Controlling the Covalent Reactivity of a Kinase Inhibitor with Light

Abstract

Covalent kinase inhibitors account for some of the most successful drugs that have recently entered the clinic and many others are in preclinical development. A common strategy is to target cysteines in the vicinity of the ATP binding site using an acrylamide electrophile. To increase the tissue selectivity of kinase inhibitors, it could be advantageous to control the reactivity of these electrophiles with light. Here, we introduce covalent inhibitors of the kinase JNK3 that function as photoswitchable affinity labels (PALs). Our lead compounds contain a diazocine photoswitch, are poor non-covalent inhibitors in the dark, and becomes effective covalent inhibitors after irradiation with visible light. Our proposed mode of action is supported by X-ray structures that explain why these compounds are unreactive in the dark and undergo proximity-based covalent attachment following exposure to light.

Graphical Abstract

To increase the tissue selectivity of kinase inhibitors, we introduce the combination of covalent inhibitors, targeting a cysteine in the vicinity of the ATP binding site of the c-Jun N-terminal kinase 3 (JNK3) and diazocine photoswitches. X-ray crystallography shows that our lead inhibitor is a poor non-covalent inhibitor in the dark and undergoes proximity-based covalent attachment following irradiation with visible light.

The development of protein kinase inhibitors is one the most active areas of medicinal chemistry. Early on, selectivity amongst the more than 500 members of the human kinome was considered a major challenge since the ATP binding site of protein kinases is highly conserved. In recent years, however, a wide range of selective inhibitors have emerged that have put such concerns largely to rest. One strategy that has proven to be particularly successful is the covalent engagement of native cysteines that are present in the vicinity of the ATP binding site with suitable electrophiles. Several covalent kinase inhibitors that target the “cysteinome”, such as afatinib, ibrutinib, osimertinib, and neratinib, are now clinically approved (Figure 1A).^[1] Most of them feature an acrylamide “warhead” that is poorly reactive towards water and other weak nucleophiles but undergoes rapid bioconjugation following non-covalent binding and placement into the trajectory of a suitable cysteinate anion.^[2,3]

Figure 1.

A) Representative FDA-approved covalent kinase inhibitors binding irreversibly to specific cysteine residues. B) Photoactivation through caging. C) Photoactivations through photoswitching.

To target covalent inhibitors to kinases expressed in specific cells and tissues, one could take advantage of the temporal and spatial precision that light affords. This could be done, for instance, with a caging group that masks the part of the molecule that binds noncovalently (Figure 1B).^[4,5] Alternatively, the covalent reactivity could be concealed with a photoreactive group that delivers a suitable electrophile upon irradiation. Although caging groups for electrophiles, such as α-haloketones,^[6] have been reported, they have not been applied, to the best of our knowledge, to covalent kinase inhibitors. A third alternative for the design of light activatable covalent inhibitors is shown in Figure 1C. It involves a photoswitch that keeps the electrophile away from the nucleophile in the dark form but allows for proximity-based rate enhancement and efficient bioconjugation after photoisomerization. Photoswitchable tethered ligands (PTLs) and photoswitchable affinity labels (PALs) have been reported for the optical control of a wide variety of enzymes and receptors that feature engineered cysteines or native electrophiles, respectively.^[7]

Non-covalently binding photoswitchable kinase inhibitor have been explored, but their development has proven to be challenging.^[4,8–13] This prompted us to focus on covalent inhibitors, which have advantages in terms of residency time and selectivity. Amongst these, compounds that target the cysteinome have proven to be most effective, due to the nucleophilicity of this residue and its occurrence close to the ATP binding site in a variety of important kinases.^[3]

We report here the development of covalent inhibitors for the MAP kinase JNK3 that function as PALs targeting a native cystine close to the ATP binding site (Cys154). Our lead compound contains a diazocine photoswitch, is a comparatively weak noncovalent inhibitor in the dark, and becomes an effective covalent inhibitor of JNK3 upon irradiation with 400 nm light.

The MAP kinase c-Jun N-terminal kinase 3, abbreviated as JNK3, is mainly expressed in the central nervous system, testes, and pancreatic β-cells, in contrast to its ubiquitously expressed isoforms JNK1 and JNK2.^[14] JNK1–3 are key signaling enzymes in the cellular stress response and JNK3 has been targeted for the treatment of neurodegenerative diseases, including Alzheimer’s, Huntington’s and Parkinson’s disease, and type 2 diabetes.^[14,15]

The development of reversible and selective JNK inhibitors that bind non-covalently has proved to be difficult.^[16] However, JNK3 features Cys154 in the vicinity of the ATP binding site that has been exploited for the covalent attachment of inhibitors with improved kinome selectivity.^[17,18] So far, no JNK inhibitor has progressed to clinical testing. Pharmacological intervention is further complicated by the nature of JNK signaling, where the time course of either kinase activity or inhibition is crucial in controlling the cellular outcome, which can range from cell survival to apoptosis.^[19,20,15] Given the importance of spatiotemporal control of the JNK pathway,^[21,22] implementation of light-sensitivity into our JNK inhibitors represents a compelling strategy to further improve their utility.

Our photoswitchable inhibitors are based on the known ligand 1, which combines a pyridinylimidazole hinge-binding motif with a diarylamide spacer that connects to a cysteine-reactive acrylamide (Figure 2A). It is important to preserve the pyridinylimidazole moiety binding the ATP-pocket, since incorporation of a photoswitch into this crucial motif would negatively influence potency and selectivity of our inhibitor.^[23] Therefore, we replaced the central diarylamide of 1, which does not engage in hydrogen bonding, as shown by a recent structure, with an isosteric azobenzene^[17].

Figure 2.

Design and synthesis of the covalent photoswitchable inhibitors. A) Azologization of the known covalent inhibitor 1 and sign inversion to yield diazocine 3. B) Additional photoswitchable JNK3 inhibitors prepared. C) Synthesis of diazocine 3. D) Absorption spectra of inhibitors 2 and 3 in the dark or after 400 nm irradiation for 3 minutes in DMSO and thermal characterization of 2 and 3 in a 1:1 mixture of DMSO and PBS (pH 7.4).

This afforded the photoswitchable JNK inhibitor 2, which was designed to be reactive in the dark adapted trans state and remain a non-covalent inhibitor when irradiated to preferentially adopt the cis form. To ensure that the photoswitchable inhibitor would be unreactive in the dark and would become active only after irradiation, we next implemented a sign-inversion approach,^[24] replacing the azobenzene moiety with a diazocine photoswitch. Diazocines are known to be more stable in the bent cis form and isomerize to the thermodynamically less stable elongated trans form upon isomerization with violet-blue light.^[25] This strategy led to the light-activated JNK3 inhibitor 3, which is shown both in its cis form and its trans form in Figure 2A. Molecular docking confirmed that both isomers should be able to bind to the ATP-pocket of JNK3 but only the metastable trans form would be able to reach and covalently bind to Cys154. In addition to this, we designed a version with an (E)-dimethylaminocrotyl amide electrophile, which is known to improve solubility and reactivity, yielding inhibitor 4. The half-life, absorption spectra and photostationary state (PSS) are strongly dependent on azobenzene substituents, which are in part dictated by the requisite pharmacophore and the appended electrophile. Therefore, we also explored a version with a meta-meta substitution pattern in the diazocine, i.e. compound 5. Finally, we designed a control compound 6, which bears a propionamide and is therefore not cysteine-reactive.

The representative synthesis of inhibitor 3 is shown in Figure 1C. It started with ethylene dianiline 7, which underwent oxidative cyclization to afford diazocine 8.^[26] Slow addition of the oxidant favors the desired intramolecular Baeyer-Mills reaction and avoids overoxidation (see Supporting Scheme 1).^[26] Desymmetrization through Buchwald coupling with tert-butyl carbamate yielded 9, which underwent deprotection and acylation to afford acrylamide 10. Another Buchwald coupling with the known pyridinylimidazole ligand 11^[18,23] yielded the desired photoswitchable inhibitor 3. Covalent inhibitors 4 and 5, as well as the propionamide control compound 6 were prepared in a similar fashion (see Supporting Information).

With compound 2-6 in hand, we turned to their photophysical characterization (Figure 2D, S1). In keeping with the general photophysical behavior of substituted diazocines, the absorption bands are not well separated and in particular, amino substituted diazocines show only partial cis-to-trans isomerization at the 400 nm PSS.^[26] However, this is not a functional flaw for a PAL since the PSS is a dynamic equilibrium and continued pulse irradiation will therefore enrich the covalently bound inhibitor over time. The meta-meta-substituted diazocine 5 showed lower extinction coefficients than the corresponding para-para substituted diazocines 3 and 4 (Figure S1).

Relaxation to the thermodynamically favored cis state could be achieved via irradiation with longer wavelengths (500 – 600 nm) or thermally. The latter occurs on a timescale of minutes to hours in DMSO or mixtures of DMSO and PBS (Figure 2D, S1). In aqueous environments, amino substituted azobenzenes such as 2 relax within seconds whereas the corresponding diazocines are substantially more stable in their trans states.^[24,27]

Following their photophysical characterization, the inhibitors were tested for their ability to inhibit JNK3 in vitro (Figure 3, Supporting Table 1). This was determined through measuring the phosphorylation of the immobilized kinase substrate ATF-2 at different inhibitor concentrations after incubation for 50 minutes. Kinase activity was determined with an ELISA assay, following a previously published method.^[28] Photoswitches were either pulse-irradiated every 5 s for 100 ms or kept in the dark during the 50 min reaction time. As expected, the parent inhibitor 1 did not respond to 400 nm irradiation. No light-dependency was observed for its azostere 2, which had reduced potency compared to 1. The lack of response to irradiation was in line with the fast relaxation of a red-shifted p-amino substituted azobenzene and the reactivity of the acrylamide in the thermodynamically stable trans form of the azobenzene.

Figure 3.

In vitro characterization. Quantitative ELISA assay of differential ATF-2 phosphorylation by JNK3 for 50 min at 37 °C in the dark or with pulsed irradiation (100 ms every 5 s) and inhibitors: A) 1 B) 2 C) 3 D) 4 E) 5 F) 6.

Photoswitchable inhibitor 3, which incorporates a para, para-substituted diazocine, was a weak inhibitor of JNK3 in the dark (IC₅₀ = 646 nM) but showed much stronger inhibition at all tested concentrations after pulse-irradiation with 390 nm light (IC₅₀ = 21.4 nM). Compound 4, which bears a dimethylaminocrotyl amide as electrophilic warhead, showed improved solubility but showed JNK3 inhibition with similar potency in the dark (IC₅₀ = 507 nM) and after irradiation (IC₅₀ = 31.9 nM). The meta, meta-substitution pattern on the diazocine in 5 gave similar results compared to 3, with a pronounced increase in JNK3 inhibition upon isomerization to its more potent trans form after 400 nm irradiation. Interestingly, once isomerization and potential covalent attachment took place, inhibition by compounds 3 and 5 was found to be irreversible. Use of an inactivating 565 nm pulse after the activating light pulse did not restore kinase activity (Figure S2).

Thus, it appears that binding to the ATP binding site was too tight to pull the pyridinylimidazole from the active site. Indeed, non-covalent inhibitors that contained this motif did bind to JNK3 and p38α in the low nanomolar concentration range.^[23] Alternatively, covalent attachment of the inhibitor could inhibit kinase function even if the ligand does not occupy the ATP-pocket. The non-covalent, propionamide control compound 6 only showed a small increase in JNK3 inhibitory activity upon 400 nm irradiation. This small activity difference was probably due to an increased steric clash of the cis isomer with the glycine-rich loop which was in agreement with a non-covalent mechanism of inhibition where isomerization only changes positioning of the propionamide.

We demonstrated cellular target engagement of the inhibitors using the NanoBRET tracer displacement assay^[29] and determined the IC50s to be in the high nanomolar to low micromolar range (Figure S3). The observed improvement in potency after 400 nm irradiation was not as prominent as in the in vitro assay. However, the potent cellular activity of the inhibitors harboring the diazocine moiety demonstrated target engagement, suggesting that these inhibitors represent lead structures for further development, including in vivo studies.

To confirm covalent labelling, we performed mass spectrometry of JNK3 and inhibitor 4 incubated under pulsed 400 nm irradiation (Figure S4). A mass shift corresponding to the molecular weight of 4 was observed, demonstrating covalent adduct formation in a 1:1 ratio.

To gain further insight into the mode of action of our compounds, we carried out structural studies using X-ray crystallography. To this end, we soaked crystals of JNK3 with inhibitor 4, which showed improved solubility, whilst pulse-irradiating with 400 nm light. We collected several datasets after different irradiation schedules and observed clear electron density that supported a non-covalent as well as covalent binding mode of 4 (Figure 4A, B and C).

Figure 4.

X-ray analysis of 4 bound to JNK3. A) Structural overlay of 1 and 4 covalently bound to JNK3. B) The bent cis isomer cannot reach Cys154 and only binds non-covalently. C) The extended isomer binds covalently to Cys154. The crystal structures have been deposited to the wwPDB with accession codes 7ORE and 7ORF.

Both isomers have the ATP-mimetic pyridinylimidazole moiety in the same position which was interacting with the hinge region, the adenine pocket, and the hydrophobic pocket in a similar manner to the same moiety of the starting compound 1 (Figure 4A). However, the electrophilic warhead reached Cys154 only in the elongated trans isomer but not in the bent cis form present in the dark (Figure 4 and Supporting Figure S5). This enabled covalent adduct formation, which was not feasible with the cis isomer. As such, inhibitor 4 operated in the manner indicated in Figure 1C.

In summary, we developed covalent kinase inhibitors that can be activated by irradiation and function as photoactivated affinity labels. Notably, our approach has advantages over caged inhibitors, since it largely retains the properties of the parent drug, whereas a caging group adds significant molecular weight and can negatively impact cellular uptake and distribution. Additionally, the uncaging reaction generates byproducts, whereas our inhibitors just undergo a reversible isomerization. While we were unable to show reversibility on target, an unbound inhibitor can still revert to its inactive form, whereas an uncaged inhibitor will always remain active.

Our most successful inhibitor incorporates a new class of photoswitch, viz. diazocines, a moiety that is increasingly used to convert compounds that are constitutively inactive in the dark into light-activated ones. X-ray crystallography provides further insight into the mode of action of a photoswitchable inhibitor bound to its target.^[30,31] Our strategy could be easily transferred to other kinase targets that harbor a cysteine in positions close to the ATP binding site once suitable inhibitors and linkers have been developed and may therefore serve as a prototype for future photoswitchable covalent kinase inhibitors.