Co-crystal structure of PI3Kγ in complex with an initial organoruthenium inhibitor lead compound
In order to identify initial lead inhibitors for PI3K from a pyridocarbazole organometallic ruthenium-based scaffold, a library of 75 compounds (22
) with a diverse configuration of ligands around the ruthenium metal () were screened against the human PI3Kα isoform using a fluorescence polarization-based kinase assay (Echelon Biosciences). From this library, we identified the racemic compound DW12, which bears an additional hydroxyl group on the indole group as the most potent compound with an IC50
value of around 1 μM using a Kinase Glo assay (, S1
). DW12 is also known to be a highly potent inhibitor for the protein kinases GSK3 (26
) and PIM1 (23
). In order to understand its mode of inhibition and to provide a structural platform for structure-based inhibitor optimization to improve potency as well as selectivity, we co-crystallized one enantiomer, named DW2, with human PI3Kγ. The PI3Kγ/DW2 co-crystals formed in the space group C2 with one protein/inhibitor complex per asymmetric unit cell, and the structure was determined by molecular replacement using the human unliganded PI3Kγ structure as a search model. The strong electron density signal of the ruthenium atom and the proximal outline of the inhibitor allowed for the unambiguous placement of the DW2 inhibitor into the PI3Kγ active site (). The structure was refined to 2.8 Å resolution to excellent refinement statistics and geometrical parameters (). Overall, the protein component of the PI3Kγ/DW2 structure adopts the same conformation as the unliganded PI3Kγ protein and the conservation of the overall structure suggests that the organoruthenium inhibitor binds to the PI3Kγ ATP binding pocket without significant alteration of the native PI3Kγ conformation ().
Initial lead inhibitor identification and crystal structure of PI3Kγ in complex with DW2
Statistics of crystallographic data collection and structural refinement.
Detailed view of the interactions between the PI3Kγ kinase domain and DW2
As designed, the DW2 inhibitor occupies the ATP binding pocket of the kinase domain with about a 2.4 Å displacement towards solvent of the DW2 inhibitor relative to ATP (). The maleimide moiety of the DW2 inhibitor largely overlaps with the adenosine moiety of ATP. The DW2 maleimide moiety and derivatized pyridocarbazole makes extensive hydrogen bonding interactions with both protein side chain and main chain atoms of the PI3Kγ kinase domain (). In particular, the N12 atom of the maleimide nitrogen forms water mediated hydrogen bonds with the main chain carbonyl of residue Glu880; and the DW2 carbonyl oxygens, O1 and O2, of the maleimide moiety form hydrogen bonding interactions with the side chain oxygen of residue Glu964 and the main chain amide nitrogen of residue Val882, respectively. The O4 atom of the DW2 phenol moiety also hydrogen bonds to the main chain carbonyl of Val882 and main chain amide of residue Asp884. Several van der Waals contacts stabilize the interaction between PI3Kγ and DW2 including contacts between the inhibitor ring system with the side chains of residues Trp812, Met953, Ile881, Ile831, Met804, Ile963, Thr887, Val882, Phe961 and Ser806. The ruthenium coordination center of the DW2 inhibitor adopts a pseudo-tetrahedral geometry with the CO ligand pointing towards the C-lobe of the kinase domain and the cyclopentadiene (Cp) ligand pointing to the P-loop of the kinase. The CO ligand of the ruthenium makes hydrogen bonding contacts with the main chain amide nitrogen and side chain hydroxyl of residue Thr887, while the Cp ligand of the ruthenium makes van der Waals contacts with the aliphatic portions of Lys890 and Lys802. A surface representation of the DW2 inhibitor shows favorable shape complementarity with the ATP binding pocket of PI3Kγ, further supporting the favorable binding properties of the DW2 inhibitor with PI3Kγ ().
Structure based design of second generation PI3K inhibitors
A superposition of the PI3Kγ/DW2 structure with previously determined PI3Kγ structures with conventional organic inhibitors such as staurosporine, myricetin, LY294002 (28
), AS605240 (29
) and PIK90 (30
), reveals some unique features underlying the binding of DW2 to PI3Kγ. While the maleimide, pyridine and indole moieties of the DW2 inhibitor largely overlap with other inhibitors, the phenol moiety extends into a hinge region between the N-lobe and C-lobe of the kinase domain, representing a novel region of PI3K “inhibitor space” (). The observation that the phenol ring of DW2 makes complementary interactions with the PI3K active site is particularly interesting, since the HB12 homologue missing this hydroxyl group does not inhibit PI3K at a comparable inhibitor concentration (). This structural evidence suggests that this phenol moiety of DW2 be retained in a subsequent round of inhibitor optimization.
Structure based design of optimized inhibitor and structure activity relationship correlation
Most notably, the DW2 inhibitor appears to be pushed out of the ATP binding pocket relative to the other inhibitors (). This causes the maleimide nitrogen of DW2 to be too far away to directly contact the protein within the ATP binding pocket. Instead, the maleimide nitrogen makes water mediated hydrogen bonding interactions with the protein backbone. This finding is in contrast to the previous co-crystal structure of organoruthenium inhibitors with similar scaffolds in complex with protein kinases such as GSK3 (unpublished data) and PIM1 (23
), in which the maleimide nitrogen is in direct contact with the protein pocket, leaving essentially no space for additional modification on the imide nitrogen. Based on this observation, we hypothesized that a derivatization of the maleimide nitrogen, preferably to a hydrophobic group to complement a largely hydrophobic protein pocket in this region, might enhance protein interaction as well as improve kinase selectivity towards PI3K. Indeed, we observed a slight enhancement in inhibitory potency for PI3K going along with a dramatic improvement in selectivity, when a methyl group was introduced on the maleimide nitrogen ().
An overlay of DW2 with other PI3K inhibitors also shows that the other inhibitors occupy an overlapping region of PI3K inhibitor space that is not occupied by the DW2 inhibitor. In particular, the other PI3K inhibitors occupy a tunnel like pocket (called pocket A hereafter) that extends into a space underneath the P-loop which is further defined by the α3 helix, the β4-β7 strands and the DFG motif of the activation loop (). We attempted to derivatize the pyridine moiety which is placed in proximity to pocket A. Intriguingly, a fluorine substitution on the 3’ position of pyridine moiety (EAd125) significantly increased its inhibitory potency against PI3K ().
The Cp and CO ligands of the ruthenium atom occupy a novel region of PI3K inhibitor space relative to other PI3K inhibitors and serves to cap off the relatively hydrophobic PI3Kγ ATP binding pocket from solvent (). The proximity of the Cp group to solvent suggests that more hydrophilic capping groups might enhance inhibitor binding. An exhaustive screening against a library of compounds with modifications on the Cp ligand of the parental EAd125 compound revealed several compounds that contained hydrogen bonding donors in this region of the molecule (Figure S2
The most potent compounds from this series of DW2 analogues, E5 was separated into the pure enantiomers, named E5E1 and E5E2 and their PI3Kγ inhibition properties were characterized. As illustrated in , E5E1 and E5E2 displayed comparable IC50 values of 41 nM and 39 nM, respectively.
E5E2’s activity for PI3K and kinase specificity screen
Kinase specificity of the E5E2 inhibitor
In order to further characterize the role of methylation of the maleimide nitrogen on this new class of organoruthenium inhibitor’s in mediating inhibition of the PI3K lipid kinases over protein kinases, we assayed the E5E2 inhibitor against five human protein kinases (MST1, PAK1, BRAFWT
, GSK3α and PIM1) representing four major kinase families of STE, TKL, CMGC and CAMK, respectively. This kinase panel screen revealed that the E5E2 inhibitor had significant selectivity towards PI3K over these protein kinases. In particular, the E5E2 inhibitor did not show appreciable inhibition for BRAFWT
, PAK1 and MST1 and only showed about 50% inhibition towards the GSK3α and PIM1 kinases at an inhibitor concentration of 4 μM (). Together, these results reveal that the E5E2 inhibitor exhibits nearly a 100 fold selectivity for PI3Kγ over these representative protein kinases. The selectivity of E5E2 is quite remarkable considering that GSK3α and PIM1 are common targets for this class of organoruthenium inhibitors (25
). For comparison, we assayed a previously characterized PI3K inhibitor, LY294002 (15
), against the same panel of kinases used in this study (Figure S3
). Interestingly, we found that LY294002 was both less potent against PI3Kγ and less specific for PI3Kγ relative to the other kinases profiled in this study (compare and S3
Cellular activity of E5E1 and E5E2
Melanomas are a tumor known to have high constitutive activity in the PI3K/AKT pathway (33
). To probe E5E1 and E5E2’s ability to inhibit proliferation of melanoma cells, we examined the cellular response of E5E1 and E5E2 in a melanoma cell line. We found that both E5E1 and E5E2 were able to block the constitutive activity of AKT in a concentration dependent manner (). Similar levels of phospho-AKT inhibition were also seen following treatment with the PI3K inhibitor LY294002 (30 μM) . In contrast, no effects were seen upon the activity of the MAP kinase pathway, as shown by the lack of effect on phospho-ERK expression (). Inhibition of the PI3K pathway in melanoma is also known to inhibit cell growth in 2D cell culture (35
). In agreement with these findings, treatment of the 1205Lu melanoma cell line with increasing concentrations of E5E1, E5E2 and LY294002 was associated with an inhibition of cell proliferation (). There appeared to be a close correlation between the concentrations of E5E1/E5E2 required to inhibit phospho-AKT and those required to inhibit proliferation. In contrast, treatment of primary human fibroblasts with LY294002 (10 μM), E5E1 (10 μM) or E5E2 (10 μM) did not inhibit cell growth, demonstrating a lack of a general cytotoxic effect (data not shown). In 3D cultures of melanoma cells, infection with an adenovirus encoding for a dominant negative construct of PI3K is known to block cell migration (35
). In agreement with these previous results, we found that the invasion of melanoma cells grown as 3D collagen-implanted spheroids, could be completely blocked following treatment with either E5E1 or E5E2 (both 30 μM) (). Taken together, we show that the PI3K inhibitor E5 works both in vitro
and in vivo
E5E1 and E5E2 inhibit PI3K activity in vivo
Crystal structure of PI3Kγ in complex with E5E2
To understand the structural basis for selective PI3Kγ inhibition by E5E2 () and also to confirm the design strategies we inferred from structural information of the PI3Kγ/DW2 complex, we co-crystallized the PI3Kγ/E5E2 complex and determined its structure by molecular replacement with the PI3Kγ extracted from the PI3Kγ/DW2 complex as a search model. The refined structure in the absence of the E5E2 inhibitor showed strong difference electron density for the inhibitor allowing for the unambiguous modeling of the inhibitor into the ATP binding site of PI3Kγ (). The E5E2 inhibitor showed more clearly defined electron density than the DW2 inhibitor and also refined to lower B-factors, consistent with the more potent PI3Kγ inhibition of the E5E2 inhibitor relative to DW2. An overlay of the E5E2 and DW2 inhibitors in the PI3Kγ active site reveals that the two inhibitors bind in almost identical positions except that E5E2 is bound about 0.7 Å more deeply in the active site pocket than DW2 ().
Structure of the E5E2 inhibitor in complex with PI3Kγ
A detailed analysis of the PI3Kγ/E5E2 complex shows that the hydroxyl group of the ring system mediates hydrogen bonding interactions with the Val882, Asp884 and Ala885 main chain atoms that are analogous to the interactions that are mediated by the DW2 inhibitor (). Although the fluorine substitution on the pyridine moiety of the pyridocarbazole does not make direct interactions with the protein, the highly electron rich fluorine could mediate long range electrostatic interactions with the highly polar proximal surface of the inhibitor binding cavity formed by side chains Lys833, Asp964, Asp950, Lys808, Asp836 and Lys807. The dihydroxy-t-butyl (DHtB) group on the amide linkage of the Cp ring sticks into the space sandwiched by the main chain carbonyl and side chain of Ala805 as well as the side chain of Lys890. This places the two hydroxyl groups of the DHtB moiety in a favorable position to form hydrogen bonding interactions with several main chain and side chain atoms in this region, although the electron density map corresponding to the DHtB moiety is not well enough defined to assign details of the interaction. The close van der Waals surface complementarity between the derivatized DHtB group and the adjacent protein also facilitates binding. Moreover, the polar nature of the DHtB group on the rather hydrophobic Cp ring ligand also likely makes more complimentary interactions with the solvent channel into the inhibitor binding pocket relative to the underivatized Cp ring.
Perhaps the most interesting feature of the PI3Kγ complex with E5E2 relative to its complex with the DW2 inhibitor is the methyl derivatization on the maleimide moiety, which only slightly increases inhibitory potency but significantly increases binding selectivity (). In the PI3Kγ/E5E2 structure this methyl group on the maleimide nitrogen is within van der Waals contact distance with a tyrosine residue (Tyr867) that is conserved among Class I PI3K enzymes but not present in protein kinases. Therefore, the methyl group of the E5E2 inhibitor is well accommodated in this pocket backed by Tyr867 of PI3Kγ, which is in contrast to the limited space present in the same region of organometallic inhibitors complex structures with protein kinases such as GSK3 and PIM1. Taken together, this analysis suggests that the maleimide methyl group of E5E2 inhibitor might be a particularly important kinase inhibitor specificity determinant of this inhibitor.
Structural basis of PI3K specificity of the E5E2 inhibitor
An overlay of the PI3Kγ/E5E2 structure with several other kinase structures in addition to GSK3 and PIM1 including BRAF, STK10 (shares more than 65% sequence homology with MST1) and PAK1 reveals that while each of the kinases share a high degree of structural homology within the inhibitor binding site, there is significant difference in the Tyr867-bearing loop preceding the β6 strand. In particular, Pro866 of PI3Kγ faces away from the methylated maleimide of E5E2, while the corresponding hydrophobic residues of the other kinases are in a position that would clash with the methyl group on the maleimide of the inhibitor (). This modeling is consistent with the poor inhibitory properties of E5E2 for these other kinases.
Structural comparison of PI3Kγ/E5E2 with other protein kinases
Based on the observations described above, we hypothesized that Tyr867 and Pro866 mediate important roles in the PI3K selectivity of the E5E2 inhibitor. To test this hypothesis, we carried out a site directed mutagenesis of the BRAF kinase directed at increasing its sensitivity to the E5E2 inhibitor. Specifically, we mutated Leu514 of BRAF (analogous to Tyr867 of PI3Kγ) to the less bulky valine and alanine residues and compared the ability of these mutants to be inhibited by the E5E2 inhibitor. Consistent with our hypothesis, the E5E2 inhibitor shows an IC50 of about 10 μM for the BRAF mutants, while the E5E2 inhibitor shows no detectable inhibition against wild-type BRAF at an inhibitor concentration of 10 μM (). Taken together, these results suggest that the specific methylation on E5E2 inhibitor complements with a unique conformation centered around Tyr867 on PI3Kγ to mediate selective inhibition against PI3K lipid kinases over protein kinases. Furthermore, E5E2 also serves as an excellent lead compound for the further development of potent and specific organoruthenium PI3K inhibitors.
A molecular switch controlling organometallic inhibitor selectivity for lipid kinases
The idea of using a pyridocarbazole organometallic scaffold as a template for designing protein kinase inhibitors stemmed from a chemical mimetic strategy based on the property of staurosporine as a nonselective kinase inhibitor and the versatility of ligand shuffling around the metal for specificity towards the intricate differences in kinase ATP binding pockets. A close examination of protein kinase structures in complex with staurosporine revealed a strikingly unified feature of two invariant direct hydrogen bonding interactions between the lactam moiety of staurosporine and the protein. The lactam moiety of staurosporine is exploited and converted into a maleimide moiety in the original design of the initial inhibitor derived from an organometallic scaffold to confer inhibitory potency. Structural studies centering on this new class of highly specific organoruthenium inhibitors, particularly in complex with two representative protein kinases GSK3 (unpublished data) and PIM1 (23
), have confirmed that the two invariant hydrogen bonds are largely conserved in the binding of organoruthenium inhibitor in the protein kinase pocket. In the follow-up studies, it was discovered that a methyl group addition to the maleimide moiety nearly abolished the inhibitory activity. From the structural point of view, this is reasonably anticipated because a methyl group would disrupt the hydrogen bonding interactions mediated by the maleimide imide as well as introduce additional steric clashes with the protein.
Despite the overall structural homology of PI3K lipid kinases with protein kinases, its mode of staurosporine binding is quite divergent. In particular, the direct hydrogen bonding interactions described above for protein kinases are not present in PI3Kγ. This observation is mirrored in the structure of PI3Kγ bound to the organoruthenium DW2 inhibitor. We took advantage of this feature by adding a methyl group on the maleimide imide (compound E5E2 and DW12Me) to occupy this space. As expected, these inhibitors were potent for PI3K but not for several protein kinases. From our structural analysis of the complex between PI3Kγ and the E5E2 inhibitor, we identified a Tyr867 residue which plays a central role in accommodating the methylated imide and modeled the structural basis for why protein kinases cannot accommodate the methylated imide. We tested this modeling by producing a site directed mutant of the BRAF protein kinases that re-sensitizes it to the E5E2 inhibitor. Our results highlight the importance of this methyl group on the maleimide moiety of our new class of organoruthenium inhibitor as a selectivity switch between PI3K lipid kinases and protein kinases.
Future design of improved PI3K inhibitors using an organoruthenium scaffold
In this study, we have identified mid-low nanomolar organoruthenium with significant selectivity towards PI3K, yet other structural aspects of our study can still be used as a guide to further improve PI3K potency and isoform selectivity. First, the fluorine atom of the pyridocarbazole ring faces the A pocket in the PI3K active site that would better accommodate a larger R group that could mediate direct hydrogen bonds. Second, a shallower pocket proximal to a maleimide carbonyl group (B pocket) formed by residues Asp964, Ile963, Tyr867 and Ile879 might accommodate a larger R group (Figure S4a
). Both modifications are likely to increase PI3K potency and specificity. When overlaid with the recently published PI3Kα crystal structure (36
), the E5E2 inhibitor of the PI3Kγ/E5E2 structure would be predicted to make analogous interactions with PI3Kα. Nearly 70% of the residues of PI3Kγ that make van der Waals contacts with the E5E2 inhibitor are conserved both in sequence and structure in PI3Kα (Figure S4b, c
). This observation is consistent with the finding that the PI3Kγ and α isoforms are close “pharmologues” exemplified by their similarity for their sensitivity to isoform specific PI3K inhibitors (30
). However, there are regions of divergence directly outside the E5E2 binding site that might be exploited for the design of organoruthenium inhibitors with PI3K isoform selectivity (Figure S4c, d, e
). In particular, the variations in the solvent exposed region of the ATP binding pocket provide an opportunity to identify a Cp derivatization that might distinguish between PI3K isoforms. In our preliminary studies we have shown that the E5E2 inhibitor is about 5-fold more potent for PI3Kγ over PI3Kα (data not shown) suggesting that the E5 inhibitor scaffold might indeed be exploited to develop isoform specific PI3K inhibitors.
Taken together, we have identified a new class of mid-low nanomolar organoruthenium inhibitors against PI3Kγ that also show significant selectivity against other protein kinase families. Crystallographic analysis of these inhibitors bound to PI3Kγ reveals the mechanism of inhibition and identifies a structural switch that confers significant selectivity against PI3Ks over protein kinases. Our structural analysis also suggests further modifications to increase PI3K potency and PI3K isoform selectivity.