Conformational plasticity of the ULK3 kinase domain

The human protein kinase ULK3 regulates the timing of membrane abscission, thus being involved in exosome budding and cytokinesis. Herein, we present the first high-resolution structures of the ULK3 kinase domain. Its unique features are explored against the background of other ULK kinases. An inhibitor fingerprint indicates that ULK3 is highly druggable and capable of adopting a wide range of conformations. In accordance with this, we describe a conformational switch between the active and an inactive ULK3 conformation, controlled by the properties of the attached small-molecule binder. Finally, we discuss a potential substrate-recognition mechanism of the full-length ULK3 protein.


Introduction
ULK1 was the first subfamily member to be described 7 , since then this isoform has been subjected to intense research. It has been found to be a key regulator of autophagy, in that it links nutrient sensing by the mammalian target of rapamycin complex 1 (mTORC1) to the induction of the early autophagosome, a complex of ULK1 and its binding partners ATG13, ATG101 and FIP200. 8 This complex initiates autophagy, a survival mechanism comprising the orderly degradation of cellular components. ULK1 and the homologous ULK2 are thereby linked to the development of human diseases such as tumorigenesis 9 and neurodegeneration 10 .
In addition to modulating the timing of membrane abscission, ULK3 was suggested to play a role in several signalling pathways. Though it is not required for autophagy induced by low nutrient levels, studies on the Drosophila ULK3 ortholog Another drosophila Unc-51-like kinase (ADUK) demonstrated its requirement for autophagy induced by chemical stress. 11 In human fibroblasts, overexpression of ULK3 induced autophagy and senescence, suggesting that the role of ULK3 in stress-induced autophagy is conserved from fly to man. 12 ULK3 has also been described as a regulator of hedgehog signalling by binding to the suppressor of fused (Sufu), a protein required for negative regulation of GLI proteins. 13 Furthermore, ULK3 was reported to directly phosphorylate GLI2. 14 In genome-wide association studies (GWAS), genetic alterations in ULK3 have been linked to blood pressure and hypertension 15,16 and mood states 17 .
To date, eight high-resolution structure models of the related ULK1/2 kinase domains have been made available in the PDB. [18][19][20][21] All of them are similar in that the kinase domains adopt the canonical active-like conformation, and in that a broad-specificity type-1 inhibitor is bound in the active site. [18][19][20][21] Notably, nothing is known about ULK1/2 plasticity and about possible inactive conformations. In contrast to the other ULK subfamily members, ULK4 is a pseudokinase. Several sequence motifs that are typically found in kinases and that are indispensable for catalytic activity are not conserved in ULK4. Two ULK4 structure models have recently been published, one showing ULK4 in complex with its physiological cofactor ATP 22 , and one in complex with a low-affinity inhibitor 23 . Again, there is no indication for domain plasticity -ATP binds tightly to the pseudoactive site and condenses ULK4 as also shown in MD simulations. 22 In contrast, structural information on the ULK3 and STK36 kinase domains is still lacking. The understanding of their complex functional roles both in physiology and in pathophysiology will be aided by high-resolution structure models that provide insight into their regulation as well as by selective chemical probes that can be used in cellular model systems or in vivo to study ULK function. 24,25 Here, we report screening data of ULK3 against a set of clinical kinase inhibitors as well as crystal structures of ULK3 in complex with type-1 and type-2 inhibitors providing a basis for the rational development of more selective ULK3 inhibitors targeting the active as well as the inactive state. Due to the diverse functions of ULK3 in membrane Downloaded from http://portlandpress.com/biochemj/article-pdf/doi/10.1042/BCJ20210257/915717/bcj-2021-0257.pdf by guest on 30 June 2021 Biochemical Journal. This is an Accepted Manuscript. You are encouraged to use the Version of Record that, when published, will replace this version. The most up-to-date-version is available at https://doi.org/10.1042/BCJ20210257 abscission, stress induced autophagy as well as regulation of hedgehog signalling, selective chemical probes will represent valuable tools for the exploitation of this interesting kinase as a potential drug target.

Methods
Cloning. The DNA coding for a His 6 -tag, a TEV cleavage site and the ULK3 residues 2 to 277 was synthesised (Genscript, Figure S1) and cloned into the expression vector pET-28a, using the NcoI and XhoI restriction sites. The plasmids for ULK1 1-283 expression 19 and ULK2 1-277 expression 20 were kind gifts from Kevan Shokat's group at UCSF.
Expression and purification. The respective expression plasmid was transformed into BL21(DE3) Competent E. coli (NEB), and expression was performed as previously described 26 . For ULK3 2-277 purification, bacteria were then resuspended in lysis buffer (50 mM HEPES pH 7.4, 500 mM NaCl, 20 mM imidazole, 0.5 mM TCEP, 5% glycerol) and lysed by three passages through the high-pressure cell breaker. The lysate was cleared by centrifugation and loaded onto a Ni NTA column. After vigorous rinsing with lysis buffer, the His 6 -tagged protein was eluted in lysis buffer containing 300 mM imidazole. While the protein was subjected to dialysis to remove the imidazole, the N-terminal tag was cleaved by TEV protease.
Contaminating proteins, the cleaved tag and TEV protease were removed by another Ni NTA step. Finally, ULK3 2-277 was concentrated and subjected to gel filtration using an AKTA Xpress system combined with a HiLoad 16/600 Superdex 200 pg gel filtration column. The elution volume 94.4 mL indicated the protein to be monomeric in solution. The final yield was 15 mg ULK3 2-277 /L TB medium. ULK1 1-283 purification 19 and ULK2 1-277 purification 20 was performed as previously described.
Differential scanning fluorimetry (DSF). The assay was performed according to a previously established protocol. 27 Briefly, a solution of 2 μM ULK protein in assay buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 0.5 mM TCEP, 5% glycerol) was mixed 1:1000 with SYPRO Orange (Sigma). The compounds to be tested were added to a final concentration of 10 μM. 20 μL of each sample were placed in a 96-well plate and heated gradually from 25°C to 96°C. Fluorescence was monitored using an Mx3005P real-time PCR instrument (Stratagene) with excitation and emission filters set to 465 and 590 nm, respectively. Data was analysed with the MxPro software.

Isothermal titration calorimetry (ITC).
Measurements were performed at 20°C on a MicroCal VP-ITC (GE Healthcare). ULK3 2-277 was dialysed overnight into assay buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 0.5 mM TCEP). The syringe was loaded with 87 μM ULK3 2-277 , the cell was filled with assay buffer containing 8 μM of the respective inhibitor. Every 5 minutes, 10 μL of the protein solution was injected into the cell for a total of 29 injections. The heat flow data was analysed with the MicroCal ORIGIN software package employing a single binding site model.  Table S1).
Crystallisation of the ULK3-momelotinib complex. 100 nL of a solution containing the protein-ligand complex (15 mg/mL ULK3 2-277 , 500 μM momelotinib) were transferred to a 3-well crystallisation plate (Swissci), mixed with 50 nL precipitant solution (0.1 M HEPES pH 7.5, 0.2 M MgCl 2 , 30% 2-propanol) and incubated at 4°C. Crystals appeared after 1 day and did not change appearance after 4 days. Data collection and analysis was performed as described above. Model and structure factors have been deposited to the PDB with the ID 6FDZ (crystallographic parameters in Table S1). ULK3 2-277 phosphomapping. 100 μL of a ULK3 2-277 solution (50 μM) were mixed with 100 μL of an ATP/Mg 2+ solution (2 mM ATP, 5 mM MgCl 2 ). The reaction mixture was incubated at 37°C for 30 minutes to allow for autophosphorylation. Following this, the ULK3 protein was subjected to methanol/chloroform precipitation, digestion with elastase and analysis by liquid chromatography tandem mass spectrometry (LC-MS/MS) as described previously. 32

Results
The ULK3 sequence indicates an active protein kinase. The kinase domain in the ULK3 N terminus contains all 5 conserved kinase motifs necessary for catalytic activity, but with a number of minor sequence variations: In the glycine-rich loop, one of the conserved glycine residues is exchanged for an alanine in ULK3. While the VAIK motif in the β3 strand is intact, the HRD motif is altered to HLD in ULK3, importantly still harbouring the catalytic aspartate. Also present are the asparagine in +5 position from the HRD and, finally, the DFG motif in the base of the activation segment ( Figure 1B). The ULK3 secondary structure elements (predicted with the PSIPRED online tool 33 ) corresponded perfectly to the well-known kinase topology. In the human kinome, ULK1 is the closest homolog of ULK3. Both kinase domains share a sequence identity of 40%. The most obvious difference is a 7 amino acid insertion in the loop connecting the β7 and β8 strands that is only present in ULK1. Interestingly, the ULK1 and ULK3 activation loops share no sequence similarity whatsoever. Even potential phosphorylation sites are shuffled, indicating different regulation mechanisms for ULK1 and ULK3. Striking sequence variability is also observed in the αG helices of both kinases. On top of this variability, the position of the ULK3 αG helix is shifted by two residues compared to ULK1 ( Figure 1B).
Producing the recombinant ULK3 kinase domain. The ULK3 residues 2 to 277 covered the kinase domain as estimated from comparison with ULK1. However, the expression of ULK3 2-277 in E. coli led to highly phosphorylated protein, indicating the ULK3 kinase to be catalytically active even without priming phosphorylations and accessory factors. Due to the phosphorylations at multiple and most likely random sites during expression, the protein was unstable, and we were unable to purify it to homogeneity. The coexpression of λ phosphatase allowed for the expression of non-phosphorylated ULK3  . In contrast to the protein expressed in absence of λ phosphatase, this protein was stable even at high protein concentrations and turned out to be suitable both for structural work and for inhibitor binding studies. As references, the ULK1 and ULK2 kinase domains were expressed and purified as described in recent publications. 19,20 Structure of the ULK3 active state. To gain insight into the ULK3 regulatory mechanism, and to explore how to target the ULK3 kinase domain with small-molecule binders, we determined the 3D structure of ULK3 in complex with a tight-binding reversible inhibitor (the identification of bosutinib as ULK3 inhibitor will be discussed in a later section). As expected, ULK3 presented the canonical bilobal kinase fold, with the inhibitor occupying the cleft between the two lobes ( Figure 1D). In active kinases, several core residues assemble into an internal scaffolding network referred to as hydrophobic spines. 34 As a consequence of this three-dimensional alignment, the active site residues are oriented for catalysis. In ULK3, both the catalytic and the regulatory spine were fully assembled. The catalytic spine comprised (top to bottom) A42 from the VAIK motif, the inhibitor's aromatic hinge binder instead of the ATP purine rings, L144 and I143 from the β7 strand and, finally, I202 that mounted the spine to the rigid αF helix. The regulatory spine contained (again top to bottom) L77 from the αC β4 loop, L66 from the αC helix, F158 from the DFG motif and H135 from the HRD motif ( Figure 1D). Another kinase feature that is linked to activity regulation is the activation loop.
For many kinases, a phosphorylated activation loop is firmly attached to the kinase C lobe and serves as a docking site for potential protein substrates, while a non-phosphorylated activation loop is more flexible and restricts the access of potential protein substrates to the active site. Thereby, activation loop phosphorylation switches the kinase from its enzymatically inactive to its active form. 35 The ULK3 activation loop was not phosphorylated. Nevertheless, it was firmly attached to the ULK3 C lobe. It is worth mentioning that the electron density of the activation loop was clearly defined, and that its B factors were similar to the average B factors of the kinase domain. While the kinase core with the intact spines and the attached activation loop indicated an active kinase conformation, the K-E salt bridge that usually anchors the αC helix to the β3 strand in active kinases was not formed. However, the K44 to E62 distance of 7.0 Å resulted mainly from the inhibitor displacing the K44 side chain and pushing it away from E62. Comparison of the ULK3 and ULK1 catalytic domains. Though both kinase domains share only 40% sequence identity, their overall 3D appearances were remarkably similar (PDB ID 5CI7 19 , Figure 1C). The RMSD value between all Cα atoms was 2.4 Å. A striking difference was observed in the β3 αC linker that was extended in ULK3. This might allow the ULK3 αC helix to adopt more pronounced out conformations.
And as already spotted in the sequence alignment ( Figure 1B), the insertion between the β7 and β8 strands contained additional residues in ULK1. The functional consequence of this insertion was, however, not obvious. Surprisingly, the ULK1 and ULK3 activation loops adopted identical conformations despite their very different primary sequences. We concluded that this particular activation loop conformation is conserved in ULK1 and ULK3, but the mechanisms to stabilize and to destabilize it differ between the two kinases. Finally, the orientations of the αG helices varied substantially. While the ULK1 αG helix was rather attached to the C lobe, the ULK3 αG helix pointed outwards. Thereby, they likely formed distinct docking sites for their respective sets of protein substrates. These observations highlighted once again that despite all similarities, even closely related kinases can differ in activity regulation and substrate recognition.
ULK3 is a highly druggable kinase target. In order to identify chemical starting points for the development of ULK3 binders, we screened a library of clinical and pre-clinical kinase inhibitors (purchased from Selleckchem, Table S2). We performed differential scanning fluorimetry (DSF) assay, a versatile method to determine the temperature at which a given protein denatures from heat. Tight binding of a small molecule generally stabilizes the protein thus leading to a higher melting temperature (T M ). Notably, kinases are particularly sensitive to this assay format. 36 In addition to the ULK3 kinase domain, we also tested the homologous ULK1 and ULK2 kinase domains for comparison. In the absence of inhibitors, the melting temperatures of the three proteins were similar, with ULK1 (46°C) being slightly less stable than ULK2 and ULK3 (both 49°C). As expected, some of the tested compounds stabilized the ULK proteins, a list of all compounds with the corresponding T M shifts is given in Table S2. Out of 150 compounds, 7 compounds increased ULK1 T M by >8 K. For ULK2, it was only 3 compounds, and 10 compounds were identified for ULK3, suggesting excellent druggability. The best hit for ULK1 and ULK2 was dabrafenib (with a ΔT M of 12 and 11 K, respectively) in line with recent potent inhibitors that have been developed for this target 37 sharing the aminopyrimidine hinge binding motif. For ULK3, the most potent hit was the pan-FGFR inhibitor LY2874455 38 (ΔT M of 13 K). ULK1 and ULK2 were stabilized by a similar set of compounds, while the stabilization patterns for ULK1 and ULK3 differed substantially (the cross-correlations of the T M shifts are depicted in Figure 2A). In terms of isoform specificity, the compounds foretinib and SU6668 stood out.
Foretinib, a broad-spectrum tyrosine kinase inhibitor, stabilized ULK3 by 11 K, but ULK1/2 only by <4 K suggesting that ULK3 differs from the other ULK family members by accommodating this bulky type-2 kinase inhibitor. The oxindole SU6668 (orantinib), that was previously reported to inhibit ULK3 39 , was confirmed in our screen. It stabilized ULK3 by 8.6 K, but ULK1/2 only by <2 K. Taken together, this demonstrated that ULK3 inhibitors without cross reactivity towards the other ULK proteins can be  Figure 2C). Interestingly, ULK3 was stabilized by both type-1 kinase inhibitors (such as lestaurtinib and bosutinib) and type-2 inhibitors (such as foretinib and momelotinib). In contrast, only weaker interactions with type-2 inhibitors were observed for ULK1/2 suggesting that the DFG-out conformation is less favoured in these kinases.
Hit validation with isothermal titration calorimetry (ITC). Since DSF results do not only depend on affinity, but also on the respective inhibitor binding mode and on the stability of the apo protein, we confirmed inhibitor binding affinity with an orthogonal assay. We employed ITC with the identified inhibitors foretinib, SU6668, bosutinib and momelotinib ( Figure 2B). The binding of foretinib to the ULK3 kinase was characterized by a dissociation constant (K D ) of 20 nM. The binding was enthalpically driven (ΔH= -7.3 kcal·mol -1 ) with a substantial entropic contribution to binding (-TΔS= -3.0 kcal·mol -1 ). The affinity of bosutinib to ULK3 was lower (K D = 48 nM). This was in good agreement with the DSF results. In comparison to foretinib, bosutinib binding was more entropically driven (-TΔS= -6.2 kcal·mol -1 ), with a significant enthalpic contribution (ΔH= -3.6 kcal·mol -1 ). The ITC titrations for SU6668 and momelotinib were noisy (data not shown) and thus not suitable for calculation of binding constants and thermodynamic parameters. This was most likely due to poor compound solubility. A comparison of DSF and ITC results is shown in Figure 2C.
Bosutinib binding mode to ULK3. The BCR-ABL inhibitor bosutinib was identified as a ULK3 kinase binder in DSF assay and confirmed by ITC. The active-like ULK3 structure described above ( Figure 1D) was obtained by co-crystallization of ULK3 and bosutinib. The inhibitor was ATP competitive, occupying the solvent exposed front pocket (Figure 3A), in a similar fashion to what was observed with the tyrosine kinase ABL 40 .
Bosutinib was bound as a type-1 inhibitor with the base of the ULK3 activation segment in the DFG-in conformation. The quinoline nitrogen formed a hydrogen bond with the backbone of C94 in the ULK3 hinge. The other interactions were exclusively mediated by the substituents of the bosutinib benzene ring: the ortho-chlorine to the backbone of A42 in the β3 sheet, the para-chlorine to D157 in the DFG motif, and the methoxy group to the conserved N142 prior to the β7 sheet ( Figure 3A). Notably, bosutinib binding prevented the K44 E62 salt bridge from being formed, mainly by displacing the K44 side chain.
Momelotinib binding induces the DFG-out conformation. All DSF hits with ΔT M >6 K were probed in ULK3 co-crystallization studies. Of these, only momelotinib led to crystals with satisfying diffraction properties (resolution <3 Å). Momelotinib was originally developed to target JAK1/2. In our structure model, it bound to ULK3 as a type-2 inhibitor with the base of the activation segment in the DFG-out conformation, but without occupying the DFG-out allosteric pocket ( Figure 3B). Instead, one of the momelotinib biphenyl rings recruited the DFG F158 sidechain to form a π-stacking interaction thus stabilizing the DFG-out conformation. Similar to bosutinib, momelotinib was anchored via a hydrogen bond to the C94 backbone in the ULK3 hinge. We also observed its nitrile pointing towards the ULK3 glycine-rich loop and interacting with G23 and A26 backbone atoms. And finally, the momelotinib amide oxygen tightly bound the conserved K44 in the β3 sheet, providing additional stability to the complex (Figure 3B). We are not aware of any other structure model showing momelotinib bound to a kinase. XL019, a compound that resembles momelotinib, has high selectivity for JAK2, but unlike momelotinib, it stabilized its protein target in the DFG-in conformation. 41 The surprising binding mode and its favourable kinome-wide selectivity make momelotinib an interesting starting point for the development of a ULK3 chemical probe. One possible strategy is here to exploit the pocket resulting from the DFG-flip that is likely not present in other momelotinib targets. Another option is to merge bosutinib and momelotinib features in one molecule to obtain inhibitors with improved affinity and selectivity.
Structural flexibility in the ULK3 kinase domain. Our diverse ULK3-inhibitor complexes revealed the ULK3 propensity to undergo substantial conformational changes. When bosutinib was bound, the kinase domain adopted an active-like αC-in DFG-in conformation. Both E62 from the αC helix and D157 from the DFG motif pointed towards the ATP binding site. Furthermore, the activation loop was firmly attached to the kinase C lobe, and the regulatory spine was fully assembled (Figure 4A). In contrast, when momelotinib was bound, the αC helix was rotated outwards by 45°, and the DFG F158 was flipped out of its hydrophobic pocket to form a π-stacking interaction with the inhibitor. In this inactive αC-out DFG-out conformation, neither E62 nor D157 were close to the ATP binding site. The activation segment was not attached to the C-lobe but it occupied the space vacated by the displaced αC helix. As a consequence of all these structural rearrangements, the regulatory spine was completely broken (Figure 4B). In addition to their role as enzymes, kinases can also function as scaffolds and allosteric modulators. If this is the case for ULK3, our differentially binding inhibitors are expected to have a differential impact on non-catalytic ULK3 signalling.
The ULK3 kinase is constitutively active. Regulation of kinase activity by activation loop phosphorylation is a common theme in protein kinases. 35 Accordingly, autophosphorylation of the activation loop residue T180 boosts the kinase activity of the ULK3 homolog ULK1. 19 The regulatory mechanism for ULK3 is different though. In order to identify potential autophosphorylation sites, we incubated unphosphorylated ULK3 2-277 with ATP/Mg 2+ for 30 minutes at 37°C, followed by elastase digestion and LC-MS/MS analysis ( Figure S2). We found that ULK3 was heavily autophosphorylated with all 10 phosphosites being located in flexible loop regions on the surface of the protein, such as S55 in the β3 αC linker, S146 in the β7 β8 insert and S215 in the αF αG loop (Figure 4C). There was no indication any of these phosphosites were physiologically relevant. The activation loop, however, was spared from the in vitro autophosphorylation.
We concluded that ULK3 possessed constitutive kinase activity even in its unphosphorylated form. This observation is also supported by the ULK3-bosutinib structure model. The unphosphorylated activation loop was firmly attached to the kinase C lobe, the lack of a stabilizing phosphoresidue was compensated for by an unusual salt bridge (Q162 to R174, Figure 4D). Taken together, we conclude that the activation loop conformation is conserved between ULK1 and ULK3, but the regulatory mechanism is not. However, these observations concern the kinase domain only. In the full-length ULK3, there are additional levels of activity regulation, also linked to substrate recognition.
Modelling of the ULK3-IST1 complex. Several proteins have been described as cellular ULK3 kinase substrates. 4, 14 A particularly compelling substrate recognition mechanism can be hypothesized for the ESCRT-III component IST1: ULK3 bears two MIT domains in its C terminus, while IST1 contains two MIT interacting motifs (MIMs). The binding of the second MIM to the second MIT domain was demonstrated experimentally 4 , furthermore, the binding of both IST1 MIMs to the ULK3 MIT tandem is sterically possible.
This binding mode places the IST1 polymerization domain next to the ULK3 kinase domain. The interaction of ULK3 and IST1 was modelled and refined using a simple molecular dynamics tool (phenix.dynamics 42 ). In the resulting model, the ULK3 activation loop and αG helix served as docking patch for the IST1 polymerization domain, and the IST1 phosphoacceptor residue S152 was in close proximity to the ULK3 active site (Figure 4E). According to this substrate-recognition mechanism, the cellular ULK3 might be localized to regions of high ESCRT-III activity, scanning for free MIMs, and phosphorylating the CHMP and IST1 polymerization domains. 4 Like this, free CHMP and IST1 are silenced, resulting in a more stringent abscission checkpoint.

Discussion
Domain plasticity is a well-established concept in the kinase field. 43 Its fundamental principle is that kinases toggle between an active conformation, which is very similar in all kinases, and one or several inactive conformations, which can differ substantially between individual kinases. In general, the kinase structural elements with the highest degree of freedom are the glycine-rich loop, the αC helix and the activation loop including the DFG motif in its base. All these elements are highly dynamic in ULK3, establishing ULK3 as an interesting example for a kinase with pronounced domain plasticity. In addition to the active-like ULK3 conformation, we observed a DFG-out αC-out inactive conformation. Importantly, we have identified smallmolecule binders that trap ULK3 in either the active-like (such as SU6668 and bosutinib) or the inactive conformation (such as foretinib and momelotinb). In solution, we expect the ULK3 kinase domain to adopt an ensemble of diverse conformations. The secondary structure elements are formed also in the apo protein as confirmed with circular dichroism (CD) spectroscopy ( Figure S3). When adding an inhibitor, one particular conformation is stabilized. Accordingly, our crystal structures provide two of many possible snapshots of the ULK3 kinase domain in solution. At this point, we can only speculate about the atomistic mechanism by which the active-like ULK3 conformation transforms into the inactive conformation. Likely, the mechanism is initiated by rotating out the αC helix to creating space for the DFG-flip and the HRD-flip as observed for other kinases. 44,45 An important question here is whether there are stable intermediates that possibly can be targeted by yet to discover inhibitors. These hypothetic intermediates might involve a Until now, structural data on the ULK3 full-length protein is elusive. Either the kinase and C-terminal MIT1/2 domains act as pearls-on-a-string, with the MIT1/2 domains just fishing for potential substrates, or they contribute to stabilizing the inactive kinase conformation as autoinhibitory domains. As mentioned above, active ULK3 modulates the timing of membrane abscission, resulting in a more stringent abscission checkpoint. 4 The spaciotemporal regulation of ULK3 activity is therefore expected to be tight. However, the cellular determinants of this regulation are so far unknown: ULK3 might be the client of an allosteric protein binder that stabilizes one specific ULK3 conformation. There is also the possibility of ULK3 post-translational modifications (PTMs) we are not aware of at this point. And finally, the binding of a physiological substrate such as monomeric IST1 (Figure 4E) might induce and stabilize the active ULK3 conformation and trigger phosphorylation. Thus, only monomeric IST1 is phosphorylated, and unspecific ULK3 activity is prevented.
The next steps towards a better understanding of ULK3 regulation and mechanism of action will likely include the identification of cellular ULK3 interaction partners, structural work on the ULK3 full-length protein and the development of chemical tools for functional studies on the endogenous ULK3 kinase. Our study lays the foundation for this future work.
The coordinates and structure factors of the kinase-inhibitor complexes have been deposited to the PDB and are available under https://www.rcsb.org/structure/6FDY and https://www.rcsb.org/structure/6FDZ, respectively. Figure 1A) The ULK family of protein kinases has derived from subsequent gene duplications. All five family members contain the kinase domain in their N terminus and long C-terminal extensions. CTD-like -carboxy terminal domain-like domain; M -microtubule interacting and trafficking molecule (MIT) domain; ARMarmadillo repeat domain; IDR -intrinsically disordered region. The percentages describe the identities with the ULK1 kinase domain sequence. The blue brackets indicate the ULK3 expression construct used in this study. B) Alignment of ULK1 and ULK3 kinase domain sequences. β strands are highlighted in yellow, α helices in blue, the activation loops in pink, and conserved kinase motifs are in bold. Please note the unusual insertions between the β7 and β8 strands and the sequence variability in the activation loops / the αG helices. C and D) Comparison of the ULK1 (PDB ID 5CI7) and ULK3 kinase domains (PDB ID 6FDY). Regions with obvious differences are labelled in orange. Both kinases adopt the active-like conformation with fully assembled hydrophobic spines. Figure 2A) 150 clinical kinase inhibitors were tested for their ability to stabilise ULK1, ULK2 or ULK3 in differential scanning fluorimetry (DSF). Several of the inhibitors induced melting temperature (T M ) shifts. The cross-correlation plots reveal that ULK1 and ULK2 were targeted by a similar subset of inhibitors, while the inhibitors stabilizing ULK1 and ULK3 differed. B) The binding of some of the DSF hits to ULK3 was validated by isothermal titration calorimetry (ITC). Foretinib and bosutinib bound ULK3 with K D values of 20 and 48 nM, respectively. C) Summary of DSF and ITC assay results for the best ULK3 binders. Foretinib and SU6668 showed some selectivity for ULK3 over ULK1/2.  Figure 3A) Binding mode of bosutinib to the ULK3 active site. The quinoline nitrogen anchored the inhibitor to C94 in the hinge. The substituents of the terminal 6-ring formed hydrogen bonds to the A42 backbone oxygen in the β3 strand, to D157 in the DFG motif, and to the conserved N142 prior to the β7 strand (PDB ID 6FDY). B) Momelotinib was anchored to C94 in the ULK3 hinge, too, but the other inhibitor interactions with ULK3 were different from what was observed for bosutinib. Its nitrile pointed towards the ULK3 glycine-rich loop, forming interactions with G23 and A26 backbone atoms. One of the biphenyl rings recruited the DFG F158 sidechain for a π-stacking interaction, thus stabilising ULK3 in the DFG-out conformation. And finally, the momelotinib amide oxygen tightly bound the conserved K44 in the β3 strand (PDB ID 6FDZ). For clarity, parts of the β1 and β2 strands were removed from the models. Figure 4A) Domain plasticity of the ULK3 kinase. In the active-like conformation, the regulatory spine was fully assembled, and the activation loop attached (PDB ID 6FDY). B) In the inactive conformation, the regulatory spine was broken, and the activation loop displaced (PDB ID 6FDZ). C) The recombinant ULK3 kinase domain was constitutively active and readily autophosphorylated. The phosphosites (indicated in pink) were embedded in flexible loop regions. Please note that the activation loop was spared from autophosphorylation. D) Despite their differing sequences, the ULK1 and ULK3 activation loops adopted very similar conformations. They were closely attached to the C lobe, allowing for substrate docking. In ULK1, this conformation was stabilized by T180 phosphorylation. In ULK3, no phosphorylation was necessary for the formation of the Q162 to R174 salt bridge, promoting a constitutively active conformation. E) Model of the ULK3-IST1 complex. The IST1 MIM helices bind to the ULK3 MIT tandem, the IST1 polymerization domain binds to the ULK3 kinase domain. The phosphoacceptor S152 is proximal to the kinase active site. Modelling done with phenix.dynamics. 42 Figure S2) ULK3 phosphosite mapping by mass spectrometry. The autophosphorylated ULK3 kinase domain was digested with elastase and analysed by LC-MS/MS. The mass spectrometry dataset was evaluated using the software PEAKS (v8, Bioinformatics Solutions Inc) as described previously. 46 Peptides identified with high confidence are shown in blue, and phosphorylated S/T/Y residues are highlighted in orange. Notably, no extensive phosphorylation was detected in the ULK3 activation segment (residues D157 to E183). In contrast, several serine residues situated in flexible loop regions such as S55 in the β3 αC linker, S146 in the β7 β8 insert, and S215 in the αF αG loop, were highly phosphorylated, reflected by the number of detected phosphopeptides.   Figure S3) Circular dichroism (CD) spectra of the ULK3 kinase domain. The minima at 208 nm indicated the presence of α helices as expected for a kinase. Interestingly, no pronounced changes in secondary structure elements were induced by inhibitor binding. Subtle differences in amplitude were likely due to protein precipitation during sample preparation.