Scaffold-Hopping Strategy on a Series of Proteasome Inhibitors Led to a Preclinical Candidate for the Treatment of Visceral Leishmaniasis

There is an urgent need for new treatments for visceral leishmaniasis (VL), a parasitic infection which impacts heavily large areas of East Africa, Asia, and South America. We previously reported on the discovery of GSK3494245/DDD01305143 (1) as a preclinical candidate for VL and, herein, we report on the medicinal chemistry program that led to its identification. A hit from a phenotypic screen was optimized to give a compound with in vivo efficacy, which was hampered by poor solubility and genotoxicity. The work on the original scaffold failed to lead to developable compounds, so an extensive scaffold-hopping exercise involving medicinal chemistry design, in silico profiling, and subsequent synthesis was utilized, leading to the preclinical candidate. The compound was shown to act via proteasome inhibition, and we report on the modeling of different scaffolds into a cryo-EM structure and the impact this has on our understanding of the series’ structure–activity relationships.


■ INTRODUCTION
Bites from infected sandflies are known to transmit Leishmania parasites, a protozoan parasite. There are more than 20 different species of Leishmania, which cause a group of diseases that manifest in three main clinical forms: cutaneous, mucocutaneous, and visceral leishmaniasis (VL, also referred to as kala-azar). 1−3 There is also the complication of VL, called post kala-azar dermal leishmaniasis. While the other three forms cause major health problems such as nonhealing ulcers and scarring, VL is the most severe and becomes lethal if not treated. Recurrent epidemics of VL in East Africa (Ethiopia, Kenya, South Sudan, and Sudan) have caused high morbidity and mortality in those countries and together with Brazil, India, and Somalia accounted for more than 90% of the new cases reported in 2018. 4 VL is usually due to infection with either Leishmania donovani or Leishmania infantum. The infection causes irregular bouts of fever, substantial weight loss, swelling of the spleen and liver, and anemia. Current available treatments are not adequate due to high toxicity, resistance issues, treatment cost and duration, or the need for hospitalization to administer the treatment, a particularly relevant aspect given the settings where the disease is endemic. Indeed, the only effective oral treatment for VL is miltefosine, although it requires 28 days of dosing, and contraception is necessary for women both during and after treatment due to its teratogenicity. For reasons that are poorly understood, clinical outcomes for these drugs seem to vary from region to region of the world, with much reduced cure rates being observed in East Africa and South America. Given these factors, the need for safer, oral, short-course, and low-cost medicines for the treatment of VL is urgent.
There is some cause for optimism, as for the first time, there are a number of compounds in late preclinical or early clinical development, as indicated on the Drugs for Neglected Diseases Initiative (DNDi) website. 5 We recently reported a leishmanial CRK-12 inhibitor as a preclinical candidate; 6,7 Novartis (LXE408, compound 2) 8, 9 and ourselves (GSK3494245/ DDD01305143, compound 1) 10 have both reported proteasome inhibitors ( Figure 1); DNDi have other compounds in clinical development, notably an oxaborole and a nitroimidazole, although the mechanism of action (MoA) of these has not been reported. 5 In all these cases, the optimization was run phenotypically, with the MoA only being elucidated during or after the medicinal chemistry program. It must be noted 10 that there is a lack of validated drug targets for Leishmania.
As we reported recently, compound 1 was the product of a program which started from substituted 2-phenyloxazolo[4,5b]pyridine 3 which was identified through phenotypic screening against the related pathogen Trypanosoma cruzi, the causative agent of Chagas' disease ( Figure 1). Subsequently, compound 3 was screened against the physiologically relevant intramacrophage form of L. donovani, where it showed weak activity. In parallel to our efforts, N-(4-chloro-3-(oxazolo[4,5-b]pyridin-2-yl)phenyl)furan-2-carboxamide (the 4-chloro phenyl analogue of 3) had also been identified by the groups of Buckner, Gelb, Baell, and Guy as a compound with antikinetoplastid activity, with their optimization being subsequently reported. 8,11−15 As previously described, an initial scaffold hop from 3 led to a 2-phenylimidazo [1,2-a]pyrimidine and optimization of the 6position of the bi-cycle led to a 6-morpholino substituent which increased the potency and metabolic stability. Alongside this, the fluorination of the phenyl ring increased potency and replacement of the furanyl amide with a pyrrolidinyl urea improved solubility (similar changes had also been observed on related scaffolds by other groups 13,15 ). This led to compound 4, which showed efficacy in a mouse model of VL. 10 The key to the success of this compound series was a further set of scaffold hops from 2-phenylimidazo[1,2a]pyrimidine, and this strategy, along with the discussion of the structure−activity relationships (SARs) around the 2amino group, is described in detail herein. Although the series was developed phenotypically, it was subsequently found to act principally through the inhibition of the β5-subunit of the proteasome, specifically at the "chymotrypsin" site, and the relationship between the antiparasitic SAR and the identified target is also discussed in detail. 10

■ RESULTS
Compounds were initially screened against L. donovani in an intracellular assay (L. donovani cultured in differentiated THP1 cells, referred to as the "INMAC assay" 16 ), where we targeted pEC 50 values >5.8, alongside stability in mouse liver microsomes of <5.0 mL/min/g (Cl i ) and kinetic aqueous solubility >100 μM (measured by chemiluminescent nitrogen detection, CLND). We had successfully used these criteria in the hit to lead stage to identify compounds for in vivo proof-of-concept studies in another program, which subsequently led to the identification of preclinical candidate DDD853651/ GSK3186899 active against CRK12. 7 As reported previously 10 and shown in Table 1, in vitro profiling of 4 shows that it met these targets, with an INMAC pEC 50 value of 6.2, Cl i of <0.5 mL/min/g in mouse liver microsomes, and aqueous solubility of 150 μM, alongside high mouse and human plasma stability (100% compound remaining after 3 h). A pharmacokinetic (PK) study was thus performed, dosed at 10 mg/kg po and 3 mg/kg iv. An oral dose of 10 mg/kg showed 4 to have blood levels above EC 90 for around 2 h (pEC 90 5.7, 780 ng/mL), which would translate to coverage above 6 h for a 50 mg/kg dose, assuming a linear dose increase (see Supporting Information). With the limited knowledge of the PK/PD drivers for the treatment of VL, we considered this to be a suitable profile for the progression of 4 into a previously reported mouse efficacy model of VL, 6,7,17,18 where it was dosed twice daily at 50 mg/kg ip in order to further maximize exposure and give the best chance of achieving proof-ofconcept.
After dosing twice daily for 5 days, a reduction in parasitemia of 98% was observed (Table 1). This established the proof-of-concept for this series in VL and, in fact, met our criteria for a preclinical development candidate. 7 Unfortunately, despite reasonable kinetic aqueous solubility, 4 was poorly soluble in FaSSIF (Fasted State Simulated Intestinal Fluid) 19,20 media and this, alongside a positive result in the Ames assay, 21−23 as reported previously, 10 which indicated Journal of Medicinal Chemistry pubs.acs.org/jmc Article potential genotoxicity issues, precluded further development of this particular compound.
Having achieved the in vivo proof-of-concept, the series moved into the lead-optimization phase, with an initial aim to identify a compound with a suitable profile for rodent toxicology studies. As this would require us to achieve an exposure significantly higher than the minimum efficacious dose, we continued to target compounds with pEC 50 values >6 and now aimed for FaSSIF solubilities >100 μM and to maintain Cl i < 1 mg/mL/g. We carried out a systematic exploration of the molecule, initially preparing further analogues of 4 with variations to the pendant substituents of the 2-phenylimidazo[1,2-a]pyrimidine core. However, this failed to identify any better molecules (data not shown).
We next focused our attention on the core phenylsubstituted bi-cycle. Initially, we investigated changes to the phenyl ring and, as shown in Table 2, modifications to this part of the scaffold led to analogues that maintained stability in mouse liver microsomes and also showed improved solubility in both aqueous (CLND) and FaSSIF media. Unfortunately, these changes, such as moving or removing fluorine (5 and 6, respectively), introducing alternative substituents (e.g., OMe analogues 7 and 8), or replacing phenyl with six-membered heterocycles (e.g., pyridine 9, pyrimidines 10 and 12, pyrazine 11, and pyridazine 13), all led to a >10-fold loss of potency. We noted that similar findings were also reported by Tatipaka et al. 13 and Ferrins et al. 14 on the related scaffolds. Analogue 4, with the fluorine ortho to the bi-cyclic substituent, appeared to be optimal and our subsequent knowledge of the binding mode of these compounds allowed us to rationalize this (seen later). We were interested to note that the changes made did have a big impact on FaSSIF solubility, with methoxy analogue 8, pyrazine 11, pyridazine 13, and pyrimidines 10 and 12 all having FaSSIF solubilities well above 100 μM. Also, 10−13, all had intrinsic clearances of <0.5 mL/min/g. Encouraged by this, we turned our attention to the core bi-cycle, surmising that changes to the pattern of hydrogen bond acceptors (HBAs) could lead to similar improvements in solubility.
In order to define a set of alternative bi-cycles for synthesis, we first looked at the SAR from our early hit discovery efforts, on scaffolds with no 6-substituent (Table 3). On comparing with initial hit 3, we noted that a scaffold which removed the HBA in the 8-position (e.g., 14) showed no activity in the INMAC assay. We therefore focused our scaffold-hopping efforts on bi-cycles, which contained a HBA in the 8-position, and then prioritized them based on cChromLogD (Chrom-LogD 24 is the chromatographically determined log D and cChromLogD is the calculated version of this), cpK a (calculated using ACD pK a predictor), calculated aqueous solubility (using Gastroplus v9.7), and synthetic tractability (based on literature precedent and in-house knowledge). Without a robust model for activity in the Ames test, this was not included in the design process, the intention being to profile compounds in the assay when they demonstrated suitable efficacy in the VL mouse model to warrant further study.
A total of 28 alternative scaffolds were evaluated in silico, all maintaining a HBA in the 8-position (for comparison purposes, the numbering for the imidazo[1,2-a]pyrimidine scaffold in Figure 2 is used in this section), plus the correct orientation of the morpholinyl and phenyl substituents. We explored HBDs and HBAs in various positions around the bicycle and also examined the effect of reversing the 6,5-fused system so that morpholine was attached to the five-membered ring and phenyl to the six-membered ring. Finally, we also profiled a fused 5,5 bi-cycle.
As we reported for a previous series, we had focused on reducing lipophilicity as a key strategy for improving solubility. 7 In that case, we targeted a PFI 25 <7 (PFI = property forecast index, ChromLogD + #Ar). Within the current series which contained three aromatic rings, this would equate to a cChromLogD below 4, so we utilized this as a key property for the designed scaffolds. We also targeted scaffolds which had a higher predicted solubility than 4 and scaffolds which increased cpK a . The increased cpK a would indicate potential for improved solubility, as well as allowing possible salt formation. We then further prioritized based on the potential synthetic tractability (based on literature precedent and in-house synthetic expertise), as with each scaffold requiring a bespoke synthesis, there would be a limited chemistry resource to develop completely novel routes.
The initial scaffolds all examined the nature of the atom in the 1-position of the bi-cycle (S-1 to S-16) (Table 4). Scaffolds S-1−S-8 all examined the removal of the HBA, where the best predicted profiles were obtained for S-1, S-4, S-5, and S-6, all predicted to have cChromLogD below 4 and either higher cpK a , improved solubility, or both when compared to 3 (Table  4). Scaffolds S-9 and S-10 would demonstrate whether a hydrogen bond donor (HBD) was acceptable in position 1. Although S-9 gave a more direct comparison to 3, S-10 proved to be more synthetically tractable. On this basis, S-1 and S-10 were synthesized with the desired substituents, S-5 was synthesized as the non-fluorinated analogue, and S-9 as the phenyl analogue. Although S-4, S-6, and S-11 had good in silico profiles, the syntheses of S-4 and S-6 were not completed when it became apparent that the removal of HBA at the 1position would lead to an inactive compound, and for S-11, there would be a possible tautomerization of the fivemembered ring, making any data generated difficult to interpret so this scaffold was not synthesized (see Figure 3). S-12 would further probe the nature of the heteroatom in the 1position and had a very good in silico profile. Scaffolds S-13−S-16 all examined the replacement of the nitrogen in the 1position with oxygen. All these compounds showed some advantage over 3 and so were targeted for synthesis, although S-15 was synthesized without the fluoro substituent and S-16 was not synthesized due to lack of a tractable route.
Scaffolds S-17−S-21 all retained the HBA nitrogen in positions 1 and 8 and examined the effect of varying the heteroatoms in the other positions around the rings. All of them showed some potential advantage over 3 and so were selected for synthesis (although a tractable synthesis of S-21 was not successfully developed).
Scaffolds S-22−S-27 all examined the reversal of the 6,5fused scaffold to a 5,6-fused scaffold. S-22 had a cChromLogD below 4 and higher predicted pK a compared to 3 and so was selected for synthesis. While S-23−S-25 all had higher predicted solubility, they proved to be synthetically challenging    and we only developed a tractable synthetic route to S-24 with phenyl, rather than the morpholine substituent. S-26, which replaced the nitrogen in the five-membered ring with oxygen, had a poor in silico profile and so was not synthesized. cChromLogD calculated using GSK predictive software. Colored as green <4 and amber >4. cpK a calculated using ACD labs pK a calculator for the most basic nitrogen on the bi-cycle. Colored as green if >compound 4 (4.4) and amber if <compound 4. cAq. Sol (pH 7.4) mg/mL is GastroPlus calculated aqueous solubility at pH 7.4, colored green if >compound 4 and amber if ≤compound 4. Synthesis color scheme: green = desired compound synthesized, orange = alternative analogue synthesized (e.g., no F or phenyl instead of morpholinyl), red = not synthesized (no successful route or abandoned due to SAR learnings). Although S-27 had predicted improved solubility, it was not synthesized due to our understanding that the N-methyl group would not be tolerated. Finally, we profiled some 5,5-bi-cycles, as exemplified by S-28. The only advantage this might confer would be a cChromLogD below 4, as the predicted solubility was lower than 3, and the core did not contain the basic nitrogen. Because of this, and also because the vectors of the substituents would be very different to 3, as well as a lack of literature precedent for synthesis, we elected not to synthesize this analogue.
This led to a set of 10 scaffolds synthesized with the desired substituents, plus a further 4 which were synthesized with alternative substituents (Table 5). From this, we gained some key SAR learnings. Scaffolds S-1 (compound 15) and S-5 (compound 16) showed that removing the HBA nitrogen from position 1 led to inactive compounds in the INMAC assay.   compounds with a "reversed" bi-cycle (S-22 and S-24, compounds 1 and 27, respectively) proved to be very interesting. Compound 27 was only synthesized as the phenyl analogue; although this showed similar potency to 3, it had very poor FaSSIF solubility. On the other hand, 1 was only slightly less potent than 3 and had the highest FaSSIF solubility of all the scaffold hop compounds. Based on these findings, compounds 23 (S-17) and 1 (S-22) were investigated further (23 was selected over 24 as it proved much more straightforward to scale-up the synthesis to get suitable quantities for in vivo studies). The progression of 23 was subsequently stopped as the aniline resulting from the hydrolysis of the urea was positive in the Ames assay, 10 indicating a potential genotoxicity issue. This led us to focus on 1 as the most promising scaffold for further development; although it was slightly less potent than 23, it was the compound with the best solubility profile and had high stability in mouse liver microsomes.
To summarize the SAR, a HBD was essential in positions 8 (N) and 1 (N or O), whereas a HBD in the 7-position was detrimental to activity ( Figure 4). A 6-substituent, particularly morpholine, was essential for activity and metabolic stability. 10 In the other positions around the ring (e.g., 3, 4, 5, 7, and 9) heteroatoms were tolerated but not essential to activity. Regarding the central phenyl ring, 4-fluoro boosted activity and 1-urea gave a better balance of potency, solubility, and metabolic stability than the corresponding amides (data reported previously). 10 Finally, it proved possible to reverse the central bi-cycle, a change which had a small detrimental effect on potency that was compensated for by the improvement in FaSSIF solubility.
A key aim of the scaffold-hopping exercise was to identify compounds with improved solubility, in both aqueous and FaSSIF media. Of the 10 scaffolds synthesized with the same substituents as 4, only three showed an improvement in solubility in both media (24,25, and 1), with 23 showing an improvement only in FaSSIF solubility (S-18, S-19, S-22, and S-17, respectively). Compounds 23, 24, and 25 all had better in silico predictions for solubility than for compound 4, whereas 1 was predicted to be similar. However, compound 1 had much better solubility than compound 4, and of these four compounds, it had the highest FaSSIF solubility. Interestingly, scaffold S-22 was predicted to have a much higher pKa than S-17 and S-18, indicating that for compound 1 a significantly higher proportion is likely to be protonated at physiological pH. Therefore, while the in silico profiling helped to triage the number of scaffolds for synthesis, it was clear that a large number of scaffolds needed to be synthesized in order to identify the optimal scaffolds for progression.
Because reversing the imidazo[1,2-a]pyrimidine scaffold (compounds 1 and 27) was liable to orient the pendant amine and urea substituents slightly differently, we further explored the SAR around the 3-position of the bi-cycle, where morpholine was substituted ( Table 6). The replacement of morpholine of 4 with phenyl (28) gave an expected increase in potency alongside a decrease in solubility, while switching to pyridyl (29) maintained the improved potency alongside an increase in FaSSIF solubility compared to 28. However, the Cl i was above our targeted level of <1 mL/min/g for a preclinical candidate. We were interested in exploring saturated substituents; isopropyl 30 showed a similar potency to morpholine but with poorer FaSSIF solubility, and isobutyl 31 showed a small improvement in potency but with poorer metabolic stability. In order to improve solubility, we explored the introduction of a basic center with methylene-linked morpholine 32 and directly linked morpholine 33. While this was successful in delivering compounds with improved FaSSIF solubility compared to 1, it unfortunately led to a loss of potency.
We also further explored N-linked amines, where piperidine 34 showed good potency but with poor FaSSIF solubility. Piperazines 35−37 were inactive, although 36 and 37 did show enhanced FaSSIF solubility. As we had seen success with methylmorpholines in other series, 6,7 we synthesized rac-2methyl morpholine 38 and the enantiomeric 3-methylmorpholines 39 and 40. All of these compounds showed similar potency to 1 and improved FaSSIF solubility, although all the three were less metabolically stable. Likewise, dimethylmorpholines 41, 42, and 43 were synthesized, all showing similar potency to 1 but with poorer metabolic stability, although 42 and 43 did show improved FaSSIF solubility. Spiro analogue 44 again was less metabolically stable than 1, and secondary amines 45 and 46 showed a drop in potency compared to 1.
From all the in vitro profiling, none of the changes to the morpholine ring led to compounds with an improved overall profile; therefore, 1 was identified as the most suitable compound for further in vitro and in vivo studies. As reported previously, 10 a PK study of 1, dosed at 25 mg/kg po, supported its progression into the infected mouse model, where it showed a candidate level efficacy against VL at this dose. When dosed orally twice a day for 10 days, 1 had an ED 90 of 16 mg/kg and ED 99 of 30 mg/kg. The compound had suitable PK for oral delivery and a good predicted safety margin of at least 37-fold when dosed in rats at 300 mg/kg. Critically, unlike compounds 4 and 23, both 1 and the aniline derivative potentially released by hydrolysis of its urea were negative in the Ames assay, as reported previously. 10 Alongside the previously reported safety profiling, 1 was also screened against a representative set of three kinases (LCK, PI3Kγ, and AuroraB) and was inactive against them all (pIC 50 <4.5). The compound is now being advanced for the first time in human studies for VL.
Modeling. In parallel to these phenotypic optimization studies, attempts to identify the target of these series were carried out as previously reported. 10 It was found that these compounds act through the inhibition of the chymotrypsin-like activity catalyzed by the β5 subunit of the L. donovani proteasome, demonstrating good selectivity over the human enzyme. A high-resolution cryo-EM structure of compound 1 bound to the Leishmania tarentolae proteasome 10 revealed a binding site that lies between the β4 and β5 subunits (Figure Table 6. Effect of Modifications or Replacements to the Morpholine Group on Activity and Physicochemical Properties a INMAC is the intramacrophage assay carried out in THP-1 cells with L. donovani amastigotes. Data are the result of at least three independent replicates and standard deviations are ≤0.4. b Aq. solubility is the kinetic solubility measured by CLND. c FaSSIF solubility is the fasted state simulated intestinal fluid solubility. d Cl i is the mouse liver microsomal intrinsic clearance. e Data reported previously. 10 f ND means not determined. 5) and exploits an induced cavity that is lined on one side by β4 residues that are divergent between humans and kinetoplastid protozoan. The cryo-EM structure of a related compound in the complex with the Leishmania tarentolae proteasome has also been recently published by Novartis. 9 Description of the Binding Mode of Compound 1. A homology model of compound 1 bound to the L. donovani 20S proteasome β4 and β5 subunits was generated using the L. tarentolae 20S proteasome cryo-EM structure as a template. Compound 1 binds at the interface between the β4 and β5 subunits close to the catalytic Thr100 residue ( Figure 6A), which is the first residue of the β5 subunit. The pyrrolidine carboxamide sits in the most buried part of the binding site, a mainly hydrophobic cavity important for selectivity against the human orthologue. The cavity is defined by Ile27, Ile29, and Phe24 from the β4 subunit and by Phe225, Val227, and Thr235 from the β5 subunit. The urea carbonyl oxygen is hydrogen bonded to the Gly228 backbone nitrogen, and the urea nitrogen is hydrogen bonded to the hydroxyl of the Tyr212 side chain. The phenyl ring is placed on top of Gly197 of the terminal part of the β-strand 7 of the β5 subunit with the fluorine substituent facing the Ser195 side chain hydroxyl. The "top edge" (see Figure 2) of the imidazopyrimidine bi-cyclic system is mainly solvent exposed with the six-membered ring sitting on top of Gly146 and the sp 2 nitrogen of the fivemembered ring interacting with the donor NH group of Ser229. The nitrogen on the "bottom edge" of the pyrimidine ring does not make any specific interactions; however, it contributes to the charge delocalization, resulting in a favorable charge interaction with Thr100 (see later). The morpholine group is largely solvent exposed and is directed toward the β hairpin motif critical for the recognition of bortezomib (β strands 3 and 4 of the subunit β5), establishing a hydrogen bond with the Gly122 backbone NH ( Figure 6A). (Bortezomib is a clinically used inhibitor of the human proteasome.) Computational Analysis of Binding Interactions. The fragment molecular orbital method (FMO-MP2) 27 was used to calculate the interaction energy between compound 1 and the proteasome and to gain a deeper understanding of the molecular recognition event. Being a quantum-mechanical (QM)-based method, FMO-MP2 is capable of detecting and accounting for nonclassical interactions that are poorly parameterized in molecular mechanics-based force fields (e.g., CH−π, halogen−π, cation−π interactions, and nonclassical hydrogen bonds) resulting in a more accurate binding energy. To overcome the high computational costs associated with the QM calculation, the system is fragmented into smaller parts and QM calculations are performed on each individual fragment pair to derive pairwise interaction energy (PIE) between each of the fragments and the ligand. By combining the PIE of the ligand and the fragments, it is possible to derive the total interaction energy of the ligands with the target. The decomposition analysis of the PIE (PIEDA) 28 allows the derivation of four different energy terms (electrostatic, charge transfer, dispersion, and exchange-repulsion) that provide a deeper residue-by-residue insight into the nature of the ligand interaction with the target. The electrostatic and charge transfer terms are dominant in H-bonds, polar (favorable and unfavorable) interactions, and salt bridges. The dispersion term is more prominent in hydrophobic and van der Waals interactions. The exchange-repulsion term describes the stericrepulsion between electrons of different atoms accounting for steric clashes.
The FMO-PIEDA analysis shows that, out of the 40 residues in a 5 Å range from the ligand, there are 13 contributing significantly (PIE < −3 kcal/mol) to the overall calculated binding energy ( Figure 6B,C). The initial work focused on analyzing the H-bonding and hydrophobic interactions revealed by the cryo-EM structure; the contribution to the binding energy from the residues directly involved in hydrogen bonding with the ligand was confirmed. The Gly228 hydrogen bond to the urea carbonyl is the strongest H-bonding interaction with a calculated electrostatic energy term of −15.9 kcal/mol (PIE −14.3 kcal/mol).
In the cryo-EM structure of the L. tarentolae structure, a CH 2 group next to morpholine oxygen is adjacent to NH of Gly 122, which is a part of a semi-flexible loop. However, in the model of the L. donovani enzyme, the morpholine twists slightly, allowing N−H of Gly 122 to establish a strong hydrogen bond with the morpholino oxygen atom (PIE −9.0 kcal/mol, electrostatic term −9.0 kcal/mol). For Tyr212, the PIE energy is −8.7 kcal/mol which is the result of both the hydrogen bonding (electrostatic) with the urea NH and the favorable hydrophobic interaction with the pyrrolidine ring (dispersion). The FMO-PIEDA result also shows that the Ser229 interaction (PIE −6.26 kcal/mol) is mainly due to the favorable interaction between the β carbon atom of the side chain (dispersive interaction) with the heterocyclic scaffold of the ligand rather than the hydrogen bonding of the nitrogen in the 5-membered ring of imidazopyrimidine with the backbone NH. This is consistent with the poor geometry observed for the hydrogen bond present in both the L. tarentolae cryo-EM structure and the L. donovani model. The favorable contribution of hydrophobic contacts is also significant for Gly197 (−4.2 kcal/mol) and Gly146 (−6.8 kcal/mol) as both residues are in contact with the central phenyl ring. The contact between the pyrrolidine carboxamide and the Phe24 side chain has a contribution of −9.2 kcal/mol. Importantly, in addition to the interactions that could be identified from a visual inspection of the cryo-EM structure, the FMO-PIE analysis highlighted other residues that establish nonintuitive interactions and are important in the molecular recognition event. Thr100 (PIE −30.3 kcal/mol), Asp214, and  (Figures 2 and 7). A similar effect is demonstrated by the presence of the fluorine atom on the phenyl ring, which increases potency by approximately 10-fold (see compounds 4 and 6). As previously discussed, 10 the ESP of the ligand shows that the presence of the fluorine atom on the phenyl ring causes an accumulation of the positive charge on positions 3 and 4 of the fluorophenyl ring giving rise to another strong dipole−charge interaction with the negatively charged Asp214 and Asp215. This effect of fluorine appears to be more important to binding than its interaction with Ser195. Val227 also provides a significant contribution to binding energy (PIE −12.7 kcal/mol) with the decomposition analysis indicating an equal contribution from the dispersive term (related to the hydrophobic side chain) and electrostatic term (related to the positively charged patch localized on the pyrrolidine ringsee Figure 7). Therefore, FMO-PIE analysis showed that the binding of compound 1 to the proteasome is governed by a complex mixture of specific hydrogen bonds, stacking interactions, optimal electrostatic complementarity, and longrange electrostatic interactions between the protein and the ligand.
Computational Analysis of Structure Activity Relationships. To analyze the effect of the imidazopyrimidine Figure 7. Molecular ESP maps for some of the studied analogues. Protein surfaces were generated using the APBS plugin for PyMol and colored by calculated charges (red −5 kbT/ec, blue +5 kbT/ec). The ligand surfaces were generated using Jaguar in Schrodinger and colored by the electrostatic potential (red −80 kcal/mol, blue +75 kcal/mol). scaffold replacement in the series related to compound 1, all the synthesized scaffold hop compounds (Table 5) were modeled in the L. donovani 20S proteasome model and studied by FMO-PIEDA and ESP. The compounds tended to be either active (pEC 50 5.8−7.3) or inactive (pEC 50 < 4.3). Derivatives with a fluorine substituent on the phenyl ring ortho-to the bicycle were consistently more active than the hydrogen analogues. We and others have noted a correlation of activity between the inhibition of proteasome and activity against axenic Leishmania parasites. 8,10 While there could be subtleties in INMAC activity due to physicochemical properties of compounds, this data allows an interpretation of the SAR.
The modeling data generated confirmed the important role that the electrostatic complementarity between the ligands and the proteasome binding site plays in the optimization of the interaction. The ESP maps (Figure 7 and Supporting Information), clearly show that even if the hydrogen acceptor interacting with the Ser229 backbone nitrogen is conserved, where there is a lack of the negative charge accumulation on the bottom edge (positions 1 and 8) of the imidazopyrimidine scaffold (as observed for compounds 15, 18, and 19), these compounds are inactive (Table 5). This is further supported by the FMO-PIEDA analysis results (Figure 8), indicating that one of the strongest interactions that compound 1 establishes is with Thr100, an interaction primarily electrostatic in nature, which contributes with −30.3 kcal/mol. Compound 27 (Table  5) is also characterized by a strong interaction between Thr100 (FMO-PIE −30.3 kcal/mol) and the negatively charged bottom face (positions 1 and 8) of the imidazotriazine scaffold. The scaffold of compound 27 also seems to enhance the partial positive charge on 3-and 4-positions of the phenyl ring and strengthen the interaction with Asp214 and Asp215 (see the ESP, Supporting Information).
The "reversed" 5−6 bi-cyclic system compounds 1 and 27 not only are characterized by a particularly favorable accumulation of the negative charge on position 1 of the bicyclic scaffold but also by a slightly different placement of the substituent in position 3 in relation to the 6−5 bi-cyclic systems of the other analogues. An inspection of the bound conformations shows that the vector of the substitution in the 3-position is similar between the 5−6 and 6−5 bi-cycles ( Figure 9), but in the 5−6 system, the substituent is about half a bond longer than in the 6−5 system. This results in the morpholine ring being pushed closer to Gly122 on the β strands 3 and 4 of the subunit β5 and might explain the lower potency of the compounds with a "reversed" scaffold. Docking studies suggest that in order to avoid a steric clash with Gly122 and the beta-hairpin involved in bortezomib binding, compound 24 appears to adopt a higher energy conformation with the morpholine ring almost perpendicular to the bi-cyclic scaffold. Due to this steric requirement, and together with an unfavorable electrostatic distribution on the bi-cyclic scaffold, resulting in the lack of a partial negative charge interacting with Thr100 (Supporting Information), compounds 16 and 17, where morpholine is replaced by a phenyl ring, failed to dock and are reported as inactive.
All the different cores required the bespoke synthesis, as detailed in Schemes 2− 9 below (the synthesis of compounds 1, 3, 4, and 23 was described elsewhere 10 ).
Compounds 15 and 27 are prepared, as outlined in Scheme 2. Compound 53 could be brominated with trimethyl(phenyl)ammonium tribromide to yield either monobromo 54 or dibromo 55. The thermal cyclization of 54 with pyridazine 56 led to 15, whereas 55 could be cyclized with 1-aminoguanidine to give 57 which was further cyclized with 2-bromo-1,1diethoxyethane to 58. Bromination to give 59 was followed by Suzuki coupling with benzeneboronic acid to yield 27.
Compound 24 (Scheme 7) was prepared starting from 3amino-5-morpholinopyridazine which was aminated on the 2position to give diamino compound 79. Alongside this, ethyl 5amino-2-fluorobenzoate was treated with pyrrolidine-1-carbonyl chloride, with subsequent ester hydrolysis giving 80, which was converted to acid chloride 81 by treatment with sulfonyl chloride. Compounds 79 and 81 were then cyclized to give 24.
Compound 25 was synthesized according to Scheme 8, whereby 2-chloro-5-bromopyrimidine was treated with hydrazine to give 82, which was condensed with 2-fluoro-5nitrobenzaldehyde to give 83 and then cyclized to 84. Thermal rearrangement led to 85, with nitroreduction giving 86. This was cross-coupled with morpholine and then treated with pyrrolidine-1-carbonyl chloride to give 25.
To explore the SAR around the 3-position of the 6phenylimidazo[1,2-a]pyrimidine scaffold, two approaches were taken; either an initial cyclization, followed by functionalization of the 3-position of the bi-cycle or cyclization of a precursor with the 3-substituent in place. The synthesis of 28 was previously described, 29 and a similar route was used to synthesize the pyridyl analogue 29, as shown in Scheme 9, where the previously reported 2-aminopyrimidine 87 was cyclized with 2-bromo-1,1-dimethoxyethane to give 88, which could be brominated to 89. 10 The Suzuki coupling of 89 with 2-pyridineboronic acid then led to 29. Alternatively, to access alkyl substitutions, 87 could be cyclized directly with a suitable α-bromoaldehyde to give 30 and 31.
Alternatively, to synthesize C-linked morpholine analogues 32 and 33, the routes shown in Scheme 10 were utilized. Intermediate 90 could undergo a Mannich reaction to give 91, with nitro reduction and subsequent urea formation giving 32. A Vilsmeier−Haack reaction on 90 led to formylated 92. Its treatment with a SnAP reagent 30 led to the carbon-linked morpholine substituent 93, which could be Boc-protected to give 94, the nitro group reduced, and converted to pyrrolidinyl urea 95, with Boc deprotection and subsequent methylation giving 33.
The 3-amino analogues were synthesized according to Schemes 11 and 12. Where applicable, the previously published route for compound 1 was utilized (Scheme 11), whereby the relevant amine was condensed with glyoxal and benzotriazole to give 1,2-bis electrophiles 96a-i, which were subsequently cyclized with 87 under Lewis acid-promoted conditions to give compounds 1, 34−38, and 41−43. In cases where this was unsuccessful, due to the relevant 96 not forming, the alternative route 31 in Scheme 12 was utilized, whereby 2-chloropyrimidine analogue 97 was treated with relevant glycinamide 98a−e to give 99a−e which was cyclized to 100a−e. Cbz-Deprotection, followed by generation of the urea thus led to compounds 39, 40, and 44−46.
■ CONCLUSIONS 2-Phenylimidazo[1,2-a]pyrimidine 3 was a suitable starting point for a phenotypic lead-optimization program for VL. Changes to the central phenyl ring demonstrated that it was possible to significantly improve FaSSIF solubility but failed to yield compounds with a suitable balance of potency, metabolic stability, and FaSSIF solubility; we, therefore, embarked on a scaffold-hopping exercise. A round of design and synthesis led to a set of 14 compounds with different cores, giving us an understanding of the requirements for potency, where HBA's at positions 1 and 8 of the bi-cycle were critical for activity. While the scaffolds, which fulfilled these requirements, notably 20−26 and the "reversed" scaffolds 1 and 27, all met our target potency in the INMAC assay, the solubilities, in both aqueous and FaSSIF media, were highly variable in ways that were difficult to predict.
From this, compounds 1 and 23 were identified as the most promising for progression, although the development of 23 was halted due to a genotoxicity issue. Attempts were made to optimize the morpholine substituent in compound 1 with a variety of different analogues and replacements. While some of these showed greater potency, this came at the cost of solubility and/or microsomal stability. Compound 1 was selected for further profiling, showing that neither the parent, nor the aniline potentially released by the hydrolysis of the urea were positive in the Ames assay. After safety profiling, 1 was selected as a preclinical candidate for VL, as previously reported. 10 Modeling studies were carried out using the cryoEM L. tarentolae proteasome structure that we had previously reported. This was used to generate a model of the L. donovani proteasome. The structure was not available until after the chemistry program had finished. By performing a pair interaction energy decomposition analysis (PIEDA) using the fragment molecular orbital method (FMO), we were able to predict which are the important interactions between the protein and ligand on a residue by residue basis. Important interactions were with Thr100, Gly122, Asp214, Asp215, Val227, and Gly228. The presence of a negative charge on the The relative importance of the protein−ligand interactions would not have been identified unless such a sophisticated analysis had been carried out. We also analyzed the ESP of the ligands. When combining this with the FMO analysis, we were able to rationalize the SAR that we have seen. We suggest that this approach could be used going forward in predicting whether modifications to compounds are likely to improve binding. This project demonstrates the utility of Cryo-EM costructures for rationalizing protein−ligand interactions. In this case, we were able to obtain structures of sufficient resolution to be able to understand in a detailed manner the protein− ligand interactions. As well as being able to rationalize the SAR, as we have previously reported, we were able to use the structural information to rationalize the selectivity of our compounds for the parasite proteasome compared to the human proteasome, as reported previously. 10 The proteasome represents an exciting new drug target for VL and our lead compound, 1, is being progressed toward human studies.

■ EXPERIMENTAL SECTION
Chemistry. Chemicals and solvents were purchased from the Aldrich Chemical Company, Fluka, ABCR, VWR, Acros, Fluorochem, and Alfa Aesar and were used as received. Air-and moisture-sensitive reactions were carried out under an inert atmosphere of argon in oven-dried glassware. Analytical thin-layer chromatography (TLC) was performed on precoated TLC plates (layer 0.20 mm silica gel 60 with fluorescent indicator UV254, from Merck). Developed plates were air-dried and analyzed under a UV lamp (UV254/365 nm). Flash column chromatography was performed using prepacked silica gel cartridges (230−400 mesh, 40−63 mm, from SiliCycle) using a Teledyne ISCO CombiFlash Companion, or CombiFlash Retrieve. 1 H NMR and 13 C NMR spectra were recorded on a Bruker AVANCE DPX 500 spectrometer ( 1 H at 500.1 MHz, 13 C at 125.8 MHz). Chemical shifts (δ) are expressed in ppm recorded using the residual solvent as the internal reference in all cases. Signal splitting patterns are described as singlet (s), doublet (d), triplet (t), quartet (q), multiplet (m), broad (b), or a combination thereof. Coupling constants (J) are quoted to the nearest 0.1 Hz. LC−MS analyses were performed with either an Agilent HPLC 1100 series connected to a Bruker Daltonics MicrOTOF or an Agilent Technologies 1200 series HPLC connected to an Agilent Technologies 6130 quadrupole LC/ MS, where both instruments were connected to an Agilent diode array detector. The mobile phase was water/acetonitrile + 0.1% HCOOH or water/acetonitrile + 0.1% NH 3 ; linear gradient 80:20 to 5:95 over 3.5 min, and then held for 1.5 min; flow rate 0.5 mL min −1 . All intermediates had a measured purity ≥90% and all assay compounds had a measured purity of ≥95% as determined using this analytical LC−MS system (TIC and UV). High-resolution electrospray measurements were performed on a Bruker Daltonics MicrOTOF mass spectrometer. Microwave-assisted chemistry was performed using a Biotage initiator microwave synthesizer. The synthesis of all intermediates, and spectral data for final compounds, is included in the Supporting Information.