Structural insights into inhibitor regulation of the DNA repair protein DNA-PKcs

The DNA-dependent protein kinase catalytic subunit (DNA-PKcs) has a central role in non-homologous end joining, one of the two main pathways that detect and repair DNA double-strand breaks (DSBs) in humans1,2. DNA-PKcs is of great importance in repairing pathological DSBs, making DNA-PKcs inhibitors attractive therapeutic agents for cancer in combination with DSB-inducing radiotherapy and chemotherapy3. Many of the selective inhibitors of DNA-PKcs that have been developed exhibit potential as treatment for various cancers4. Here we report cryo-electron microscopy (cryo-EM) structures of human DNA-PKcs natively purified from HeLa cell nuclear extracts, in complex with adenosine-5′-(γ-thio)-triphosphate (ATPγS) and four inhibitors (wortmannin, NU7441, AZD7648 and M3814), including drug candidates undergoing clinical trials. The structures reveal molecular details of ATP binding at the active site before catalysis and provide insights into the modes of action and specificities of the competitive inhibitors. Of note, binding of the ligands causes movement of the PIKK regulatory domain (PRD), revealing a connection between the p-loop and PRD conformations. Electrophoretic mobility shift assay and cryo-EM studies on the DNA-dependent protein kinase holoenzyme further show that ligand binding does not have a negative allosteric or inhibitory effect on assembly of the holoenzyme complex and that inhibitors function through direct competition with ATP. Overall, the structures described in this study should greatly assist future efforts in rational drug design targeting DNA-PKcs, demonstrating the potential of cryo-EM in structure-guided drug development for large and challenging targets. Cryo-electron microscopy structures of DNA-dependent protein kinase catalytic subunit bound to ATPγS and four inhibitors (wortmannin, NU7441, AZD7648 and M3814) provide molecular details and insights useful for drug design.

Article structural guidance for future drug design targeting DNA-PKcs, demonstrating the great potential of cryo-EM in structure-based drug discovery.

DNA-PKcs in complex with ATPγS-(Mg 2+ ) 2
DNA-PKcs purified from HeLa cell nuclear extract was incubated with ATPγS and loaded onto a previously prepared grid with a support film of homemade single-layer graphene oxide (Extended Data Fig. 1). The overall resolution of the DNA-PKcs-ATPγS complex is 3.40 Å with the local resolution of the catalytic core region better than 3 Å, showing clear density for the modelling of ATPγS with two Mg 2+ ions and demonstrating a similar structure to that of the homologue mTOR ( Fig. 1a and Extended Data Fig. 2a) 12 . The p-loop (residues 3729-3735) interacts closely with the phosphate groups of ATPγS. Central to this is the interaction between the β-phosphate and the main chain NH group of Arg3733, together with the side chain of Ser3731. Lys3753 interacts with the α-phosphate of ATPγS. Asn3927, Asp3941 and Glu3756 together coordinate the Mg 2+ ions. The γ-phosphate of ATPγS points towards   (Fig. 1b) 13 , and coordinated movements of α-helices in the head unit of the protein (Fig. 1c). Notably, the PRD (residues 4009-4039) does not interfere with the interaction of DNA-PKcs with ATP and Mg 2+ . However, the α-helix of the PRD (residues 4009-4023) of DNA-PKcs is moved away from the substrate-binding site following ATP binding, partly relieving the blockage by PRD of the substrate-binding site required for subsequent catalysis (Fig. 1c).

DNA-PKcs in complex with inhibitors
DNA-PKcs was incubated with several DNA-PKcs inhibitors with different specificities (wortmannin, NU7441, AZD7648 and M3814), to  Table 1). The overall resolutions of the structures range from 2.96 Å to 3.33 Å. In all the structures, the kinase core has the highest local resolution, allowing the inhibitors to be unequivocally modelled (Fig. 2a). Wortmannin is one of the early DNA-PK inhibitors used for kinase inhibition 14 . In the cryo-EM structure, wortmannin packs on one side, mainly against the N-lobe of DNA-PKcs (Leu3751, Ile3803 and Trp3805), while on the other side it packs against the C-lobe (Met3929 and Asp3940) (Fig. 2b). One edge of the compound is facing the deep pocket of the ATP-binding site, and the opposite edge is exposed to solvent. The primary amine (ε-amino group) of Lys3753 forms a covalent C-N bond with C19 in the furan ring of wortmannin ( Fig. 2a), irreversibly inhibiting kinase activity. The ATP-binding site of DNA-PKcs has close structural complementarity to wortmannin. The C14 methyl resides in a pocket formed by Leu3751, Ile3803 and Trp3805, while the C10 methyl fits into a pocket formed by Met3729, Pro3735 and Leu3751 (Fig. 2b). There is no density corresponding to the acetoxy group attached to C11 of wortmannin, as was observed in the published structure of wortmannin in complex with porcine PI3Kγ (Extended Data Fig. 3a) 15 .
Compared with wortmannin, NU7441 is a more selective inhibitor, developed from the earlier inhibitors NU7026 and LY294002, which were derived from the broad-spectrum protein kinase flavonoid inhibitor quercetin 16 . The chromone core and morpholino ring of NU7441 bind in the centre of the deepest pocket formed by Leu3751, Tyr3791, Ile3803, Trp3805, Leu3806, Ile3940 and Met3929, within which O1 and O27 of NU7441 form hydrogen bonds with the peptide backbones of Asp3941 and Leu3806 (Fig. 2c). The dibenzothiophene group interacts with the N-lobe and docks in the hydrophobic groove of Met3729, Pro3735 and Leu3751 (Fig. 2c).
AZD7648 is a recently developed DNA-PK inhibitor shown to be highly selective and to enhance the efficacy of doxorubicin and IR 11,17 . Moreover, AZD7648 was identified as a potential combinational therapy with the PARP inhibitor olaparib and is currently under clinical trial for advanced malignancies (trial identifier, NCT03907969) 11 . The developers of AZD7648 took advantage of a previously published crystal structure of AZD7648 bound to PI3Kγ 17 . In the DNA-PKcs-AZD7648 complex, the compound binds the hinge loop (Fig. 2d). The triazolopyridine moiety with a methyl group lies in the deep hydrophobic pocket formed by Tyr3791, Ile3803, Leu3806 and Ile3940. The purine moiety docks in the narrow tunnel formed by Trp3805, Leu3806 and Met3929. Similarly to the PI3Kγ-AZD7648 complex, the N3 hydrogen of AZD7648 (the aniline NH) forms bonds with the backbone oxygen of Glu3804 and N7 accepts the hydrogen from the backbone NH group of Leu3806 (ref. 17 ). N6 in the triazolopyridine moiety also binds to the backbone NH group of Asp3941. There is a further hydrogen bond in the case of DNA-PKcs-AZD7648: N5 forms a bond with the primary amine group of Lys3753. Comparison of the two structures shows a 90° rotation of the indole ring for Trp812 in PI3Kγ relative to Trp3805 in DNA-PKcs, which in DNA-PKcs not only provides a better hydrophobic binding surface but also aligns nicely to the AZD7648 purine ring, leading to a π stacking interaction that enhances affinity (Extended Data Fig. 3b).
M3814 (peposertib or nedisertib) is another recently developed DNA-PK selective inhibitor 18,19 . Preclinical studies have revealed its synergy with radiotherapy and chemotherapies using doxorubicin and etoposide 20 . The compound was shown to be well tolerated with modest efficacy in unselected tumours in phase 1 clinical trial results (trial identifier, NCT02316197) 19 ; M3814 is currently in four phase 2 clinical trials targeting different cancers (trial identifiers, NCT03770689, NCT04068194, NCT04071236 and NCT04172532). According to our structure, the quinazoline and morpholino moieties of the compound fit well into the deepest, largely hydrophobic pocket formed by the side chains of Leu3751, Tyr3791, Val3793, Ile3803, Trp3805, Leu3806, Ile3938 and Ile3940 and the main chains of Glu3804 and Asp3941 (Fig. 2e). The chloro-fluorobenzene ring rotates by ~60° and points towards the N-lobe, facing the side chains of Met3729, Ser3731, Pro3735, Leu3751 and Lys3753. The remaining moiety containing the pyridazine ring then rotates back to be almost parallel to the quinazoline plane, lying in the groove of Met3729, Trp3805, Thr3811, Asn3926 and Met3929.

Comparison of binding modes
All the inhibitors studied target the ATP-binding groove of DNA-PKcs, overlapping with the ATPγS-binding site (Fig. 3). Among them, wortmannin has the maximum overlap. Compared with ATPγS, wortmannin has greater complementarity to the binding site (Fig. 3a), binding deeper into the ATP adenine-moiety pocket, with its two protruding methyl groups fitting better into the hydrophobic pocket on the surface of the N-lobe. NU7441 binds quite differently from ATPγS to the ATP-binding groove. The chromone core and the morpholino ring of NU7441 bind to the pocket occupied by the adenine moiety and α-phosphate group of ATPγS (Fig. 3b). The dibenzothiophene moiety overlaps with the ATPγS ribose and inserts into the groove between the p-loop and the hinge loop, which is formed by the 120° inward swinging of Met3729.
In the case of AZD7648, the triazolopyridine moiety with a methyl group docks deeper than the adenine moiety of ATPγS in the same pocket (Fig. 3c). While the tetrahydropyran ring occupies the same space as the ATPγS ribose, the purine ring of AZD7648 fits nicely into the empty groove between Trp3805 and Met3929.
M3814 has the least overlap with ATPγS in the binding groove. The quinazoline and morpholine parts reach deeper into the pocket of the adenine moiety and α-phosphate group of ATPγS (Fig. 3d). Moreover, when the compound binds, the side chain of Trp3805 moves inwards, changing the open adenine-binding site to an enclosed hydrophobic pocket, in which the morpholine ring of M3814 is fully accommodated. The chloro-fluorobenzene ring overlaps with the ATPγS α-phosphate and points towards the N-lobe, causing an uplift of the p-loop and the region around Lys3753. The pyridazine ring is adjacent to the ATPγS ribose but closer to the C-lobe, where Asn3926 rotates towards M3814 to form a small groove for docking of the drug.
The different inhibitors result in varying conformational changes that each modify the surface properties to optimize binding interactions with the ligand. There are two key regions of conformational change. One is the p-loop, where the conformation is guided by the residues with their side chains facing the ATP-binding groove. In apo DNA-PKcs, the position of the p-loop is restrained by the flanking β-sheets and the electrostatic interaction between Arg3733 and Asp3587 (Fig. 4a). The conformations of Met3729, Ser3731 and Pro3735 change according to the different chemical properties of the ligands, resulting in an up-down movement of the p-loop like a 'spring leaf'. This movement induces corresponding conformational changes in the flanking β-sheets that are passed on to the hydrophobic core of the DNA-PKcs head region (Fig. 4b). The resulting conformational changes of DNA-PKcs, especially the movement of the PRD, appear concerted (Fig. 4c). The other key region of conformational change is at Trp3805 on the hinge loop, which guards the entrance to the main pocket (Fig. 4b). The indole side chain of Trp3805 contributes to the stacking effects and hydrophobic interactions with ligands, and its movement impacts the architecture of the binding site.

Ligand regulation and future development
Understanding of the mechanism of DNA-PKcs in DNA repair has recently been advanced through structural studies using crystallography and cryo-EM 13,21-26 . However, technical challenges and limitations have been the major hurdle in biochemical and structural investigations. The structure of DNA-PKcs-ATPγS-(Mg 2+ ) 2 described here shows how ATPγS and two Mg 2+ ions occupy the ATP-binding groove, coordinating the p-loop, activation loop and catalytic loop. The presence of ATPγS-(Mg 2+ ) 2 results in movement of the PRD to partly relieve the blockage of the substrate-binding site before catalysis. This is consistent with the observation that DNA-PKcs can be active in vitro in the absence of assembly of the DNA-PK holoenzyme and explains why the kinase activity of the C terminus of DNA-PKcs is not stimulated by Ku70-Ku80 and DNA 27 . Together with the structures of the DNA-PKcsinhibitor complexes, this reveals that PRD conformation can be regulated by spring leaf-like p-loop movements. Moreover, we obtained a structure for the DNA-PK-ATPγS-(Mg 2+ ) 2 complex at a medium resolution of 4.3 Å (Extended Data Fig. 4a, b). In this structure, the kinase domain of DNA-PKcs fits nicely into the corresponding region of the map of DNA-PK-ATPγS-(Mg 2+ ) 2 , indicating that the binding modes of the ligands in the single polypeptide chain of DNA-PKcs and the holoenzyme are similar. Moreover, we conducted electrophoretic mobility shift assays (EMSAs) to confirm that the inhibitors do not stop formation of the DNA-PK holoenzyme (Extended Data Fig. 4d). In addition, structural and single-molecule studies have demonstrated that binding of ligands does not affect dimerization of the long-range NHEJ complex involving DNA-PKcs 24,28 . Therefore, DNA-PKcs inhibitors should not structurally disturb high-order complex formation. They function via direct ATP competition.
Current drug candidates have been successfully developed from large-scale screening targeting the ATP-binding site, but molecular details of the modes of action of such candidates have been unclear 4,10 . In our structures, all inhibitors investigated are less elongated than ATP and do not extend beyond the position of the α-phosphate group. However, the inhibitors more effectively target the deep hydrophobic pocket where the adenine moiety of ATP is located. The inhibitors all have large hydrophobic groups that bind deeply into this pocket, achieving high occupancy. It appears that the occupancy level of the pocket is related to selectivity: NU7441, AZD7648 and M3814 display higher occupancy than wortmannin. Comparisons between the older-generation inhibitor NU7441 and the newer-generation inhibitors AZD7648 and M3814 indicate that selectivity can also be improved by exploring the entrance tunnel between the N-and C-lobes and the peripheral region on the C-lobe (Fig. 3c, d). Comparison of the structures of DNA-PKcsligand complexes with those of mTOR-ligand complexes demonstrates that inhibitor specificity is related to the p-loop conformation and composition (Extended Data Fig. 2a-c). The ligand-binding pockets of DNA-PKcs and mTOR are highly similar, with the main differences Article lying in the p-loops (Extended Data Fig. 2b). In DNA-PKcs, the p-loop is closer to the C-lobe than it is in mTOR. Together with the extended side chain of Met3729, the p-loop of DNA-PKcs creates a narrower path to the binding pocket, the effect of which can be visualized through comparison of ATPγS-(Mg 2+ ) 2 binding modes ( Supplementary Fig. 2a, b). While the adenine moieties overlap nicely in the two structures, the ribose moiety and phosphate groups have different binding modes due to p-loop differences (Extended Data Fig. 2b).
Future inhibitor development can also be guided using our in-house hotspot-mapping program 29 . Hotspots can be defined as areas within the protein that make an essential contribution to the overall binding of small molecules 30 . The predicted hotspot map of the catalytic core of DNA-PKcs exhibits a large apolar region at the centre, juxtaposed with polar hydrogen-bond donor and acceptor regions (Extended Data  Fig. 5a). The hydrophobic core overlaps with the adenosine moiety of ATPγS as well as the central heterocyclic scaffold of the inhibitors described in this study. Strong polar regions can be observed adjacent to this hydrophobic core, close to the p-loop, and overlap with the ATP phosphate group-binding site (Extended Data Fig. 5a). A second polar hotspot region is seen at the edge of the ATP pocket close to the catalytic loop. Of note, the more selective inhibitors examined in this study, AZD7648 and M3814, have functional groups that engage this region (Extended Data Fig. 5a, b). The polar contacts mediated by the side chain of Thr3811 and the backbone carbonyl of Asn3926 here allow compounds AZD7648 and M3814, respectively, to adopt a binding mode distinct from those of ATP, wortmannin and NU7441 by extending outside the ATP-binding groove (Extended Data Fig. 5b). It is tempting to speculate that interactions at this part of the ATP pocket, close to the catalytic loop, could further contribute to selective inhibition of DNA-PKcs in addition to the p-loop conformation. Therefore, introducing functional groups to engage Asn3926, Asn3927, Thr3811 and Thr3809 in this region of the active site may potentially help further enhance selective inhibition of DNA-PKcs.
Furthermore, now that it is possible to obtain high-resolution structures for DNA-PKcs-ligand complexes routinely (Extended Data Fig. 6), structure-guided drug discovery can exploit other specific pockets of DNA-PKcs in the development of allosteric inhibitors or proteolysis-targeting chimeric (PROTAC) drugs with enhanced potency and selectivity 31,32 . Degradation of DNA-PKcs eliminates not only its kinase activity but also the stage provided for downstream NHEJ and DNA damage response signalling 13,32,33 . However, PROTAC drugs cannot be guaranteed to be beneficial as previous studies have shown that inhibiting kinase activity has more potent effects than knockdown or knockout 34 . Nevertheless, with optimized sample preparation and the reported cryo-EM structures, future drugs could be further developed to target allosteric sites instead of, or in addition to, the conserved kinase catalytic core, to achieve better specificity for DNA-PKcs.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-021-04274-9.