Synthesis and Biological Activity of Conformationally Restricted Indole-based Inhibitors of Neurotropic Alphavirus Replication: Generation of a Three-Dimensional Pharmacophore

Abstract

We have previously reported the development of indole-based CNS-active antivirals for the treatment of neurotropic alphavirus infection, but further optimization is impeded by a lack of knowledge of the molecular target and binding site. Herein we describe the design, synthesis, and evaluation of a series of conformationally restricted analogues with the dual objectives of improving potency/selectivity and identifying the most bioactive conformation. Although this campaign was only modestly successful at improving potency, the sharply defined SAR of the rigid analogs enabled the definition of a three-dimensional pharmacophore, which we believe will be of value in further analog design and virtual screening for alternative antiviral leads.

Graphical Abstract

Neurotropic alphaviruses are NIAID Category B Priority pathogens that infect the central nervous system (CNS) and have the potential to cause severe and debilitating disease in humans and animals.¹ Among these, the mosquito-borne equine encephalitis viruses (Eastern, Western, and Venezuelan) are endemic to the Americas and constitute an emergent public health risk due to the potential for human mortality² and neurological damage,^3–5 in addition to disruption of agriculture.^6–7 Despite this risk, prophylactic therapies are inadequate; current vaccines lack wide availability and exhibit limited efficacy in humans,^8–9 and recent vaccine candidates have demonstrated protection in animal models only.^10–13 Options for treating infection with small molecule drugs are similarly minimal and are unable to compensate for vaccine deficiencies. For example, the efficacy of the current standard of treatment, ribavirin, is constrained by poor CNS exposure.^14–16 Remarkably, equine encephalitis viruses have received scant historical attention to address these shortcomings.¹⁷ Only a handful of reports have been published in recent years detailing the discovery of drug-like small molecules^18–20 and natural products,^21–22 so there remains an urgent need to identify and develop effective therapies that can intercede with infection.

We have previously reported the discovery²³ and optimization^24–25 of indole-based alphavirus RNA replication inhibitors, e.g. 1 and 2 (Figure 1), and, more recently, anthranilamide- based inhibitors possessing improved pharmacokinetic properties.²⁶ Unfortunately, the molecular target remains elusive − a consequence of the phenotypic nature of our primary antiviral assays − and thus the architecture of the binding site that could guide design of more potent inhibitors is unknown. An approach that is often effective in such a scenario is the synthesis and evaluation of rigid analogues.^27–30 Of the multiple conformations that a drug can adopt, only a small subset are expected to have maximal activity if the binding site is well defined.³¹ It can therefore be informative to restrict lead molecules into specific conformations in an effort to pinpoint the most active ones.³² In addition to the obvious benefit of potentially defining a 3-dimensional (3D) pharmacophore, this strategy can also improve both potency and selectivity versus off-target effects.^33–35

Figure 1.

Indole-based alphavirus replication inhibitor leads.

In the present work, we thus pursued two central objectives: (1) to conformationally restrict rotatable bonds in our leads 1 and 2, and (2) to develop a 3D pharmacophore model based on the resulting SAR that could expedite future drug design. Toward this end, we describe three different approaches to conformational restriction: 1) replacement of small functionality with sterically demanding functionality; 2) bridging of nonadjacent atoms; and 3) freezing of ring conformations. We also strove to minimize the number of additional atoms used in locking conformations because higher molecular weight could impede CNS permeability.

Analogs were synthesized according to one of two general procedures (Scheme 1). In all cases where chiral centers were present, the analogs were prepared and tested as racemates. General Procedure A was used to incorporate methyl groups onto the indole and to modify the benzyl group of lead 1. N-(4-Chlorobenzyl)indole-2-carboxylic acids 30a-c³⁶ were condensed with ethyl isonipecotate or ethyl nipecotate and the respective esters were subsequently hydrolyzed to afford carboxylic acid intermediates 31a−c. These acids were condensed with a variety of amines (see Supporting Information) to produce analogs 3 − 16 (Tables 1 and 2).

Scheme 1.

General Procedure A reagents and conditions: (a) ethyl isonipecotate or ethyl nipecotate, EDC, HOBT, TEA, DCM, r.t.; (b) NaOH, H2O, THF, r.t.; (c) R³R⁴NH, EDC, HOBT, TEA, DCM, r.t.; General Procedure B reagents and conditions: (d) various reactions − see Scheme 2 and SI; (e) 30a, EDC, TEA, HOBT, DCM, r.t. PG = protecting group; G = reactive group.

Table 1.

Modification of the indole of 1.

No	R1	R2	R3	IC50 (μM)a	CC50 (μM)b
125	H	H		15.6 ± 1.8	>100
3	CH3	H		61 ± 39	>100
4	H	CH3		>100	>100

^aInhibition of luciferase expression in WEEV replicon assay.

^bCell viability determined by the MTT reduction assay. Values are mean of at least n=3 independent experiments.

Table 2.

Modification of the benzyl amide of 1.

No	R	IC50 (μM)a	CC50 (μM)b
125		15.6 ±1.8	>100
5c		30.6 ± 7.3	>100
6		>100	>100
7		>100	>100
8		>100	>100
9		>100	>100
10		>100	>100
11		>100	>100
12		0.53 ±0.04	>100
13		1.7 ± 0.1	>100
14		7.1 ± 0.9	>100
1525		15.2 ± 4.7	62 ± 13
16		>100	>50

^aInhibition of luciferase expression in WEEV replicon assay.

^bCell viability determined by the MTT reduction assay. Values are mean of at least n=3 independent experiments.

^cBased on nipecotic acid central ring (racemate).²⁵

Alternatively, General Procedure B was employed to explore changes to the piperidine and amide moieties of lead 2 (Scheme 1). Piperidines, piperazines, and bicyclic amines were functionalized through a variety of reactions to afford intermediates with the general structure 34 (Scheme 2 and Supp. Info). Following deprotection, these intermediates were condensed with N- (4-chlorobenzyl)indole-2-carboxylic acid 30a to afford analogs 17– 29 (Table 3).

Scheme 2.

Reagents and conditions: (a) EDC, HOBT, Et₃N, DCM, r.t.; (b) HCl, dioxane, r.t.; (c) COCl₂, Et₃N, toluene, DCM, 0 °C; (d) Na⁰, naphthalene, THF, r.t.; (e) PhLi, t-BuOH, LiBr, THF, −78 °C-r.t.; (f) n-BuLi, THF, r.t.

Table 3.

Modification of the piperidine amide of 2.

No	R	IC50 (μM)a	CC50 (μM)b
224		0.53 ± 0.08	65.0
17		>50	>50
18		>50	>50
19		17.6c	66.7c
20		24.7 ± 3.5	77.3
21		24.4 ± 4.0	78.0
22		3.9 ± 0.4	74.2
23		>50	>50
2424		1.6 ± 0.8	47.9
25		24.4 ± 6.3	61.0
26		24.6 ± 1.6	>100
27		0.96 ± 0.06	50.9
28		10.9 ± 0.6	56.5
29		0.49 ± 0.09	41.1

^aInhibition of luciferase expression in WEEV replicon assay.

^bCell viability determined by the MTT reduction assay. Values are mean of at least n=3 independent experiments.

^cValue is n=1 experiment.

The majority of intermediates of general structure 34 were amides and as such were prepared by condensation of an amine with a carboxylic acid followed by acid-promoted Boc-deprotection (e.g. 37 and 42, Scheme 2; for others, see SI). Urea 40 was prepared by condensation of two different amines with phosgene and was likewise deprotected. N-Tosyl-protected spirocyclic amino acid 43^37–39 was similarly condensed with 4-(2-aminoethyl)-pyridine 39 but proved challenging to deprotect. Samarium (II) iodide either failed to remove the tosyl under neutral conditions or afforded reductive cleavage of the amide in the presence of tertiary amines,^40–41 and other methods^42–43 were similarly unsuccessful. Only sodium naphthalenide cleanly cleaved the S-N bond to afford amine 44.

Distinct from the other intermediates, alkenes 47 and 48 were synthesized via a traditional Wittig reaction and the Schlosser modification thereof,⁴⁴ respectively, from N-boc-4-formylpiperidine 45 and phosphonium bromide 46⁴⁵ (Scheme 2). Although a large number of highly E- and Z-selective olefinations exist,⁴⁶ this divergent approach allowed for convenient and rapid access to both alkenes from the same reactants.All new analogues were evaluated for antiviral activity in a phenotypic cell-based assay against the WEEV genome replicon as previously described.²⁵ Analogues were also tested for cytotoxicity under similar conditions using a colorimetric MTT assay. The data appear in Tables 1–4.

As noted earlier, definition of a 3D pharmacophore was our primary goal through the evaluation of conformationally-restricted analogs. Our initial efforts focused on the restriction of the indole substituents of the lead 1²⁵ (Table 1). Utilizing a non-covalent steric approach, two compounds were designed on the notion that strategically placed methyl groups would impede bond rotations of adjacent sidechains with minimal impact on molecular weight. Both the 7- methylindole 3 and the 3-methylindole 4 possessed substantially reduced potency, indicating that the methyl groups may be preventing rotatable access of the N-benzyl group and the 2-carboxamide to active conformations.

We next turned to restricting the flexibility of the N-benzyl amide on the isonipecotic acid frame of 1 (Table 2). Previously prepared analogs²⁵ 5 and 15 are included in the table for comparison. Remarkably, the majority of new analogs (6 − 11) exhibited no activity whatsoever, suggesting that the binding pocket for the N-benzyl amide is well defined and intolerant of many rigid conformations. One compound (12), however, proved to possess markedly better potency than all others in this series, which we do not believe can be attributed to simple increased lipophilicity or chain extension because analogs of similar ClogP (14) and length (15²⁵ and 16) were less potent (Table 2). Interestingly, the racemic methyl analog 13 was also quite active; however its lower potency relative to 12 is further evidence that the amide binding site is well defined and not tolerant of even modest additional bulk or altered conformation. Importantly, the increased potency of 12 has apparently been achieved without a commensurate increase in cytotoxicity vs 1. Finally, the dramatic loss of potency of analogs 7 and 8, each a single methylene shorter than the corresponding active analogs 12 and 14, respectively, suggests that some degree of flexibility is required to engage the binding site. Overall, the results in Table 2 are consistent with an amide binding pocket in the unknown molecular target that is very well defined and discriminating.

Our attention then turned towards the central piperidine ring and the piperidine-4-carboxamide. Compound 2²⁴ was employed as the lead for this series due to its equipotency with 12 and greater ease of analog synthesis. Biological results are summarized in Table 3. The carboxamide imparts significant conformational bias to the linker between the piperidine and the pyridine, so we initially explored replacing it with E or Z olefins 17 and 18, respectively, which are conformationally locked bioisosteres for the two possible rotamers of the carboxamide.⁴⁷ Remarkably, both analogs lost all activity, indicating that the caboxamide itself is likely making at least one key binding interaction. To explore this further, inverse amide 19 and shifted amide 20 were prepared, each proving to be over 30-fold less active than 2. This supported that a carboxamide bound directly to the piperidine 4-carbon through the carbonyl is a critical part of the pharmacophore. We therefore retained this motif through the remainder of our investigation.

Urea analog 21 and unsaturated amide 22 were examined next, each of which retained all of the atoms of the carboxamide and its position relative to the piperidine, but would be expected to have some minor conformational differences. The simple unsaturated amide retained more activity than the urea, but still lost nearly 10-fold potency. Methyl analogs 23 and 24 were particularly informative. While the previously prepared N-methyl analog 24²⁴ only lost 2-fold potency, suggesting that the N-H is not functioning as a hydrogen bond donor, the new α-methyl analog 23 lost all activity, indicating that the carbonyl likely plays a much more significant role. The α-methyl group is either occluding the amide carbonyl from binding (sterically or conformationally) or is impeding favorable chair conformations of the piperidine. Finally, since we had shown with 24 that the N-H was not critical for activity, we prepared the highly rigid bicyclic analog 25. This analog lost most potency, obviously indicating that it is not locked in the optimal conformation. Collectively, the results for 17 − 25 highlight the importance of a sterically unhindered and flexible carboxamide to the pharmacophore.

For the next three analogs 26 − 28, we retained the “flexible” carboxamide moiety in the linker, and focused on freezing the conformational mobility of the piperidine ring. While the exo-bicycle 26 and the spirocycle 28 lost over 20-fold potency, the tricyclic analog 27 retained almost all of the potency of lead 2, suggesting that the locked equatorial-equatorial relationship between the 1- and 4-substituents may reflect the unknown bioactive conformation, Unfortunately, no activity improvement was realized by this conformational restraint.

Conversely, we examined an azapane analog 29 that we expected would be more flexible than piperidine 2 but about the same overall length. Interestingly, this compound was equally potent with the lead despite being a racemate. Depending on the eudismic ratio (not determined), the active enantiomer could potentially be two-fold more active, marking this compound as the most active compound to-date in the WEEV replicon assay. The simplest explanation for this positive result is that a larger population of azapane conformers are available that are favorable to binding than are available to the piperidine.

The replicon activity of azapane 29 was confirmed in an antiviral assay in FMV-infected BE(2)-C cells. At 5 μM, 29 demonstrated a nearly two-fold improvement in infected cell viability (36.1 ± 4.0 % versus negative control 19.6 ± 4.0 %), similar to the two-fold enhancement observed for lead 2 at 25 μM. At higher concentration improvement was more modest (22.9 ± 4.9 % cell viability at 25 μM), possibly due to concentration-dependent cytotoxicity.

The primary objective of this work was to establish a 3D pharmacophore model that could facilitate future optimization and enable virtual screening to identify alternative templates. We first noted that a number of small changes in the structure resulted in large differences in biological activity, suggesting that an “activity cliff” analysis⁴⁸ could be useful for identifying key pharmacophoric elements. Employing DataWarrior⁴⁹, we identified pairs of compounds exhibiting both high 2D structural similarity and divergent potency and ranked them according to structure activity landscape index (SALI)⁵⁰ scores (Table S1 in Supporting Information). Based on the pairs having the greatest SALI values, there appear to be three pharmacophoric features most sensitive to structural changes: the distance between the terminal aryl group and the piperidine (e.g. 16 vs 15, 12 vs 6, 12 vs 14), the flexibility of the linker to the terminal aryl group (7 vs 12, 8 vs 14) and the piperidine-4-carboxamide carbonyl on the piperidine (2 vs 23, 2 vs 17/18, 2 vs 20, 2 vs 19).

In order to generate a hypothetical 3D pharmacophore, we used the Pharmacophore Elucidation tool in MOE 2019.0102.⁵¹ The approach for Pharmacophore Elucidation is to exhaustively search for all pharmacophore queries (arrangements of molecular features such as aromatic rinds, hydrophobic groups, hydrogen bond acceptors/donors, etc.) that induce good overlay of most of the active molecules and separate actives from inactives. It was assumed that all or most of the active compounds bind to the same biological target in a similar way and can therefore be overlaid.

“Active” compounds were defined as those with IC₅₀ < 5 μM, affording a set of 7 actives and 22 inactives. These sets were used to generate a collection of pharmacophore queries such that all or most of the active compounds satisfy the queries. The generated queries were then scored by the accuracy of separating actives and inactives. The overall accuracy = m/N where N is the number of compounds in the input database and m is the sum of the number of actives that match the query and the number of inactives that do not match the query. Queries were also scored by degree of overlap of the actives with the pharmacophoric features. The overlap is between 0 and n where n is the number of actives. The larger the overlap the better the alignment. The overlap is calculated by the heavy atom root mean square deviation among the aligned compounds.

The program generated a database of 8850 queries from our analogue set. The output database contained the conformations of the high scoring alignment induced by the query, features of each query, the number of actives that match the query, the score of the alignment, the accuracy of the query in separating actives and inactives, the accuracies of the query in matching the actives/inactives, etc. The queries were sorted by the accuracies, and the top queries were examined for the alignment of actives (see Table S2 in Supporting Information for the top 10 queries.) One query (H H H a a) was selected as the best on the aggregate basis of accuracy, overlap, and agreement with the activity cliff analysis. The selected query has 5 features: 3 hydrophobic groups and 2 hydrogen bond acceptors (defined as the postulated locations of potential hydrogen bond donors).

The overall accuracy of this query is 0.93 out of 1.0; all the actives except for one (27) match this query, all of the inactives except for one do not match this query, and the overlap of the actives is 5.22 out of 6.0. The alignment of the six most active compounds overlaid on top of the pharmacophoric query is shown in Figure 2. This query contains the terminal hydrophobic group, which is consistent with the activity cliff analysis above. It also contains two hydrogen bond acceptors, one of which is the piperidine-4-carboxamide which was also shown to be important in the activity cliff analysis. A detailed description of how the pharmacophore elucidation was performed is provided in the Supporting Information.

Figure 2.

The overlay of the six active analogs with the pharmacophoric query. Green represents location of hydrophobic groups and cyan indicates hydrogen bond acceptors (represented by optimal location of potential hydrogen bond donors in binding site).

In conclusion, we have explored a variety of conformationally restricted analogs of our WEEV antiviral leads 1 and 2, with several objectives in mind: 1) improving antiviral potency; 2) increasing selectivity vs off-targets that could reduce cytotoxicity; and 3) beginning to define a 3D pharmacophore for the unknown molecular target to facilitate future design, including virtual screening to potentially identify new lead chemotypes. Both covalent and non-covalent approaches to restricting rotational freedom were examined. As expected, many rigid analogs completely lost activity, suggesting that the binding site is well defined and highly discriminating based on molecular shape. Significant improvements in antiviral potency were realized with terminal amide analogs of 1 (e.g. 12) without increasing cytotoxicity, suggesting some improvement in selectivity for the unknown viral target. Modification of the carboxamide and piperidine of more potent lead 2 were less successful at improving potency but helped to identify important pharmacophoric elements. The piperidine ring had little tolerance for conformational rigidity, with only the locked 1,4-diequatorial analog 27 retaining most of the potency of lead 2. Conversely, increasing flexibility, as in azapane analog 29, actually improved potency slightly. We applied computational modeling to identify the most important SAR (activity cliffs) and to subsequently define a common pharmacophore which rationalized most of the SAR. Key elements of this pharmacophore include an optimum linker length between the indole and the terminal aromatic group of the amide N-substituent, as well as the requirement for a carboxamide with precise placement on the linker and an unhindered carbonyl. We believe this pharmacophore could be useful for designing new compounds and may enable virtual screening to identify novel viral replication inhibitors.