Structure–Activity Relationships and Molecular Modeling of Sphingosine Kinase Inhibitors

The design, synthesis, and evaluation of the potency of new isoform-selective inhibitors of sphingosine kinases 1 and 2 (SK1 and SK2), the enzyme that catalyzes the phosphorylation of d-erythro-sphingosine to produce the key signaling lipid, sphingosine 1-phosphate, are described. Recently, we reported that 1-(4-octylphenethyl)piperidin-4-ol (RB-005) is a selective inhibitor of SK1. Here we report the synthesis of 43 new analogues of RB-005, in which the lipophilic tail, polar headgroup, and linker region were modified to extend the structure–activity relationship profile for this lead compound, which we explain using modeling studies with the recently published crystal structure of SK1. We provide a basis for the key residues targeted by our profiled series and provide further evidence for the ability to discriminate between the two isoforms using pharmacological intervention.


■ INTRODUCTION
The lipid kinase, sphingosine kinase (SK), plays a myriad of roles in regulating cell survival, growth, and migration of mammalian cells through its product, sphingosine 1-phosphate (S1P). S1P is a ligand for five cell-surface G-protein-coupled receptors and for several intracellular targets, such as histone deacetylases 1 and 2 (HDAC1/2, which regulate gene expression). 1 There are two isoforms of sphingosine kinase, SK1 and SK2, that are encoded by different genes and that exhibit distinct biochemical properties, substrate and inhibitor sensitivities, and subcellular distribution. SK1 and SK2 have redundant roles to some extent as knockout of both genes is embyronically lethal in mice, whereas animals survive when either gene is removed alone. 2 Moreover, there is evidence that both SK1 and SK2 play a similar role in certain cancers. 3,4 Surprisingly, SK1 and SK2 may have opposing roles in inflammation, being largely proinflammatory and antiinflammatory, respectively. On the other hand, a proinflammatory role for SK2 in other cell types has also been reported. 3 The differing roles of SK1 and SK2 in inflammatory disease and supporting data from knockout mice make a compelling case for the development of isoform-selective inhibitors in order to elucidate the functions and roles of each isozyme and for designing drugs for therapeutic intervention in pathophysiological processes such as inflammatory diseases and cancer.
Various SK inhibitors have been identified (see Figure 1 for examples). Enzyme kinetic studies show that many of these are competitive with sphingosine (Sph). The initial SK inhibitors were Sph analogues, such as D,L-threo-dihydrosphingosine which has a K i of ∼5 μM. 3 N,N-Dimethylsphingosine also inhibits both SK isoforms. The first SK1-selective inhibitor to be reported was the water-soluble Sph analogue SK1-I (also known as BML-258; K i ∼ 10 μM). 5 The most potent nanomolar SK1-selective inhibitor is PF-543. 6 However, this compound is also a substrate for SK1, thereby compromising data interpretation. Analogues of the immunosuppressive agent FTY720 (Gilenya) have been prepared using a synthetic route starting with 4-octylphenethyl alcohol and were found to be SK1-selective inhibitors (e.g., RB-005). 7 FTY720 is an inhibitor of SK1 8 and also inhibits other sphingolipid-metabolizing enzymes, such as ceramide synthase and S1P lyase. 9, 10 We also recently showed that SK1 contains an allosteric site. 11 Replacement of the amino group in (S)-FTY720vinylphosphonate with an azido group changes this compound from an allosteric inhibitor to an activator of SK1. 12 Therefore, allosteric inhibitors of SK1 are also an exciting option for future study. With regard to SK2-selective inhibitors, ABC294640, 13 (R)-FTY720 methyl ether (ROMe), 14 K145 (3-(2-aminoethyl)-5- [3-(4-butoxylphenyl)propylidene]thiazolidine-2,4dione), 15 and SLR080811 ((S)-2-[3-(4-octylphenyl)-1,2,4oxadiazol-5-yl]pyrrolidine-1-carboximidamide) 16 have K i values in the range of ∼1−10 μM. Taken together, it is clear that there is still a need to develop new SK1 and SK2 inhibitors, both to increase the number of selective tools that can be used to interrogate the biology of these enzymes and to increase the possibility of having useful new therapeutic agents to treat disease.
In a recent study, we showed that RB-005 is a highly selective SK1 inhibitor. 7 Herein, we describe the synthesis of analogues of the known SK1 inhibitors RB-005, FTY720, 8 SKi, 17 compounds 36a 18,19 and 82, 20 BML-258, 5 CB5468139, 21 and SLR080811 16 ( Figure 2). These new analogues, which were designed to possess some degree of structural similarity to the known inhibitors, were evaluated through enzyme activity studies as     Chemical Synthesis. Linker Length. The synthetic routes we employed to prepare compounds with one-, three-, and fourcarbon tethers are displayed in Scheme 1. The sole carbon atom of the hydroxymethyl group of 4-iodobenzyl alcohol provided the tether in RB-023, whereas the three carbons of propargyl alcohol were the source of the tether in RB-024. In each of these compounds, a Sonogashira reaction followed by catalytic hydrogenation of the alkyne intermediate (1 and 3) and a S N 2 reaction of a mesylate intermediate derived from 2 and 4 with 4hydroxypiperidine gave the desired compounds. We used 4-(4octylphenyl)butan-1-ol (5) 22 as the starting material for the preparation of RB-025.
Modifications of the 4-Alkylphenyl and the Piperidyl Groups. Scheme 2 shows the synthetic pathways employed to prepare RB-026−RB-033. After a Sonogashira reaction was used to install an alkynyl group with 6−12 carbons, catalytic hydrogenation of 6 and 7 afforded alkyl derivatives 8 and 9, and S N 2 displacement as in Scheme 1 gave the desired compounds in good yield. To assess the role of the hydroxyl group in RB-005, RB-026, and RB-028 in inhibition of SK, we replaced this group with an azido group via mesylation of the alcohol and reaction with sodium azide in DMF to obtain 10, RB-029, and RB-030. Reduction of the azide afforded the amino derivatives, RB-031, RB-032, and RB-033.
Fluorination of RB-005 with diethylaminosulfur trifluoride (DAST) gave RB-034 in 90% yield (Scheme 3). The 4-keto derivative RB-035 was synthesized by oxidation of the 4hydroxyl group of RB-005 with pyridinium chlorochromate. Reaction of mesylate 11 7 with 4-methoxypiperidine provided RB-036. Similarly, reaction of commercially available chiral Benzamide Derivatives. Benzamide-containing analogues RB-044−RB-050 were prepared from 4-iodobenzoic acid as outlined in Scheme 4. Alkyne intermediate 12 was reduced to carboxylate 13, which was converted to the acyl chloride with thionyl chloride and then treated with the desired cyclic amine in the presence of potassium carbonate.
Quaternary Ammonium Derivatives. To synthesize RB-052, which contains a pyridinium ion in the polar headgroup, we used the halogens in 1-bromo-4-iodobenzene to carry out two separate Sonogashira reactions, as shown in Scheme 5. First, we used 1-octyne to prepare alkyne 14, which reacted with 3ethynylpyridine to give dialkyne 15. Reduction of 15 yielded RB-051, and N-alkylation with methyl iodide (5 equiv) in acetonitrile afforded the N-methylpyridinium salt RB-052.
Effects on SK1 and SK2 Activity and SAR. We previously demonstrated the importance of the 4-hydroxypiperidinyl group in the selective inhibition of SK1, 7 which was subsequently confirmed by Gustin et al., who generated chiral piperidyl analogues bearing hydroxyl and hydroxymethyl groups. 20 To  . Effect of inhibitors on SK1 or SK2 activity. SK1 activity was measured using 3 μM Sph and 250 μM ATP. SK2 activity was assayed using 10 μM Sph and 250 μM ATP (n = 3 for each compound; results are expressed as % of control ± SD). RB series compounds were used at 50 μM. BML-258 (50 μM) inhibited SK1 activity by 74.5 ± 3.3% (n = 3). The control is 100% and equals activity against Sph alone. examine SAR among a panel of related compounds, we prepared a series of analogues bearing a 4-hydroxypiperidinyl group but varied the linker length between the aryl group and the piperidine. We also assessed the role of the alkyl substituent in the aryl group. To evaluate compound selectivity against SK1 or SK2, the assays were performed (see Supporting Information) using Sph at concentrations of 3 and 10 μM (the K m values of SK1 and SK2, respectively), which corresponds to 50% substrate saturation and enables a qualitative estimation of selectivity by comparing the % inhibition of each kinase using a fixed concentration of inhibitor. We consider this approach to be an appropriate comparison of selectivity, since both enzymes exhibit 50% occupancy with the substrate. Compounds that were found to be effective inhibitors were then analyzed in more detail by performing dose−response curves.
We have previously shown that RB-005 (the "parent compound") is a selective inhibitor of SK1 and exhibits an IC 50 = 3.6 ± 0.38 μM at 3 μM Sph (which corresponds to the K m of SK1) and reduces SK1 activity by ∼90% at 50 μM RB-005. 7 The effect of linker length on potency was assessed by comparing the % inhibition of SK1 and SK2 obtained with RB-023 (which has a one-carbon tether), RB-024 (three-carbon tether), and RB-025 (four-carbon tether). The linker length did not significantly alter the ability of RB-023, RB-024, and RB-025 to inhibit SK1 activity ( Figure 3). RB-023−RB-025 also retained selectivity for SK1 over SK2.
The aliphatic chain at the para position of the benzene ring of FTY720 is C 8 H 17 , which is known to be optimal for the action of FTY720 on its targets such as S1P receptors. 23 Knott et al. 24 reported that the ability of quaternary ammonium salts with a phenyl-substituted cyclohexylamine scaffold to inhibit SK2 was affected by the alkyl chain length. To examine the role of the alkyl substituent on the benzene ring of RB-005, and thus the lipophilicity of the molecule, we compared the inhibitory activity of RB-026 (which has a methyl group as the alkyl substituent), RB-027 (which has a n-hexyl group), RB-005 (which has a noctyl group), and RB-028 (which has a n-dodecyl group). SK1 inhibition was decreased by more than 6-fold in RB-026 compared with RB-023. The almost complete lack of inhibition displayed by RB-026 against SK1 indicates that a larger alkyl group than a methyl group is required for inhibitory activity. We also evaluated the effect of alkyl chain length in the lipophilic tail of compounds in which the 4-hydroxypiperidinyl group was replaced by a 4-aminopiperidinyl group (see below). Changing the n-octyl group of RB-032 to a methyl or n-dodecyl group gave RB-031 and RB-033, respectively, and eliminated the inhibitory activity toward SK1. These results confirm the critical requirement for the n-octyl group.
Next, we probed the role of the 4-hydroxyl group of RB-005 by replacing it with an azido, amino, fluoro, keto, or methoxy group (RB-029−RB-036). Azido replacement (RB-029, RB-030) reduced SK1 inhibition markedly, while replacement of the 4hydroxyl group with an amino group (RB-032) diminished the potency of SK1 inhibition ( Figure 4). The isoform selectivity of SK1 over SK2 was retained for RB-032, suggesting that the amino group replacement maintains efficient binding to SK1. Replacement of the 4-hydroxyl group of RB-005 with a fluoro (RB-034) or methoxy group (RB-036) eliminated inhibitory activity against SK1, while replacement with a keto group to produce RB-035 increased inhibition of SK2 and maintained inhibition of SK1 but eliminated the isoform selectivity.
To examine the role of the piperidyl group in inhibition of SK, we replaced it with a pyrrolidine ring; the hydroxyl-containing substituent was retained (as either a chiral hydroxyl or a chiral hydroxymethyl group), but its orientation was varied, as shown in compounds RB-037−RB-043. RB-037 and RB-038 retained inhibitory activity against SK1 despite having opposite configurations at C-3 of the pyrrolidin-3-ol group. Stereoisomers RB-040 and RB-042, which differ in the length of the aliphatic chain (C 8 H 17 vs C 12 H 5 ) but possess the R configuration at C-2 of the 2-hydroxymethylpyrrolidinyl group, were equipotent inhibitors of SK1 and SK2 ( Figure 3 and Figure 5). The corresponding S enantiomers RB-041 and RB-043 were much less active ( Figure 3). To establish whether RB-041 and RB-043 were capable of inhibiting SK1 and SK2 activity in a concentrationdependent manner, we used a higher concentration of each (100 μM, compared to the 50 μM concentration data shown in Figure  3), and found that the inhibition of SK1 and SK2 with RB-041 was 72.2 ± 5.9% and 45.7 ± 2.6%, respectively, whereas with RB-043 the inhibition of SK1 and SK2 was 49.9 ± 6.2% and 49.7 ± 7%, respectively. These findings indicate that RB-041 and RB-043 can inhibit SK1 and SK2 but that the sensitivity of inhibition compared with RB-040 and RB-042 is considerably reduced. Interestingly, the S enantiomers RB-041 and RB-043 are substrates for SK2 (see Supporting Information, Figure S1).
To further examine the influence of the length of the alkyl substituent on the benzene ring on SK activity, we assessed the extent of SK inhibition afforded by pyrrolidine derivatives RB-039, RB-042, and RB-043. The ability of the compound to inhibit SK1 is abolished in RB-039 and RB-043, which have a methyl and a n-dodecyl group in the lipophilic tail, respectively.
An amidine or proline headgroup incorporated into a benzamide-containing scaffold was shown to provide potent SK1 inhibitors, 18,19 as in compound 36a ( Figure 1). When we replaced the methylene linker between the aryl group and the heterocycle with a keto group to produce the benzamide analogues RB-044−RB-050, inhibition of SK1 was effectively abolished (Figure 3), as were the pyridine derivatives (RB-048 and RB-051). The report that quaternary ammonium salts are selective SK2 inhibitors 25 prompted us to prepare RB-052, RB-053, RB-060, RB-061, and RB-062, which were ineffective as SK inhibitors, although RB-053 demonstrated a moderate selectivity for SK2 ( Figure 3). We also prepared aliphatic quaternary ammonium salts (not shown) that were inactive. SK inhibitors containing a central thiazole group have been reported (e.g., SKi and compound 82, Figure 1). 1,2,3-Triazoles are mimics of thiazoles and are easily prepared by Cu(I)catalyzed azide/alkyne click chemistry. In our series of triazole The results are expressed as % of control ± SD; n = 3. The control is 100% and equals activity against Sph alone. RB-032 inhibits SK1 activity with IC 50 = 16.9 ± 1.6 μM. RB-005 inhibits SK1 activity with IC 50 = 3.6 ± 0.36 μM. 7  analogues of RB-005 (RB-054−RB-065), we found that RB-065 was a highly selective SK1 inhibitor, whereas the other 10 triazole analogues, all of which lack the 4-hydroxypiperidinyl group, were inactive ( Figure 3).
Modeling the Inhibitors in the Atomic Structure of SK1. The crystal structures of human SK1 in a complex with ADP and SKi were determined recently. 26 We demonstrate here that the chemical modifications of the highly selective SK1 inhibitor RB-005 produced SAR that can be explained using this crystal structure. Figure 6A displays the result of a modeling analysis of RB-005 in the active site of human SK1; the piperidyl hydroxyl group is hydrogen-bonded to D81, and the protonated amine of the headgroup forms a salt bridge with the carboxylate of D178. Fluoro (RB-034) or methoxy (RB-036) containing compounds do not exhibit inhibitory activity against SK1 ( Figure  3), as these groups no longer have hydrogen bond donating capacity, suggesting that the interaction of the 4-hydroxypiperidinyl group of RB-005 is with a hydrogen bond acceptor in the protein. The lack of SK1 inhibitor activity of the azide-containing compounds (RB-029, RB-030) supports this possibility. One of the two oxygens of the carboxylate ion of D81 is hydrogenbonded to the backbone NH of L116, and the other is hydrogenbonded to the backbone NH of A115; these interactions prevent the carboxylate of D81 from being catalytic and shift the catalytic role to D178 (Figure 7). The latter carboxylate oxygen also forms a hydrogen bond to the hydroxyl group of the inhibitor. If these compounds formed hydrogen bonds with the side chain of S168, then RB-029, RB-034, and RB-036 could also bind to the donor/acceptor hydroxyl group of S168 or water and therefore act as inhibitors. Since RB-029, RB-034, and RB-036 cannot form hydrogen bonds with D81 and are not inhibitors, we propose that the key interaction of the hydroxyl group of RB-005 ( Figure 6A), RB-025 ( Figure 6B), and RB-028 ( Figure 6C) is with D81 and not with S168. In contrast, modeling of RB-035 (which contains a 4-keto group instead of a 4-hydroxyl group, yet maintains inhibition of SK1) suggests that the carbonyl group can form a hydrogen bond with the hydroxyl group of S168 and water (see Supporting Information, Figure S2).
The inhibitory effect of RB-032 ( Figure 6D) and its absence for RB-033 ( Figure 6E) can be explained by protonation of the primary amine. When the protonated primary amine rather than the piperidyl group forms a salt bridge with D178, the inhibitors are pushed deeper into the J channel, which was identified by Wang et al. 26 as the region that accommodates the alkyl chain of Sph. Since RB-032 has a shorter alkyl chain than RB-033, it can be accommodated in the substrate pocket, whereas the dodecyl group of RB-033 cannot fit into the channel formed when the protonated primary amine forms a salt bridge with D178. Therefore, RB-033, which is inactive, does not bind to D81, D178, S168, or L268, indicating that these are key amino acid residues required for binding SK inhibitors.
The salt bridge between the NH 3 + group of RB-032 and D178 ( Figure 6D) is the only polar interaction with the protein, which might explain the lower SK1 inhibitor activity when compared with RB-005; the latter forms a salt bridge and hydrogen-bonds with D81. Interestingly, the R-enantiomers RB-040 and RB-042 are equipotent inhibitors of SK1 and SK2, while the Senantiomers RB-041 and RB-043 are weak substrates for SK2, implying that the spatial orientation of the hydroxyl group in RB-041 and RB-043 required for catalysis is different in SK2 compared with SK1. The protonated amine in RB-040 ( Figure  6F) and RB-042 ( Figure 6G) can form a salt bridge with D178 and can also form a hydrogen bond with the carbonyl oxygen of L268. Both inhibitors can orientate the hydroxymethyl group of the pyrrolidine (R enantiomer) to also form a hydrogen bond with the side chain of D81. The protonated amino group of RB-041 and RB-043 can form a salt bridge with D178 but, because of the orientation of the hydroxymethyl group of the pyrrolidine (S enantiomer), cannot form a hydrogen bond between their hydroxyl group and D81, as found in our modeling study. Instead, the hydroxymethyl group could form a hydrogen bond to D178. As the experimental evidence shows that RB-041 and RB-043 do not inhibit SK1, this suggests that dynamic factors (accessing the binding site), which are not taken into account by docking studies, prevent the binding of these compounds.
RB-044−RB-050 are ineffective inhibitors of SK1. There are three possible explanations: first, the nitrogen in an amide cannot be protonated, thus preventing salt bridge formation. Second, the link between nitrogen and phenyl is constrained and planar compared with a methylene group, which prevents optimization of the hydrogen bonding network with the hydroxyl group. Third, the carbonyl group of the amide would be proximal to the side chain of D178, which would result in electrostatic repulsion.
The pyridinium salts RB-052 and RB-053 and the quaternary ammonium salts RB-060, RB-061, and RB-062 were also ineffective SK1 inhibitors. The absence of a hydroxyl group in these compounds rules out hydrogen bonding with D81 or D178. The triazole moiety in RB-065 forms a hydrogen bond with T196 ( Figure 6H) and, furthermore, adds a kink in the chain that helps orientate the alkyl group into the J channel, which may account for its SK1 inhibitory activity.
Our modeling studies suggest that RB-005 ( Figure 6A), RB-025 ( Figure 6B), and RB-028 ( Figure 6C) interact with D81 and not with S168. These findings are consistent with D81 not acting as a base, because RB-005, RB-025, or RB-028 are not substrates for SK1. As depicted in Figure 7, modeling of Sph into the catalytic site of SK1 suggests that S168 can form hydrogen bonds with the NH 3 + group of Sph and, via a water molecule, with the secondary hydroxyl group of Sph. The water molecule also hydrogen-bonds with A339, thereby linking this amino acid residue to the secondary hydroxyl group of Sph (Figure 7). Thus, D178 functions as the deprotonating base in this model to enable nucleophilic attack by Sph on the γ-phosphate group of ATP, with subsequent transfer of this phosphate to Sph.

■ CONCLUSION
In this study we have identified a series of SK1-selective inhibitors and have used molecular modeling to define their interactions with the catalytic site of the enzyme. These studies reveal a substantial flexibility in the catalytic site in terms of binding SK1 inhibitors. For instance, RB-005 is proposed to interact with D81 and D178, while RB-025 appears to interact with D81 and L268. The findings obtained from the modeling study fully account for the SAR of the inhibitors and explain why some of these compounds are inactive. These findings also reveal the architecture of the SK1 catalytic site and suggest a major role for D178 as the deprotonating base that facilitates phosphorylation of Sph by ATP. In summary, the novel information presented here should enable development of new SK1 inhibitors with improved potency and selectivity. Similarly, resolution of the atomic structure of SK2 (yet to be achieved) along with information provided herein will enable better insights into the molecular basis of the selectivity of these inhibitors for SK1 over SK2.
■ EXPERIMENTAL SECTION Docking Studies. The crystal structure of SK1 in complex with Sph (PDB entry 3VZB) was used for docking studies. Chain A of the complex was kept along with a single water molecule found to be tightly bound to the complex which hydrogen-bonds to the side chain hydroxyl group of S168, the backbone −NH of G342, and the secondary hydroxyl group of Sph (water number 680). Hydrogen atoms were added to the protein and water using Accelrys Discovery Studio 3.1 (Accelrys Software, San Diego, CA), and all of the inhibitors presented in this study were docked using GOLD 5.1 for Windows (Cambridge Crystallographic Data Centre, Cambridge, U.K.). Default software settings were used, keeping ChemPLP as a scoring function after redocking Sph in place in the 3VZB crystal structure as well as the SKi inhibitor in the 3VZD PDB entry (rmsd of 1.8 and 0.2 Å, respectively) as validation.
Synthesis. General Methods. All chemicals were reagent grade and used as purchased. Reactions were run under nitrogen and were monitored by TLC using silica gel 60 F 254 aluminum-backed plates. Flash column chromatography was performed on silica gel grade 60 (230−400 mesh). THF was distilled over sodium/benzophenone immediately prior to use. Dichloromethane was distilled over CaH 2 , and Et 3 N was distilled over KOH pellets. All other solvents were of anhydrous quality and were used as received. 1 H NMR and 13 C NMR spectra were recorded on a Bruker Avance I spectrometer, and chemical shifts are reported in δ units relative to deuterated solvents, which served as internal references, at 400 and 100 MHz, respectively. Highresolution mass spectra were recorded at the CUNY Mass Spectrometry Facility on an Agilent Technologies G6520A Q-TOF mass spectrometer using electrospray ionization (ESI). Microwave reactions were performed in a Biotage Emrys Creator synthesizer. HPLC was carried out using a reverse-phase column with a gradient of acetonitrile/water from 50/50 to 90/10, with detection at 214 and 254 nm. Elemental analysis was performed at Columbia Analytical Services, Tucson, AZ. All compounds were ≥95% pure as determined by examining their HRMS and 1 H NMR spectra.