Computer-aided design, synthesis, and biological evaluation of [4.3.0] bicyclic prolyl oligopeptidase and fibroblast activation protein-α dual inhibitors

We have previously described several different chemical series of bicyclic prolyl oligopeptidase (POP) inhibitors as probes for neurodegenerative diseases that demonstrated nanomolar activity in vitro and submicromolar activity in cellulo. The more recent implication of POP in cancer, together with homologous fibroblast activation protein α (FAP), implicated in tumor growth, led us to consider developing POP/FAP dual inhibitors as a promising strategy for the development of cancer therapeutics. We report herein docking-guided design of a new bicyclic scaffold and synthesis of both covalent and non-covalent bicyclic inhibitors. Biological evaluation of first-of-their-kind [4.3.0] bicyclic compounds confirmed that reactive groups, or covalent warheads, are required for inhibitor activity. This work ultimately led to a dual inhibitor equipotent to the only anti-POP/FAP drug that ever-reached clinical trials.

tissues, 27 making it an extremely valuable target for therapeutic intervention against refractory tumors, and inhibitor development has in fact already started ( Figure 2). [28][29][30]

Figure 1. Selected POP and FAP inhibitors
In recent medicinal chemistry endeavors, selective inhibition of one enzyme over the other has been pursued. As illustrated in Figure 2 selectivity is very sensitive to minor structural changes.
For example, while compound 5 is highly FAP selective, the analogue 6 is highly POP selective.
This D-Ala-induced selectivity for FAP has been further observed recently. 31 Nevertheless, Christiansen et al. suggested that targeting both FAP and POP blocks stromal invasion and angiogenesis, respectively, and may alter cancer growth. 32 They designed a pseudopeptide dual inhibitor which was found to block tumor growth in mice. These findings suggest that dual inhibition is a promising strategy, though this large, non-drug-like molecule was unsuitable for further consideration. 33 Consequently, although an overview of the literature suggests that selectivity may be easier to achieve than dual inhibition, the latter may be an ideal strategy for designing and developing anti-cancer therapeutics.  28,30,34,35 In 2009, our own group reported a series of [3.3.0] bicyclic POP inhibitors based on compound 2 which were found to be cell-permeant and potent in the sub-micromolar range. This series of nitrile-containing compounds were designed to act as covalent inhibitors targeting the catalytic serine in the POP active site. 19 Interestingly, a few years later, KYP-2047 was co-crystallized with POP, demonstrating the covalent nature of the binding of nitrile derivatives in the active site of POP. 36 However, the series of inhibitors based on compound 2 was halted after metabolism studies revealed it to be metabolized into complex mixtures via oxidation of the sulfur. 37 A few years later, compound 3 was discovered via virtual screening and docking-guided optimization, This inhibitor exhibited a POP inhibitory activity five times more potent than that of our first hit 2 and was active in low-micromolar concentrations on human glioblastoma and endothelial cancer cells. 20 In addition, we found that the introduction of the [4.3.0] bicyclic molecular scaffold improved the metabolic stability of our inhibitors. 20 Five years ago, we also reported the structure-based design and synthesis of a novel class of POP inhibitors based on a hexahydroisoindole scaffold, such as 4 ( Figure 1). A docking study guided the selection of structures (both in terms of stereo-and regiochemistry) for synthesis.
Following the synthesis of the best virtual candidates, in vitro assays revealed that one member of this chemical series, compound 4, was more active than any of our previous inhibitors, exhibiting a Ki of 1.0 nM. Additional assays also showed that the scaffold of this potent inhibitor, in contrast to the series based on compound 2, is highly metabolically stable. 38 However, upon in vitro testing of 3 and 4 against recombinant FAP, they were completely inactive. Analysis of docking poses revealed a lack of stabilizing interactions with the two glutamic acid residues in the active site of FAP (Glu203 and Glu204).
With this information in hand, we became interested in the design of dual POP/FAP inhibitors.
We report herein our successful efforts in the development of dual inhibitors based on an improved bicyclic core.

Results and discussion
Computer-aided design. With our first three series of POP inhibitors illustrated by compounds 2, 3 and 4, we have demonstrated the accuracy of our docking program FITTED [39][40][41] in predicting binding modes of POP covalent inhibitors. When 2 and stereoisomers of 2 (adhering to the [3.3.0] bicyclic system) were evaluated, we found that the stereochemistry corresponding to that of Damino acids was optimal (hydrogen atom highlighted in blue in Figure 3, compound 2). The resultant stereochemistry upon cyclization (hydrogen atom highlighted in green in Figure 3) at the ring junction fortunately imposed a shape that fit best in the binding site. In this previous report, 19 computational studies also indicated that this [3.3.0] bicyclic system was less optimal for binding to POP, and that a [4.3.0]-ring system with a specific stereochemistry should exhibit better affinity ( Figure 3, compound 2a). 22 Unfortunately, our synthetic efforts were vain, as the epimer at the ring junction (hydrogen atom highlighted in green in Figure 3, compound 2a) was the only isomer observed experimentally but was not predicted to bind optimally in the active site of POP. Further computational predictions indicated that the affinity for POP could be improved by increasing the size of the western ring and inverting two stereocenters, both the carbon at the cyclic fusion (C7a in 2, C8a in 2a) and the carbon alpha to the cyclic amide (C6). To do so, we decided to prepare a first series of analogues built around a [4.3.0]-ring system similar to that of 2a but which could be accessible synthetically. After several rounds of virtual modifications and docking predictions, inhibitor structure 10a was discovered. As can be seen in Figure 4, the predicted binding mode of nitrile 10a is highly favored, featuring the same key interactions as potent aldehyde 1. This prediction encouraged us to pursue the synthesis of compound 10a and other analogues.
Our previous inhibitor 2 and this newly-designed scaffold resemble previously-reported potent inhibitor 1 (Figure 4). The bicyclic scaffolds 2 and 10 were introduced by virtually rigidifying 1 and introducing heterocyclic alkanes to both optimize the docking pose and ensure synthetic feasibility. The valine-based side chain of Talabostat ( Figure 2), a POP-FAP inhibitor that reached Phase III clinical trials, 42 inspired the introduction of methyl groups into 10a and incorporation of the boronic acid warhead, leading to 12c and 13b. The complete list of new analogues selected for synthesis is provided in Table 1. Many POP inhibitors feature nitriles, activated nitriles (i.e., with proximal fluorine atoms), or boronic acids, the latter two of which are more electrophilic and lead to more potent FAP inhibition. Our current version of our docking program FITTED does not consider either the reactivity of the catalytic residue nor the reactive warhead. Nevertheless, a computational study from our group on the reactivity of the catalytic serine residues in both POP and FAP suggests that the catalytic serine in POP is more nucleophilic than that in FAP and that, as a result, nitrile derivatives are unlikely to act as potent covalent inhibitors of FAP, while boronic acids are promising alternatives. 43,44 As a result, in our quest to develop dual POP/FAP inhibitors, the boronic ester or acid derivatives were also considered.
Boronic acids have been widely used in medicinal chemistry, notably as warheads of reversible covalent inhibitors of proteases, 45,46 including two approved drugs (Bortezomib and Ixazomib for the treatment of relapsed multiple myeloma and mantle cell lymphoma). 47 In addition, boronic acids are remarkably stable despite their high reactivity and consistently display very low toxicology profiles. 48,49 Consequently, we designed our bicyclic boropeptides to be structurally close to Talabostat (Figure 4), a multi-target inhibitor, which entered Phase III clinical trials for the treatment of advanced non-small cell lung cancer. 42 However, Talabostat displayed a loss in efficacy in vivo believed to be a result of a reversible intramolecular cyclization into an inactive cyclic adduct. 50 The constrained scaffold of our designed boronic acids 10f, 11b, 12c, and 13b would circumvent this cyclization.
Furthermore, in both POP and FAP, the boronic acid motif may act as a transition state analogue, forming both hydrogen bonds (with His680 and Tyr473 and with His734 and Tyr571, respectively) and covalent bonds with the catalytic triad (Ser554 and Ser624, respectively) in a tetrahedral configuration, as opposed to the trigonal planar configuration conferred by nitrilecontaining inhibitors ( Figure 4). Nonetheless, the design of FAP/POP dual inhibitors remains challenging due to the difference in polarity between the active sites. While three hydrophobic or hydrogen bond donor residues contribute the necessary interactions for high inhibition of POP (aromatic interactions with Phe173 and hydrogen bonding with Trp595 and Arg643), high inhibition of FAP relies on interaction with hydrogen bond acceptors Glu203 and Glu204 in the hydrophilic pocket ( Figure 5).  n.b. the hydrolyzed boronic esters (boronic acids) were docked.
Upon docking to POP, the N-Cbz boronic ester derivative 10c was observed to fit very well into the active site of POP ( Figure 6A), while docking to FAP gave unfavorable proposed binding modes, as the carboxybenzyl group is too large to fit into the active site (not shown). This compound is therefore expected to be selective for POP. After virtual optimization of the amide side chain, the acetyl group turned out to be an excellent compromise for the design of potent dual inhibitors, as key interactions were conserved. The N-acetyl group may act as a hydrogen bond donor in FAP (with Glu203 or 204) and as a hydrogen bond acceptor in POP (with Trp595). The docking-predicted binding mode of the N-acetyl analog in both POP and FAP is shown in Figure   6. Furthermore, in order to evaluate the impact of the covalent warhead, the non-covalent analog 10b was also prepared.

Biological evaluations
The non-covalent, carbonitrile, and boronic ester/acid bicyclic series were tested in vitro for inhibition of POP activity. The results of these assays can be found in Table 3.  candidates. As observed previously by our group, 44 it is likely that the nitrile is not properly oriented to react covalently with the catalytic serine in POP, and therefore binds non-covalently.
Dose-response curves of the most potent POP bicyclic inhibitors can be found in Figure 8A. stereochemistry was optimal to inhibit the enzymes. Furthermore, the bicyclic compounds are likely to be more metabolically stable 20,38 and more specific to our enzymatic targets in vivo. 55 Dose-response curves comparing of our top linear peptidic inhibitors can be found in Figure 8B.
The compounds predicted to be the most promising against FAP by docking, the N-acetyl bicyclic derivative and three of the free amines, were next tested against FAP. One of the Cbzcontaining bicycles was also tested on FAP and displayed no inhibitory activity (see Supporting Info), confirming the need for smaller side chains in FAP inhibitors we proposed previously. 1 The results of the FAP assay are displayed in Table 4 and Figure 9. In vitro results indicate that free amine boronic acid 10f exhibits nanomolar activity in FAP and low micromolar activity in POP, making it a promising dual inhibitor for future development. However, N-acetyl boronic ester derivative 12c exhibits submicromolar activity in both enzymes and comparable potency to failed clinical trial candidate Talabostat against POP (Figure 1), making it a very promising drug candidate for future studies.

Conclusion.
Our group's research has previously led to potent bicycle-based POP inhibitors, revealing that the introduction of bicyclic scaffolds can enhance the metabolic stability of these inhibitors. 20,38 In the shift toward POP-FAP dual inhibitors, we then aimed to improve these bicyclic scaffolds while simultaneously constraining the known inhibitor Cbz-Pro-Prolinal 1 and failed drug candidate Talabostat. Our results indicate that we were not only able to obtain potent compounds using our computationally guided optimizations of known inhibitors, but that we were able to use this method along with synthetic developments to produce an inhibitor with comparable potency to a drug that reached Phase III clinical trials. Currently, we are carrying out cell-based assays to assess the activity of our leads in cellulo, as well as performing further experiments to optimize the activity and pharmacokinetic properties of our leads including 10a, 10c, and 12a as selective POP inhibitors and 10f and 12c as dual inhibitors.

Experimental Section.
In Vitro Assays. POP in vitro assays were performed as previously published by our group. 20,38,43 The POP batch used in these assays exhibited a Km of 141.2 µM and kcat of 21.2 s -1 .
The FAP assay was performed using the FAP Assay Kit from BPSBioscience. 57 The FAP batch used in these assays exhibited a Km of 33 µM.
Synthesis. All commercially available reagents were used without further purification. All reactions, unless otherwise indicated, were carried out in flame-dried flasks under argon atmosphere with anhydrous solvents. FTIR spectra were recorded using a Perkin-Elmer Spectrum One FT-IR. 1 H, 13

(R)-2-methyl-N-((R)-1-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)pent-4-en-1yl)propane-2-sulfinamide (19)
Tricyclohexylphosphonium tetrafluoroborate (63 mg, 0.1 eq) was dissolved in toluene (2.1 mL), and the solution was stirred rapidly. Copper (II) sulfate pentahydrate (43 mg, 0.1 eq) and water (0.9 mL) were added, turning the reaction light blue. Benzylamine (37 mg, 0.2 eq) was added, turning the mixture dark blue. The mixture stirred at room temperature for 10 minutes, and was then cooled to 0°C. The sulfinylimine 15a (320 mg, 1 eq) in toluene (2.1 mL) was added, followed by B2pin2 (651 mg, 1.5 eq), and the reaction mixture turned turquoise. The reaction was kept at 0°C for 15 minutes, then was warmed to room temperature and stirred overnight, after which the reaction turned dark brown. The mixture was diluted with ethyl acetate and quenched with saturated NaHCO3. The biphasic mixture stirred for 30 minutes. The product was then extracted from the aqueous layer with ethyl acetate, and the organic layer was washed with saturated NH4Cl, saturated NaHCO3 (copiously) and brine, dried over Na2SO4, and concentrated in vacuo to give the crude product as a brown oil, which was clarified with charcoal to give a clear oil (385 mg, 71%  1-((3aS,4S,6S,7aR) General peptide coupling procedure A. The protected amino acid (1 eq) was suspended in DCM (0.1 M), and HOBt•H2O (1.2 eq) was added, followed by EDC•HCl (1.2 eq). The reaction stirred at 0℃ for 1h. The amine salt (1 eq) was then added, followed by Et3N (3 eq). The reaction stirred at 0℃ for 1h, then at room temperature overnight. Water was added, and the product was extracted with DCM. The combined organic layers were washed with saturated NH4Cl, saturated NaHCO3, and brine, dried over Na2SO4, filtered, and concentrated in vacuo to give the crude product. The crude material was purified by flash chromatography on a silica gel column to give the corresponding dipeptide.
General peptide coupling procedure B. The protected amino acid (1 eq) was suspended in DCM (0.1 M), and Et3N (5 eq) was added, followed by Piv-Cl (1.1 eq). The reaction stirred at 0℃ for 1h. The amine salt (1 eq) was then added. The reaction stirred at 0℃ for 1h, then at room temperature overnight. Water was added, and the product was extracted with DCM. The combined organic layers were washed with saturated NH4Cl, saturated NaHCO3, and brine, dried over Na2SO4, filtered, and concentrated in vacuo to give the crude product. The crude material was purified by flash chromatography on a silica gel column to give the corresponding dipeptide.
The reaction stirred at 0℃ for 1h, then at room temperature overnight. Water was added, and the product was extracted with DCM. The combined organic layers were washed with saturated NH4Cl, saturated NaHCO3, and brine, dried over Na2SO4, filtered, and concentrated in vacuo to give the crude product. The crude material was purified by flash chromatography on a silica gel column to give the corresponding dipeptide.
General peptide coupling procedure D. The protected amino acid (1 eq) was dissolved in DMF (0.3 M), and the solution was cooled to 0C. HATU (1.2 eq) was added, followed by the amine (1 eq), then Et3N (10 eq). The reaction stirred at room temperature overnight. Water was added, and the product was extracted with DCM. The combined organic layers were washed with saturated NH4Cl, saturated NaHCO3, and brine, dried over Na2SO4, filtered, and concentrated in vacuo to give the crude product. The crude material was purified by flash chromatography on a silica gel column to give the corresponding dipeptide.

methanobenzo[d][1,3,2]dioxaborol-2-yl)pent-4-en-1-yl)amino)propan-2-yl)carbamate (22d).
The product was synthesized following general coupling procedure C, using Boc-D-Ser as the amino acid and 20 as the amine. The crude product was purified by silica gel (eluent 60:40 EtOAchexanes) to give the product as a white solid (57%     (enough to reach a pink color). The solution was cooled to -78°C, and N2 gas was bubbled into the solution for 5 minutes, followed by ozone (~80% ozone output). When the solution turned dark blue, ozone addition was stopped, and N2 was bubbled until the solution was colorless. Polymer-bound triphenylphospine (1.5 eq, ~3mmol/g loading, CAS 39319-11-4) was added, and the mixture stirred for 5 minutes at -78°C, then at room temperature overnight under argon atmosphere, after which the mixture became slightly opaque. TFA (1.5 eq) was added at room temperature, and the mixture was stirred for 2h. The mixture was filtered through Celite®. The solid was rinsed with DCM, and the filtrate was concentrated in vacuo. The residue was redissolved in EtOAc and washed with saturated NH4Cl and brine. The organic layer was dried over Na2SO4, filtered, and concentrated in vacuo to give the crude product, which was purified by flash chromatography on a silica gel column to give the pure product.
The mixture was filtered through Celite®. The solid was rinsed with DCM, and the filtrate was concentrated in vacuo. The residue was redissolved in EtOAc and washed with saturated NH4Cl and brine. The organic layer was dried over Na2SO4, filtered, and concentrated in vacuo to give the crude product, which was purified by flash chromatography on a silica gel column to give the pure product.