Targeting Conformational Flexibility of a Reactive Intermediate to Enhance Selectivity of a GABA Aminotransferase Inactivator

Abstract

Currently, mechanism-based inactivators (MBIs) are the only available therapeutic option to target γ-aminobutyric acid aminotransferase (GABA-AT). However, off-target activity against homologous enzymes is a well-recognized challenge for the clinical use of MBIs. For example, CPP-115, a MBI of GABA-AT that completed a Phase I clinical trial, also inactivates ornithine aminotransferase (OAT). Here, we present a comprehensive investigation of an OAT-specific inactivation mechanism for CPP-115 by integrating biochemical experiments, X-ray crystallography, and computational simulations. Unlike in GABA-AT, where CPP-115 forms a noncovalent tight-binding adduct only, a covalent adduct was additionally observed with human OAT (hOAT). Notably, the crystal structures of CPP-115-treated hOAT at different mechanistic stages indicate that the conformational transition of a key intermediate is a prerequisite for the covalent addition pathway. Based on this finding, to selectively reduce the off-target activity, a proof-of-concept molecule that regulates the intermediate conformational flexibility was designed and synthesized. The resulting inactivator achieved greatly enhanced GABA-AT selectivity over OAT and demonstrated therapeutic efficacy in an inflammatory pain animal model. Our strategy in this study, targeting dynamics of a reactive intermediate based on a precise mechanistic understanding, serves as a general design principle for fine-tuning the selectivity of MBIs, particularly for other aminotransferases.

Graphical Abstract

Introduction

CPP-115 is a drug that successfully completed a Phase I clinical trial and is being taken by a child with infantile spasms.¹ It adopts a distinguishing mode of action as a mechanism-based inactivator (MBI) of γ-aminobutyric acid aminotransferase (GABA-AT),^2–4 which exploits the enzyme machinery for the target-specific inactivation.^5,6 GABA-AT, a pyridoxal 5’-phosphate (PLP)-dependent enzyme, converts the inhibitory neurotransmitter GABA into succinic semialdehyde (Figure 1A) via tautomerization of the external aldimine and subsequent hydrolysis (Figure 1B; for the full mechanism, see Scheme S1).^7,8 GABA-AT has been linked to various neurological conditions, such as epilepsy,^9,10 infantile spasms,^1,11 and neuropathic pain,¹² as its potential therapeutic indications.

Figure 1. Enzymatic catalysis and overall structures of GABA-AT and OAT.

(A) GABA-AT catalytic cycle. (B, C) Schemes of key steps for the GABA degradation mechanism by GABA-AT (B) and the reported inactivation mechanism of GABA-AT by CPP-115 (C). (D) OAT catalytic cycle. (E) Schematic representation of superimposed crystal structures of hOAT (PDB 1OAT) and GABA-AT (PDB 1OHV) and active site residues in stick form. (F) Scheme of key steps for the OAT-specific inactivation mechanism of CPP-115 proposed in this study. (G) GABA-AT/OAT selectivities of existing GABA-AT or OAT inactivators. For structures of compounds noted on the x-axis and their kinetic data, Figure S1 and Table S1. a¹³; b¹⁴; c¹⁵; d¹⁶; e¹⁷; f¹⁸; g¹⁹.

We have reported that CPP-115 follows the first few steps of the GABA catalytic pathway upon binding to GABA-AT, proceeding to the external aldimine (1; Figure 1C) and its tautomer, the external ketimine (2; Figure 1C).²⁰ However, unlike the GABA catalysis that releases succinic semialdehyde at this stage (Figure 1B), the δ-difluoromethylene group forms a reactive Michael acceptor.²⁰ It then no longer follows normal enzyme catalysis, but rather forms a tight-binding adduct without any covalent linkage to the enzyme (3; Figure 1C), via water-mediated attack and subsequent fluoride release (Figure 1C, for the full mechanism, see Scheme S1), effectively inactivating the enzyme.²⁰

According to this mechanism, CPP-115 remains inert until its activation by the enzyme’s substrate-specific catalytic machinery,^5,6 as many other PLP enzyme-targeting MBIs do.^21–23 Although CPP-115 can escape from the GABA-AT active site via its turnover pathway, the δ-difluoromethylenyl group has already been converted into an unreactive carboxylate after the turnover,²⁰ which secures its warhead-related safety. This intrinsic inertness can distinguish MBIs from other reactive molecules, such as affinity-labeling agents, which easily develop side effects due to their non-specific reactivity.^5,6 Moreover, this fragment-sized molecule can efficiently target the highly specific, small-sized binding pocket of GABA-AT, which might be inaccessible to conventional reversible binders. This may explain why MBIs are such an important option targeting GABA-AT.

Inactivating GABA-AT is an effective strategy to raise GABA levels in the central nervous system,²⁴ rather than directly administering GABA itself, which is impermeable to the blood-brain barrier.^25,26 Therefore, we have reported CPP-115’s outstanding efficacy in the in vivo models of cocaine addiction³ and infantile spasms,¹¹ based on its 187-fold higher effective enzyme inactivation compared to vigabatrin (Sabril^®),² the only FDA-approved GABA-AT inactivator.⁸ CPP-115 also demonstrated an improved in vivo safety profile over vigabatrin,^3,4 while vigabatrin has shown severe adverse effects, well-known as vigabatrin-induced retinal toxicity.²⁷ Furthermore, CPP-115 showed its ability to increase GABA levels in the human brain²⁸ and has a favorable safety profile in a Phase I clinical trial (NCT01493596),²⁰ suggesting this molecule could replace vigabatrin as a safer GABA-AT-targeting medication.

However, CPP-115 exhibits off-target activity against a homologous PLP-dependent enzyme, ornithine aminotransferase (OAT).¹³ Despite the safety advantage of an MBI because of the unreactive warhead,^5,6 off-target activity mainly poses a challenge in its clinical development, since it also can inactivate other enzymes that use similar substrates.²⁹ OAT, which is classified in the same enzyme subgroup as GABA-AT,³⁰ degrades ornithine via the same mechanism (Figure 1D; for the full mechanism, see Scheme S1).^8,31 Based on their substrate, mechanism, and active site similarity (Figure 1E),³² we have reported that existing GABA-AT inactivators also inactivate human OAT (hOAT) and have poor selectivities (Figure 1G).^13–19 Therefore, reducing the off-target activity of CPP-115 against OAT is necessary for the development of next-generation GABA-AT inactivators.

Herein, we present comprehensive mechanistic investigations of CPP-115 with hOAT, unveiling its OAT-specific inactivation mechanism (Figure 1F). Guided by key mechanistic features, the structure of CPP-115 was rationally engineered to selectively reduce the OAT-specific pathway, thereby yielding the most selective GABA-AT inactivator reported to date (Figure 1G). Additionally, in a mouse model of inflammatory pain, the newly developed inactivator suppressed mechanical hypersensitivity without significant sedative effects, supporting its potential as a preclinical candidate.

Results

Structural Characterization of hOAT Inactivation by CPP-115 Using X-ray Crystallography

To structurally investigate the final adduct upon inactivation of hOAT by CPP-115, X-ray crystallography was performed following treatment of hOAT with excess CPP-115 (Figure 2A). After the enzymatic activity was completely abolished by 16 h of incubation, the sample was presumed to be the final adduct-bound complex. Using this pre-inactivated hOAT, crystals were grown under previously reported hanging drop conditions.¹⁹ To capture an early-stage intermediate and investigate its conformational flexibility, a short-soaking crystallography was conducted. In this experiment, preformed hOAT crystals were soaked with excess CPP-115 and α-ketoglutarate for 5 minutes (Figure 2A). Structures from two different pre-inactivated crystals and one short-soaked crystal were solved via molecular replacement using the previously reported hOAT structure (PDB 1OAT).³³ Refined models from the pre-inactivated crystals revealed electron densities in the active site with two structurally distinct final adducts: a tight-binding adduct without covalent linkage to the enzyme (Figure 2B and 2E, Figure S2, PDB 9Y3K) and a covalent adduct linked to *Thr322 (asterisk denotes the adjacent subunit) and Lys292 (Figure S2, PDB 9Y3M), at 2.15 and 1.73 Å resolutions, respectively. In the short-soaked crystal, the electron density indicated that the reactive ketimine intermediate (2) was complexed in the active site of hOAT (Figure 2D and 2G, Figure S2, PDB 9Y53, 1.95 Å resolution).

Figure 2. Crystal structures of hOAT with CPP-115.

(A) Schematic representation of procedures for the preparation of each crystal. (B–D) Entire architectures of refined structures: hOAT-3 (B, PDB 9Y3K), hOAT-4 (C, regenerated model from PDB 9Y3M), and hOAT-2 (D, PDB 9Y53). Ligand atoms are shown in CPK form. (E–G) Active sites of hOAT-3 (E), hOAT-4 (F), and hOAT-2 (G). Ligand and active site residues around the inactivator ring are shown in stick form. The phosphate groups were omitted in this figure for clarity. For crystallographic data, see Table S2.

In the crystal structure showing the noncovalent adduct, the observed electron density (Figure S2) supports the formation of a dicarboxylate tight-binding adduct (3, via pathway a1 in Scheme S2), similar to that observed in the crystal of GABA-AT inactivated by CPP-115 (Figure S3).²⁰ The PLP part of the observed ligand maintained key interactions with Gly142, Val143, Phe177, Asp263, Ile265, and Gln266 (Figure S2) as previously reported. The intact α-carboxylate forms a strong hydrogen bond with the side chain of Tyr55 (2.7 Å), which is a key interaction for ornithine recognition by hOAT. The δ-carboxylate, generated by hydrolysis of the δ-difluoromethylenyl group, interacts with Gln266 (2.7 Å, hydrogen bond) and Arg413 (4.5 Å, electrostatic interaction).

In contrast, the electron density of the covalent adduct (Figure S2) was inconsistent with the possible final adduct (4 in Figure 1E, via pathway a2 in Scheme S2). Specifically, the electron density corresponding to the γ-δ bond of the inactivator ring was deficient, while the density still indicated a covalent addition with *Thr322 and Lys292 through the δ-moiety (Figure S2). Given the theoretical bond dissociation energy (BDE) of the γ-δ bond, approximately 30 kcal/mol in the model system (~0.05 Hartree at the level of B3LYP-D3/6–31G**, Figure S4), this electron deficiency may be attributed to the cleavage of the γ-δ bond of 4 by specific radiation damage, resulting in 4* (Figure S4). For comparison, disulfide bonds, known to be susceptible to radiation damage even under cryogenic conditions (BDE = 50–70 kcal/mol^34,35), are reported to gradually diminish upon exposure to only a few MGy of accumulated radiation.³⁶ Given the lower BDE of the γ-δ bond, it is plausible that this bond can be cleaved by the radiation damage. Thus, to enable further structural analysis, the putative intact covalent adduct complex was modeled by regenerating the γ-δ bond and subsequent energy minimization (Figure 2C and 2F).

The early-stage intermediate complex from the short-soaking (2) showed a consistent PLP binding mode and α-carboxylate interaction (Figure S2). To assess the ligand conformational flexibility, the isotropic displacement (B-factor) of each ligand was collected and normalized by the average atomic B-factor of each PLP part (Figure 3A). Across all ligands, the normalized B-factors indicated that the thermal shifts of the inactivator ring atoms are generally higher than those of the PLP atoms, except for N9’ in 2 and 4, which is directly linked to the PLP ring (Figure 3A). Notably, the inactivator atoms in the hOAT-2 complex showed substantially larger displacements than the others (Figure 3A), implying higher conformational flexibility compared to the other final adduct complexes. Among these, the atoms corresponding to the δ-difluoromethylenyl group (C10’, F10”−1, and F10”−2 in Figure 3A) showed the highest normalized B-factors, indicating a dynamic motion of this moiety. Furthermore, the normalized B-factor correlated well with the atomic deviations between the ligands in hOAT-2 and hOAT-4 complexes, but not with those between hOAT-2 and hOAT-3 (Figure 3B), suggesting the potential association between the flexibility of 2 and the mechanism of covalent bond formation toward 4.

Figure 3. Conformational flexibility of the key intermediate of CPP-115 (2) in hOAT.

(A) Atom numbering and normalized atomic isotropic displacement (B-factor) of the PLP and the inactivator (Inact) part of the ligands in each crystal structure. (B) Correlation between the normalized B-factor and ligand atom distances between 2 and 3 (left), and between 2 and 4 (right). Light- and dark-colored circles represent the atoms in the PLP and inactivator part, respectively. (C) Detailed views of the ligand, Lys292, and *Thr322 in hOAT-2 (left) and hOAT-4 (right) complexes. The phosphate groups were omitted for clarity. (D) Scheme of key steps for the suggested mechanism of covalent adduct (4) formation.

Further Mechanistic Characterization of hOAT Inactivation by CPP-115

To further support the mechanisms inferred from the X-ray crystallography data, additional mechanistic experiments were conducted. First, the catalytic turnover number of CPP-115 in hOAT was determined, defined as the number of substrate molecules that an enzyme molecule can process, or equivalently, the number of inactivator molecules required for complete inactivation of an enzyme molecule.⁵ The turnover number (10.44; x-intercept in Figure 4A) and partition ratio (9.44; the ratio of reaction rate of catalytic turnover to inactivation, in other words, the number of turnover events leading to one inactivation event) of CPP-115 in hOAT were determined by plotting the remaining enzymatic activity after treating the enzyme with various equivalents of CPP-115 (Figure 4A).⁵ Several selected hOAT samples with various residual activities, obtained by treating with 2, 3, 5, 10, and 30 equivalents of CPP-115, were dialyzed in a PLP-containing buffer to assess irreversible binding. As a result, no enzymatic activity was recovered after 48 h of dialysis (Figure 4B), indicating that the final products of CPP-115 bind irreversibly to hOAT.

Figure 4. Mechanistic investigation of CPP-115 with hOAT.

(A) Turnover number determination. (B) Dialysis assay result. (C) ¹⁹F NMR spectra of unmodified and CPP-115-treated hOAT, and after its denaturation. Sodium fluoride (NaF) was used as an internal standard. (D) Schematic representation of procedures for preparation of intact protein MS sample. (E) Intact protein mass spectrum. (F) deconvoluted mass spectrum. (G) Scheme of possible tautomerization and hydrolysis from 4 under denaturing conditions.

The equivalents of released fluoride ions from each pathway can be estimated by combining the partition ratio with the total fluoride release during inactivation.¹⁷ Based on the determined partition ratio, each theoretical total fluoride ion release during one inactivation event, including contributions from 9.44 turnover events, was calculated for all possible combinations of fluoride release (Table 1). The total fluoride ion release during one inactivation event of CPP-115 with hOAT was measured using a fluoride ion-selective electrode. Among the theoretical fluoride ion release, the release of 11.44 equivalents of fluoride ions most closely aligns with the experimental release (11.8 ± 1.1 equiv.), suggesting that 2 and 1 equivalents of fluoride ion are released from the inactivation and turnover pathways, respectively (Table 1).

Table 1.

Theoretical and experimental release of fluoride ion from CPP-115 in hOAT

	Fluoride ion release (equiv.)
During inactivation (n)	0	0	0	1	1	1	2	2	2
During turnover (m)	0	1	2	0	1	2	0	1	2
Theoretical fluoride ion release (n + m × partition ratio)	0.00	9.44	18.88	1.00	10.44	19.88	2.00	11.44	20.88
Experimental fluoride ion release	11.8 ± 1.1

To further confirm the release of fluorine atoms during the inactivation pathway, ¹⁹F nuclear magnetic resonance (NMR) spectra of completely inactivated hOAT were acquired using the same protocol as in our recent report with (R)-3-amino-5,5-difluorocyclohex-1-ene-1-carboxylate (Figure S5).¹⁹ In both intact and denatured samples of completely inactivated hOAT treated with an excess of CPP-115, no additional fluorine peaks were detected compared to the spectrum of the native, unmodified enzyme (Figure 4C). Thus, the ¹⁹F NMR spectra supported the absence of fluorine atoms in the final adduct of the CPP-115 inactivation pathway, corroborating the release of 2 equivalents of fluoride ions per inactivation event, as predicted from the fluoride release assay.

Because the crystal structure with the covalent adduct is believed to be damaged by radiation, intact protein mass spectrometry (MS) was performed under denaturing conditions with inactivated hOAT treated with CPP-115, and the covalent adducts were detected. Intact protein MS is a powerful technique for detecting covalently bound adducts resulting from mechanism-based inactivation, as it enables quantification of mass shifts upon covalent modification of the protein, while discarding noncovalent binders (Figure 4D).^17,37 The deconvoluted mass spectrum of CPP-115-treated hOAT revealed a peak matching the mass of hOAT monomer without modification (46140.6 Da; Figure 4E and 4F), consistent with the vehicle-treated, thus unmodified hOAT (46138.4 Da; Figure S6). In addition to the unmodified monomer peak, several peaks with mass shifts of +136, 157, and 294 Da were observed (Figure 4F). By comparison, in our previous report,²⁰ CPP-115-treated GABA-AT did not exhibit any additional peaks under the same conditions, indicating the absence of covalent adduct formation with GABA-AT.

Among the observed mass shifts, the +136 and +157 Da species (Figure 4F) matched the theoretical mass shifts (136.01 and 154.02 Da; 4’ and 4b’ in Figure 4G) for the proposed double-addition product 4 under denaturing conditions, within the possible error range. In addition, the +294 Da mass shift, approximately matching the sum of 136 and 157 Da, suggests the possibility of a double modification on Thr322 and Lys292 within a single monomer chain (Figure 4D and 4F). Also, repeated experiments consistently detected mass peaks for unmodified and modified hOAT, yielding an average mass shift of +156.8 ± 3.1 Da, in agreement with 4b’ (Figure S6).

Evaluating Conformational Flexibility of a Key Intermediate by Metadynamics Simulation

To evaluate the feasibility of the conformational flexibility in hOAT and GABA-AT, particularly the torsional rotation required for the covalent addition at the δ-difluoromethylenyl group of 2, metadynamics simulations were performed using the docking-derived hOAT- and GABA-AT-2 complexes (Figure S7). In mechanistic investigations of MBI, computational simulations have proven effective for characterizing structural features that cannot be captured experimentally.^14,38–40 For metadynamics simulation, the distance from the threonine side chain oxygen atom (O_*Thr) to the terminal carbon of the δ-difluoromethylenyl group (C_δ’) (CV1 axis in Figure 5) and the dihedral angle along the C4-C4’ bond (Dihedral_C4-C4’) of 2 (CV2 axis in Figure 5) were defined as collective variables (CVs) to assess their impact on the free energy of the complex (Figure 5A). Metadynamics simulation enables efficient sampling of conformational transitions by applying bias to avoid previously visited CVs.⁴¹ Thus, the metadynamics approach is particularly well-suited for this study, where experimental results already suggest and illustrate the actual conformational change of the intermediate, which is required for defining CVs and determining their desired ranges during simulation. The accumulated bias potential required to overcome energetically unfavorable states was further used to estimate the free energy as a function of the CVs, resulting in the construction of relative free energy surfaces (FES) for both enzyme complexes (Figure 5B and 5C).

Figure 5. Metadynamics simulations of reactive intermediate 2 in hOAT and GABA-AT.

(A) Defining CVs used in metadynamics simulations. Dihedral_C4-C4’ is defined as positive for counterclockwise rotation when viewed from the inactivator ring side. In this panel, an example conformation of 2’ was generated by manually adjusting Dihedral_C4-C4’. (B) FES of the hOAT-2 complex. (C) FES of GABA-AT-2 complex. 2_OAT and 2_GABA-AT denote the initial conformations in each enzyme. 2’_OAT and 2’_GABA-AT represent the near-attack conformation-like state. 2^‡_GABA-AT indicates the high-energy conformational intermediate along the transition pathway. (D) Free energy profile along the conformational transition of 2_GABA-AT into 2’_GABA-AT. (E–G) Representative conformations of 2_GABA-AT (E), 2^‡_GABA-AT (F), and 2’_GABA-AT (G), illustrating the steric clash with Ile72 along the torsional rotation. The phosphate groups in the intermediate were omitted for clarity.

The resulting relative FES for 2 in hOAT indicated that the initial stabilized state, which presumably corresponds to the initial conformation of 2, shows a Dihedral_C4-C4’ near 0° and an O_*Thr-C_δ’ distance around 4 Å (2_OAT in Figure 5B and S8). As the inactivator ring undergoes torsional rotation with increasing Dihedral_C4-C4’, the O_*Thr-C_δ’ distance decreases accordingly (Figure 5B and S8). As predicted by the manual manipulation (Figure 5A), conformations with ~3.0 Å O_*Thr-C_δ’ distance are achieved at approximately +90° Dihedral_C4-C4’ (3’_OAT in Figure 4B and S8). The associated free energy fluctuation along this conformational transition is minimal for all replica runs (~1 kcal/mol), hence, exhibiting a shallow free energy landscape between the two conformational states (Figure 5B and S8).

In contrast to the relatively shallow FES enabling torsional rotation of 2 in hOAT, the FES in GABA-AT exhibited a markedly different profile characterized by a steep energy barrier along the transition between the two states, 2_GABA-AT and 2’_GABA-AT (Figure 5C). Specifically, a transition between the initial stable conformation 2_GABA-AT (Dihedral_C4-C4’ = ~25°, O_*Thr-C_δ’ distance = ~5 Å) and the near attack conformation-like conformer 2’_GABA-AT (Dihedral_C4-C4’ = ~85°, O_*Thr-C_δ’ distance = ~3 Å) requires passing through a high-energy conformational intermediate 2^‡_GABA-AT (Dihedral_C4-C4’ = ~70°, O_*Thr-C_δ’ distance = ~4 Å), showing 3–9 kcal/mol higher free energy than 2_GABA-AT throughout the replica runs (Figure 5D and S8).

To investigate the detailed structural basis of the energy barrier observed in GABA-AT, representative conformations of 2_GABA-AT, 2^‡_GABA-AT, and 2’_GABA-AT were extracted from the metadynamics trajectories and analyzed (Figure 5E–5G). In the initial stable conformer 2_GABA-AT, the α-carboxylate of the inactivator part interacts with Arg192, stabilizing the complex upon its generation (Figure 5E). This interaction is maintained in the energetically unfavorable conformer 2^‡_GABA-AT despite the torsional rotation (Figure 5F). However, the δ-difluoromethylenyl group clashes with the side chain of Ile72 (Figure 5F), which is non-conserved in OAT. This steric hindrance displaces the Ile72 side chain and likely contributes to the energy barrier during the torsional rotation observed in the FES, as estimated by the metadynamics simulation. Only when this unfavorable torsional rotation occurs, can the carboxylate reorient to interact with Arg445, thus allowing the O_*Thr-C_δ’ distance to be close enough for the *Thr353 addition (Figure 5G).

The same simulation using GABA-AT with an in silico mutation of Ile72Gly further supports the interpretation that Ile72 contributes to the energetic instability of 2^‡_GABA-AT. Although steric hindrance partially remains due to the shifting of the *Ile105 side chain, which appears to contact the Ile72 side chain in wild-type GABA-AT, the energy barrier was significantly reduced by an I72G substitution (Figure S9).

Design, synthesis, and in vitro evaluation of a new inactivator to enhance the GABA-AT/OAT inactivation ratio

To attenuate the OAT-specific inactivation pathway of CPP-115, we rationally designed a novel inactivator, which can modulate the torsional rotation of its reactive intermediate, a prerequisite for the covalent addition pathway. A new inactivator (5, (S)-MeCPP-115; Figure 6B) was designed to sterically repulse with the side chain of Tyr55 in the binding site (Figure 6A) during the intermediate rotation (Figure 6B and 6C); thus, this can selectively interrupt the covalent addition mechanism with OAT.

Figure 6. Design strategy and enzyme activity of 5.

(A) Binding site view of hOAT-2 crystal structure showing protein surface around the inactivator ring. The ligand, Tyr55, and Tyr85 are shown in stick form. (B) Structures of CPP-115, 5, and the corresponding reactive intermediates, and their VdW surfaces. (C) Schematic representations showing free rotation of 2 and ‘torsional brake’ model of the reactive intermediate generated by 5 in hOAT. (D) Concentration-dependent enzyme inactivation assay results with hGABA-AT. (E) Concentration-dependent enzyme inactivation assay results with hOAT. (F) Time-dependent enzyme inactivation assay results of 5 with hGABA-AT. (G) Time-dependent enzyme inactivation assay results of 5 with hOAT.

The newly designed compound was synthesized from CPP-115 to experimentally validate whether α-substitution could attenuate its potency toward OAT and thereby enhance selectivity for GABA-AT. Following our established protocol for α-phenylselenylation using appropriate protecting groups,¹⁴ 5 was synthesized in three steps (Figure S10). However, due to the non-enantioselective nature of the α-substitution step via an enolate intermediate, the undesired opposite diastereomer was also obtained. The two diastereomers were successfully separated by flash column chromatography, and the absolute stereochemistry of each diastereomer was determined by NOESY analysis (Figure S10). The desired diastereomers were identified by the observed spatial proximity between the γ-proton and the proton of the newly introduced methyl group, whereas the undesired diastereomer lacked such NOE correlation (Figure S10).

To evaluate the selectivity of the newly synthesized inactivator (5), its activities against hGABA-AT and hOAT were compared to those of the parent molecule, CPP-115. 5 inhibited both enzymes in a concentration-dependent manner (Figure 6D and 6E); however, its potency against hOAT (IC₅₀ = 1.2 mM) was more than 10-fold lower than that of CPP-115 (IC₅₀ = 0.083), while exhibiting comparable potency against hGABA-AT (IC₅₀ = 11 and 5.2 μM for CPP-115 and 5, respectively). To further assess time-dependent inactivation, we determined the observed inactivation rate constants (k_obs) from residual enzymatic activity over time at various concentrations (Figure 6F and 6G). Curve fitting of the k_obs values enabled estimation of the inactivation efficiency, expressed as the ratio of maximal inactivation rate (k_inact) to the inactivator concentration required for half-maximal rate (K_I). 5 showed a k_inact/K_I value of 50 min⁻¹mM⁻¹ for hGABA-AT (Figure 6F) and only 0.24 min⁻¹mM⁻¹ for hOAT (Figure 6G), indicating 210-fold selectivity. In comparison, CPP-115 exhibited k_inact/K_I values of 27 min⁻¹mM⁻¹ for hGABA-AT (Figure S11) and 0.83 min⁻¹mM⁻¹ for hOAT.¹² Furthermore, 5 showed only weak inhibitory activities against other aminotransferases, L-Ala aminotransferase and L-Asp aminotransferase (Figure S11), even at the highest concentration of 10 mM.

5 alleviates in vivo mechanical hypersensitivity induced by CFA in a dose-dependent manner

The link between inflammatory pain and reduced GABAergic inhibition in the central nervous system has been reported.⁴² Since the complete Freund’s adjuvant (CFA) model is widely used to mimic peripheral sensitization and central hyperexcitability in inflammatory pain,⁴³ 5 ((S)-MeCPP-115 in Figure 7) was treated in the CFA mouse model (for experimental design, see Figure 7A). To assess mechanical hypersensitivity, the paw withdrawal threshold was measured in response to mechanical stimuli using an electronic von Frey anesthesiometer. As expected, CFA produced robust mechanical hypersensitivity in the ipsilateral paw (the one that was CFA injected), which was manifest as a decrease in the threshold that elicited paw withdrawal. Notably, CFA-induced mechanical hypersensitivity was attenuated by intraperitoneal (i.p.) treatment of 5 in a dose- and time-dependent manner (Figure 7B). 5, at doses of 10 and 30 mg/kg, increased mechanical paw withdrawal thresholds overall compared to saline controls, whereas lower doses (1 or 3 mg/kg) did not alter CFA-induced mechanical hypersensitivity (Figure 7B). 5 (10 and 30 mg/kg) produced peak anti-allodynic efficacy relative to saline from 0.5 to 1 h post-injection. None of the treatments altered paw withdrawal thresholds in the contralateral paw (the non-inflamed one) (Figure 7C).

Figure 7. (S)-MeCPP-115 (5) attenuates CFA-induced inflammatory nociception in a dose-dependent manner in male mice.

(A) Schematic representation of the experimental design and timeline. (B) 5 (10 and 30 mg/kg, i.p.) dose-dependently reduced CFA-induced mechanical hypersensitivity in male mice compared with saline-injected controls. (C) Neither treatment nor dose altered withdrawal thresholds in the non-inflamed (contralateral) paw. Data are expressed as mean ± SEM (n = 6 per group). Two-way ANOVA followed by Dunnett’s post hoc test; brackets indicate main drug effect. *5 10 mg/kg, i.p. vs. saline; ^#5 30 mg/kg, i.p. vs. saline.****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05. For statistical analyses, see Table S3.

Additionally, neither 5 (10 and 30 mg/kg, i.p.) nor control GABA-AT inactivator OV329 (see Figure S1 for structure; 30 mg/kg, i.p.) altered locomotor activity in otherwise naive mice. Neither compound altered the total distance traveled, horizontal activity, rest time, or movement time in the activity meter (Figure S13). In addition, neither 5 nor OV329 altered the time spent in the center of the open field or ambulatory velocity (Figure S13).

Discussion

At the beginning of this investigation, several possible OAT inactivation pathways for CPP-115 were hypothesized (Scheme S2), based on experimentally supported inactivation mechanisms of GABA-AT^8,20,44–48 and OAT.^15–19,49 After forming the external aldimine (1), CPP-115 undergoes γ-deprotonation to yield a transient quinonoid intermediate, which can diverge into two distinct pathways (a and b in Scheme S2) depending on the re-protonation site, like vigabatrin (Scheme S2).⁴⁴ In pathway a, C4’ re-protonation generates a ketimine intermediate (2), forming a Michael acceptor that enables nucleophilic addition at its δ-difluoromethylene moiety. In pathway a1, water-mediated addition and subsequent second water attack lead to hydrolysis of the δ-difluoromethylenyl group to form the dicarboxylate adduct (3), consistent with CPP-115’s mechanism in GABA-AT.²⁰ Alternatively, if *Thr322 acts as a nucleophile, 2 proceeds via pathway a2, involving sequential addition of *Thr322 and Lys292, to yield covalent adduct 4. Similar double-addition pathways were previously reported for the mechanism of CPP-115’s analogues OV329¹⁸ and (S)-3-amino-4-(difluoromethylenyl)cyclohex-1-ene-1-carboxylic acid¹⁷ in hOAT, including activation of *Thr322 by proximal catalytic lysine (Lys292) for nucleophilic attack (Scheme S3). Pathway b (Scheme S2) leads to an enamine mechanism via tautomerization⁸ by re-protonation of the δ-difluoromethylenyl group in the quinonoid. Subsequent degradation of the tautomer affords an enamine species, which can inactivate the enzyme by forming a ternary adduct. Among the suggested possible pathways of CPP-115 in OAT, pre-inactivated crystal structures correspond to the products of pathways a1 and a2, and further mechanistic experiments also support those two pathways. Dialysis assay results matched the irreversible binding of final adducts, a tight-binding adduct (3) and a covalent adduct (4). The fluoride ion release assay and ¹⁹F NMR spectra supported the release of two equivalents of fluoride ion during inactivation, as described in both pathways a1 and a2. Most importantly, intact protein MS demonstrated the formation of covalent adduct 4.

The most remarkable difference from the inactivation mechanism in GABA-AT is that CPP-115 forms two distinct final products in hOAT: a noncovalent tight-binding adduct (3) and a covalent adduct (4), whereas only the noncovalent adduct 3 was confirmed in GABA-AT.²⁰ With the other δ-difluoromethylene-containing MBIs of OAT, it has been suggested that conformational flexibility of the Michael acceptor intermediate might be necessary for the *Thr322 addition step (Scheme S3).^17,18 Upon binding, the δ-difluoromethylenyl group is believed to be positioned away from *Thr322. Thus, to enable the *Thr322 addition, the intermediate must undergo sufficient torsional rotation and adopt a near attack conformation, placing the nucleophile (*Thr322) and the acceptor (the δ-difluoromethylenyl group) within a range for covalent bond formation.

A comparison of the hOAT-noncovalent adduct (3) and the hOAT-intermediate (2) complexes reveals that the inactivator ring conformation of 2 closely resembles that of 3 (compare Figure 2E and 2G), suggesting that pathway a1 can proceed without a substantial conformational rearrangement. In contrast, the large distance between the δ-difluoromethylenyl group of 2 and *Thr322 (6.4 Å, Figure 3C) indicates that 2 must undergo a conformational transition to enable the covalent addition via pathway a2 (Figure 3D). Furthermore, given the relatively high thermal shift of the inactivator part in hOAT-2 complex (Figure 3A), the atoms in the inactivator part of 2 are considered more disordered in the crystal, suggesting 2 is more flexible than 3 and 4. This flexibility correlates well with the ligand atom deviation between hOAT-2 and hOAT-4 complexes (Figure 3B), implying the putative motion of 2 allows nucleophilic attack of *Thr322. The manually adjusted conformation of 3 (Figure 5A) suggests that the near attack conformation enabling *Thr322 nucleophilic attack is likely achieved when Dihedral_C4-C4’ adopts near +90°. Since this conformational transition of 2 is a prerequisite for pathway a2 but not for a1, analyzing the dynamics of the reactive intermediate (2) could provide a structural basis for the pathway a1/a2-differentiation. Accordingly, the conformational flexibility of 2 may underlie the observed mechanistic difference of CPP-115 in the two enzymes, specifically in determining whether pathway a2 is allowed (OAT) or not (GABA-AT).

Subsequent metadynamics simulations with reactive intermediate 2 complexed in GABA-AT and hOAT showed distinct free energy profiles during the conformational transition of 2 in the two homologous enzymes. Based on the simulation with hOAT, the torsional rotation of 2 required for *Thr322-mediated nucleophilic attack was energetically accessible (Figure 5B), indicating dynamic equilibrium between 2_OAT and 2’_OAT and the feasibility of both pathways a1 and a2 (Scheme S2) in hOAT. In contrast, simulation of 2 in GABA-AT showed a steep free energy barrier (3–9 kcal/mol) associated with the high-energy conformer 2^‡_GABA-AT. Based on the Boltzmann equation, the population ratio of 2^‡_GABA-AT/2_GABA-AT is estimated to be 10⁻³–10⁻⁶, implying that the transition toward 2’_GABA-AT required for pathway a2 is unfavorable in GABA-AT. A similar simulation, showing a reduced energy barrier in the I72G mutant model of GABA-AT, further supports the role of GABA-AT-specific residue Ile72 in restricting conformational accessibility for threonine addition and, consequently, differentiating the mechanisms of CPP-115 with GABA-AT and OAT. Thus, reducing this torsional rotation by structural modification of CPP-115 selectively downregulates its reactivity toward OAT, without altering its potency against GABA-AT.

To target the dynamic nature of the reactive intermediate (2), the surrounding binding site residues were analyzed in the complex, including the side chains of Tyr55 and Tyr85, and the main chains of *Gly320, *Ser321, and *Thr322. As a result, Tyr55 and Tyr85 were identified as possible targets that can evoke steric hindrance with additional substitution (Figure 6A). However, conformational flexibility and displacement of Tyr85 have been reported in the previous crystal structures (Figure S14), which may facilitate recognition of smaller substrates such as α-ketoglutarate during the second half of the OAT catalytic cycle (Figure 1B). This flexibility of Tyr85 side chain may also allow torsional rotation of the intermediate (‘Free rotation’ in Figure 6C). In contrast, Tyr55, which forms key interactions with the carboxylate and α-amino group of the natural substrate ornithine, exhibits a highly conserved conformation across all reported hOAT crystal structures (Figure S14). Accordingly, we designed an α-substituted derivative of CPP-115 (Figure 6B) to induce steric repulsion with the conformationally rigid Tyr55 during the torsional rotation of the key intermediate (‘Torsional brake’ model in Figure 6C), thereby selectively interrupting the OAT-specific pathway. Although our previous quantum mechanics-based calculations indicated the γ-deprotonation as the theoretical rate-limiting step,⁵⁰ we hypothesize that the ‘targeting conformational flexibility’ strategy might be significant enough to make the *Thr322 addition step the rate-limiting step.

As intended, the newly synthesized compound (5) demonstrated enhanced selectivity for GABA-AT over OAT (selectivity hGABA-AT/hOAT = 210, Figure 6D–6G), compared to CPP-115 (selectivity = 33; Figure S11). Since the slightly increased GABA-AT activity (k_inact/K_I = 27 and 50 min⁻¹mM⁻¹ for CPP-115 and 5, respectively) also partially contributed to its enhanced selectivity, the impact of the methyl group on the intermediate conformation in GABA-AT will be investigated by determining the complex structure in the future. In addition, 5 also showed weak inhibitory activities against L-Ala aminotransferase and L-Asp aminotransferase (Figure S12). These results establish 5 as the most selective GABA-AT inactivator reported to date.

Because MBIs typically derive their unique advantages by resembling the substrate, they can access small catalytic pockets, which are often inaccessible to conventional reversible binders. However, the substrate resemblance that makes MBIs powerful, ironically makes it more challenging to differentiate their activities from homologous enzymes. This selectivity concerns may have contributed to the limited consideration of MBIs in the context of contemporary drug discovery campaigns,²⁹ even with their safety obtained from the unreactive warhead.^5,8 To address this, in this study, we employed an integrated strategy combining comprehensive experimental investigation with computational analysis to precisely state the mechanistic difference between two homologous enzymes with a MBI clinical drug. Based on this insight, a proof-of-concept molecule was rationally designed, which showed superior selectivity relative to any other reported GABA-AT inactivator (Figure 1G) and demonstrated in vivo therapeutic potential as a preclinical candidate (Figure 7). Therefore, this approach to target the dynamics of a reactive intermediate may provide a generalizable strategy for developing selective MBIs, particularly for PLP-dependent enzymes. Additionally, for future directions, we have synthesized a series of α-substituted analogues, and an investigation of their kinetic profiles, mechanistic pathways, and crystal structures with the enzymes is currently in progress.⁵¹

Supplementary Material

Supplementary figures, schemes, and tables, experimental materials and methods, computational details, supplementary ¹H and ¹³C spectra, and crystallographic data (PDF)