Design, Synthesis and Biological Evaluation of Neogliptin, a Novel 2-Azabicyclo[2.2.1]heptane-Based Inhibitor of Dipeptidyl Peptidase-4 (DPP-4)

Compounds that contain (R)-3-amino-4-(2,4,5-trifluorophenyl)butanoic acid substituted with bicyclic amino moiety (2-aza-bicyclo[2.2.1]heptane) were designed using molecular modelling methods, synthesised, and found to be potent DPP-4 (dipeptidyl peptidase-4) inhibitors. Compound 12a (IC50 = 16.8 ± 2.2 nM), named neogliptin, is a more potent DPP-4 inhibitor than vildagliptin and sitagliptin. Neogliptin interacts with key DPP-4 residues in the active site and has pharmacophore parameters similar to vildagliptin and sitagliptin. It was found to have a low cardiotoxic effect compared to sitagliptin, and it is superior to vildagliptin in terms of ADME properties. Moreover, compound 12a is stable in aqueous solutions due to its low intramolecular cyclisation potential. These findings suggest that compound 12a has unique properties and can act as a template for further type 2 diabetes mellitus drug development.


Introduction
Pharmacological therapy for type 2 diabetes mellitus (T2DM) patients has developed considerably over the past decades involving several new strategies [1][2][3]. Many of these strategies involve more patient-friendly ways of drug administration. Several decades ago, insulin injections were the only way to overcome hyperglycaemia, but now many oral antihyperglycaemic agents are available that are administrated either in monotherapy or as combinational drug therapy. Currently, one of the most recent and promising methods to treat T2DM is to use dipeptidyl peptidase 4 (DPP-4) inhibitors, which are also called gliptins [4][5][6][7][8][9]. They prevent the degradation of glucagon-like peptide-1 (GLP-1) and glucose-dependent insulinotropic peptide (GIP), stimulate insulin synthesis, suppress glucagon secretion, inhibit appetite, reduce body weight, slow gastric emptying, and can

Molecular Modelling
We initially analysed analogous compounds that contain the same beta-amino acid ((R)-3-amino 4-(2,4,5-trifluorophenyl)butanoic acid) and bicyclic amino moiety (2-azabicyclo[2.2.1]heptane) with a nitrile substituent. We performed molecular modelling predictions and compared the predicted DPP-4 inhibition activity of these compounds. We looked particularly at the presence of spontaneous intramolecular cyclisation previously described for the acylated cyanopyrrolidine derivatives [27]. This intramolecular cyclisation leads to the formation of substituted diketopiperazines with no affinity to DPP-4 ( Figure 1). To avoid these negative consequences, we had an idea to modify the cyanopyrrolidine moiety by adding the second aliphatic cycle to make this part less flexible. The same unsymmetric ring system described in [28] for ledipasvir demonstrated beneficial SAR properties compared to pyrrolidine-based analogues. We also decided to prefer betaamino acid (the same as the one in sita-and evogliptin) instead of alpha derivatives because of the enlargement of the aliphatic chain that connects cyanopyrrolidine moiety with the amino group that is crucial in protein-ligand interaction leads to energetically unfavourable cyclisation reaction profiles. The presence of chiral centre in 3-position of 2-aza-bicyclo[2.2.1]heptane results in two isomeric structures of exoand endo-configuration ( Figure 2). Next, we analysed protein-ligand binding modes and spatial orientation of reference compounds in terms of functional groups needed for correct ligand-protein interaction (Figure 3a,b). Amino acids in the active site of DPP-4 form different subsites (S2, S1, S'1, and S'2) that are usually occupied by different residues of the substrate molecule. The S1 pocket is formed predominantly by S630, D708, H740 (a catalytic triad), Y631, V656, W659, Y662, Y666, V711. Meanwhile, the S'2 pocket consists of R125, E205, E206, S209, F357. Critical residues of S'1 and S'2 binding pockets are Y547 and W629, respectively. Sitagliptin (S) and vildagliptin (V), the two earliest DPP-4 inhibitors approved, occupy the active centre differently (Figure 3). Due to the structural dissimilarity, they have differences in DPP-4 binding site filling efficiency (S > V), ability to form pi-stacking interaction with aromatic groups (S > V), hydrophobic contacts intensity (S > V), and hydrogen bonding intensity (S < V).
Our new compound 12a has shown the following protein-ligand interaction profile: Residues in S'1 and S'2 binding pockets are involved in hydrophobic interactions. Trifluorophenyl aromatic ring forms significant hydrophobic site-specific interaction in S1 and S2 pockets using crucial π-π stacking interactions with tyrosine sidechains and hydrogen bonds with E205/E206. Trifluorophenyl ligand moiety partially occupies the S1 pocket, thus forming π-π stacking with Y666/Y662 and H740. Additional aliphatic cycle allows 12a to form hydrophobic interactions with residues in the S2 extensive region.
Also, 12a contains the cyanopyrrolidine moiety and, thus, mimics the penultimate proline of DPP-4 substrates; its nitrile moiety is targeted to occupy the S'1 binding pocket. This interaction mediates the formation of a hydrogen bond with Arg669 in the S'1 pocket. Here, the nitrile group is a non-covalent interactant [29] and an important part of the site-specific interaction machinery (as in vildagliptin, Figure 3).   (Table 2). Although 12a has GlideScore and Emodel values within the same range as for reference structures, its ∆Gbind value is preferred, indicating that the protein-ligand complex of DPP-4 with 12a is energetically more favourable. Moreover, the predicted parameter of potential cardiotoxicity (QPlogHERG, blocking of HERG K+ ion channels) for 12a is the safest. Sitagliptin has the highest risk of potential cardiotoxicity (the reference value less than -5 may result in high cardiotoxicity). Some reviews and clinical trials showed conflicting findings of this effect [30][31][32]. Predicted cardiotoxicity of 12a is comparable to vildagliptin, which does not cause side effects associated with heart failure [33]; 12a is also predicted to have a 1.5-fold reduced cardiotoxicity than sitagliptin. Thus, the new compound 12a unifies the binding parameters of two known active DPP-4 inhibitors, and at the same time, is predicted to be less toxic. As can be seen in Figure 4, non-active compound 12b binds in reversed, incorrect mode. This fact is linked with steric incompatibility with the active cleft of DPP-4. The same applies to compounds 12c,d.
Additionally, this fact is confirmed with MM-GBSA free energy calculations (Table 2), compounds 12b-d showed a significant increase in free Gibbs (∆G) energy, which would result in the unfavourability of ligand-protein complex with DPP-4 (also confirmed by docking results). We observed a loss in site-specificity that resulted in the absence of some protein-ligand interactions that are typical for reference compounds (see Figure 3). Table 2. Scoring results of re-docked reference structures, comparison with best-fitting compound 12a and non-active 12b-d. Higher risk of potential cardiotoxicity is highlighted in yellow and lower risk of potential cardiotoxicity is highlighted in green.
Here we also provide a scheme for the stereoselective synthesis of Sand R-exo-2azabicyclo[2.2.1]heptane-carbonitriles starting from S and R-1-phenylethylamine, respectively. This synthetic route appears to be more promising than the necessity of isolating each resulting diastereomer from the mixture on the final step.
For the synthesis of pure 3S-and 3R-exo-2-azabicyclo[2.2.1]heptane-carbonitriles 9a and 9b, we used the intermediates 6a and 6b, correspondingly. The 3S-acid 6a was prepared from the precursor 14a as described in the literature [38,[41][42][43][44][45][46] and shown in Figure 8. The use of (R)-1-phenylethylamine 13a followed by two-step hydrogenation and saponification afforded the only isomer 14a with the desired S-configuration of carbon in 3-position. Subsequent Boc-protection step afforded to the pure isomeric intermediate 6a [42]. The intermediate of 3R-configuration, 6b, was prepared from 13b by the analogy with 6a [47]. The use of another enantiomeric starting reagent (S)-1-phenylethylamine 13b afforded the pure compound 6b of 3R-Exo-configuration [45]. The following steps to prepare 9a and 9b were carried out by the analogy with the scheme shown in Figure 7.  The obtained individual isomeric intermediates Sand R-exo-2-azabicyclo[2.2.1]heptanecarbonitriles 9a and 9b were used to synthesise the corresponding individual compounds 12a and 12b.

Structure and Purity Confirmation
The structures of key target final compounds 12 were unambiguously confirmed by 1H, 13C{1H}, 13C apt, COSY, HSQC, and NOESY NMR spectroscopy and liquid chromatography-mass spectrometry (LC/MS) methods. For structure confirmation of their precursors, 11, 1H NMR spectroscopy was applied, and in some cases, if necessary, 13C NMR spectra were also registered. In each case from compounds 11 and 12, chromatograms obtained by LC/MS analysis revealed the only molecular ion peak of 100% purity (see Supplementary file S1_NMR. Supplementary file S2_mass_spectra).
The exact assignment of individual proton signals in 1H NMR spectra of the final compounds 12 and their precursors 11 seemed not to be an easy task due to the overlapping and complex multiplicity of most signals. The doubling of some proton signals further complicated the task-however, total intensities of the same signals precisely correlated with each other.
Signal doubling of the identical type protons is in good accordance with the existence of equilibrium E/Z-rotameric forms described in the literature. It is well known (see, for example, [48]) that the introduction of the N-acyl function in any 2-substituted pyrrolidines always leads to the formation of E/Z-rotameric forms (Figure 9a). The main reason for the stabilisation of rotamers is the hindered rotation around the N(1) -CO amide bond, and an additional reason is a steric factor [proximity of the substituent at C(2) in pyrrolidine]. In most cases, the existence of equilibrium forms can be seen in 1H NMR spectra by doubling the corresponding proton signals (see, for example, [49]). All the synthesised inhibitors belonging to the class of N-acyl-2-substituted pyrrolidines should exist in solutions in E/Z-rotameric forms, as it has been shown for 12a, as an example ( Figure 9b).
Indeed, the presence of rotamers of all the synthesised compounds 11a-d and 12a-d was confirmed by registration of 1H NMR spectra with doubling proton signals at room temperature (see Section 3.5 and Supplementary Materials) as it is shown for 12a as an example ( Figure 10). The presence of rotameric forms was also evident in the 13C{1H} NMR spectra (see Section 3.5 and Supplementary file S1_NMR). The exact proton and carbon nuclei signals assignment was performed using 13C atp, COSY, HSQC, HMBC NMR spectroscopy (see Supplementary file S1_NMR).
In order to prove the existence of E/Z-rotameric forms in all the synthesised compounds, three representatives, 12a, 12b, and their mixture 12a,b, were chosen. Registration of 1H NMR spectra of these compounds in the temperature range 30-140 • C unambiguously confirmed the E/Z-rotamerism. For example, fragments of 1H NMR spectra of 12a in a dimethyl sulfoxide (DMSO)-d6 solution at different temperatures are shown in Figure 11. Each methine proton signal, H-1 and H-3 (numeration in Figure 10), of the azanorbornane fragment was registered in the temperature range 30-90 • C as two broadened singlets in the ratio of~85:15. The energy barrier of free rotation around the amide bond is large enough, and it is possible to overcome it only at a temperature of more than 140 • C. At a temperature of nearly 140 • C, a partial coalescence of the abovementioned signals was observed. Coalescence of H-1 and H-3 proton signals in 1H NMR spectrum of 12b was also found at a temperature above 140 • C (see Supplementary materials). The ratio of E/Zrotameric forms of 85:15 in DMSO-d6 solution at room temperature calculated based on 1H NMR data was almost the same for all the synthesised compounds, 11 and 12. Therefore, the data set of results obtained by NMR and LC/MS methods unambiguously confirmed that the final inhibitors 12a and 12b used for in vitro activity assay were chemically and stereochemically pure. The compounds 12a,b and 12c,d were diastereomeric mixtures and did not contain any side components.

Inhibitory Activity Evaluation
The DDP-4 enzyme inhibitory activities of the synthesised compounds were assessed using protocols similar to those described in the literature [50]. For IC50 values determination, inhibitory assays were carried out using recombinant DPP-4 enzyme D4943, chromogenic substrate Gly-Pro-pNA and buffer system (50 mM Tris-HCl, 50 mM NaCl, 0.01% Triton, pH = 7.6). After short incubation (37 • C for 30 min), the absorbance at 405 nm was measured. The procedure was initially optimised using reference compounds. Each inhibitor was analysed in the dilution range from 10 −4 to 10 −11 M. Certified samples of commercial drugs-inhibitors of DPP-4 (vilda-, sita-, alo-and linagliptin) with known IC50 were used as reference compounds. It was necessary to ensure that IC50 values estimated during the experiments fit into the ranges of values known for them from the literature and were reliably reproduced. Only after that precaution, synthesised compounds were tested. Correspondingly, the inhibition activity analysis of related enzymes (DPP-8 and DPP-9) was performed by fluorescent method with DPP-8 assay Kit and DPP-9 assay Kit. We analysed every compound in the dilution range 10 −2 to 10 −8 M (Table 3). Table 3. DPP-4-, DPP-8-, and DPP-9-inhibitory activities of target compounds 12a-d.

Molecular Modelling
The docking procedure was performed using the Schrödinger Glide module in standard precision mode. The docking grid was calculated according to native ligands dimensions using available PDB models (1X70 and 6B1E). In order to perform ligand docking, protein structures were superimposed, and coordinates of the docking grid were tied to the ligand centroid. The docking area was limited per reference ligand size, with 7 Å as a buffer zone. Grid spacing was set 0.375 Å, VdW radii cut-off 0.8 Å. Several optional constraints were added: nitrile group orientation (reference-vildagliptin), hydrophobic attraction-halogen-substituted moiety (sitagliptin). Docking solutions generation was performed using the Glide module of Schrödinger Suite (version 2020-3) in standard precision mode with 0.8 Å VdW radius and with previously mentioned optional constraints. Docking protocol was validated by redocking of reference compounds. The geometry of the best-fitting ligand was shown on ligand interaction diagrams. For each inhibitor, 45 docking solutions were generated, the best 15 were used for binding mode analysis. GlideScore and EModel values-controlled target affinity. Optimal binding poses were selected per cluster RMSD less than 1.5 Å. Binding pose and calculated parameters of reference ligand were taken as a control. Free Gibbs energy (∆G) was calculated using the MM-GBSA method, implemented in Schrodinger Suite v.2020-3, module Prime. All results were processed using Maestro molecular modelling interface (Schrodinger Suite v.2020-3). All proteinligand complexes were prepared and refined using Schrodinger Protein Prepwizard. This procedure was essential to fix missing amino acid sidechains, incorrect bond orders, and correct protonation states. Optimal binding poses were selected by cluster RMSD less than 1.5 Å. Binding parameters of the reference ligand were selected as a control. To prove the structural novelty of identified hits, we used the Tanimoto coefficient that shows similarity scores between library compounds and already known drugs. A score below 0.5 is a good sign of low similarity. We used nine marketed gliptins (sita-, vilda-, alo-, lina-, saxa-, teneli-, trela-, evo-and omarigliptin) as reference compounds. The search for hits (top-scoring compounds with residues involved in the binding site of the enzyme with an increase in field regions) [51] was carried out in a library of chemical structures with a high level of similarity (Tanimoto coefficient more than 0.8), built on 3-azabicyclo[2.2.1]heptene-2carbonitrile scaffolds. Potential toxicity evaluation of ligands was carried out using the QikProp module, using 2D/3D-QSAR descriptor combinations analysis.

Chemical Synthesis
All starting reagents were bought from reliable commercial vendors, mostly Sigma-Aldrich, Merck, and Acros, and used without further purification. Intermediates and final compounds were isolated using column chromatography on silica gel. Compounds were only used for biological evaluation if the purity was ≥95%.

LC/MS, NMR, and Elemental Analysis
Liquid chromatography-mass spectrometry (LC/MS), NMR spectroscopy, and elemental analysis methods were applied to confirm the structure and purity of all synthesised compounds.
The structures of key target final compounds were unambiguously confirmed by 1H, 13C{1H},13C apt, COSY, HSQC, and NOESY NMR spectroscopy. For structure confirmation of intermediates, 1H NMR spectroscopy was applied, and in some cases, if it was necessary, 13C{1H} NMR spectra were also registered. NMR spectra were registered on spectrometers Bruker DRX 400 (400. 13  Elemental analysis was performed on Vario MICRO cube CHNS analyser (Elementar Analysensysteme GmbH, Hanau, Germany).

2-tert-
The synthesis was made according to the procedure described in [34][35][36]. To a mixture of a saturated solution of ammonium chloride (39.3 g) and a toluene solution of ethyl glyoxylate (50%, 150 g) freshly prepared cyclopentadiene (64.7 g). The reaction mixture was stirred for 12 h at room temperature, then extracted with a mixture of methyl tert-butyl ether (MTBE) and petroleum ether (PE) in a 1:3 ratio. The aqueous layer was alkalised with NaOH aq. solution (50%) up to pH 8.0-9.0, extracted with MTBE and dried over anhydrous sodium sulphate. After solvent evaporation, the mixture of 3a-d was obtained as a yellow oil (53%, 67 g) and used without any purification.

(3S,R)-exo-2-(tert-Butoxycarbonyl)-2-azabicyclo[2.2.1]heptane-3-carboxylic acid (6a,b)
The synthesis was made according to the procedure described in [37,39]. A mixture of ester 5a,b (14.8 g), and LiOH (monohydrate, 6.58 g) in a water-methanol solution was stirred overnight at room temperature. TLC monitoring showed SM remained. Another 1.5 equivalents of LiOH were added, and the mixture was kept under stirring at 40-50 • C for 2 h. Then methanol was evaporated, the mixture was diluted with water, extracted with EtOAc, then the aqueous layer was acidified with citric acid to pH 4.0 and extracted with DCM. After drying with sodium sulphate and removing the solvent, the desired product 6a,b (11.9 g, 89%) was obtained.

(3S)-exo-2-(tert-Butoxycarbonyl)-2-azabicyclo[2.2.1]heptane-3-carboxylic acid (6a)
The synthesis was made according to the procedure described in [42,46]. To EtOH (15 mL) solution of compound 14a (0.9 g, 6.4 mmol) BOC2O (1.6 g, 7.4 mmol) was added. The mixture was stirred at RT for 24h and evaporated. Water (15 mL) was added to the residue. The product was extracted with EtOAc; then, the aqueous layer was acidified with citric acid to pH 4.0 and extracted with DCM. After drying with sodium sulphate and removing the solvent, the desired product 6a (1 g, 61%) was obtained. 1H NMR and LC-MS are the same as described in 6a,b synthesis.

tert-butyl (3S,R)-exo-3-cyano-2-azabicyclo[2.2.1]heptane-2-carboxylate (8a,b)
Trifluoroacetic anhydride (14.4 g) was added dropwise to a suspension of amide 7a,b (10.3 g) in anhydrous THF, at a temperature not higher than 4 • C, within 10 min. After the TLC, it became clear that the starting amide was still present, and another portion of trifluoroacetic anhydride (9 g) was added. The mixture was kept undercooling for 3 h, then ammonium hydrogen carbonate (45 g) was added portion-wise. The mixture was subjected to column chromatography. Eluent: a mixture of PE:

(3S,R)-exo-2-azabicyclo[2.2.1]heptane-3-carbonitrile (9a,b)
To a solution of BOC-protected nitrile 8a,b (7 g) in acetonitrile (30 mL), p-TSA (12 g, two-fold excess) was added. The mixture was stirred overnight. Acetonitrile was evaporated, the residue was triturated with diethyl ether (3-4 treatments with decantation). The residue was dissolved in DCM and saturated with ammonia from a balloon. The precipitated ammonium salt of p-TSA was filtered. The filtrate was evaporated, and the residue was purified by column chromatography. Eluent: DCM after extraction of aqueous ammonia in the ratio of 1:10. The desired product 9a,b (3.

Conclusions
Here we show that pseudo peptides containing trifluoro-substituted aromatic βamino acid and the proline functional analogue (with a labile nitrile group) inhibit DPP-4. Interestingly, the Exo-isomeric derivatives within all these groups were more potent against DPP-4 than the corresponding endo-analogues. Moreover, we found that the S-configuration of the amine part was the most favourable in all the experiments. The maximum similarity score of 0.38 was obtained for compounds 12a and evogliptin among all the pairs (a synthesised compound + already known drug). Therefore, all our synthesised compounds can be considered significantly different compared to marketed gliptins.
The IC50 of compound 12a is in the same range as for reference compounds, and this is consistent with literature data [9,48,[52][53][54][55]. As for DPP-8, 12a showed feeble inhibitory activity in the concentration range 10 −3 -10 −6 M, and for a homologous enzyme DPP-9 in concentration 10 −3 M. This is significantly below the estimated therapeutic range and is consistent with literature data for already known DPP-4 inhibitor compounds [55]. This makes it possible to predict the absence of side effects associated with undesirable inhibition of homologous enzymes DPP-8 and DPP-9.
We revealed and proved the presence of E/Z-rotameric forms in a solution. The energy barrier of free rotation around the amide bond is large enough, and it is possible to overcome it only at a temperature of more than 140 • C. At a temperature close to 140 • C, a partial coalescence of the abovementioned signals was observed. The initial studies of 12a suggest that this compound could be less toxic and more stable in aqueous solutions than marketed gliptins.