overall protein as a sum of the individual amino acids or a self-parameterizing force field, which build upon the aforementioned force fields by randomly changing a given parameter such as van der Waals radii and reminimizing the model to determine if an improvement in the model was obtained (Liu et al., 2001; Krieger et al., 2002).
The final step in homology modeling is the validation of the model, which identifies errors inherent to all homology models. As homology modeling is an inherently iterative process, errors in model design are easily propagated and can result in serious deficiencies in the final model design. Commonly observed errors include sequence misalignments, incorrectly 25 assigned geographical sequences of the target protein, unacceptable side-chain conformations and abnormal bond lengths or angles (Morris et al., 1992; Czaplewski et al., 2000; Czaplewski et al., 2003; Krieger et al., 2003; Saxena et al., 2013). Validation of the model can involve the entire model or distinct subdivided regions of the model, with each approach generating scoring functions that evaluate template alignment, protein stereochemistry and protein misfolding, in addition to a plentitude of other parameters (Sippl, 1995; Hooft et al., 1996; Marti-Renom et al., 2000; Hillisch et al., 2004). The scoring functions can generally be divided into either statistical-based or physical-based energy functions (Sippl, 1995; Lazaridis and Karplus, 1999; Al-Lazikani et al., 2001; Xiang, 2006). The former scoring functions incorporate the well-characterized properties of amino acids in a given structure while the latter is based on calculating the conformational free energy of the overall protein structure (Xiang, 2006).
Sequence Verification and Expression of CYP26B1.
To express recombinant CYP26B1, the human CYP26B1 cDNA was obtained from OriGene Technologies (Rockville, MD) (catalog number TC120799). Upon sequencing of the obtained clone, two single nucleotide polymorphisms were discovered that differed from the sequence reported in NCBI (Q9NR63). The two SNPs were an A191G conversion resulting in an H>R amino acid change and a G788A conversion resulting in a G>S amino acid change (CYP26B1*1, Figure 9-1). To determine which of the possible SNPs would be reflective of the CYP26B1 sequence in the human population, genomic DNA was extracted from 12 human livers from the University of Washington human liver bank and the two sections of the CYP26B1 gene were sequenced in all 12 donors. In brief, genomic DNA (50 ng) was amplified by PCR using either forward (5’-TCTTTGAGGGCTTGGATCTG-3’) and reverse (5’-GGCAGAGAGGGAAGG-3’) primers for the A191G SNP or forward (5’ GACAAAGGGGAGAGGTGTCA-3’) and reverse (5’-GTAGAAATGGCTGGGCACAT-3’) primers for the G778A SNP at concentrations of 400 nM. The primers and template DNA together with a Ready-to-Go bead (puReTaq Ready-to-Go PCR beads, Amersham Biosciences, Piscataway, NJ) were mixed in a final volume of 25 μL and PCR amplification was done as follows: after an initial denaturing step at 94°C for 4 min, amplification was performed for 32 cycles of denaturation (94°C for 30 s), annealing (55°C for 20 s), and extension (72°C for 30 s), followed by a final extension at 72°C for 30 s. PCR products were analyzed by gel electrophoresis, spin column-purified to remove unincorporated nucleotides and primers using the QIAquick® PCR Purification Kit (Qiagen Inc., Hilden, Germany) and sequenced for the forward and reverse direction on an ABI Prism 377Xl DNA Sequencer (Applied Biosystems, Foster City, CA) with the ABI Prism® BigDyeTM Terminator Cycle Sequencing Ready Reaction Kit (PerkinElmer, Waltham, MA). After the wild-type sequence was confirmed, the CYP26B1 coding sequence from the original clone was amplified while adding a 6xHis tag with a TEV cleavage site to maintain similarity with the commercially available clone as previously described (Topletz et al., 2012). CYP26B1 protein was expressed using the Bac-to-Bac baculovirus expression system (Life Technologies, Grand Island, NY) in Sf9 cells according to the manufacturer’s instructions as described previously (Topletz et al., 2012). Sf-900 II SFM liquid media (Life Technologies, Grand Island, NY) supplemented with 2.5% fetal bovine serum was used and during protein expression ferric citrate (0.2 mM) and δ-aminolevulinic acid (0.3 mM) were added to the media 24 hours post-infection to facilitate heme synthesis. The cells were harvested 48 hours post infection, washed once in PBS with 1 mM PMSF, pelleted and stored at –80˚C. Membrane fractions containing CYP26B1 were prepared by centrifugation as described previously (Topletz et al., 2012) and P450 content determined via CO-difference spectrum.
IC50 Determination for Retinoic Acid Receptor Agonists.
Six retinoic acid receptor agonists were assessed for in vitro inhibition of CYP26A1 and CYP26B1 catalyzed 9-cis-4-hydroxyretinoic acid formation. Various concentrations of each inhibitor (0 – 100 µM) were incubated with 5 pmol CYP26A1 or CYP26B1, 10 pmol cytochrome P450 reductase, and 100 nM 9-cis-retinoic acid in 100 mM potassium phosphate buffer (pH 7.4). Incubations were initiated by the addition of 1 mM NADPH (final concentration) and quenched after 2 minutes (CYP26A1) or 5 minutes (CYP26B1) with 5 volumes of ethyl acetate containing acitretin as an internal standard. All samples were evaporated to dryness under a gentle stream of N2, reconstituted in methanol and assayed for 9-cis-4-hydroxyretinoic acid concentrations by HPLC-UV as previously described (Thatcher et al., 2011). All IC50 determinations were conducted in triplicate.
Homology models of CYP26A1 and CYP26B1 were constructed using Prime (Schrodinger LLC, New York). The amino acid sequence of human CYP26A1 was obtained from the NCBI protein server (GenBank ID: 2688846) and the CYP26B1 amino acid sequence was obtained as described above. CYP120 (crystal structure, pdb 2VE3) was used as the template for both homology models. Compared to CYP120, CYP26A1 had 33% sequence identity and 53% positive sequence coverage while CYP26B1 had 34% sequence identity and 54% positive sequence coverage. The heme prosthetic group was added to each homology model and ligated to Cys442 (CYP26A1) or Cys441 (CYP26B1), followed by energy minimization prior to ligand docking using OPLS_2005 force field constraints as defined within the MacroModel algorithm (Schrodinger LLC, New York). In order to flexibly dock at-RA, tazarotenic acid and tazarotenic acid sulfoxide, a ligand grid (12 x 12 x 12 Å) for which the center of mass of each ligand would be constrained to was defined and centered approximately 2 – 3 Å above the heme iron using Glide (Schrodinger LLC, New York). Structural rationalization of each homology model was performed through the evaluation of Ramachandran plots and model assessment of odd bond lengths and angles (Figure 9-2). Determination of model flexibility was assessed by comparison of helical versus loop motifs and by prediction of 2° structure characteristics using PSIPRED (University College London, UK) and SSPro (Schrodinger).
Tazarotenic Acid Phenotyping.
To assess the relative contribution of CYP26A1 and CYP26B1 to the in vitro oxidative metabolism of tazarotenic acid, metabolite formation was monitored across a panel of drug metabolizing enzymes. Previously reported studies to characterize the enzymes responsible for the metabolism of tazarotenic acid were conducted at substrate concentrations of 1 – 200 µM (Attar et al., 2003). As total circulating plasma concentrations of tazarotenic acid are approximately 1 – 280 nM following typical doses of tazarotene, current studies were conducted using clinically relevant substrate concentrations. In vitro incubations consisted of 5 nM recombinant enzyme, 50 nM purified human reductase and 100 nM tazarotenic acid (final concentrations) in 100 mM potassium phosphate buffer (pH 7.4). Following a three minute pre-incubation at 37 ºC, reactions were initiated with the addition of 1 mM NADPH (final concentration). For incubations utilizing flavin-containing monooxygenase enzymes, the pre-incubation step consisted of enzyme and NAPDH followed by initiation with substrate due to the known instability of FMOs at 37 ºC in the absence of cofactor (Foti and Fisher, 2004). Incubations (50 µL, final volume) were carried out for 30 minutes at 37 ºC before being quenched with 3 volumes (v/v) of ice-cold acetonitrile containing tolbutamide as an internal standard. Samples were vortexed and centrifuged at 1240 x g for ten minutes before being transferred for LC-MS/MS analysis. Data was expressed as the percent of total metabolite formed across the panel of enzymes for each individual metabolite.
Analysis of tazarotenic acid and its metabolites was conducted using LC-MS/MS. The analytical platform was comprised of an Applied Biosystems API4000 fitted with an electrospray ionization source (Applied Biosystems, Foster City, CA). Liquid chromatography and sample introduction was achieved using two LC-20AD binary pumps with an in-line DGU-20A5 solvent degasser (Shimadzu, Columbia, MD) and a LEAP CTC HTS PAL autosampler (CTC Analytics, Carrboro, NC). An injection volume of 10 μL was used for all analyses. For enzyme kinetic experiments, chromatographic separation was achieved using 0.1% formic acid (v/v) in water (mobile phase A) and 0.1% formic acid in methanol:acetonitrile (1:1; mobile phase B) on a Synergi 2.5 µm Hydro RP 100 Å (50 x 2.0 mm) column (Phenomenex, Torrance, CA). Gradient conditions consisted of 2.5% B (0 – 0.4 minutes), 2.5% B – 95% B (from 0.4 – 1.4 minutes), 95% B (from 1.4 – 2.5 minutes) and re-equilibration at 2.5% B for 0.5 minutes. For metabolite identification experiments, the same mobile phase system was used with a Kinetex 2.6 µm C18 100 Å (100 x 2.1 mm) column (Phenomenex, Torrance, CA). A gradient of 2.5% B (0 – 3 minutes), 2.5% B – 95% B (from 3 – 14 minutes), 95% B (from 14 – 17 minutes) followed by re-equilibration at 2.5% B for 3 minutes was used to achieve chromatographic separation of all analytes. Initial metabolite identification experiments used full scan analysis from 100 – 800 amu followed by analysis of the corresponding product ion spectra for each observed analyte. Subsequent LC-MS/MS analyses utilized multiple reaction monitoring (MRM) for each analyte. MRM transitions (positive ionization mode) were as follows: tazarotenic acid (m/z 324.2 / 294.3), tazarotenic acid sulfoxide and hydroxytazarotenic acid (m/z 340.3 / 280.3), tazarotenic acid sulfone (m/z 356.3 / 276.3) and the internal standard tolbutamide (m/z 271.2 / 91.1). Generic parameters applied to all MS analyses included the curtain gas (12 arbitrary units), CAD gas (medium), ion spray voltage (5000 V), source temperature (500 °C) and ion source gas 1 and gas 2 (30 arbitrary units, each).
Mass spectrometry data was evaluated using Analyst (version 1.5; Applied Biosystems, Foster City, CA). Analyte concentrations were determined by comparing peak areas in unknown samples to those obtained from standard curves with analytical standards (dynamic range: 1 – 2000 nM; weighting: 1/x). Parameter fitting for IC50 and enzyme kinetic data was performed using Graphpad Prism as described below (version 6.03; Graphpad Software Inc., San Diego, CA).
IC50 values for retinoic acid receptor agonists in the 9-cis-4-hydroxyretinoic acid assay were determined by nonlinear regression using Equation 1. In the following equation, 100%*(Vi/V) represents the percent activity remaining for a given inhibitor concentration, [I], (Vi/V)max*100% is the maximum observed activity with no inhibitor present, and (Vi/V)min is the remaining enzyme activity at infinitely high concentrations of inhibitor.
Equation 1 100% * vi Enzyme kinetic parameters (Km and Vmax) were estimated through nonlinear regression analysis using the Michaelis-Menten model as shown in Equation 2. In the equation below, Km denotes half the substrate concentration ([S]) at maximal reaction velocity (Vmax).
Table of contents :
1. Table of Contents
1. Table of Contents
3. Funding Sources
4. List of Tables
5. List of Figures
6. List of Abbreviations
7. Chapter I: Introduction to Retinoic Acid Signaling and Cytochrome P450 26
7.2. Retinoic Acid Signaling.
7.3. Cytochrome P450.
7.4. Role of CYP26.
7.5. CYP26 Pharmacology.
7.6. CYP26 Homology Modeling.
8. Aims and Scope.
9. Chapter II: Identification of Tazarotenic Acid as the First Xenobiotic Substrate of Human Retinoic Acid Hydroxylase CYP26A1 and CYP26B1
9.2. Materials and Methods.
9.2.2. Sequence Verification and Expression of CYP26B1.
9.2.3. IC50 Determination for Retinoic Acid Receptor Agonists.
9.2.4. Homology Modeling.
9.2.5. Metabolic Profiling.
9.2.6. Enzyme Kinetics.
9.2.7. Tazarotenic Acid Phenotyping.
9.2.8. LC-MS/MS Analysis.
9.2.9. Data Analysis.
9.3.1. Homology Modeling.
9.3.2. Metabolic Profile.
9.3.3. In Vitro Enzyme Kinetics.
9.3.4. Tazarotenic Acid Phenotyping.
10. Chapter III: Comparison of the Ligand Binding Site of CYP2C8 with CYP26A1 and CYP26B1: A Structural Basis for the Identification of New Inhibitors of the Retinoic Acid Hydroxylases
10.2. Materials and Methods.
10.2.2. Homology Modeling and Computational Docking Simulations.
10.2.3. In Vitro Inhibition Assays.
10.2.4. Spectral Binding Determination.
10.2.5. Assessment of In Vitro Free Fraction.
10.2.6. In Vitro Stability of Candesartan Cilexetil.
10.2.7. Calculation of Cmax,u / IC50.
10.2.8. Liquid Chromatography – Mass Spectrometry Analysis.
10.3.1. Evaluation of tazarotenic acid sulfoxide formation as a probe substrate of CYP26.
10.3.2. In Vitro Inhibition Screening and IC50 Determination.
10.3.3. Computational Docking Simulations
10.3.4. Spectral Binding Studies.
10.3.5. Calculation of Cmax,u / IC50.
11. Chapter IV: Contribution of CYP26 to the Metabolism and Clearance of Retinoic Acid Receptor Agonists and Antagonists
11.2. Materials and Methods.
11.2.2. In Vitro Clearance of Retinoic Acid Receptor Agonists and Antagonists by Recombinant CYP26s, CYP2C8 and CYP3A4.
11.2.3. Adapalene Phenotyping.
11.2.4. Metabolite Identification of Adapalene and Des-Adamantyl Adapalene in Recombinant CYP26s.
11.2.5. Computational Docking of Adapalene and Des-Adamantyl Adapalene in CYP26A1, CYP26B1 and CYP26C1 homology models.
11.2.6. LC-MS/MS Analysis.
11.2.7. Data Analysis.
11.3.1. In Vitro Clearance of Retinoic Acid Receptor Agonists and Antagonists by Recombinant CYP26s, CYP2C8 and CYP3A4.
11.3.2. Metabolite Identification of Adapalene and Des-Adamantyl Adapalene in Recombinant CYP26s.
11.3.3. Computational Docking of Adapalene and Des-Adamantyl Adapalene in CYP26A1, CYP26B1 and CYP26C1 homology models.
11.3.4. Adapalene Phenotyping.
12. Chapter V: General Conclusions