Abstract
Two novel homodimer metabolites were identified in rat samples collected during the in vivo study of GDC-0994. In this study, we investigated the mechanism of the formation of these metabolites. We generated and isolated the dimer metabolites using a biomimetic oxidation system for NMR structure elucidation to identify a symmetric dimer formed via carbon-carbon bond between two pyrazoles and an asymmetric dimer formed via an aminopyrazole-nitrogen to pyrazole-carbon bond. In vitro experiments demonstrated formation of these dimers was catalyzed by cytochrome P450 enzymes (P450s) with CYP3A4/5 being the most efficient. Using density functional theory, we determined these metabolites share a mechanism of formation, initiated by an N-H hydrogen atom abstraction by the catalytically active iron-oxo of P450s. Molecular modeling studies also show these dimer metabolites fit in the CYP3A4 binding site in low energy conformations with minimal protein rearrangement. Collectively, the results of these experiments suggest that formation of these two homodimer metabolites is mediated by CYP3A, likely involving activation of two GDC-0994 molecules by a single P450 enzyme and proceeding through a radical coupling mechanism.
SIGNIFICANCE STATEMENT These studies identified structures and enzymology for two distinct homodimer metabolites and indicate a novel biotransformation reaction mediated by CYP3A. In it, two molecules may bind within the active site and combine through radical coupling. The mechanism of dimerization was elucidated using density functional theory computations and supported by molecular modeling.
Introduction
GDC-0994 (structure shown in Fig. 1A) is a selective, potent, orally bioavailable, small molecule inhibitor of extracellular signal-regulated kinase (ERK)1 and ERK2. It has proven antitumor activity in RAS- and RAF-mutant human tumor cell-line models in vitro and in vivo (Blake et al., 2016). Based on its mechanism of action and the high frequency of mitogen-activated protein kinase pathway activation in human cancers, GDC-0994 was in development as an anticancer therapeutic with potential in a broad array of cancer indications (Ren et al., 2015; Kirouac et al., 2017) including its clinical testing in patients with locally advanced or metastatic solid tumors (Varga et al., 2020). While thoroughly evaluating the metabolic profile of GDC-0994 in rats, the nonclinical toxicological species that supported entry into clinical development, we identified two novel dimer metabolites (M13 and M14) that formed via cytochrome P450 (P450) oxidation.
P450 enzymes catalyze a wide range of reactions for the metabolism of xenobiotics. The most common reactions include carbon hydroxylation, heteroatom oxygenation, dealkylation, and epoxidation (Ortiz de Montellano and De Voss, 2005). However, a variety of uncommon reactions by P450s have been observed. These reactions are of particular interest because their products are often not easily predicted (Guengerich and Munro, 2013). From a research perspective, these case examples provide interesting opportunities to probe the diversity of P450 function. From a drug discovery and development perspective, an unexpected metabolite adds to uncertainties and associated risks of a candidate molecule upon its first administration to human subjects (Schadt et al., 2018). Therefore, understanding reaction mechanisms, structural requirements, and specific isoform(s) responsible for these reactions is critical to anticipate and mitigate metabolic liabilities.
Of the P450 isoforms found in humans, CYP3A4 is most frequently identified as the principalmetabolizing enzyme of approved drugs (Cerny, 2016). CYP3A4 has a large and flexible active site that accommodates a variety of ligands, including large substrates such as cyclosporine A (molar mass 1203 g/mol). Understanding the diversity of reactions that CYP3A4 can catalyze provides a foundation to understand the breadth of possible metabolic products it can generate and feed back to the design and optimization of new molecular entities.
In this report, we describe the identification and structure determination of two distinct homodimers of GDC-0994 and provide in vitro evidence that CYP3A catalyzed their formations. We also embarked on a computational study using density functional theory (DFT) to determine the likely mechanism of dimer formation and molecular modeling to evaluate the feasibility of dimer formation within CYP3A4.
Materials and Methods
Chemicals and Reagents.
[14C]GDC-0994 was synthesized by Selcia Limited (Essex, UK) with radiochemical purity >98.5% (specific activity 52.23 mCi/mmol), and nonradiolabeled GDC-0994 was synthesized at Genentech (South San Francisco, CA). Liver microsomes and recombinant P450 enzymes were purchased from BD Gentest (San Jose, CA) or Bioreclamation IVT (Westbury, NY). Horseradish peroxidase was obtained from Sigma-Aldrich (St Louis, MO). Myeloperoxidase from human leukocytes was purchased from EMD Millipore Co. (Temecula, CA). All chemical reagents and solvents were purchased from commercial sources at the highest purities possible.
In Vivo Metabolite Identification Study.
Male and female Sprague-Dawley rats were orally administered a single dose of [14C]GDC-0994 at 50 mg/kg (100 µCi/kg) in 1% carboxymethylcellulose, 0.5% Tween, and 5 mM citrate solution. In vivo study design, sample collection, and radioanalysis are described in Supplemental Materials. Metabolite profiling and identification was performed using liquid chromatography coupled with simultaneous radiodetection and high-resolution mass spectrometric analysis. Method details are also provided in Supplemental Materials.
Metabolite Generation Using Biomimetic Enzymes.
BMO Production Kit (HCK1001-03-027-A7-10) (HepatoChem Inc., Beverly, MA) was used to generate, scale up, and isolate two metabolites of interest, M13 and M14. Standard protocols provided by the vendor were followed for screening, optimization, and production of metabolites. The incubations were performed with 6.25 mM of GDC-0994 free base. The dimer metabolites were purified by semipreparative liquid chromatography using a Hypersil Gold C18 column (150 × 10 mm, 5 µm particle size; Thermo Scientific, San Jose, CA) with mobile phases of 0.1% formic acid in water and acetonitrile at a constant flow rate of 4.6 ml/min. The incubation products were separated by gradient elution, and fractions were collected and surveyed by liquid chromatography–mass spectrometry for the analytes of interest. The fractions containing M13 and M14 were then separately combined and evaporated to dryness under vacuum.
Structure Determination by NMR.
All NMR experiments were collected on an Avance II spectrometer (Bruker BioSpin, Billerica, MA) operating at 500 MHz for 1H, 125 MHz for 13C equipped with 5 mm TCI cryoprobe with Z-gradient. Each metabolite sample was dissolved in 180 µl of DMSO-d6 (Cambridge Isotope Laboratories, Tewksbury, MA). One-dimensional 1H and 13C NMR were collected at ambient temperature and 335 K, respectively. Chemical shifts were referenced to tetramethylsilane at δ = 0 ppm. Two-dimensional correlation spectroscopy, heteronuclear single quantum correlation spectroscopy, heteronuclear multiple bond correlation spectroscopy, and rotating frame nuclear Overhauser effect spectroscopy were collected at 335 K.
In Vitro Metabolism.
GDC-0994 (10 µM) was incubated with liver microsomes (male rat, female mouse, male beagle dog, male cynomolgus monkey, or 200 mixed sex donor pooled human, 0.5 mg/ml) in potassium phosphate buffer (100 mM, pH 7.4) containing magnesium chloride (3 mM) supplemented with 1 mM NADPH in an incubation volume of 0.5 ml. Incubations with individually expressed recombinant P450 enzymes (Supersomes) were completed similarly with GDC-0994 (1 µM) and a constant amount of P450 (40 pmol/ml). Inhibition experiments were conducted with human liver microsomes using standard concentrations of P450 isoform-selective chemical inhibitors: 1-aminobenzotriazole (ABT; 1 mM, pan-P450), furafylline (10 µM, CYP1A2), 2-phenyl-2-(1-piperidinyl)propane (20 µM, CYP2B6), montelukast (2 µM, CYP2C8), sulfaphenazole (10 µM, CYP2C9), 3-benzylnirvan (2 µM, CYP2C19), quinidine (1 µM, CYP2D6), ketoconazole (1 µM, CYP3A4), or troleandomycin (20 µM, CYP3A4). In the incubations with ABT and troleandomycin, human liver microsomes were preincubated with inhibitors for 15 minutes before GDC-0994 was added to the incubations. At the end of incubation, reactions were stopped, and proteins were removed by adding acetonitrile and centrifuging.
GDC-0994 (100 µM) was incubated with horseradish peroxidase (40 U/ml), NaCl (150 mM), and H2O2 (100 µM) in 100 mM phosphate buffer pH 7.4 at 37°C for 1 hour. GDC-0994 (100 µM) was separately incubated with myeloperoxidase (0.1 mg/ml), NaCl (150 mM), and H2O2 (100 µM) in 100 mM phosphate buffer pH 6.0 at 37°C for 1 hour. Reactions were quenched with three volumes of acetonitrile. The supernatant was diluted with three volumes of water and injected for liquid chromatography–mass spectrometry analysis.
Computational Methods.
All DFT computations were performed using Gaussian09 (Frisch et al., 2013). Geometry optimizations and frequency calculations were performed at the B3LYP (Becke, 1993) level using LANL2DZ for Fe and 6-31G(d) on all other atoms. Frequencies were used to check for the presence of a local minimum (zero negative frequencies) or a transition structure (exactly one negative frequency, corresponding to the desired transformation). Frequencies were also used to compute enthalpy and free energy corrections at a standard state of 1 atm and 298.15 K. An additional correction for entropy was used to apply a quasiharmonic approximation, as discussed by Truhlar (Zhao and Truhlar, 2008; Ribeiro et al., 2011). Single point energy calculations were computed for each structure at the B3LYP-D3(BJ) (Grimme et al., 2011) level using LANL2DZ for iron and 6-311+G(d,p) on all other atoms and the SMD (Marenich et al., 2009) implicit solvent models for water.
Molecular Modeling.
Molecular modeling studies were performed using MOE (Chemical Computing Group ULC, 2018) with the MMFF94x forcefield (Halgren, 1996a,b,c,d; Halgren and Nachbar, 1996; Halgren, 1999a,b), and docking was performed with Glide (Schrödinger LLC, 2016). Conformational analyses of the dimer structures were performed with the sdfMMConfAnalysis module of the Chemalot software package (Lee et al., 2017).
Results
Identification of Metabolites.
GDC-0994 was extensively metabolized in rats (Supplemental Fig. 1). A total of 14 metabolites were tentatively identified (Supplemental Fig. 2), primarily consisting of oxidation, N-dealkylation, glucuronidation, or a combination of these reactions. Two distinct radiopeaks, M13 and M14, which eluted later than unchanged GDC-0994, were observed in bile and feces profiles (Supplemental Fig. 1). The protonated molecular ions observed in M13 and M14 were identical [m/z 879.2343, C42H35N12O4Cl2F2 (−0.1 ppm)], which corresponded to two molecules of GDC-0994 connected by loss of two hydrogens, suggesting they are homodimer metabolites. The MS2 and MSn product ion spectra of GDC-0994, M13, and M14 are shown in Fig. 1. M13 and M14 shared highly similar MS2 and MS3 fragmentation spectra (Fig. 1, B and C), which showed product ions at m/z 707 (loss of chloro fluoro phenyl ethanol), m/z 689 (loss of water from m/z 707), and m/z 535 (loss of chloro fluoro phenyl ethanol from m/z 707). However, their MS4 spectra were dramatically different, implying the connectivity of the dimers was the structural distinction between M13 and M14. Heterodimers composed of 1 molecule of GDC-0994 and 1 molecule of the N-desmethyl metabolite (M12) were only detected in bile by mass spectrometry in significantly lower quantities than M13 or M14. Homodimers of M12 were not detected in any matrices, presumably due to its low overall concentration.
Dimer Metabolite Generation and Structure Elucidation.
To fully characterize the structures of the proposed GDC-0994 dimers, we attempted to scale up the metabolites with various in vitro systems. The product yields in microsomal or expressed P450 incubations were too low to efficiently obtain sufficient amounts of product. Thus, metabolite generation was surveyed and scaled using organometallic catalysis purchased from HepatoChem. This tool provides biomimetic catalysts to mimic the suite of P450 enzymes present in human hepatocytes, offering the researcher a unique approach to quickly find reaction conditions that generate target drug metabolites and scale them to provide sufficient quantities for structural characterization or pharmacological potency testing (Chen and White, 2007; Cusack et al., 2013). From the biomimetic screening panel, M13 and M14 were generated to greatest extent by metalloporphyrin FeTDClPP and H2O2 as oxidant. This catalytic system was used to generate sufficient quantities of metabolites for isolation by liquid chromatography and fractionation for definitive structure characterization by NMR.
Based on NMR studies, M13 is a symmetrical dimer, whereas M14 is an asymmetric dimer of GDC-0994 (Fig. 2). At room temperature, proton NMR spectra of both M13 and M14 showed broad resonances due to restricted rotation. Thus, both M13 and M14 were heated to 335 K to obtain well resolved one-dimensional proton NMR and two-dimensional data. Proton NMR of M13 gave only one set of resonances, which was comparable to those observed for GDC-0994. The distinct differences were that one pyrazole proton H16 at 6.3 ppm was missing and the other pyrazole proton H17 became singlet by losing coupling with H16. There was also an extra quaternary carbon on carbon-13 NMR of M13. These structural features were consistent with two GDC-0994 molecules forming a symmetrical dimer at C16 position. The new carbon-carbon bond formed in M13 was labeled as C16-C47. For M14, proton NMR gave two sets of resonances consistent with GDC-0994 with two resonances missing, one pyrazole proton H16 from one GDC-0994 molecule, and H45 from N-H of another GDC-0994 molecule. This evidence concludes that a carbon-nitrogen bond was formed between two GDC-0994 molecules to give an asymmetric dimer with same molecular weight as M13. The new carbon-nitrogen bond formed in M14 was labeled as C16-N45.
In Vitro Formation of Dimer Metabolites.
In vitro studies demonstrated dimer formation with liver microsomal incubations for rats, mice, rabbits, dogs, monkeys, and humans, albeit all to low extent. With human liver microsomes, the formations of the dimers were time and NADPH dependent and were abolished when P450 enzymes were inactivated by preincubating with ABT or ketoconazole, pan-P450 and CYP3A inhibitors, respectively (Supplemental Fig. 3). Incubations with singly expressed recombinant P450s showed CYP3A4 and 3A5 were the most productive in generating M13 and M14.
Incubations of GDC-0994 with horseradish peroxidase in the presence of Cl− and H2O2 produced both M13 and M14 with less than 5% yield (Supplemental Fig. 4), providing evidence of a radical initiation of the mechanism. Dimers were not detected in the incubations of GDC-0994 with myeloperoxidase.
Mechanism of Dimer Metabolite Formation.
Density functional theory was used to computationally evaluate two mechanisms of dimer formation (Fig. 3, Pathways A and B), using a truncated model of GDC-0994 (X). Computations show that the preferred mechanism (Fig. 4) is Pathway B, involving dual radical formation followed by radical coupling. Pathway A is disfavored due to the high energetics of radical addition to a neutral monomer (Fig. 5).
Docking of either dimer product (M13 or M14) to CYP3A4 (Protein Data Bank code 6BD5) did not yield reasonable substrate poses. Suspecting the protein conformation captured in the crystal structure does not adequately capture the flexibility of the binding pocket, we built custom models for each dimer-protein complex using a protocol similar to that described by Hayes et al. (2014). Binding pocket residues of the enzyme were mutated to alanine (details in Supplemental Information), and the core of each dimer structure (X1X2 in Fig. 3 and the N-linked dimer analog) were positioned such that the dimer bond was in proximity of the iron-heme. We then gradually and iteratively grew each ligand to its full form in the active site by: 1) adding compound substituents to the core in conformations that avoided steric clashes with the backbone and mutated residues of the protein, 2) minimizing the resulting ligand in situ, 3) removing the protein and minimizing the ligand in vacuo to remove ligand strain, and 4) reintroducing the protein and minimizing side chains and backbone with harmonic constraints in the presence of the fixed ligand. After the dimer structures were completely assembled, the mutated sidechains were restored to the native structure in rotamer states that accommodated the dimer-protein complex. Lastly, each completed and restored complex was minimized one additional time to produce the final models (Fig. 6). The final models had minor differences from the original PDBs, with root mean square deviation values of 1.1 and 0.48 Å for the CYP3A4-M13 complex and CYP3A4-M14 complex, respectively. Both M13 and M14 are in low energy conformational states within the complexes (<2 kcal/mol from the conformational global energy minimum), so ligand strain is minimal.
Discussion
The P450-mediated bioactivation of a small molecule to generate a radical reactive intermediate and initiate dimerization was an unexpected biotransformation route. Some preceding examples of similar reactions have been described in the literature, and diverse mechanistic routes for dimer formation have been proposed. Gillam and coworkers reported P450-mediated oxidation of indole to indoxyl, which was subsequently oxidized and dimerized to form indigoids (Gillam et al., 2000; Isin and Guengerich, 2007). Two diastereoisomeric thiophene S-oxide dimers were observed in thiophene metabolism in vitro and in vivo in rats, where the combination of molecules was rationalized to occur via a Diels-Alder reaction (Treiber et al., 1997). Similar dimerization was reported for ticlodipine (Dansette et al., 2005). Dimerization through free radical intermediates has been described for capsaicin and CJ-047710 (Emoto et al., 2007; Reilly et al., 2013). Homodimerization of raloxifene was also reported to occur via a CYP3A4-mediated oxidative reaction (Davis et al., 2011). Several overlaps exist in the observations for CJ-047710, raloxifene, and GDC-0994, suggesting these dimerization reactions may share common mechanistic features. After identifying that the dimers of GDC-0994 were formed via CYP3A4 using standard in vitro techniques, we sought to determine the mechanism of their formation to glean a deeper understanding of how these and other dimers are formed.
Two possible mechanisms for the formation of the observed dimers were evaluated with DFT (Pathways A and B, Fig. 3). Aromatic C-H abstraction is not a part of either proposed mechanism. C-H hydrogen abstraction from pyrazole was ruled unfavorable given the high bond dissociation energies of the C-H bonds, which were previously computed to be between 118.7 and 122.2 kcal/mol (Feng et al., 2003). In contrast, the bond dissociation energy for aniline N-H has been computed to be significantly lower at 91.4 kcal/mol (Song et al., 2003). The proposed mechanisms in Fig. 3 are consistent with existing precedence for P450-mediated aromatic couplings (Isin and Guengerich, 2007; Makino et al., 2007; Woithe et al., 2007; Belin et al., 2009; Grandner et al., 2016).
Both Pathways A and B initiate with N-H abstraction from model substrate X by the iron-oxo complex for P450 (cpd I) to provide a radical species (X1˙) and iron-hydroxo (cpd II). This radical species (X1˙) is delocalized between the nitrogen and other atoms in conjugated pyrazole via resonance, and thus the radical can form a new bond from the initial nitrogen involved in the abstraction or carbon in the pyrazole ring. For subsequent steps in Fig. 3, only C-C coupling is shown for simplicity, but C-N coupling is proposed to occur in the same manner. In Pathway A, radical addition of X1˙ to X2 occurs, and subsequent N-H abstraction by cpd II from X1X2˙ forms the final product. The alternate Pathway B describes forming a second radical species (X2˙) via N-H abstraction, facilitated by cpd II, and radical coupling of X1˙ and X2˙ for the formation of the final dimer product. Thus, Pathways A and B both describe two hydrogen abstraction steps and a coupling reaction but only diverge in the order of the reaction steps.
The proposed mechanisms were computed using DFT and compared to determine the intrinsically favored pathway (Fig. 4). N-H abstraction by cpd I from X1 has a barrier of only 6.7 kcal/mol (2-TS, Fig. 4). This very low barrier indicates abstraction would occur rapidly at room temperature upon appropriate substrate binding. For radical additions of X1˙ to ground-state X2 (Pathway A) to form either C-N or C-C coupled product are >30 kcal/mol and prohibitively high at room temperature (Fig. 5). Accounting for reduced entropic costs due to precomplexation of X1˙ and X2 to CYP3A, this barrier is still rather high and indicates a disfavoring of this radical addition. In the subsequent steps for Pathway B (Fig. 4), ligand exchange to allow the N-H of X2 to coordinate to the iron-hydroxo cpd II, formed by the first abstraction, is favored by >5 kcal/mol (3 to 4, Fig. 4). N-H abstraction from X2 by cpd II is nearly barrierless at 1 kcal/mol (4 to 5-TS, Fig. 4). This barrier is low due to the release of water in the transition state. Radical combination of the delocalized radicals after formation via the two hydrogen abstractions would be rapid, further facilitated by the forced close proximity in the active site. The mechanism is the same for each metabolite, as formation of the two distinct products is facilitated by the delocalized nature of the radicals formed from N-H abstraction (X1˙ and X2˙).
The molecular modeling studies generated two dimer-CYP3A4 structures that indicate that M13 or M14 formation within the active site of the enzyme is sterically feasible with minimal rearrangement of the protein. Although we do not claim that these models are definitive binding modes, they provide supportive evidence for the mechanism we propose for homodimer formation.
In summary, during the in vivo metabolite profiling of GDC-0994 in rats, we discovered two novel dimer metabolites. In vitro studies indicated these metabolites were generated across several animal species and human, with CYP3A4/5 being the most efficient. Biomimetic catalysis produced sufficient quantities of the metabolites to resolve the structures and lead to the identification of a symmetrical and asymmetric homodimer. Computational studies determine that the dimers likely form via tandem N-H abstractions from monomer GDC-0994 units and subsequent radical coupling. Molecular modeling indicates that the CYP3A4 pocket is large and flexible enough to accommodate the dimer in the active site. All together, these studies provide evidence that large dimers can be formed by P450s and the mechanistic analysis can inform other drug discovery programs dealing with metabolism of similar substrates.
Acknowledgments
We thank Michael Drummond from Chemical Computing Group for sharing his knowledge and insights about structural aspects of cytochrome P450 ligand binding.
Authorship Contributions
Participated in research design: Takahashi, Grandner, Bobba, Liu, Beroza, Ma.
Conducted experiments: Takahashi, Bobba, Liu, Ma.
Contributed to computational studies: Grandner (DFT), Beroza (modeling).
Performed data interpretation: Takahashi, Grandner, Bobba, Liu, Beroza, Zhang, Ma.
Contributed to the writing of the manuscript: Takahashi, Grandner, Bobba, Liu, Beroza, Ma.
Footnotes
- Received November 20, 2019.
- Accepted March 4, 2020.
↵1 Current affiliation: Drug Metabolism and Pharmacokinetics, Denali Therapeutics, South San Francisco, California.
↵2 R.H.T., J.M.G., and S.B. contributed equally to this work.
↵This article has supplemental material available at dmd.aspetjournals.org.
Abbreviations
- ABT
- 1-aminobenzotriazole
- DFT
- density functional theory
- ERK
- extracellular signal-regulated kinase
- P450
- cytochrome P450
- Copyright © 2020 by The American Society for Pharmacology and Experimental Therapeutics