Inhibitory Kappa B Kinase α (IKKα) Inhibitors That Recapitulate Their Selectivity in Cells against Isoform-Related Biomarkers

IKKβ plays a central role in the canonical NF-kB pathway, which has been extensively characterized. The role of IKKα in the noncanonical NF-kB pathway, and indeed in the canonical pathway as a complex with IKKβ, is less well understood. One major reason for this is the absence of chemical tools designed as selective inhibitors for IKKα over IKKβ. Herein, we report for the first time a series of novel, potent, and selective inhibitors of IKKα. We demonstrate effective target engagement and selectivity with IKKα in U2OS cells through inhibition of IKKα-driven p100 phosphorylation in the noncanonical NF-kB pathway without affecting IKKβ-dependent IKappa-Bα loss in the canonical pathway. These compounds represent the first chemical tools that can be used to further characterize the role of IKKα in cellular signaling, to dissect this from IKKβ and to validate it in its own right as a target in inflammatory diseases.


■ INTRODUCTION
Nuclear factor-κB (NF-κB) transcription factors are central coordinators of the innate and adaptive immune response and play key roles in cancer development and progression. 1,2 NF-κBs also have a major role in controlling the ability of both preneoplastic and malignant cells to resist apoptosis and support tumor angiogenesis and invasiveness. 1,2 The signaling pathways that mediate the activation of the different NF-κB complexes are therefore attractive targets for new chemotherapeutic interventions.
The NF-κB pathways, which are regulated by the inhibitory κB kinases (IKKs), are elevated when homeostasis is disrupted. This is represented by an increase in constitutive IKKα/β activity leading to enhanced NF-κB expression and subsequent nuclear localization. The IKKs are upstream regulators of the NF-κBs, which exist as either homo-or heterodimers bound to inhibitory kappa Bs (IκB's). 1,2 The activation of these IKK complexes dictates the phosphorylation, targeted ubiquitination, and proteolytic removal of IκBs in the canonical pathway and the phosphorylation and processing of high molecular weight NF-κB proteins (p100) in the noncanonical pathway. 1,2 This in turn allows NF-κB complexes to translocate to the nucleus and bind specific promoter regions of their targeted genes. Studies 3,4 have indicated that IKKα and IKKβ play key but divergent roles in the regulation of global NF-κB signaling and many aspects of cellular transcription. IKKβ controls the canonical pathway via activation of p65 RelA−p50 heterodimers, 5−7 and its inhibition leads to a reduction in proinflammatory gene expression in several cell types. This is relevant to cancer because several pro-inflammatory species associated with tumor development and progression are encoded by genes regulated through the IKKβ-NF-κB axis. 3,4,6 IKKα has been shown to have a minor role in the canonical pathway 4,6 but is pivotal in the noncanonical pathway, catalyzing the phosphorylation and proteolytic processing of p100 NF-κB2 which in turn liberates distinct NF-κB p52/RelB dimers and initiates transcription of a specific subset of genes. IKKα and IKKβ have specific cellular functions, 3,8,9 and the selective inhibition of one isoform over the other may provide a useful and novel therapeutic strategy in cancer and inflammatory diseases.
Over the past 15 years, many inhibitors of IKKβ have been reported, 10−13 primarily toward developing clinical agents to treat inflammatory conditions such as asthma. However, recent studies suggest there may be significant toxicity and side effects associated with IKKβ inhibition, including the development of inflammatory skin disease and sensitization of colonic epithelium to a range of insults. 14 In addition, IKKβ knockout mice display severe liver dysfunction. 15 Intestinal and liver toxicity have also been an issue in several clinical trials of IKKβ inhibitors which may further limit their clinical applications. Some IKKα inhibitors have been described in the patent literature but with little detail regarding activity and specificity. 16 Asamitsu et al. 17 reported that the natural product, noraristeromycin (NAM), inhibits IκBα phosphorylation and degradation upon TNFα stimulation and prevents p65 phosphorylation through selective IKKα inhibition. However, the pharmacodynamic readouts used were reporters for both IKKβ and IKKα activity in cells and do not focus on specific biomarkers of the IKKα-controlled noncanonical pathway such as p100 phosphorylation and subsequent processing to p52.
Given the growing evidence that IKKα has an important role in a number of cancers, 18−20 selective IKKα inhibitors are required in order to fully understand and validate its role in cancer development and progression, particularly in prostate, 19,21 breast, 22−24 and pancreatic 25−27 cancers. Herein we describe the design, synthesis, and evaluation of a series of 4-substituted 2-amino-5-cyanopyrrolo [2,3-d]pyrimidines as part of our program to develop isoform selective IKKα inhibitors. We present a comparison of the kinase domains of IKKα and IKKβ based on molecular dynamics simulations to explore differences in conformational flexibility that would enable small molecule inhibitors to discriminate between the two isoforms. Finally, we report the first example of IKKα-selective compounds that recapitulate activity in cells against isoform-related pharmacodynamic readouts.
■ RESULTS AND DISCUSSION Strategy for Discriminating between IKKα and IKKβ. To date, no group has been able to successfully crystallize IKKα and report a high resolution structure of sufficient detail to guide structure-based inhibitor design. To explore differences between the two IKK isoforms, we built a homology model of

Journal of Medicinal Chemistry
Article the IKKα kinase domain based on the crystal structure of IKKβ (chain B, residues 1−309, PDB entry 4KIK), 28 keeping the inhibitor (KSA700 in the pdb file) and water molecules found within 6 Å of the protein−inhibitor complex ( Figure 1). Both IKK kinase domains were solvated and then subjected to extended molecular dynamics, with an average structure generated for the last 21 ns (IKKα) or 26 ns (IKKβ) and subsequently minimized.
When superimposing the presimulated structures of both IKK isoforms, it was striking to see that regions making up the ATP-binding pocket were essentially identical (Figure 2, left). However, analysis of descriptors of motion extracted from their MD trajectories such as residual fluctuation revealed dynamic differences between the two isoforms that could potentially be exploited in an inhibitor design program.
Residual fluctuations obtained from the MD trajectories highlighted areas of the IKKs that acted differently during the simulations (Figure 2, right). Overall, the two isoforms behaved very similarly but IKKα appeared more flexible in several key areas around the ATP binding site, particularly at the G-loop (residues 22−27) above the site entrance and the loop located just adjacent to the hinge (residues 155−159 in IKKα (VGGKI) and residues 156−160 in IKKβ (GEQRL)). Two residues could account for the differences observed with the G-loop: Pro52 Figure 3. Superimposition of IKKα (white) and IKKβ (blue) highlighting the differences near/in the ATP binding site (marked by the staurosporine analogue as a stick model) between the two isoforms. The expanded area shows equivalent residues in IKKα (Asn28, Leu48, and Thr52) and IKKβ (Asn28, Gln48, and Pro52) engaged in different interactions/structural effects, resulting in Asn28 being available to interact with putative ligands in the binding pocket of IKKα but not IKKβ. (left) Loop conformation located below the active site in IKKα (white) and IKKβ (blue) and its relationship with α-helix 3 residue Asp127 (IKKα)/Asp128 (IKKβ). In IKKβ, Arg159 makes a reciprocal hydrogen bond dimer interaction with Asp128, whereas in IKKα Lys158 has no close interactions with Asp127. (right) Side chain amine nitrogen (Lys158 (IKKα)/Arg159 (IKKβ)) to side chain acid oxygen (Asp127 (IKKα)/Asp128 (IKKβ)) distance throughout the equilibrated phase of the simulation.

Journal of Medicinal Chemistry
Article and Gln48 in IKKβ (Thr52 and Leu48 in IKKα) induce a tension at the tip of the first α-helix through proline's intrinsic structure and the engagement of the glutamine side chain in a reciprocal H-bond dimer arrangement with the side chain amide of Asn28 ( Figure 3). This asparagine is located at the end of the G-loop, and its interaction with Gln48 restrained the G-loop movement by more than 1 Å in IKKβ when compared with IKKα. A different dynamic behavior was observed with IKKα, as the equivalent residues do not impose restraints on the G-loop movements: Thr52 has no rigid turn restriction like Pro52 in IKKβ, and the side chain of Leu48 seeks a hydrophobic environment and will not engage in H-bond formation with Asn28, leaving the side chain of this latter residue free to make interactions in the ATP binding site of IKKα (in contrast to being sequestered by Gln48 as in IKKβ) ( Figure 3). The other significant difference was related to the VGGKI loop (residues 155−159) in IKKα (residues GEQRL: 156−160 in IKKβ). In this case, one residue is responsible for the change observed in residual fluctuation: Lys158 in IKKα is replaced by Arg159 in IKKβ. The slightly longer arginine side chain and its

Journal of Medicinal Chemistry
Article bifurcate H-bonding capacity forms a reciprocal H-bond dimer with the side chain carboxylate of Asp128 in IKKβ that was maintained throughout the simulation, while the equivalent lysine in IKKα never engaged in a strong interaction with the equivalent Asp127 in IKKα ( Figure 4). This interaction in IKKβ is responsible for tethering the 156−160 loop to α-helix 3 (Asp128 is located in the middle of this helix), thus reducing its flexibility compared with IKKα.
With respect to putative inhibitor binding, the implications for isoform selectivity of these two sets of differences in the ATP site are 2-fold: first, it should be possible to design small molecules that target the free Asn28 side chain amide presented at the back of the IKKα pocket that is otherwise engaged in IKKβ (Figure 3), and second, because the ATP binding pocket has greater flexibility in IKKα, it has the potential to accommodate larger substituents, particularly below the G-loop.
Initial Hit Identification. To identify hits that could be developed to exploit the differences between the two isoforms, we screened our in-house compound library compiled of fragments designed to target the common hinge-binding motif found in protein kinases. Using a DELFIA kinase assay kit with minor modifications 29 to measure IKKα and IKKβ inhibitory activity, we identified 2-amino-4-chloro-5-cyanopyrrolo[2,3-d]pyrimidine 4 as our primary hit (Table 1). On the basis of the assumption that the 2-aminopyrrolo [2,3-d]pyrimidine core of compound 4 was binding at the hinge of IKKα, we varied substitution at the 4-position widely. Differences in bulk and polar functionality were incorporated to exploit the greater

Journal of Medicinal Chemistry
Article flexibility in the IKKα ATP binding site and the position of the Asn28 side chain at the back of the pocket.
Chemistry. The general route used to prepare N 4 -substituted 2-amino-5-cyanopyrrolo [2,3-d]pyrimidines for assessment began with the preparation of 1 using a published procedure. 30 1 was then protected as either its diacetyl derivative 2 or pivaloyl derivative 3, which were subsequently converted to the aryl chlorides 4 and 5 (Scheme 1) and used to prepare an initial series of compounds (4−31) (Table 1) by nucleophilic aromatic substitution with a range of primary and secondary amines. The second series of aryl cross coupled compounds exploring the 4-phenyl 2-amino-5-cyanopyrrolo[2,3-d]pyrimidines was prepared by Suzuki−Miyaura couplings (43−62) with 5. In general, the use of the N-pivalamide 5 gave higher yields of the final compounds compared to the unprotected form 4.
To explore polar functionality further in the N 4 -substituted 2-amino-5-cyanopyrrolo [2,3-d]pyrimidines, a series of amino alcohol derivatives (23−31) was prepared as analogues of 16. Compounds 23−29 were prepared from commercially available amines using the procedure detailed in Scheme 1. To incorporate 4-substituents with more than one hydroxyl group in the cyclopentane ring as in the putative IKKα selective inhibitor NAM, 17

Journal of Medicinal Chemistry
Article the unstable halide 33, which was immediately treated with N,N-dibenzylamine. The resulting allylic amine 34 underwent syn-dihydroxylation to afford the diol 35 33 with excellent diastereoselectivity, which was then protected as the dibenzoate 36. The amine was subsequently deprotected by hydrogenation to yield 37, which was coupled with 5. Global deprotection gave the desired final product 30 (Scheme 2).
To prepare the triol 31, the allyl acetate 38 was reacted with sodium di-tert-butyl iminodicarbonate 34,35 to produce 39, which was oxidized to the syn-diol 40 followed by benzoate protection of the hydroxyl functionalities to give 41 and removal of the BOC groups under acidic conditions to yield the amine 42. This was then coupled with 5 followed by basic hydrolysis to remove the protecting groups to give the desired final product 31 (Scheme 3).
An obvious progression from this series was to replace the bulky saturated ring with a phenyl ring bearing a selection of polar o-, m-, or p-substituents. This series was prepared by reacting 5 with a variety of boronic acids or esters under Suzuki− Miyaura coupling conditions to afford the products 43−62 (Scheme 1; Table 2).
To assess the importance of the 2-amino and 5-cyano groups for activity and selectivity, we purchased (63) or prepared compounds (64−71) without either or both substituents (Scheme 4, Table 3).
Structure−Activity Relationship Studies. In the N 4 -substituted series (Table 1) several compounds had similar K i values against IKKα compared to the initial hit 4. For example, compounds with aliphatic substituents retained potency and selectivity against IKKα and had the potential to be improved through further derivatization. Inhibitory activity for the secondary amines (7, 8, and 9) suggests the presence of a restricted lipophilic pocket in IKKα, which could be responsible for a 5-fold increase in potency from the methylamine analogue 7 to the cyclohexylamine derivative 9. However, the inactivity of the compound with the bulkier cyclohexylmethylamino group (10) suggested that this lipophilic pocket was limited in size. In contrast, the more planar phenyl and benzyl analogues (11 and 14) were found to be less active than 9, and the insertion of polar substituents in the para position of the aniline group produced no significant improvement in activity (12 and 13). Conversely, the insertion of polar groups in more flexible side chains was tolerated (16), which suggested the presence of a polar region adjacent to the lipophilic pocket.
Although compounds with a piperidyl side chain (17 and 18) were found to be active, other tertiary amines at position 4 were not tolerated (19−22), which could be due to additional heteroatoms not being accepted in a lipophilic area of the binding directly proximal to position 4 of the pyrrolo[2,3-d]pyrimidine scaffold. By contrast, the activity of 16 suggested that a flexible ethyl chain is able to direct the heteroatom away from this lipophilic region and possibly engage an adjacent polar region more effectively.
Removal of the hydrogen bond donor (HBD) from the 4-amino substituent by replacing the secondary cyclohexylamino group in 9 with the tertiary N-methylcyclohexylamine (10) abrogated activity. However, the activity of piperidyl analogue 17 suggests that a HBD at the 4-position is not an absolute requirement for inhibition of IKKα. The inactivity of compound 10 can be explained by rotation of the N-methyl (and hence the cyclohexyl group) substituent orthogonal to the pyrrolo [2,3-d]pyrimidine moiety to reduce steric clash with the 5-cyano substituent and generate a conformation that is no longer compatible with the binding site.
On the basis of the activity of the 4-hydroxyethylamino derivative 16, we prepared a second series of amino alcohol derivatives (Table 1). Extending the alkyl chain of 16 from ethyl to propyl was detrimental to activity (23), whereas the cyclic analogues produced more interesting results. The (R)-pyrrolidin-3ol enantiomer 25 was active and selective against IKKα, whereas the (S)-enantiomer 24 was inactive, suggesting the directionality of the hydroxyl group had a major influence on activity and selectivity. On the other hand, the (R)-and (S)-pyrrolidin-2-yl methanol enantiomers 26 and 27 shared the same activity and selectivity against IKKα, implying that free rotation around the methyl alcohol is sufficient to allow equivalent interactions.
The trans-aminocyclohexanol derivative 28 was also active and selective against IKKα, but the cyclohexylamino ethyl alcohol 29, designed as a hybrid of compounds 8 and 16, proved to be inactive. Like compound 10, this is probably due to the aliphatic ring and the ethyl chain being twisted 90°away from the nitrile axis to adopt a more energetically favored conformation that is incompatible with the binding site. The diol 30 was active and selective but did not offer significant improvement. The triol 31, prepared as an analogue of NAM, proved to be as active as 30 but less selective. Overall, no significant improvement in binding or selectivity was gained by the introduction of more than one hydroxyl group in this series of secondary cyclic amine derivatives.
The direct attachment of an aromatic ring to position 4 of the pyrrolo [2,3-d]pyrimidine scaffold exemplified by compound 43 retained the potency and selectivity of the original hit 4 ( Table 2). In general, compounds with para substituents in the phenyl ring had better activity against IKKα and produced our first nanomolar IKKα inhibitors in 47 and 48. Although activity against IKKβ was evident for these two compounds, their higher potency for IKKα ensured significant isoform selectivity was maintained. Compounds with meta substituents were low micromolar inhibitors of IKKα (for example 53, 56, and 57) but were equipotent against IKKβ, apart from 55, which was inactive against both isoforms. As with compounds 10 and 28, the inactivity of the ortho-substituted derivatives in this series can be attributed to enforced conformational rotation of the phenyl ring to a more orthogonal relationship with the pyrrolo [2,3-d]pyrimidine scaffold, which introduces steric clashes in the binding site. This is less pronounced with an o-fluoro substituent and, consequently, 59 had similar potency to the unsubstituted 4-phenyl derivative 43.
Compound 48 (proprietary code SU909), with a para-primary sulfonamide, represents the most potent IKKα inhibitor reported to date that has significant selectivity over IKKβ, although a limited kinase profile using 40 kinases representative of the
To explore the role of the primary sulfonamide, the 4-methyl sulfone analogue 49 and the secondary sulfonamides 50 and 51 were prepared. Notably, 49 and 50 had reduced potency against IKKα, but both were more active against IKKβ, reversing the isoform selectivity for the first time in this series. Similarly, when the orientation of the sulfonamide with respect to the phenyl ring (52) was reversed, a reduction in potency against IKKα occurred but an improvement against IKKβ was again evident, inverting isoform selectivity.
The absence of a 5-cyano group in 48 (compound 64) reduced activity against IKKα more than 30-fold (Table 3), although there was a slight improvement against IKKβ. Compound 65, which lacked both the 5-cyano and the 2-amino substituent, had lower activity against both isoforms. Compound 66, without the 2-amino group but with the 5-cyano, attenuated IKKβ activity but did not re-establish the potency against IKKα seen for 48. The importance of the 5-cyano group for activity against IKKα was further demonstrated by preparing analogues of selected active N 4 -substituted exemplars from Table 1: removal of the 5-cyano substituent produced inactive compounds against both isoforms (67−69), and its replacement with other electron-withdrawing substituents such as a trifluoromethyl (70) or a carboxamide (71) moiety also attenuated all IKKα inhibitory activity (Scheme 4, Table 3). Furthermore, compound 63, which represents the original hit compound (4) without the 5-cyano group, was also inactive against IKKα (Table 3).
Overall, our SAR data suggest that the 5-cyano substituent is essential for IKKα activity in this series of compounds but must be combined with a 2-amino group to promote selectivity. To enhance potency for IKKα, a 4-phenyl group bearing a specific polar para-substituent appears to be crucial.
Docking Studies. To explain the general SAR profile observed with our series and to relate these to the differences between the two IKK isoforms that our MD studies had revealed, we performed docking studies using GOLD. 36 A limited flexibility was enabled in the side chains of specific residues that had shown significant fluctuation in the MD trajectories, namely Asn28, Val29, and Lys44. In the co-crystal structure of IKKβ 28 that had formed the basis of our simulations, the staurosporine analogue interacts though H-bonds with the backbone groups of GK+1 (Glu97) and GK+3 (Cys99) in the conserved hinge region. Initially, no attempt was made to direct the ligands to interact specifically with the hinge region in our docking simulations; the only requirement was to occupy the ATP binding site. This first docking study was performed to find a common binding mode that could explain the IKKα inhibitory activity and selectivity reported for exemplars from our two series. To explore this, compounds bearing different hydrophobic substituents were selected for their similar potency and selectivity (e.g., 4, 9, and 43). The poses where the 2-aminopyrrolo [2,3-d]pyrimidine scaffold interacted with the hinge region by H-bonds were studied in detail and compared to similar hinge-binding protein kinase inhibitors. 37 For example, aminopurines are known to adopt two different binding modes in interactions with the hinge region of CDK2 via three hydrogen bonds. 38 However, in similar poses for our compounds in IKKα, the analogous triplet of hydrogen bonds was poorly supported by the IKKα hinge region in terms of bond distances and bond angles. Not surprisingly, similar poses were found when the same compounds were docked with IKKβ, which is due to the high isoform homology and subtle differences in topography in this hinge region of the ATP binding site. These poses could not account for the selectivity reported in our assays and suggested that this was not the binding mode responsible for conveying such discriminatory activity between the isoforms. However, a binding mode where the 2-aminopyrrolo [2,3-d]pyrimidine scaffold interacted with the back pocket of the IKKα active site explained the potency and selectivity displayed by our compounds more effectively. Parts A−C of Figure 6 illustrate how compounds 4, 9, and 43 were predicted to interact with IKKα, specifically targeting the carboxylate group of Glu61, the ammonium group of Lys44 and, most significantly, the side chain carbonyl group of Asn28 via four H-bonds. In IKKβ, because Asn28 is involved in a dimeric H-bond interaction with Gln48 revealed by our simulations (Figure 3), which also changes the position of Glu61, the triple H-bond interaction observed in IKKα is lost. We therefore propose that the aminopyrrolo [2,3-d]pyrimidine scaffold is anchored to the back of the ATP site specifically targeting Asn28 in IKKα, which accounts for the observed selectivity. In this orientation, the importance of the 5-cyano group when binding with IKKα can also be explained. Examination of the trajectory from an MD simulation of 47 revealed a water molecule present for the majority of the run that formed a three-centered hydrogen bond bridge between the nitrile, the side chain carboxylate of Asp102, and the backbone NH of the same residue ( Figure 6D). Overall, binding to IKKα appears to be facilitated through four centers on the 2-amino-5cyanopyrrolo [2,3-d]pyrimidine scaffold with four residues in the ATP binding site: Glu61, Lys44, Asn28, and Asp102. A similar binding orientation with IKKβ is less likely because the two of the equivalent sites are less accessible (Asn28 and Glu61), and this would account for the lack of activity in this isoform.

Article
Although the poses described for compounds 4, 9, and 43 showed no interaction with the hinge region, the equivalent poses for the nanomolar inhibitors 47 and 48 had significant H-bond interactions with the GK+3 residue via their p-hydroxymethyl and p-sulfonamide substituents ( Figures 6D  and 7). The equivalent binding pose in IKKβ did not feature the hydrogen bonds HB1 and HB2 due to local structural differences observed between the kinase isoforms and could explain the lower affinity of 48 for IKKβ. Interestingly, the presence of the methyl group in the secondary sulfonamide 50 compromised its ability to form an H-bond with the HBA carbonyl group of GK+3 because of a steric clash with the hinge region, thus reducing affinity for IKKα. The sulfonyl group could still interact with the NH of GK+3, which explains

Journal of Medicinal Chemistry
Article why 50 and the sulfone 49 both have similar potency against IKKα.
The reason for both 49 and 50 also having greater inhibitory activity against IKKβ that reverses isoform selectivity is because the polar side chain of Asn28 in IKKβ rotates out of the back pocket to H-bond with Gln48 ( Figure 3). When 49 and 50 were docked in the IKKβ structure, they commonly adopted a flipped pose, where the 2-aminopyrrolo [2,3-d]pyrimidine engaged with the GK+1 and +3 residues at the hinge through H-bonding, while the sulfonyl oxygen and the 5-cyano group H-bonded with K44 in the back pocket ( Figure 8). Binding to IKKβ by 49 and 50 in this orientation is facilitated by moving the polar Asn28 from the immediate vicinity, creating a cleft that can more easily accommodate the hydrophobic methyl moiety of 49 and 50. A similar reversal of selectivity was seen with 64, which lacks the 5-cyano substituent of 48. The drop in potency against IKKα can be attributed to the absence of a water-bridging H-bond interaction with Asp102 but also to the removal of the conformationally restraining effect of the 5-cyano group on the adjacent 4-phenyl ring which would normally optimize alignment of the p-sulfonamide H-bonding interaction with GK+3 in the hinge. The enhanced affinity for IKKβ by 64 can again be explained by the adoption of a flipped pose when binding to this isoform, which is enabled by a more flexible p-sulfonamide that can H-bond more effectively with Lys44. Just as importantly, this flexibility allows the HBD

Journal of Medicinal Chemistry
Article of the sulfonamide group to access the less accessible Glu41 at the back of IKKβ pocket.
In summary, in line with our SAR data and the hypothesis derived from our MD simulation studies, we suggest that nanomolar potency against IKKα with selectivity over IKKβ can be accomplished by having molecules that can bind to the hinge region of the ATP binding site and to the Asn28 residue at the back of the pocket that is available in IKKα but not in IKKβ ( Figure 8). Furthermore, inhibitory potency against IKKα can be enhanced when compounds can form H-bonding interactions with the three residues in the back pocket (Glu41, Lys44, Asn28) and the GK+3 and Asp102 residues in the hinge.
Demonstrating IKKα Activity and Selectivity Is Recapitulated in Cells. The objective of the cell-based assessment was to characterize whether the two most active compounds against IKKα (47 and 48) recapitulated their activity in a cellular environment. IKKα but not IKKβ plays a role in the regulation of the noncanonical NF-κB cascade which initially involves phosphorylation of p100, followed by p100 processing to p52 to regulate a number of genes that contribute to and promote cellular growth. 39 More recent studies have shown that overactivation of the noncanonical NF-κB pathway in a variety of cellular settings, e.g., prostate 19,21 and pancreas, 25−27 is inherently associated with cell survival and proliferation. The U2OS osteosarcoma cell line is representative of these proliferative characteristics and is dependent on constitutively activated IKKα signaling; 40 it was therefore selected as the system to demonstrate perturbation of IKKα activity in the cellular environment.
Pretreatment of U2OS cells with increasing concentration of 48 and 47 resulted in concentration-dependent inhibition of FCS-stimulated phosphorylation of p100 (Figures 9a and 10a). The IC 50 values for 48 and 47 were calculated as 8.8 and 13.9 μM, respectively. Two IKKβ-dependent readouts were selected to assess selectivity over IKKβ. First, TNFα stimulates IκBα degradation through activation of IKKβ and if IKKβ is inhibited, then a band corresponding to the IκBα protein should be evident by Western analysis. Second, IKKβ phosphorylates p65 on Ser536, which can also be assessed by Western blot analysis. Only at the highest concentration (100 μM) of 48 was IκBα degradation and p65 (Ser536) phosphorylation inhibited (Figure 9b). Even at this high concentration, these effects were not apparent following pretreatment with 47 ( Figure 10b).
To provide further evidence for cellular activity, mouse embryonic fibroblasts lacking the IKKβ kinase subunit were utilized to investigate the potency of both compounds against agonist-stimulated IKKα-regulated NF-κB transcriptional activation. Cells were infected with an adenovirus encoding an NF-κB-luciferase promoter gene. NF-κB transcriptional activity driven by IKKα was stimulated specifically using IL-1β, 41 and the ability of the compounds to inhibit this response was assessed. At maximal concentrations of 50−100 μM, 48 abolished transcriptional activity compared to 47 which inhibited activity by approximately 81% (% stim: Il-1β+DMSO = 100%, 48+Il-1β = −40.9 ± 4.6%, 47+IL-1β = 18.79 ± 7.2%, both **P < 0.01) ( Table 4). Both 47 and 48 demonstrated concentration-dependent inhibitory effects on FCS-induced phosphorylation of p100 by IKKα. As IKKα plays a lesser role in the canonical NF-κB cascade, 42 particularly the rapid and transient stimulated degradation of IκBα, selective IKKα inhibitors should not affect TNF-α induced IκBα degradation and p65 phosphorylation. Compound 47 did not affect either agonist stimulated IκBα degradation or p65 phosphorylation, while 48 only had a small effect on both markers at the highest concentration tested (100 μM). Taken together, these data demonstrate that 47 and 48 are the first examples of IKKα-selective compounds that inhibit agonist-stimulated NF-κB signaling of the noncanonical pathway at much lower concentrations than the canonical pathway.

Journal of Medicinal Chemistry
Article ■ CONCLUSION There are many reported inhibitors of IKKβ, primarily because it was considered a viable target in inflammatory disease. Designing inhibitors of IKKα that are selective over IKKβ has to this date not been reported despite the former isoform now being recognized as a potential target in a number of cancers. One reason for this is the high sequence-homology in the ATP-binding site of the two isoforms and the absence of any high-resolution crystal structure of IKKα to guide structure-based inhibitor design.
By employing molecular dynamics simulations on a homology model of IKKα based on IKKβ, we identified key dynamic differences at the ATP-binding site and exploited these to design the first selective inhibitors of IKKα. Our compounds demonstrated target engagement with IKKα-related pharmacodynamic markers and nonengagement with IKKβ markers in cells. These compounds therefore represent the first chemical tools that can be used to further characterize the role of IKKα in cellular signaling, to dissect this from the very similar IKKβ isoform and to validate IKKα in its own right as a target in cancers, such as prostate, breast, and pancreatic cancer.
The discovery of a 2-aminopyrrolo [2,3-d]pyrimidine chemical series also provides valuable information with respect to SAR for IKKα inhibitory activity. Substituents at position 4 of the pyrrolo [2,3-d]pyrimidine scaffold allowed for diversification, with IKKα selectivity achievable through the introduction of groups that specifically target residues in its binding site. However, further significant improvements in activity through aliphatic 4-amino substituents appear to be unlikely. Modifying the 4-phenyl substituent offers a more promising route toward generating compounds with improved activity and selectivity. Using this strategy, we are currently developing more potent IKKα-inhibitors that display functional outputs in a number of cell lines, all of which will be reported in due course.

■ EXPERIMENTAL SECTION
IKKα Homology Model. Building a homology model of IKKα: the kinase domain of IKKβ (chain B, residues 1−309, PDB entry 4KIK. 28 was used as a template to build the kinase domain of IKKα, keeping the inhibitor (KSA700 in the pdb file) and waters found within 6 Å of the protein−inhibitor complex. The residue alignment ( Figure Mod1) and homology building were performed using Discovery Studio 3.1 (Accelrys Inc., San Diego, USA) after missing residues D174, Q175, and G176 were added in an extended conformation.
Both IKK kinase domains were subjected to molecular dynamics using the AMBER12 simulation software. The inhibitor present (residue name KSA700) was kept and parametrized using antechamber and charges calculated and fitted using the AM1-BCC scheme. The two systems were placed in a periodic octahedral box and solvated with TIP3P water with outer edges 6 Å in each direction from the closest solute atom. The systems were then neutralized and physiological salt conditions applied (∼150 mM) by adding 24 Cl − and 29 Na + ions to the IKKα system and 24 Cl − and 30 Na + ions to the IKKβ system. The AMBER ff12SB was applied to all protein atoms, while gaff was used for the ligand. Parameters for the phosphoserine residues were taken from Craft and Legge. 43 Before the MD production phase, minimization and equilibration (to reach 310 K) were performed in two stages as described. 44 The NPT ensemble was used at 310 K until the systems had stabilized for at least 20 ns (50 and 100 ns simulation time for IKKβ and IKKα, respectively). All MD steps used the SHAKE algorithm 45 with a 2 fs timestep and a 10 Å cutoff for long-range electrostatic interactions. An average structure was generated (using ptraj within the AMBER suite) for the last 21 ns (IKKα) or 26 ns (IKKβ) and subsequently minimized in three steps, with the solvent, ions, and hydrogen atoms initially minimized while the protein and inhibitor were restrained by 100 kcal mol −1 Å −2 . The restraint was then removed from the protein side chain atoms, and finally the whole system was allowed to minimize until a derivative of 0.1 kcal mol −1 Å −2 was achieved. These structures were then utilized for further docking studies with the GOLD software.
General Experimental. Solvents (reagent grade or better) were purchased from Sigma-Aldrich or Fischer Scientific. Anhydrous solvents where purchased from Sigma-Aldrich. Deuterated solvents were purchased from Sigma-Aldrich. Chemicals (95% purity or above) were purchased from Acros Organics, Alfa Aesar, Apollo, Fluorochem, or Sigma-Aldrich. Solvents and chemicals were used as received without further purification or treatment.
Oxygen-or moisture-sensitive reactions were carried out under a nitrogen atmosphere.
Microwave reactions were performed with a Biotage Initiator system. High absorbance was selected for polar solvents and normal absorbance was selected for nonpolar solvents.
The progress of the reactions was monitored on Merck 60F254 TLC plates. Spots were visualized by irradiation with ultraviolet light (254/366 nm) or KMnO 4 , ninhydrin, or phosphomolibdic acid (PMA) TLC stains.
Column chromatography was performed with a Biotage SP4 system; cartridge size and eluent specified in the corresponding experiments (% is referring to the most polar solvent in the mixture), using silica gel as the stationary phase (particle size 0.040−0.063 mm, Merck or Fisher Scientific).
Specific rotations were measured in a PerkinElmer polarimeter 341 apparatus at 20°C and a wavelength of 589 nm (sodium D line) in DMSO UV spectrophotometric analysis grade. 1 H and 13 C NMR data were recorded on either a JEOL ECX-400 (400 MHz) or Bruker Avance3/DPX400 (400 MHz) spectrometers at 400.0 and 100.6 MHz, respectively. Chemical shifts (δ) are expressed in parts per million (ppm) coupling constants (J) are in hertz (Hz). Chemical shifts (δ) are reported relative to TMS (δ = 0 ppm) and/or referenced to the solvent in which they were measured. All measurements were carried out at 298 K (except when stated). Abbreviations used in the description of NMR data are as follows: app, apparent; s, singlet; bs, broad singlet; d, doublet; t, triplet; q, quartet; p, pentuplet; m, multiplet.
HR-MS was conducted using a Thermo Scientific Exactive Orbitrap mass analyzer. LR-MS was conducted using a ThermoQuest Finnigan LCQ Duo instrument. GC-MS was conducted using a ThermoQuest Finnigan Polaris Q instrument.

Journal of Medicinal Chemistry
Article production of high titer recombinant adenovirus was performed by routine methods. 47 IKKβ knockout mouse embryonic fibroblasts (MEFs; kindly provided by Prof. I. Verma, UCSD, USA) were counted when approximately 60−70% confluent in T75 cm 2 flasks and infected with adenovirus up to 300 pfu/cell −1 for 24 h in 10% DMEM. Cells were then lifted from the flasks and plated into 96-well clear bottom, black luciferase plates and allowed to settle for 24 h before serum starved overnight. Cells were then stimulated with the appropriate agonists for the indicated times and reactions terminated by addition of lysis buffer containing 0.2 mM luciferin substrate. Relative light units (RLU) were measured using a Trilux microbeta counter (luminometer).
Kinase Profiling. Kinase profiling of compound 48 was outsourced to Millipore, specifically the Kinase Profiler Customer Service for single concentration studies using duplicate assay points.