Structure guided design of potent and selective ponatinib-based hybrid inhibitors for RIPK1

Summary

RIPK1 and RIPK3, two closely related RIPK family members, have emerged as important regulators of pathologic cell death and inflammation. In the current work, we report that the Bcr-Abl inhibitor and anti-leukemia agent ponatinib is also a first-in-class dual inhibitor of RIPK1 and RIPK3. Ponatinib potently inhibited multiple paradigms of RIPK1- and RIPK3-dependent cell death and inflammatory TNFα gene transcription. We further describe design strategies that utilize the ponatinib scaffold to develop two classes of inhibitors (CS and PN series), each with greatly improved selectivity for RIPK1. In particular, we detail the development of PN10, a highly potent and selective ‘hybrid’ RIPK1 inhibitor, capturing the best properties of two different allosteric RIPK1 inhibitors, ponatinib and necrostatin-1. Finally, we show that RIPK1 inhibitors from both classes are powerful blockers of TNF-induced injury in vivo. Altogether, these findings outline promising candidate molecules and design approaches for targeting RIPK1/3-driven inflammatory pathologies.

Introduction

Receptor Interacting Protein Kinases (RIPKs) are a family of Ser/Thr and Tyr kinases with important roles in inflammation and innate immunity. The kinase activities of RIPK1 and RIPK3 were found to be critical for the activation of necroptotic cell death pathway by multiple stimuli, including TNFα family of cytokines, interferons (IFNs) and Toll-like receptor (TLR) ligands (Christofferson and Yuan, 2010; Vanlangenakker et al., 2012). Importantly, RIPK1/3 kinases have been implicated in a variety of pathologic settings that currently lack effective therapies, including stroke, myocardial infarction, retinal injuries, lethal Systemic Inflammatory Response Syndrome (SIRS) and chronic gut and skin inflammation, and acute pancreatitis (Linkermann and Green, 2014).

We have previously described the development of necrostatins, a class of efficient small molecule inhibitors of necroptosis (Fig. 1A) (Degterev et al., 2008; Degterev et al., 2005). Importantly, an optimized analog of necrostatin-1, 7-Cl-O-Nec-1 (used throughout this paper and referred to as Nec-1), displayed unusually exclusive selectivity towards RIPK1 kinase and to lack necroptosis inhibitory activity in the absence of RIPK1 (Christofferson et al., 2012; Dillon et al., 2014). Structurally, this inhibitor, as well as other necrostatins, were found to stabilize an unusual inactive αC-Glu-out/DLG-out conformation of RIPK1 characterized by the large movement of the αC helix (αC-out) from the active state in conjunction with the inactive conformation of the DLG motif (Fig. 1B) (Xie et al., 2013). DFG (Asp-Phe-Gly) (or DLG in a subset of kinases, like RIPK1) is a highly conserved tripeptide motif present in most human kinases, which changes from the inactive “DXG-out” conformation, to the active “DXG-in” state, where Asp is aligned with other residues in the active center and is involved in Mg2+ binding. In addition, Nec-1 was found to interact exclusively with the DLG-out “back” pocket of RIPK1 without contacts in the more redundant ATP binding site, likely explaining its unusually high degree of selectivity. On the other hand, extensive structure-activity relationship (SAR) analysis of Nec-1 and other necrostatins revealed that even small changes to these molecules led to the robust loss of activity and failed to identify clear directions to significantly increase affinity of these moderately potent (e.g. cellular IC₅₀ of 7Cl-O-Nec-1=210 nM) molecules (Choi et al., 2012; Jagtap et al., 2007; Teng et al., 2005; Teng et al., 2007; Teng et al., 2008). Furthermore, necrostatins could have physical limitations on maximal robustness due to the small size of the molecules and an energy penalty due to the loss of a strong Glu/Lys interaction in αC-Glu-out conformation (Fig. 1B). These shortcomings prompted us to explore additional ways to target RIPK1 that would capture the excellent selectivity of necrostatins while achieving significant increases in activity.

Figure 1. Inhibition of necroptosis and RIPK1 kinase by ponatinib.

A) Structures of necrostatins. B) Comparison of Glu-in/DLG-out (red, PDB: 4NEU) and Glu-out/DLG-out (blue, PDB 4ITH) conformations of RIPK1 kinase reveals movement of αC helix. Movement of αC helix is indicated by black arrow. Position of Nec-1 in Glu-out/DLG-out structure is shown (Nec-1 - green). Ionic bond between Glu63 and Lys45 in Glu-in conformation is indicated. C) Ponatinib inhibits recombinant RIPK1 and RIPK3 kinases in vitro. 2 μM kinases were used in the in vitro ³²P autophosphorylation assay. Nec-1 only inhibited RIPK1, while Gleevec lacked activity against both kinases. D) Gleevec does not inhibit necroptosis. FADD-deficient Jurkat cells were treated with 10 ng/ml human TNFα in the presence of 11 point dose ranges of Ponatinib and Gleevec for 24 hr. E) Ponatinib inhibits TNF-induced cell death in the presence of 100 nM TAK1 inhibitor 5z-7-oxozeaenol. MEFs were stimulated with 10 ng/ml mouse TNFα with 5z-7 for 24 hr to induce RIPK1-dependent apoptosis. To induce RIPK3-dependent necroptosis, cells were additionally treated with 50 μM zVAD.fmk. Inhibition of cell death by indicated concentrations of ponatinib, Nec-1 and RIPK3 inhibitor GSK-872 was determined. Cell viability data are presented as mean ± SD. See also Figure S1.

Results

Discovery of ponatinib as the first-in-class dual inhibitor of RIPK1 and RIPK3

We observed that Glu-in/DLG-out conformation of RIPK1 closely resembles that of Abl (Zhou et al., 2011) (Fig. S1A). Based on this similarity, we screened small panel type 2 tyrosine kinase inhibitors, many of which display potent activity against Abl kinase. The screen identified two molecules, ponatinib and DCC-2036, that efficiently attenuated necroptosis (Fig. S1B). Subsequent in vitro experiments showed that both ponatinib and DCC-2036 inhibited not only RIPK1, but also RIPK3 and another member of RIPK family RIPK2 (Canning et al., manuscript in preparation), identifying them as the first reported pan-RIPK1/2/3 inhibitors (Table 1). Both molecules efficiently inhibited RIPK1/3-dependent necroptosis in TNFα-stimulated FADD-deficient Jurkat cells with activity of ponatinib exceeding that of Nec-1 (Fig. S1B, Table 1). DCC-2036 displayed much poorer (>10-fold lower) cellular activity than ponatinib. We confirmed the in vitro activity of ponatinib by showing inhibition of RIPK1 and RIPK3 in a ³²P auto-phosphorylation assay (Degterev et al., 2008) (Fig. 1C) and of RIPK1 in an HTRF assay (Maki and Degterev, 2013) (Fig. S1C). As a negative control, a different Abl inhibitor, Gleevec (Imatinib), neither inhibited RIPK1 and RIPK3 kinases in vitro (Fig. 1C) nor prevented necroptosis (Fig. 1D).

Table 1.

Inhibition of RIPKs and necroptosis by ponatinib and DCC-2036

Compound	IC50, nM ADPGlo, RIPK2	IC50, nM ADPGlo, RIPK3	IC50, nM ADPGlo, RIPK1	IC50, nM Jurkat cell necroptosis assay (RIPK1/3-dependent)
Ponatinib	14*	1.6	12	34
DCC-2036	520	18	5.7	373
Nec-1	NI	NI	760	210

In vitro kinase assays were performed with recombinant RIPK2 (10 ng), RIPK1 and RIPK3 (20 ng) kinases using ADP-Glo assay (Promega). For necroptosis assay, human FADD-deficient Jurkat cells were stimulated with 10 ng/ml human TNFα for 24 hr. In all cases, activity of compounds was determined using 8- (HEK cells), 10- (kinases) or 11-point (necroptosis) dose response series in duplicate. Curve fitting to calculate IC50 values was performed using GraphPad software.

NI – no inhibition up to 10 μM (maximal concentration in assays).

^*Canning et al., manuscript in preparation.

Ponatinib was also effective in other paradigms of RIPK-driven cell death besides TNF-α-induced necroptosis. Ponatinib afforded potent (IC₅₀=7 nM) protection of immortalized mouse macrophages (iBMMs), undergoing TLR4-induced necroptosis (He et al., 2011) in response to LPS and the pan-caspase inhibitor zVAD.fmk (Fig. S1D). It also protected mouse embryonic fibroblasts (MEFs) stimulated with TNFα in the presence of the TAK1 inhibitor 5z-7-oxozeaenol (5z-7), a combination previously reported to induce RIPK1-dependent but RIPK3-independent apoptosis, rather than necroptosis (Fig. 1E) (Dondelinger et al., 2013). Notably, in both cases ponatinib displayed higher activity than Nec-1 and higher and broader activity than RIPK3 inhibitor GSK-872 (Kaiser et al., 2013), which did not inhibit RIPK1-dependent apoptosis (Fig. 1E).

Identification of RIPK1 kinase-selective analogs of ponatinib

Despite excellent activity against RIPK1/3 kinases, ponatinib’s relative lack of specificity limits its utility as a probe to dissect RIPK1/3-dependent signaling events and raises concerns over the safety of its use as a cytoprotective agent in clinical settings. Thus, we explored strategies to make ponatinib more selective by retaining elements of its scaffold that confer high affinity towards RIPKs, while introducing modifications enhancing selectivity towards RIPK1 and/or RIPK3. We generated a docked model of RIPK1/ponatinib based on the recently described co-crystal structure of ponatinib with a homologous kinase RIPK2 (PDB 4C8B, Canning et al., manuscript in preparation), which revealed potential differences in the binding pocket of RIPK1 vs. RIPK2/Abl around the central phenyl ring of ponatinib (Ring A) (Fig. S2). Namely, RIPK1 contains a smaller hydrophobic pocket accommodating the methyl of Ring A (Ile43, Lys45, Leu90 and Met92 (gatekeeper), Fig. 2A), compared to Abl, RIPK2 and RIPK3, which contain a smaller hydrophilic Thr gatekeeper, but a bulkier DFG motif (Fig. 2B). Notably, the combination of a DLG (rather than DFG) and a medium size hydrophobic gatekeeper (Met) is unique for RIPK1 based on human kinome alignment (http://kinase.com/human/kinome/phylogeny.html). We next tested whether these differences could be exploited to achieve selectivity between RIPK1 vs. Abl/RIPK2/RIPK3. We generated an analog lacking the Ring A methyl group (CS1, Fig. 2C), which showed reduced inhibition for all three RIPKs and Abl (Table 2), consistent with this group making positive, but not critical hydrophobic contacts in the identified lipophilic pocket. Unexpectedly, bulkier substituents in this position (CS2 – CS6) displayed an abrupt loss of activity against Abl, RIPK2 and RIPK3 (RIPK3<RIPK2<Abl) and the tert-butyl (CS6) analog retained activity only against RIPK1 (Table 2). To better understand the selectivity of these analogs, profiling was performed against a panel of 90 wild type human kinases using CS analogs, representing a gradual increase in the size of Ring A’s substituent. These data (Fig. 3A, Table S1) indicated both an increase in selectivity and a general decrease in activity with introduction of bulkier groups on Ring A, which can be expected based on the limited size of the binding pocket. CS6 displayed the highest selectivity against the kinase panel. In particular, it showed no inhibition of RIPK2, ~670-fold lower inhibition of phosphorylated Abl compared to ponatinib, but only ~10-fold reduction in activity against RIPK1 (Fig. 3A, B, Table S1). Overall, this SAR of ponatinib achieved better RIPK1 selectivity, albeit with modestly reduced activity towards RIPK1.

Figure 2. Modeling interactions of ponatinib CS analogs with RIPK1 and RIPK2 kinases.

A) Ring A of ponatinib inserts into the lipophilic pocket formed by aliphatic portions of side-chains of Ile43, Lys45, Leu90 and Met92. The backside of the molecule is aligned with the side chain of Leu157 of DLG motif. B) Alignment of Leu157 of RIPK1 DLG and Phe165 of RIPK2 DFG motifs, Phe165 is in close proximity with the ATP binding pocket moiety of ponatinib. C) Chemical structures of CS analogs of ponatinib. See also Figure S2.

Table 2.

In vitro and cellular activities of ponatinib CS analogs, necrostatins and PN10

Compound	IC50, nM ADPGlo, Abl	IC50, nM ADPGlo, RIPK2	IC50, nM HEKBlue cell assay, (RIPK2-dependent)	IC50, nM ADPGlo, RIPK3	IC50, nM ADPGlo, RIPK1	IC50, nM Jurkat cell necroptosis assay (RIPK1/3-dependent)
Ponatinib	1.4	14*	1.0*	1.6	12	34
CS1	6.7	63	3.6	7.4	42	219
CS2	5.1	45	30	49	13	48
CS3	32	1400	617	2700	33	135
CS4	31	630	472	460	19	75
CS5	181.7	5096	NI	NI	28	471
CS6	34000	NI	NI	NI	26	354
Nec-3	NI	NI	NI	NI	840	260
Nec-1	NI	NI	NI	NI	760	210
Nec-4	NI	NI	NI	NI	330	80
PN10	NI	1400	1193	NI	90	10

Experiments were performed as described in Table 1. In addition, in vitro kinase assays were performed with recombinant Abl (1 ng) using ADP-Glo assay. For RIPK2 cellular assay, human HEK cells expressing NOD2 and NFkB-SEAP reporter were stimulated for 8 hr with 1 μg/ml L18-MDP (Invivogen), followed by detection using QUANTI-Blue SEAP reagent (Invivogen).

NI – no inhibition up to 10 μM (maximal concentration in assays).

^*Canning et al., manuscript in preparation.

Figure 3. Selectivity profiling of CS analogs of ponatinib. Inhibition of a diversity set of 97 kinases (90 wild type kinase and 7 mutants, ScanEDGE, DiscoveRx) by 1 μM inhibitors.

A) Selectivity scores of CS analogs. Selectivity score values reflect number of kinases inhibited by >65% (S35), >90% (S10) or 99% (S1). B) TreeView maps of kinase inhibition by ponatinib and CS analogs. Red circles indicate kinases that were inhibited by the molecules >65%. The diameter of the red circle reversely corresponds to the percentage of kinase activity remaining in the presence of inhibitor (i.e. 0% indicates complete inhibition and corresponds to the largest size of the circle). Green circles indicate kinases that were tested but were inhibited <65%. Full data are presented in Table S1. C) MM-GBSA energy profile analysis reveals unfavorable interactions of CS6 with Abl. Energy changes between free and bound sates of CS6 and residues in the Abl and RIPK1 binding pockets were calculated as described in Methods section. Colors indicate energy changes upon small molecule binding from favorable (blue) to unfavorable (red). Side chains of gatekeeper and DXG residues are shown. D) CS6 poorly inhibits M92T/L157F mutant of human RIPK1 kinase. FLAG-tagged kinases (a.a. 1–327) were expressed in HEK293T cells, immunopricipitated using anti-FLAG beads and used in 32P autophopshosphorylation assays with indicated concentrations of ponatinib (Pon) and CS analogs. E) Inhibition of IFNγ-induced cell death by Ponatinib. RIPK1−/− MEFs were treated with 10 ng/ml IFNγ in the presence of indicated concentrations of Ponatinib, CS4 and CS6 for 24 hr. F) Ponatinib and GSK-872 inhibit poly(I:C)-induced cell death. MEFs were treated with inhibitors, 5 μg/ml poly(I:C) and 50 μM zVAD.fmk for 24 hr. Cell viability data are presented as mean ± SD. See also Figure S3, Table S1.

The selectivity of CS6 for RIPK1 appeared counterintuitive since RIPK1’s bulkier gatekeeper residue (Met) makes its pocket more restrictive (compared to Thr of Abl/RIPK2/RIPK3). Notably, the bulky T315I gatekeeper mutant of Abl was inhibited ~60–70-fold less by CS5 or CS6 compared to ponatinib (Table S1) and was not inhibited by these molecules in the ADPGlo assay (not shown), suggesting that differences in gatekeeper size per se do not explain the selectivity of the CS series towards RIPK1. Another possibility is that the bulkier and more rigid Phe of the DFG (in place of Leu157 of RIPK1 DLG, Fig. 2B) may prevent induced fit accommodating the Ring A with a substituent exceeding a specific size threshold. To further address this question, we calculated the per atom energy contribution to binding for ponatinib and CS6 in RIPK1 and Abl using a MM-GBSA approach (Beard et al., 2013; Hayes et al., 2011) with local hierarchal sampling of the residue conformations in the DXG motif, the gatekeeper residue and the ligand atoms (details in Methods section) (Fig. 3C, S3A). The results indeed indicated that CS6 had an energetically more favorable fit (indicated in blue) in RIPK1 compared to Abl (indicated in red). Furthermore, introduction of Phe residue (L157F and L157F/M92T mutants of RIPK1) rendered CS6 binding to RIPK1 energetically unfavorable (Fig. S3A). To experimentally confirm the role of the DLG, we tested the L157F mutant of RIPK1 in a ³²P autophosphorylation assay. L157F RIPK1 was inhibited poorly by all ponatinib analogs (Fig. 3D). L157F/M92T RIPK1 containing the “Abl/RIPK2/RIPK3” combination of DFG/Thr gatekeeper was inhibited by ponatinib and CS4, but no longer inhibited by CS6, similar to Abl (Fig. 3D, Table 2). Overall, these data suggested that the more flexible DLG allows RIPK1 to accommodate larger substituents attached to the Ring A of ponatinib, while the Met92 gatekeeper restricts the binding pocket, leading to the reduced inhibition of RIPK1 by CS6. These data highlighted that relatively small differences between RIPK1 and other kinases can be exploited to achieve significant gains in selectivity, however, these gains may be limited with respect to the entire kinome.

Inhibition of RIPK3-dependent cell death by ponatinib

Recent evidence suggest that in a number of situations, such as stimulation with interferons, TLR3 agonists and infection with mouse herpes virus lacking endogenous RIP inhibitor (vIRA), necroptosis may by-pass RIPK1 and proceed through direct RIPK3 activation (Dillon et al., 2014; Kaiser et al., 2013; Upton et al., 2012). Along these lines, genetic deletion of RIPK1 in MEF cells was found to promote RIPK3-dependent cell death in response to IFNγ (Dillon et al., 2014; Kaiser et al., 2014). We confirmed that activation of cell death in RIPK1−/− MEFs by IFNγ was dependent on RIPK3 by demonstrating blockade of cell death by the RIPK3 inhibitor GSK-872 (Fig. S3B). Ponatinib (Fig. 3E) efficiently inhibited this form of cell death at ~10-fold lower concentration compared to GSK-872 (Fig. S3B). Importantly, inhibition of cell death was greatly reduced with CS4 and very marginal protection was seen with CS6 (Fig. 3E), consistent with the loss of RIPK3 kinase inhibition in vitro (Table 2). Neither Gleevec nor Nilotininb, which are potent inhibitors of Abl but do not inhibit RIPK1 or RIPK3 (Fig. 1 and data not shown), inhibited IFNγ-induced cell death, excluding a role for Abl in this model of cell death (Fig. S3C). Ponatinib also displayed activity in a second paradigm of RIPK3-specific cell death induced by TLR3 agonist poly(I:C)/zVAD.fmk (Kaiser et al., 2013) (Fig. 3F). CS4 again displayed reduced activity, while CS6 provided only marginal protection on par with that demonstrated by Nec-1. Overall, these data confirmed that ponatinib can inhibit RIPK3 kinase-driven cell death and supported the selectivity of CS analogs for RIPK1 over RIPK3.

Development of Nec-1/ponatinib “hybrid” inhibitors

A comparison of the Glu-in and Glu-out conformations of RIPK1 DLG-out pocket suggested that the latter provides more space for inhibitor binding (Fig. S4A) (187ų vs. 209ų, respectively). The RIPK1•Nec-1 co-crystal structure revealed that Nec-1 assumes a “kinked” conformation (Fig. 1B) (Xie et al., 2013) in the DLG-out pocket, allowing multiple specific affinity-driving contacts within the pocket that would be precluded in a more narrow Glu-in conformation as seen in the RIPK1/ponatinib docked model (Fig. S4B). In contrast, flat hydrophobic moieties present in typical type 2 inhibitors, like ponatinib (Fig. S2), provide a good fit with the narrower Glu-in/DXG-out conformation. Thus, we hypothesized that highly selective Glu-out/DXG-out-targeting groups, such as those present in necrostatins, might provide an excellent complement to the current type 2 inhibitor scaffolds by replacing the less selective DFG pocket binding components of current type 2 inhibitors (Fig. 4A), which may allow us to capture both high activity of inhibitors, like ponatinib, and excellent selectivity of necrostatins.

Figure 4. Development of hybrid PN RIPK1 inhibitors.

A) General design of hybrid PN molecules, combining DLG-out Nec-1, Ring A-containing linker and hinge-binding fragment of ponatinib. B) Activities of PN series of compounds. IC₅₀ values against recombinant RIPK1 kinase were determined using ADPGlo assay using 6-point dose range (5 μM–20 nM) of each compound. IC₅₀ values against necroptosis were determined in in TNF-treated FADD-deficient cells. Experiments were performed independently from those presented in Table 2f. C) Comparison of the binding poses of Nec-1 (from X-ray structure, PDB: 4ITH), ponatinib (from ponatinib/RIPK1 model, Fig. 3) and PN10 (resulting from “unconstrained” Glide docking). Interaction diagrams were generated using Maestro software. PN10 forms three out of 4 targeted hydrogen bonds to Met95 of the hinge and Val76/Asp156 in the DLG-out pocket. D, E) Selectivity of PN10 towards RIPK1. PN10 was screened against ScanEDGE panel (DiscoveRx) of 97 kinases at 1 μM, described in Fig. 3. RIPK1 was the only kinase with >99%. Some inhibition of DYRK1b (79% inhibition) was also observed. No other kinase was inhibited >65% (other kinases in the panel indicated by green circles). Full data are presented in Table S1. F) PN inhibitors display increased inhibition of M92T RIPK1 mutant with Thr gatekeeper. Wild type and M92T mutant were expressed in 293T cells. Proteins were immunopricipitated using anti-FLAG (M2) magnetic beads and used in ³²P autophosphorylation assay at 10 μM. Comparable amounts of wild type and mutant kinases in kinase reactions were confirmed by Western blot. G) Strain penalties were calculated using modified MM-GBSA algorithm for select PN hybrids based on constrained Glide docking to RIPK1. Two different binding poses were observed for PN9 with different strain values. PN10 was the only inhibitor without strain penalty. Notably, M92T mutation eliminated strain penalty in case of PN13, consistent with increased inhibition of this mutant in F). See also Figure S4 and Table S1.

Accordingly, we designed and synthesized a set of ponatinib/necrostatin-1 “hybrid” inhibitors, which we termed the “PN” series. Binding site alignment of ponatinib/RIPK1 (docked model) and Nec-1/RIPK1 (co-crystal structure) revealed that the urea of Nec-1 and the amide of ponatinib both form hydrogen bonds with the backbone of Asp157 of the DLG, providing a convenient point to connect Nec-1 to Ring A of ponatinib (distance ~2.7Å, Fig. S4C). We initially designed three PN molecules (PN1-3), which showed good docking scores by GLIDE XP (Friesner et al., 2004). However, these molecules displayed lower activity towards RIPK1 compared to either Nec-1 or ponatinib (Fig. 4B). We speculated that the reduced activity might have resulted from sub-optimal geometry of the hybrids or from incompatible conformations of the hinge in the Glu-in/DLG-out and Glu-out/DLG-out conformations.

To determine whether the ponatinib portion of PN hybrids made any contribution to RIPK1 binding, we next introduced small changes to this part of the molecule. Resulting hybrids (PN4-6) revealed a sharp SAR for the hinge-binding fragment, suggesting that the ponatinib portion of PNs likely makes contacts in the ATP pocket, but that the geometry of the hybrids still was not optimal. Based on these data, we designed a panel of PNs with a broader range of linkers between the ponatinib and Nec-1 fragments. These molecules were again docked into RIPK1 using GLIDE XP (Fig. S4D) with the added constraint that molecules form hydrogen bonds to the backbone amide of Met95 of the hinge and at least two out of the three hydrogen bonds observed for Nec-1 in the DLG-out pocket Val76 (backbone carbonyl), Leu157 (backbone amide) or Ser161 (side-chain alcohol) (Xie et al., 2013)) to ensure that the hybrid retains contacts to the hinge and a binding mode for the Nec-1 substructure that is consistent with the crystallographic pose (<0.5Å). A much smaller subset of inhibitors that produced docking poses satisfying these criteria were again docked without any hydrogen bond constraints (Fig. S4D). We used two independent docking calculations to ensure that we selected molecules with the appropriate binding mode and did not bias our selection due to the initial hydrogen bonding constraints. As a result, several molecules assuming a binding pose in both docking experiments, comparable to Nec-1/ponatinib (Fig. S4E), and displaying good docking scores (Fig. S4F) were synthesized (Fig. 4B). PN12, which did not fit these criteria, was included to further characterize the effect of the linker length on activity. Excitingly, one molecule in this set (PN10) displayed excellent in vitro activity against RIPK1, exceeding that of Nec-1 (Table 2). In addition, PN10 showed better activity in necroptosis assays than either Nec-1 (~20-fold) or ponatinib (~3-fold), suggesting that we have indeed achieved a good fit for both the ponatinib and Nec-1 fragments in PN10. Most importantly, PN10 displayed excellent selectivity for RIPK1 in a 90 kinases panel screen (Fig. 4D, E). In ADPGlo and HEKBlue assays, some inhibition of RIPK2 was observed, but it was greatly reduced compared to ponatinib (Nec-1 lacks activity against RIPK2) (Table 2). Overall, these data suggested that it is possible to take advantage of the unique properties of both Glu-out/DXG-out inhibitors like Nec-1 (selectivity) and Glu-in/DXG-out inhibitors like ponatinib (binding affinity) to develop both potent and highly selective type 2 inhibitors of RIPK1 kinase.

Surprisingly, while PN10 showed improved in vitro and cellular activity against RIPK1 and necroptosis compared to other necrostatins, it was still a ~9-fold weaker inhibitor than ponatinib in vitro, despite ~3-fold better cellular activity (Table 2). This may reflect differences in the binding modes between necrostatins, including PN10, and ponatinib. We noticed that all necrostatins displayed lower activity in an in vitro kinase assay compared to cellular inhibition of necroptosis (~3–5-fold better IC₅₀ in cells than in vitro) (Table 2). In contrast, ponatinib displayed ~3-fold higher activity in vitro than in cells. We have previously optimized the length of RIPK1’s kinase domain (recombinant RIPK1, a.a. 1–327) to maximize its catalytic activity (Maki and Degterev, 2013; Maki et al., 2013) and, hence, the kinase active Glu-in conformation. The Glu-in conformation creates an additional energy barrier (due to the loss of a highly conserved Glu/Lys ionic bond in the Glu-out conformation) that must be overcome by necrostatins (Fig. 1B). This is not the case for Glu-in inhibitors like ponatinib, explaining poorer than expected performance of necrostatins in the in vitro kinase assay. Notably, another previously described Glu-out inhibitor, the Abl inhibitor PD166326 (Levinson et al., 2006), was similarly ~20-fold less active in vitro compared to cellular assays (Huron et al., 2003).

A potential steric clash with the bulky gatekeeper presents a major challenge for the design of optimal type 2 inhibitors as demonstrated by the inability of most type 2 Abl inhibitors to efficiently target the T315I gatekeeper mutant (Zhou et al., 2011). We found that the bulky Met92 gatekeeper of RIPK1 presented a similar challenge for RIPK1 and was likely responsible for the lower observed activities of PN7, PN9 and PN13 against RIPK1 in vitro and in cells (Fig. 4B) than could be expected from their high docking scores (Fig. S4F). Replacing the Met92 residue of RIPK1 with a smaller Thr residue resulted in much better inhibition by these molecules, which was now on par with PN10 (Fig. 4F). Since a noticeable difference in inhibition of wild type and M92T RIPK1 was still seen even in the case of PN10, additional optimization of PN10 inhibitory activity through improving the fit with the Met92 residue may be possible by further changes to the linker connecting the necrostatin to the hinge binding group.

Improved inhibition of necrosome formation and TNFα synthesis and reduced cytotoxicity of CS and PN molecules

Next, we selected CS4, CS6 and PN10 for additional analyses. First, we sought to further establish the mode of action of these inhibitors by evaluating inhibition of RIPK1/RIPK3 “necrosome” complex formation using a co-IP assay in MEF cells stimulated with TNFα/cycloheximide/zVAD.fmk (TCZ) (Thapa et al., 2013). All ponatinib-based inhibitors efficiently blocked cell death in this system at substantially lower concentrations than those required for Nec-1 (Fig. 5A, B) and the RIPK3 inhibitor GSK-872 (Fig. S5A). Maximal protection by ponatinib itself was somewhat weaker than by CS4, CS6 and PN10 (~60% vs. ~85%), likely reflecting off-target activities of ponatinib at concentrations approaching 1 μM. In agreement with a role for RIPK1 kinase activity in TCZ-driven necrosome formation (Cho et al., 2009), Nec-1 efficiently blocked necrosome assembly in TCZ-treated MEFs (Fig. 5C). However, necrosome assembly was not affected by two distinct RIPK3 inhibitors, GSK-843 and GSK-872 (Kaiser et al., 2013) (Fig. S5B), despite both molecules efficiently protecting cells from TCZ-induced cell death (Fig. S5A and not shown), suggesting that in TCZ-treated MEFs necrosome formation does not require the kinase activity of RIPK3. CS4, CS6 and PN10 all efficiently inhibited formation of necrosome complex at lower concentrations compared to Nec-1 (Fig. 5C). Ponatinib again displayed substantial, but lower activity, consistent with weaker inhibition of cell death. Overall, these data confirmed inhibition of RIPK1-dependent necrosome formation as a cellular target of the ponatinib-based inhibitors.

Figure 5. Inhibition of necroptosis and inflammation by ponatinib analogs.

A–B) Inhibition of TCZ-induced necroptosis in MEFs by various RIPK inhibitors. Cells were treated with 50 ng/ml mouse TNFα, 200 ng/ml cycloheximide and 25 μM zVAD.fmk for 18 hr in the presence of indicated concentrations of inhibitors. C) Ponatinib analogs inhibit necrosome formation in TCZ-stimulated MEFs. Cells were treated with 50 ng/ml mouse TNFα, 200 ng/ml cycloheximide, and 25 μM zVAD.fmk for 6 hr in the presence of indicated concentrations of inhibitors, followed by RIPK3 immunopricipitation.

D–E) Ponatinib inhibits RIPK1-dependent TNFα mRNA upregulation in FADD-deficient Jurkat (D) and iBMM (E) cells. Jurkat cells were stimulated with 10 ng/ml human TNFα for 8 hr in the presence of indicated concentrations of specific RIPK1 kinase inhibitor Nec-1, ponatinib and indicated PN and CS analogs of ponatinib. Changes in TNF mRNA relative to GAPDH were determined by qRT-PCR. iBMM cells were stimulated with 10 ng/ml LPS and 50 μM zVAD.fmk for 7 hr. F–G) Ponatinib analogs block TNFα toxicity in vivo. Survival curves are shown in F). Reduction in TNF-induced tissue injury and inflammatory response was confirmed by assessing the circulating levels of injury markers (AST, LDH, CK) and mouse IL-6 in G). n=3–5 mice per group. * p<0.05 for TNF/inhibitors vs. TNF group. H–I) Ponatinib and PN10 display higher activity than Nec-1 in vivo. Survival curves are shown in H). Reduction in TNF-induced tissue injury and inflammatory response was confirmed by assessing the circulating levels of injury markers (AST, LDH, CK) and mouse IL-6 in I). n= 3–6 mice per group. * p<0.05 for TNF/inhibitors vs. TNF group, ** p<0.015 for TNF/ponatinib and TNF/PN10 groups vs. TNF/Nec-1 group. Injury marker and IL6 levels data are presented as mean ± SD. See also Figure S5.

Activation of cell death is not the only function of RIPK1 kinase. It has also been found to promote synthesis of TNFα at the mRNA level independent of cell death regulation (Biton and Ashkenazi, 2011; Christofferson et al., 2012; McNamara et al., 2013). This and other cell death-independent pro-inflammatory activities of RIPK1 kinase are also emerging as potentially clinically relevant targets. Thus, we sought to establish whether the activity of ponatinib-based inhibitors extends beyond cell death regulation by RIPK1. TNFα stimulation of Jurkat cells and LPS/zVAD.fmk stimulation of immortalized macrophages led to robust increase in TNFα mRNA, inhibited by Nec-1. Ponatinib, CS4, CS6 and PN10 again efficiently inhibited this cell death-independent function of RIPK1 kinase (Fig. 5D, E), revealing potentially important anti-inflammatory properties of ponatinib and its RIPK1-selective derivatives.

Finally, ponatinib displayed significant cytotoxicity at concentrations >1 μM (Fig. S5C–E). We found that the increased kinase selectivity of CS molecules and PN10 also translated into substantially lower cytotoxicity (Fig. S5C–E), improving one of the significant limitations of ponatinib as a cytoprotective and anti-inflammatory agent.

Inhibition of in vivo TNFα toxicity by ponatinib analogs

To evaluate the potential therapeutic efficacy of ponatinib-based inhibitors, we examined their ability to counteract toxicity of TNFα in vivo, which was previously shown to reflect activation of necroptotic RIPK1/3 signaling (Duprez et al., 2011). Intraperitoneal injection of Nec-1 at 1 mg/kg dose provided significant protection from injury, reduced circulating levels of IL-6 and prevented death of the animals, consistent with previous reports (Duprez et al., 2011; Takahashi et al., 2012). Ponatinib completely prevented toxicity of TNFα, revealing for the first time the unexpected cytoprotective properties of this anti-cancer agent in vivo (Fig. 5F, G). PN10, CS4 and CS6 were also fully protective at 1 mg/kg dose (Fig. 5F, G). At lower 0.4 mg/kg dose, both ponatinib and PN10 displayed significantly higher activity than Nec-1 with respect to survival and injury markers (Fig. 5H, I). The improved activity of PN10 over Nec-1 demonstrated that our guided approach has indeed led to the development of significantly improved selective in vivo inhibitor of RIPK1-dependent cell death and inflammation.

Discussion

RIPK1 and RIPK3 first emerged as acting in concert in activating necroptosis (Cho et al., 2009). However, more recent genetic and pharmacologic evidence demonstrated that these two proteins may possess multiple non-overlapping functions in the regulation of inflammation (Cuda et al., 2014; Lukens et al., 2013), apoptosis (Dondelinger et al., 2013) and necroptosis (Dannappel et al., 2014; Dillon et al., 2014; Kaiser et al., 2014; Orozco et al., 2014). This array of functions has inspired us and others to pursue development of RIPK1 (Degterev et al., 2008; Harris et al., 2013) and RIPK3 (Kaiser et al., 2013) inhibitors. Our current finding that ponatinib dually targets RIPK1/3 represents a unique and important property of this molecule, making it a useful tool compound to further evaluate therapeutic consequences of inhibiting pathologic RIPK signaling, where multiple mechanisms of RIPK1/3-dependent cell death may be activated simultaneously in different cell populations, depending on the specifics of their state or individual regulation. The lack of selectivity and reported safety concerns (http://www.fda.gov/safety/medwatch/safetyinformation/safetyalertsforhumanmedicalproducts/ucm370971.htm) may exclude broad use of ponatinib as a cytoprotective and anti-inflammatory agent. However, cancer-associated inflammation could be one specific area where the ability of ponatinib to block RIPK1/3 could be of immediate value. Inflammatory mediators, including cytokines, microbial PAMPSs/DAMPs, and carcinogenic agents such as asbestos fibers promote tumorigenesis by contributing to an inflammatory microenvironment in certain human cancers (Elinav et al., 2013). As many of these pro-inflammatory agents have also been shown to activate RIPK1/3 kinases (Vanlangenakker et al., 2012), ponatinib may help reveal functions for RIPKs in cancer-associated inflammatory signaling and facilitate translation of these results into clinical benefits.

The discovery of RIPK1/3 activity of ponatinib prompted us to expand its SAR via two different approaches to achieve RIPK1 selective inhibitors. Our studies with the CS series revealed an unexpected induced fit mechanism for inhibition of RIPK1. We found that increasing the size of the phenyl ring (Ring A) substituent of ponatinib from i-propyl to t- or c-butyl led to an abrupt “activity cliff” resulting in selective inhibition of RIPK1 compared to other RIPKs and Abl (Table 2, S1). While statically RIPK1 contains a more restricted Ring A pocket as demonstrated by poor activity of ponatinib and CS molecules against L157F mutant of RIPK1 (Fig. 3D), this paradox can be explained by the greater plasticity or RIPK1 due to the presence of less bulky and more conformationally flexible DLG motif, that allows RIPK1 (but not DFG kinases Abl, RIPK2 or RIPK3) to accommodate the bulkier Ring A through induced fit. Additional specific differences in the packing of the activation loop and geometry of the Ring A pocket likely further differentiate affinities of ponatinib analogs towards RIPK2 and RIPK3 vs. Abl as we have observed that these kinases, unlike Abl are poorly inhibited even by the isopropyl CS4 analog. Overall our data reveal a possible direction for increasing selectivity of type 2 inhibitors by taking advantage of the differential flexibility of DXG pockets. Strikingly, only 4 out of 44 (9%) DFG-motif containing ponatinib-inhibited kinase targets were efficiently inhibited by CS6 (>65% at 1 μM), while this was the case for 3 out of 4 (75%) non-DFG kinases (Table S2). Thus, our findings may be particularly applicable for targeting non-DFG kinases.

While small changes to the ponatinib scaffold represent one direction for achieving additional activities for ponatinib-like molecules, we show that a fragment approach combining Nec-1 with the ATP pocket-binding moiety of ponatinib (e.g. PN10) led to a much greater improvement in selectivity and cellular activity, compared to both Nec-1 and ponatinib (Table 2). Several conclusions can be made based on the SAR of the PN series. First, despite ponatinib (and other type 2 inhibitors) binding to Glu-in/DXG-out kinase conformations, dictated by interactions of its central amide group with a Glu residue in the αC helix (Liu and Gray, 2006; Zhao et al., 2014; Zhou et al., 2011), its hinge binding fragment appears to be compatible with the Glu-out/DLG-out conformation of RIPK1 required for Nec-1 binding (Xie et al., 2013). It is unlikely that PN10 binds to the Glu-in/DLG-out conformation, as this would be expected to cause a steric clash between the Met67 residue in the αC helix of RIPK1 and the hydantoin of Nec-1 (Fig. S4B). Second, we found that multiple PN analogs, selected using constrained GLIDE XP docking, underperformed due to clashes with the Met92 gatekeeper (Fig. 4F). Furthermore, we calculated strain penalties for PN analogs using a modified MM-GBSA algorithm to understand how PN series activities can be predicted computationally. This calculation provides a cumulative index of various strains occurring in the ligand upon constrained docking to the target. We observed that PN10 was the only analog lacking strain, consistent with its highest affinity (Fig. 4G). Furthermore, strain of the less active analog PN13 was eliminated in the M92T mutant of RIPK1, consistent with comparable inhibition of this mutant by PN10 and PN13 (Fig. 4F). Overall, our data suggest that structural data available for RIPK1 in combination with GLIDE XP docking and strain analysis are sufficient for predicting RIPK1 binding properties by hybrid molecules with a high degree of fidelity. Finally, previous data showed that Nec-1 is a uniquely selective kinase inhibitor (Christofferson et al., 2012) and changes to most positions of this molecule lead to the loss of activity (Teng et al., 2005), limiting options for further optimization. Our data identify the direction for improving Nec-1 activity without sacrificing specificity. The development of PN10 captured most of ponatinib’s activity in vivo, but also provided much improved kinase selectivity, characteristic for Nec-1. High cellular activity of PN10, its selectivity towards RIPK1 and potent inhibition of TNFα toxicity in vivo warrant further evaluation of the therapeutic potential of this molecule in other pre-clinical models of human pathology, where the contribution of RIPK1 has been already established (Linkermann and Green, 2014). Our data with M92T RIPK1 also indicate that steric hindrance with the Met92 gatekeeper still exists for PN10, suggesting that further improvement of this molecule may be possible through finding a better fit between the gatekeeper and Ring A of PN10. In conclusion, we wish to highlight the potential broad applicability of this design approach to other kinases targeted by ponatinib, with the goal of obtaining “hybrids” retaining PN10’s binding mode, but containing modifications to the necrostatin moiety to fit the DXG pockets of these other kinases.

Experimental Procedures

Reagents and chemicals

All the reagents and chemicals were purchased from Sigma, Fisher, or VWR unless otherwise stated. RIPK1, actin and tubulin antibodies were purchased from Cell Signaling Technologies, RIPK3 antibody was from ProSci.

Cell lines

FADD-deficient Jurkat cells and HEK293T cells were purchased from ATCC. HEKBlue cells were purchased from Invitrogen. Immortalized bone marrow-derived macrophages (iBMM) were a gift of Dr. Kate Fitzgerald (UMass Medical Center, Worcester, MA). Cells were maintained in RPMI (Invitrogen, Jurkat cells) or DMEM (Fisher) supplemented with 10% FBS (Sigma) and 1% antibiotic-antimycotic mix (Invitrogen). RIPK1^+/+ and RIPK1^−/− MEFs were a kind gift of Drs. William Kaiser and Ed Mocarski (Emory University). RIPK3^+/+ and RIPK3^−/− MEFs were generated from timed crosses of C57Bl/6 or RIPK3^−/− mice (gift of Dr. Vishva Dixit, Genentech). MEFs were cultured in DMEM supplemented with 10% FBS (Hyclone) and 1% antibiotic-antimycotic mix.

Animals

All animal experiments were performed with approval of the IACUC of Tufts University and Fox Chase Cancer Center. Female C57BL6/J mice (6–8 weeks old) were purchased from approved vendor (Charles River labs). Animals were housed in cages with a light/dark cycle. All efforts were made to minimize the numbers of animals and their suffering.

TNF-induced injury

To study the effect of the compounds in TNF induced injury in mouse model, female C57BL6/J mice were injected i.p. With 100 μL of vehicle (25% PEG400 in PBS) or the compounds (Nec-1, ponatinib, PN10, CS4, and CS6, formulated in the above vehicle) at the doses of 1mg/kg and 0.4mg/kg. 15 min later, animals were injected i.v. through the tail vein with either 5 μg/mouse of recombinant mouse TNFα (Cell Guidance System) or sterile PBS in a volume of 100 μl. Blood samples were collected 30 hr after injection or earlier if moribund through submandibular bleed. Serum samples were submitted to IDEXX laboratories for analysis of AST, CK and LDH levels. IL-6 ELISA was performed using 25 μL of serum using mouse IL-6 ELISA kit (Meso Scale Discovery). Statistical analysis was performed using Student’s t-test (GraphPad Prism 5).