Development and Validation of [3H]OF-NB1 for Preclinical Assessment of GluN1/2B Candidate Drugs

GluN2B-enriched N-methyl-D-aspartate receptors (NMDARs) are implicated in several neurodegenerative and psychiatric diseases, such as Alzheimer’s disease. No clinically valid GluN1/2B therapeutic exists due to a lack of selective GluN2B imaging tools, and the state-of-the-art [3H]ifenprodil shows poor selectivity in drug screening. To this end, we developed a tritium-labeled form of OF-NB1, a recently reported selective GluN1/2B positron emission tomography imaging (PET) agent, with a molar activity of 1.79 GBq/µmol. The performance of [3H]OF-NB1 and [3H]ifenprodil was compared through head-to-head competitive binding experiments, using the GluN1/2B ligand CP-101,606 and the sigma-1 receptor (σ1R) ligand SA-4503. Contrary to [3H]ifenprodil, the usage of [3H]OF-NB1 differentiated between GluN1/2B and σ1R binding components. These results were corroborated by observations from PET imaging experiments in Wistar rats using the σ1R radioligand [18F]fluspidine. To unravel the binding modes of OF-NB1 and ifenprodil in GluN1/2B and σ1Rs, we performed a retrospective in silico study using a molecular operating environment. OF-NB1 maintained similar interactions to GluN1/2B as ifenprodil, but only ifenprodil successfully fitted in the σ1R pocket, thereby explaining the high GluN1/2B selectivity of OF-NB1 compared to ifenprodil. We successfully showed in a proof-of-concept study the superiority of [3H]OF-NB1 over the gold standard [3H]ifenprodil in the screening of potential GluN1/2B drug candidates.


Results and Discussion
At the outset of the studies, a five-step synthetic route was devised to prepare the dibromine-bearing precursor for the radiosynthesis of [ 3 H]OF-NB1 (7) (Scheme 1). Sonogashira coupling of substituted iodobenzene 1 and alkyne 2 yielded benzalkyne alcohol 3 in 78% yield. Alcohol 3 was reduced with Raney nickel, subsequently treated with MsCl, and the resulting intermediate was used for the N-alkylation of commercially available bezazepine derivative 6 to give the N-alkylated product 7, with a yield of 44% over three steps. The final demethylation of 7 using BBr3 gave the desired precursor 8 with a yield of 26%. Tritium-labeling of precursor 8 yielded [ 3

Results and Discussion
At the outset of the studies, a five-step synthetic route was devised to prepare the dibromine-bearing precursor for the radiosynthesis of [ 3 H]OF-NB1 (7) (Scheme 1). Sonogashira coupling of substituted iodobenzene 1 and alkyne 2 yielded benzalkyne alcohol 3 in 78% yield. Alcohol 3 was reduced with Raney nickel, subsequently treated with MsCl, and the resulting intermediate was used for the N-alkylation of commercially available bezazepine derivative 6 to give the N-alkylated product 7, with a yield of 44% over three steps. The final demethylation of 7 using BBr 3 gave the desired precursor 8 with a yield of 26%. Tritium-labeling of precursor 8 yielded [ 3 H]OF-NB1 with 99% radiochemical purity and a molar activity of 1.79 GBq/µmol (48.3 Ci/mmol).

Results and Discussion
At the outset of the studies, a five-step synthetic route was devised to prepare t dibromine-bearing precursor for the radiosynthesis of [ 3 H]OF-NB1 (7) (Scheme 1). S nogashira coupling of substituted iodobenzene 1 and alkyne 2 yielded benzalkyne alcoh 3 in 78% yield. Alcohol 3 was reduced with Raney nickel, subsequently treated with Ms and the resulting intermediate was used for the N-alkylation of commercially availa bezazepine derivative 6 to give the N-alkylated product 7, with a yield of 44% over th steps. The final demethylation of 7 using BBr3 gave the desired precursor 8 with a yield 26%.  Figure 2). Furthermore, we assessed the selectivity of the two drugs over σ1Rs using the σ1R radioligand, (+)-[ 3 H]pentazocine. The results are summarized in Table 1.  Figure 2). Furthermore, we assessed the selectivity of the two drugs over σ1Rs using the σ1R radioligand, (+)-[ 3 H]pentazocine. The results are summarized in Table 1. The results revealed a two-digit nanomolar GluN1/2B affinity for CP-101,606 when using either of the two radioligands, [ 3 H]OF-NB1 or [ 3 H]ifenprodil. Both values were, however, considerably higher than the reported literature value of 16 nM [31]. Three notable differences comparing the assay at hand from the one described in literature, are as follows: (1) the absence of other agents that block non-NMDA receptors; (2) the usage of whole rat brain homogenates; and (3) the higher temperature of 25 °C.  The results revealed a two-digit nanomolar GluN1/2B affinity for CP-101,606 when using either of the two radioligands, [ 3 H]OF-NB1 or [ 3 H]ifenprodil. Both values were, however, considerably higher than the reported literature value of 16 nM [31]. Three notable differences comparing the assay at hand from the one described in literature, are as follows: (1) the absence of other agents that block non-NMDA receptors; (2) the usage of whole rat brain homogenates; and (3) the higher temperature of 25 • C. The key reason for such a disparity is that SA-4503 is an extremely potent σ1R-selective ligand; however, ifenprodil binds to both GluN1/2B and σ1Rs indiscriminately. The K i (σ1R) value of 3.8 nM for SA-4503 matched the published value of 4.6 nM [32]. On the other hand, the K i (σ1R) value of 94 nM for CP-101,606 was considerably close to the published value of 60 nM using the σR ligand, [ 3 H](3-(3-hydroxyphenyl)N-(1-propyl)-piperidine (3-PPP) [33]. This shows that CP-101,606 potentially exhibits a σ1R binding component, thereby providing a valid explanation for the higher K i (GluN1/2B) value of CP-101,606 when using [ 3 H]OF-NB1, given its low σ1R affinity as opposed to ifenprodil.
To verify our findings of CP-101,606 possessing significant σ1R binding, we conducted in vivo PET imaging studies with [ 18 F]fluspidine, a σ1R radioligand which is known for its lack of uptake in the brain of σ1R knock-out mice [22,30]. These findings are depicted in  The key reason for such a disparity is that SA-4503 is an extremely potent σ1R-selective ligand; however, ifenprodil binds to both GluN1/2B and σ1Rs indiscriminately. The Ki (σ1R) value of 3.8 nM for SA-4503 matched the published value of 4.6 nM. [32] On the other hand, the Ki (σ1R) value of 94 nM for CP-101,606 was considerably close to the published value of 60 nM using the σR ligand, . [33] This shows that CP-101,606 potentially exhibits a σ1R binding component, thereby providing a valid explanation for the higher Ki (GluN1/2B) value of CP-101,606 when using [ 3 H]OF-NB1, given its low σ1R affinity as opposed to ifenprodil.
To verify our findings of CP-101,606 possessing significant σ1R binding, we conducted in vivo PET imaging studies with [ 18 F]fluspidine, a σ1R radioligand which is known for its lack of uptake in the brain of σ1R knock-out mice [22,30]. These findings are depicted in Figures 3 and 4.  that occupies 50% of the σ1Rs. [23].
In vivo PET results pertaining to the off-target characteristics of CP-101,606 toward σ1Rs corroborated the results obtained from the in vitro binding study. A series of doses were investigated, ranging from 0.05-15 mg/kg. Importantly, the GluN1/2B ligand, CP-101,606, displayed a dose-dependent blockade similar to that of the σ1R ligand SA-4503 [22]. This allowed us to calculate the D50 (σ1R), which indicates the administered dose of   . [33] This shows that CP-101,606 potentially exhibits a σ1R binding component, thereby providing a valid explanation for the higher Ki (GluN1/2B) value of CP-101,606 when using [ 3 H]OF-NB1, given its low σ1R affinity as opposed to ifenprodil.
To verify our findings of CP-101,606 possessing significant σ1R binding, we conducted in vivo PET imaging studies with [ 18 F]fluspidine, a σ1R radioligand which is known for its lack of uptake in the brain of σ1R knock-out mice [22,30]. These findings are depicted in Figures 3 and 4.  that occupies 50% of the σ1Rs. [23].
In vivo PET results pertaining to the off-target characteristics of CP-101,606 toward σ1Rs corroborated the results obtained from the in vitro binding study. A series of doses were investigated, ranging from 0.05-15 mg/kg. Importantly, the GluN1/2B ligand, CP-101,606, displayed a dose-dependent blockade similar to that of the σ1R ligand SA-4503 [22]. This allowed us to calculate the D50 (σ1R), which indicates the administered dose of In vivo PET results pertaining to the off-target characteristics of CP-101,606 toward σ1Rs corroborated the results obtained from the in vitro binding study. A series of doses were investigated, ranging from 0.05-15 mg/kg. Importantly, the GluN1/2B ligand, CP-101,606, displayed a dose-dependent blockade similar to that of the σ1R ligand SA-4503 [22]. This allowed us to calculate the D 50 (σ1R), which indicates the administered dose of CP-101,606 that occupies 50% of the σ1Rs (Figure 3). The D 50 (σ1R) of CP-101,606 was calculated to be 4.9 µmol/kg, which is higher than the D 50 (GluN1/2B) of 8.1 µmol/kg [26]. This strong σ1R binding behavior of CP-101,606 offsets its unwarranted reputation of being a selective GluN1/2B. There have been efforts to develop ligands with dual GluN1/2B and σ1R activity [34]. Nonetheless, for the purpose of developing selective GluN1/2B ligands, one has to consider fundamentally reformulating the development strategy.
With the aim to better understand the interactions of OF-NB1 on a molecular level, the binding mode was studied alongside the two target receptors, GluN1/2B and σ1Rs, in a retrospective in silico study. With regards to GluN1/2B, the binding of OF-NB1 was compared to ifenprodil, which was previously co-crystallized with GluN1/2B (PDB-ID: 3QEL). As shown in Figure 5A, the binding of ifenprodil can be credited to the existence of three H-bonds with Gln110 and Glu236. Additionally, the two phenyl rings of ifenprodil guided multiple hydrophobic interactions with amino acids from the two GluN subunits, such as Tyr109 (GluN1A) and Ile111 (GluN2B). While OF-NB1 was able to maintain common interactions as ifenprodil as shown in Figure 5B, it surprisingly managed to form two additional interactions. Specifically, it should be noted that OF-NB1 achieved a new H-bond with Glu106 via its secondary hydroxyl functionality. Furthermore, the distant phenyl ring of OF-NB1 makes van-der-Waals interactions with amino acids Phe176 and Leu135. These additional interactions are energetically favorable, thus contributing positively to the binding affinity of OF-NB1 toward GluN1/2B. CP-101,606 that occupies 50% of the σ1Rs (Figure 3). The D50 (σ1R) of CP-101,606 was calculated to be 4.9 µmol/kg, which is higher than the D50 (GluN1/2B) of 8.1 µmol/kg [26]. This strong σ1R binding behavior of CP-101,606 offsets its unwarranted reputation of being a selective GluN1/2B. There have been efforts to develop ligands with dual GluN1/2B and σ1R activity [34]. Nonetheless, for the purpose of developing selective GluN1/2B ligands, one has to consider fundamentally reformulating the development strategy.
With the aim to better understand the interactions of OF-NB1 on a molecular level, the binding mode was studied alongside the two target receptors, GluN1/2B and σ1Rs, in a retrospective in silico study. With regards to GluN1/2B, the binding of OF-NB1 was compared to ifenprodil, which was previously co-crystallized with GluN1/2B (PDB-ID: 3QEL). As shown in Figure 5A, the binding of ifenprodil can be credited to the existence of three H-bonds with Gln110 and Glu236. Additionally, the two phenyl rings of ifenprodil guided multiple hydrophobic interactions with amino acids from the two GluN subunits, such as Tyr109 (GluN1A) and Ile111 (GluN2B). While OF-NB1 was able to maintain common interactions as ifenprodil as shown in Figure 5B, it surprisingly managed to form two additional interactions. Specifically, it should be noted that OF-NB1 achieved a new H-bond with Glu106 via its secondary hydroxyl functionality. Furthermore, the distant phenyl ring of OF-NB1 makes van-der-Waals interactions with amino acids Phe176 and Leu135. These additional interactions are energetically favorable, thus contributing positively to the binding affinity of OF-NB1 toward GluN1/2B.  The co-crystal structure of σ1R with pentazocine was used (PDB-ID: 6DK1) for the docking of ifenprodil and OF-NB1 against σ1R. Both Schmidt et al. and the Glennon model for σ1R ligands highlighted the prominence of the presence of a positively charged nitrogen in all σ1R ligands, where it forms a salt-bridge with Glu172 [35]. Such an interaction was noted in the docking of ifenprodil, in addition to further interactions with Tyr103 and Asp126 ( Figure 6A). Conversely, OF-NB1 was only able to maintain van-der-Waals interaction with Tyr103 and even failed to interact with the crucial Glu172, despite its positively charged nitrogen ( Figure 6B). Such behavior could be explained by the odd orientation exhibited by OF-NB1 within the σ1R binding pocket. The failure of OF-NB1 to align with ifenprodil supports the observations from the in vitro saturation binding experiments. The co-crystal structure of σ1R with pentazocine was used (PDB-ID: 6DK1) for the docking of ifenprodil and OF-NB1 against σ1R. Both Schmidt et al. and the Glennon model for σ1R ligands highlighted the prominence of the presence of a positively charged nitrogen in all σ1R ligands, where it forms a salt-bridge with Glu172 [35]. Such an interaction was noted in the docking of ifenprodil, in addition to further interactions with Tyr103 and Asp126 ( Figure 6A). Conversely, OF-NB1 was only able to maintain van-der-Waals interaction with Tyr103 and even failed to interact with the crucial Glu172, despite its positively charged nitrogen ( Figure 6B). Such behavior could be explained by the odd orientation exhibited by OF-NB1 within the σ1R binding pocket. The failure of OF-NB1 to align with ifenprodil supports the observations from the in vitro saturation binding experiments.

General Methods
All non-aqueous reactions were performed under N2 atmosphere using flame-dried glassware and standard syringe/septa methods unless stated otherwise. Reactions were magnetically stirred and further monitored by thin layer chromatography (TLC) performed on Merck TLC aluminum sheets (silica gel 60 F254). TLC spots were visualized using UV light (λ = 254 nm) or through staining with KMnO4 solution. Chromatographic purification of products was performed using SiliaFlash P60 silica gel (Silicycle) for pre-

General Methods
All non-aqueous reactions were performed under N 2 atmosphere using flame-dried glassware and standard syringe/septa methods unless stated otherwise. Reactions were magnetically stirred and further monitored by thin layer chromatography (TLC) performed on Merck TLC aluminum sheets (silica gel 60 F254). TLC spots were visualized using UV light (λ = 254 nm) or through staining with KMnO 4 solution. Chromatographic purification of products was performed using SiliaFlash P60 silica gel (Silicycle) for preparative column chromatography with a particle size of 40-63 µm (230-400 mesh). Reactions at −78 • C were cooled in a dry ice/acetone bath and reactions at 0 • C were cooled in an ice/water bath. Chemicals were purchased from ABCR, Acros Organics, Amatek Chemical, Fluorochem, Merck, Perkin Elmer and Sigma Aldrich and used without further purification. Solvents for TLC, extraction and flash column chromatography were of technical grade. Extra dry solvents for all non-aqueous reactions were supplied by Acros Organics (puriss., dried over molecular sieves, water content <0.005%). Deuterated solvents (D, 99.9%) were purchased from Cambridge Isotope Laboratories, Inc. NMR spectra of compounds 3, 7 and 8 are presented in the Supplementary Material (Section S1). Tritium labeling was performed by RC Tritec AG (Teufen, Switzerland). Quality control chromatogram and mass spectrum are presented in the Supplementary Material (Section S2, Figures S1 and S2). Positron emission tomography (PET) imaging was performed in Wistar rats according to a previously published procedure [22]. PET images are presented in the Supplementary Materials (Section S3, Figure S3).

Mass Spectrometry
High-resolution mass spectra were obtained by the mass spectrometry service of the

NMR Spectroscopy
1 H, 19 F and 13 C nuclear magnetic resonance (NMR) were recorded at room temperature on a Bruker Avance FT-NMR (400 MHz) with CDCl 3 or methanol-d 4 as solvent.
Chemical shifts values (δ) are reported in parts per million (ppm) relative to tetramethylsilane (0.00 ppm) as an internal standard and the appropriate CDCl 3 solvent signals (δ H = 7.26 ppm and δ C = 77.16). For 1 H NMR spectra, resonance multiplicities are abbreviated as s = singlet, d = doublet, t = triplet, m = multiplet. Coupling constants (J) are reported in hertz (Hz).

In Vitro GluN1/2B Competitive Binding Assay
Competitive binding assays were performed as previously reported [23]. Briefly, IC 50 binding affinity assays were conducted using Wistar rat brain homogenates and HEPES buffer (30 mM HEPES, 110 mM NaCl, 5 mM KCl, 2.5 mM CaCl 2 and 1.2 mM MgCl 2 ). A dilution series of the test ligands was prepared ranging from 30 µM to 300 pM. The cold ligand was displaced with 4.7 nM of either [ 3 H]ifenprodil or [ 3 H]OF-NB1. Total binding was measured without the cold ligand, and non-specific binding was measured with 100 µM CP101,606 instead of the cold ligand. Each measurement vial contained 0.5 mg/mL of brain homogenate proteins, 20 µL of cold ligand, 10 µL of radioligand and was diluted to 200 µL with HEPES buffer. The vials were incubated at 25 • C and 110 rpm for 1 h. The mixtures were quenched with buffer and filtered through Whatman ® GF/C 25 mm filters soaked with 0.05% PEI solution. Filters were washed twice with cold buffer and placed in scintillation vials. Scintillation vials were filled with 2 mL Ultima Gold TM LSC cocktail and measured in a Beckmann LS6500 liquid scintillator counter.

In Vitro σ1R Competitive Binding Assay
The σ1R competitive binding assay was performed in line with previously reported procedures [23]. IC 50 binding affinity assays were conducted using Wistar rat brain homogenates, and a HEPES buffer (30 mM HEPES, 110 mM NaCl, 5 mM KCl, 2.5 mM CaCl 2 and 1.2 mM MgCl 2 ). Dilution series were prepared with the test cold ligands in a concentration ranging from 30 µM to 300 pM. The cold ligand was displaced with 2.5 nM of (+)-[ 3 H]pentazocine. Total binding was measured without the cold ligand, and non-specific binding was measured with 100 µM eliprodil instead of the cold ligand. Each measurement vial contained 0.75 mg/mL of brain homogenate proteins, 20 µL of cold ligand, 10 µL of radioligand and was diluted to 200 µL with HEPES buffer. The vials were incubated at 37 • C and 110 rpm for 2.5 h. The mixtures were quenched with buffer and filtered through Whatman ® GF/C 25 mm filters soaked with 0.05% PEI solution. Filters were washed twice with buffer and placed in scintillation vials. Scintillation vials were filled with 2 mL Ultima Gold TM LSC cocktail and measured in Beckmann LS6500 liquid scintillator counter. The crystal structure of NMDA-GluN1b/GluN2B dimer in complex with ifenprodil (PDB ID: 3QEL) and σ1R bound to (+)-pentazocine (PDB ID: 6DK1) were selected for docking simulations. [35,36] The simulations were performed using Molecular Operating Environment (MOE 2019.0101, Chemical Computing Group, ULC, Montreal, QC, Canada, H3A 2R7, 2021) software. After loading the structures in MOE, they were prepared for docking simulations using the default parameters in the 'QuickPrep Panel', including the removal of water molecules 4.5 Å away from the ligand pocket, adding hydrogen atoms to the protein structure, adjusting protonation states, and ensuring the energy minimization of the protein structures with an Amber10:EHT force field.

Validation of the Docking Protocol
In order to validate the docking protocol for the two selected targets implemented in the study (described below), the co-crystallized ligands were re-docked in the receptor binding pocket. The ability to reproduce the reported interactions with a minimum root mean square distance (RMSD) value between the co-crystallized pose and docked pose validate the methodology. The RMSD values obtained from the re-docking of the co-crystal ligands of both 3QEL and 6DK1 were 0.257 Å and 0.250 Å, respectively.

Ligands Dataset Curation for Docking Simulations
The 'builder program' implemented in MOE was used to model both ifenprodil and OF-NB1, where all possible conformations at the physiological pH were obtained.

Docking Simulations
Docking of the obtained conformations of both ligands (ifenprodil and OF-NB1) was carried out using placement and refinement algorithms of the MOE program. Initial docking of the molecules in the active sites used the 'Triangle Matcher' placement method and the 'London dG' scoring function. Further postplacement refinement of docking poses was achieved by using the 'GBVI/WSA dG' scoring method. The poses with minimum energy were used for visualization of the binding interactions as well as occupancy of the binding site of both receptors. Docking overlays are presented in the Supplementary Material (Section S4, Figures S4 and S5).

Conclusions
In conclusion, we successfully synthesized and evaluated a novel GluN1/2B radioligand, [ 3 H]OF-NB1, for preclinical GluN1/2B ligand development. Its superiority over the current state of the art was demonstrated in a head-to-head comparison by in vitro binding assays, and we showed that systemic errors arise from the use of an unselective radioligand, such as [ 3 H]ifenprodil, in GluN1/2B binding affinity screening assays. Our aim is to raise awareness for the need to continuously improve the ligand development toolkit. Due to the high potential of GluN1/2B antagonists exhibiting off-target effects toward σRs and vice versa, we envision that the use of [ 3 H]OF-NB1 in GluN2B-targeted drug discovery will facilitate the identification of highly selective candidate drugs with the potential to hold up to expectations in clinical trials.

Informed Consent Statement: Not applicable.
Data Availability Statement: The data generated and analyzed during our research are not available in any public database or repository but will be shared by the corresponding author upon reasonable request.