The R1275Q Neuroblastoma Mutant and Certain ATP-competitive Inhibitors Stabilize Alternative Activation Loop Conformations of Anaplastic Lymphoma Kinase*

Background: Anaplastic lymphoma kinase (ALK) plays an important causative role in some cancers. Results: Novel views of the ALK activation loop are provided by several new crystal structures. Conclusion: Certain neuroblastoma mutations and inhibitors stabilize alternative, inactive ALK conformations. Significance: Novel kinase conformations may aid the design of a new generation of selective ALK inhibitors. Anaplastic lymphoma kinase (ALK) is a receptor tyrosine kinase that, when genetically altered by mutation, amplification, chromosomal translocation or inversion, has been shown to play an oncogenic role in certain cancers. Small molecule inhibitors targeting the kinase activity of ALK have proven to be effective therapies in certain ALK-driven malignancies and one such inhibitor, crizotinib, is now approved for the treatment of EML4-ALK-driven, non-small cell lung cancer. In neuroblastoma, activating point mutations in the ALK kinase domain can drive disease progression, with the two most common mutations being F1174L and R1275Q. We report here crystal structures of the ALK kinase domain containing the F1174L and R1275Q mutations. Also included are crystal structures of ALK in complex with novel small molecule ALK inhibitors, including a classic type II inhibitor, that stabilize previously unobserved conformations of the ALK activation loop. Collectively, these structures illustrate a different series of activation loop conformations than has been observed in previous ALK crystal structures and provide insight into the activating nature of the R1275Q mutation. The novel active site topologies presented here may also aid the structure-based drug design of a new generation of ALK inhibitors.

Anaplastic lymphoma kinase (ALK) 2 is a receptor tyrosine kinase that belongs to the insulin receptor superfamily. Wildtype human ALK is a protein consisting of 1620 amino acids with an extracellular domain, a transmembrane sequence, and an intracellular region containing juxtamembrane and kinase domains. However, ALK was initially discovered as part of a fusion protein identified in a subset of anaplastic large cell lymphoma patients wherein the entire ALK cytoplasmic domain was fused to nucleophosmin (NPM) via a cytogenetic translocation (1). The NPM-ALK fusion protein possesses constitutive kinase activity resulting from NPM-derived oligomerization and subsequent ALK autoactivation, and it has strong transforming potential both in vitro and in vivo (2)(3)(4)(5)(6). More recently, additional ALK fusion proteins have been discovered in inflammatory myofibroblastic tumors (6 -8), diffuse large B-cell lymphomas (9,10), certain squamous cell carcinomas (11,12), and non-small cell lung cancer (NSCLC) (13,14). Notably, the activity of these constitutively active fusion proteins can be inhibited by small molecule inhibitors targeting the ALK kinase domain (15)(16)(17)(18). To date, several such inhibitors have been reported (15, 16, 19 -22) and one ALK inhibitor, crizotinib (PF-02341066, Xalkori), has recently been approved to treat EML4-ALK-driven NSCLC.
The role of the native ALK protein is poorly understood, although it is believed to be involved in neuronal development and neural cell differentiation (23). In mice, the extensive ALK mRNA observed in the nervous system during embryogenesis diminishes after birth and is maintained at only a low level in the nervous system thereafter (24 -26). Consistent with these observations, immunohistochemistry of adult human tissues shows a weak ALK signal only in the CNS (27). Aberrant ALK signaling can arise, however, through the amplification or mutation of the full-length protein and ALK has been identified as a driver oncogene in a subset of neuroblastomas, an aggressive form of childhood cancer that originates in the sympathetic nervous system (28 -31). Germline mutations of the ALK gene contribute to many hereditary neuroblastomas, and somatic mutations and gene amplifications contribute to a subset of sporadic neuroblastomas. Most mutations cluster to the ALK tyrosine kinase domain and the most common mutations have been shown to be activating on the basis of higher constitutive ALK phosphorylation and their transforming ability in cells (30 -33). The predominant mutations identified from patient samples and neuroblastoma cell lines are F1174L and R1275Q (34). As with the ALK fusion proteins, the neuroblastoma activating mutants are amenable to inhibition by small molecule inhibitors of the ALK kinase activity, although differential sensitivity has been observed depending on the particular inhibitor and mutant (33,35). Interestingly, the F1174L variant and the related F1174C variant have been independently identified in the clinic as a mutations conferring resistance to crizotinib treatment (36,37).
A structural understanding of inhibitor binding to ALK was recently enabled by the publication of crystal structures of the ALK kinase domain both alone and in complex with ATP-competitive inhibitors (38,39). The structures revealed that the ALK kinase domain adopts the canonical kinase-fold, but that it also contains two notable features. First, a portion of the juxtamembrane region forms a ␤-hairpin turn that packs against the ␣C-helix from the N-terminal domain of the kinase. Second, the activation loop (A-loop) forms a short, ␣-helix that packs against the ␣C-helix. This helical A-loop conformation has been observed in nearly all ALK crystal structures published to date and its conformation is incompatible with an active kinase. Importantly, all of the published ALK crystal structures use unphosphorylated protein. Interactions of the A-loop ␣-helix with both the N-terminal and C-terminal lobes of the kinase and a hydrogen bond between Tyr 1278 and Cys 1097 from the N-terminal ␤-turn motif serve to stabilize the observed conformation. The fact that Tyr 1278 is phosphorylated upon formation of fully activated ALK underscores the inactive nature of the observed structures (40,41). The fully activated ALK kinase is expected to resemble the activated form of the insulin receptor kinase (IRK), the structure of which has been reported previously using the Tris-phosphorylated IRK kinase domain crystallized with a substrate peptide and an ATP analog (42). Several structural features of the published, unphosphorylated ALK kinase domain differ from the structural template provided by the IRK ternary structure and interestingly, ALK also differs from the unphosphorylated, inactive form of IRK kinase domain (43). These differences have been described elsewhere (38,39).
Small molecule inhibition of ALK kinase activity is a promising means of treatment in NSCLC, anaplastic large cell lymphoma, neuroblastoma, and other cancers with an ALK-driven component. Based on this knowledge, we and others have pursued programs to discover novel, small-molecule inhibitors of ALK that are suitable for clinical application. During the course of these investigations, we sought to understand the structural basis for activation of ALK by the two most common neuroblastoma mutants, F1174L and R1275Q. We report here crystal structures of both mutants and we use the published structure of the ALK-crizotinib complex to show that these mutants produce no steric impediment to crizotinib binding (19). In addition, during the course of our medicinal chemistry efforts two other classes of ALK inhibitors were identified that stabilized the ALK A-loop in previously unobserved conformations. The crystal structure of ALK in complex with one such compound bound to ALK in a classic type II configuration is reported here. Type II kinase inhibitors bind to the ATP-binding site and also occupy a largely hydrophobic pocket formed by shifting the activation loop to an inactive, DFG-out conformation. In contrast, type I inhibitors occupy the ATP-binding site of the kinase in its active, DFG-in conformation (44). A second series of compounds was identified that bound to a unique, DFGshifted conformation of the enzyme (45). Also included in this report is the strategy that we used to obtain ALK kinase domain crystals. The structures reported herein provide novel views of the ALK active site that may prove useful for structure-based drug design of a new generation of ALK inhibitors, they provide a detailed view of the binding of crizotinib to the most common neuroblastoma activating mutants, and they also provide insight into the activating nature of the R1275Q mutant.

EXPERIMENTAL PROCEDURES
Molecular Biology-Plasmids encoding recombinant ALK proteins were constructed using modified pFastBac1 vectors (Invitrogen). The pFastBac1 G2T vector allows the fusion of a GST affinity tag and a thrombin protease cleavage site to the N terminus of target proteins, whereas the pFastBac HGT vector carries a His 6 and GST dual affinity tag followed by a tobacco etch virus protease cleavage site. A human ALK DNA fragment encoding amino acids 1060 -1620 was amplified from a human cDNA library (Clontech), digested with restriction enzymes BglII and EcoRI, and cloned into a pFastBac1 G2T vector between the BamHI and EcoRI sites. The resulting plasmid pFastBac1 G2T-ALK(1060 -1620) was used as a DNA template for the downstream cloning. pFastBac1 G2T-ALK(1058 -1410) and pFastBac1 HGT-ALK(1084 -1410) were made by subcloning PCR-amplified DNA fragments ALK(1058 -1410) and ALK(1084 -1410) into pFastBac1 G2T and pFastBac1 HGT vectors, respectively. Finally, C1097S, F1174L, and R1275Q substitutions were introduced into expression constructs by site-directed mutagenesis (Stratagene).
Expression and Purification-ALK proteins were expressed using a Bac-to-Bac baculovirus expression system (Invitrogen). Briefly, expression plasmids were transformed into DH10Bac competent cells (Invitrogen) and selected on LB agar plates containing kanamycin, tetracycline, gentamicin, isopropyl 1-thio-␤-D-galactopyranoside, and Bluo-gal. The resulting single white colonies were used to generate recombinant BacMid DNA, which was then used to transfect Sf9 cells to generate recombinant baculovirus stocks. After two rounds of amplification, high-titer baculoviruses were used to express recombinant ALK proteins by infecting mid-log phase Sf9 cells at a multiplicity of infection of 5. Insect cells were allowed to grow in serum-free sfx medium (Hyclone) for an additional 64 h at 27°C.
Crystallization-The ALK kinase domain was crystallized initially by hanging drop vapor diffusion at 4°C using in situ proteolysis (46). Purified ALK(1058 -1410) C1097S in 25 mM Tris-HCl, pH 8.5, 125 mM sodium chloride, 10% (w/v) glycerol, and 14 mM 2-ME was concentrated to 12 mg/ml in an Ultrafree-0.5 concentrator (Millipore). Inhibitor was added at a concentration of ϳ0.7 mM (3% (v/v) DMSO) and either trypsin or chymotrypsin (Sigma) was added to the sample at a ratio of 1 g of protease/100 g of ALK. After 30 min incubation on ice, the sample was dispensed into a 96-well hanging drop crystallization tray (Corning) using a Mosquito (TTP Laptech) and the trays were stored at 4°C. Drops consisted of 200 nl of well solution plus 200 nl of protein solution. Crystals were observed after 7 days in a drop containing chymotrypsin-treated ALK that was suspended over a well solution containing 30% (w/v) PEG 5000 MME, 0.1 M MES, pH 6.5, and 0.2 M ammonium sulfate. No crystals were observed under the same conditions with the trypsin-treated ALK sample. Later experiments showed that substituting the endoproteinase GluC (Roche Diagnostics) for chymotrypsin produced crystals under the same conditions. Protease treatment of larger batches of ALK(1058 -1410) C1097S was used to identify the constructs that produced crystals (see below).
Crystals of ALK(1084 -1410) C1097S containing either the F1174L or R1275Q mutations were grown by hanging drop vapor diffusion in 24-well trays at 4°C. All crystals grew in space group P2 1 2 1 2 1 with approximate dimensions a ϭ 51.6 Å, b ϭ 57.5 Å, and c ϭ 105.6 Å. Apo-crystals of these proteins and crystals of the wild-type protein to which 1 mM compound 2 had been added appeared in drops formed by mixing equal volumes of 11 mg/ml of protein with 18 -20% (w/v) PEG 5000 MME, 0.1 M MES, pH 6.5, 0.1-0.2 M ammonium sulfate, and 5 mM dithiothreitol. Microseeds were introduced into the drops after 24 h to produce large, single crystals suitable for data collection. Crystals of R1275Q ALK in complex with compound 1 were grown by hanging drop vapor diffusion at 4°C directly out of a 96-well PEGs screen (Qiagen). To a 11 mg/ml R1275Q-ALK solution was added 0.8 mM compound 1, 10 mM dithio-threitol, and 4 l of a concentrated microseed stock formed by crushing crystals of apo-R1275Q-ALK in 20 l of mother liquor. After thorough mixing, drops consisting of 200 nl of protein plus 200 nl of well solution were dispensed using a Mosquito (TTP Laptech). Single crystals suitable for data collection grew in condition 96 (20% PEG 3350, 0.2 M diammonium citrate). Several other drops also contained crystals of lesser quality. Prior to data collection, the crystals were transferred briefly to solutions containing the mother liquor supplemented with 20% (v/v) glycerol, and then flash-cooled in liquid nitrogen. This same cryoprotocol was used for all ALK crystals.
Data Collection and Structure Refinement-Diffraction data from the crystals described above were collected at synchrotron sources. The apo-R1275Q-ALK data were collected at beamline 5.0.1 of the Advanced Light Source, Berkeley, CA, using an ADSC Q210 CCD detector and ϭ 0.9774 Å. The apo-F1174L-ALK and R1275Q-ALK-compound 1 data sets were collected at beamline 22-ID of the Advanced Photon Source, Chicago, IL, using a MAR 300 CCD detector and ϭ 1.0000 Å. Finally, the wild-type ALK-compound 2 data were collected at the Canadian Light Source, Saskatoon, SK, Canada, on beamline CMCF1 using ϭ 0.9793 Å and a MAR 225 CCD detector. Diffraction data were processed with the HKL suite of programs (47). Phases for our initial ALK structure were obtained by molecular replacement using AMORE and the insulin receptor kinase domain (PDB 1IR3) as a search model (42,48). Subsequent structure refinements used REFMAC and model building was performed with COOT (49,50). Figures were prepared with PyMOL (Schrodinger, LLC). Data collection and structure refinement statistics appear in Table 1.
ALK Kinase Activity Assays-The catalytic activity of the intracellular domain of ALK (amino acids 1058 -1620) was measured as described previously (51). Activity measurements of the ALK mutants occurred under similar conditions, but the ATP concentration in the assays was 20 and 40 M for the F1174L and R1275Q variants, respectively, to account for differences in the apparent K m,ATP .
Proteolysis and N-terminal Amino Acid Sequencing-20-l aliquots of G2T-ALK(1058 -1410) at 0.5 mg/ml were incubated with 40 ng of sequencing grade trypsin, chymotrypsin, or endoproteinase GluC (Roche Applied Sciences) at 4°C. Aliquots withdrawn at 4, 17, and 48 h were solubilized with NuPAGE sample buffer (Invitrogen) containing 5% 2-ME and heated immediately for 10 min at 95°C followed by rapid freezing. SDS-PAGE was used to monitor the extent of proteolysis (supplemental Fig. S1). Replicates of these gels were blotted to PVDF membranes (Millipore Corp, Inc.). Bands were excised and subjected to N-terminal sequencing on an Applied Biosystems 494HT protein sequencer.

Crystallization of the ALK Kinase Domain by in Situ
Proteolysis-We employed a strategy to devise constructs for crystallization based on published structures of the kinase domains of other receptor tyrosine kinases. Several of our constructs employed the C1097S mutation to improve protein behavior, a choice inspired by a similar mutation in IRK (43). Although the expression of several of these constructs gave sol-uble protein in insect cells that could be purified to near homogeneity, we were unsuccessful in obtaining crystals (data not shown). Consequently, a soluble, well behaved construct that encompassed the juxtamembrane and kinase domain portions of the ALK intracellular domain (ALK(1058 -1410) C1097S) was subjected to crystallization with in situ proteolysis using trypsin, chymotrypsin, or GluC (46). Using this protocol, ALK crystals in complex with ATP competitive inhibitors were obtained with both the chymotrypsin and GluC-containing samples, but not with the trypsinized sample. Analysis of the proteolysis products by N-terminal sequencing identified the cleavage sites (supplemental Fig. S1). Chymotrypsin cleaves N-terminal to the kinase domain predominantly at Leu 1083 , with a minor cleavage site at Met 1089 , and it also cleaves in the activation loop at Tyr 1282 /Tyr 1283 . GluC cleaves at a single site N-terminal to the kinase domain at Glu 1077 and does not cut the activation loop. On the basis of this analysis we made ALK(1078 -1410) C1097S and ALK(1084 -1410) C1097S constructs. Both constructs expressed well and crystallized readily. All of the structures described in this work use ALK(1084 -1410) C1097S. By way of comparison, ALK kinase domain structures have also been determined using constructs with the boundaries 1072-1410, 1094 -1407, 1093-1411, and 1069 -1411 (19,38,39,52).
Crystal Structures of the Apo-F1174L and -R1275Q Neuroblastoma Mutants-The crystal structure of the wild-type, unphosphorylated ALK kinase domain revealed the exact location of the activating mutations identified in neuroblastoma, although prior efforts had deduced their location based on sequence homology to related kinases (28). All of the mutations cluster around the active site of ALK. One of the most commonly mutated residues, Phe 1174 , sits in a hydrophobic pocket at the base of the ␣C-helix, where it is involved in hydrophobic stacking interactions with other phenylalanine residues from the N-terminal ␤-turn region (Phe 1098 ), the activation loop (Phe 1271 ), and C-terminal kinase domain (Phe 1245 ) as described by others (35,38,39). The second commonly mutated residue, Arg 1275 , is found in the activation loop where it is part of the ␣-helix formed in the unphosphorylated ALK kinase domain structures. Arg 1275 forms hydrogen bonds to the backbone carbonyl of Asp 1163 on the ␣C-helix and to neighboring residue Asp 1276 , both of which likely help to enforce the observed ␣-helical A-loop conformation.
In an effort to understand how these mutations might disrupt the wild-type, inactive ALK kinase domain conformation, we crystallized and solved the structures of the two most common activating mutations in neuroblastoma, F1174L and R1275Q. Both structures used unphosphorylated protein. We purposely sought to obtain apo structures of these variants to understand the preferred protein conformation in the absence of inhibitor. Crystals of the apo-F1174L ALK kinase domain diffracted to 1.75-Å resolution and nearly the entire sequence was visible from Arg 1084 to Val 1405 , although portions of the P-loop (His 1124 -Phe 1127 ), A-loop (Arg 1279 -Gly 1287 ), and C terminus (Glu 1406 -Lys 1410 ) were missing from the electron density. The loop connecting strands ␤2 and ␤3 (Gly 1137 -Phe 1142 ) was also disordered. The structure was of high quality and the electron density maps were clearly consistent with the presence of Leu at position 1174 (Fig. 1C). It was also readily apparent that the conformations of the residues in the hydrophobic cluster surrounding Leu 1174 were not altered by the F1174L mutation (Fig.  1B). Indeed, the entire structure of ALK F1174L closely matched that of apo, wild-type ALK with an r.m.s. deviation of 0.6 Å for 284 C␣ atoms. This similarity included the ␣-helix at the beginning of the activation loop and the conformation of the ␤-turn N-terminal to the kinase domain (Fig. 1A). 3 Although the F1174L muta- tion might have been expected to disrupt the hydrophobic interactions clustered around the base of the ␣C-helix, clearly the conformational preference of the F1174L protein, at least under the crystallization conditions, was overwhelmingly similar to the wild-type protein. Other aspects of the ALK F1174L kinase domain structure also resembled features observed in the wild-type protein, notably the conformation of the DFG motif at the beginning of the A-loop, the position of the ␣C-helix, and the relative degree of closure between the N-and C-terminal lobes of the kinase. These features, together with the positioning of the C-terminal portion of the A-loop that sterically blocks the substrate binding site, have all been discussed previously as contributing to an overall inactive kinase conformation (38,39). In contrast to the F1174L ALK crystal structure, the structure of the R1275Q ALK kinase domain showed a dramatic difference in the activation loop conformation compared with the wild-type protein (r.m.s. deviation of 6.1 Å for the first nine A-loop C␣ atoms). Crystals of the apo-R1275Q ALK kinase domain diffracted to 1.70-Å resolution and once again the electron density maps were of excellent quality. Not only did they show electron density consistent with the desired Gln 1275 mutation, but they also clearly showed the activation loop in a new, more extended conformation (Fig. 1D). The small ␣-helix that had been observed in the other structures was no longer present. In the new conformation, Asp 1270 and Phe 1271 of the DFG sequence at the beginning of the activation loop are in the same position as observed previously. Gly 1272 of the DFG sequence shifts slightly, however, and the side chain of Met 1273 is oriented toward the ␣C-helix as the result of a ϳ180°rotation around its N-C␣ bond. This rearrangement places Gln 1275 adjacent to the C-terminal lobe of the kinase where its side chain hydrogen bonds to Ser 1314 and its backbone NH hydrogen bonds to the backbone carbonyl of Arg 1248 . Asp 1276 from the activation loop also forms a hydrogen bond to Arg 1248 , a residue that is notable because it is part of the HRD motif that is important in positioning the activation loop in a catalytically competent orientation (53). Although the activation loop ␣-helix is disrupted in the R1275Q structure, the kinase does not adopt an active conformation. Arg 1248 from the HRD motif does not form a hydrogen bond to the residue following the DFG sequence (the DFG ϩ 1 residue), as observed in structures of active kinases, but rather it adopts a rotamer that allows it to interact with the backbone carbonyl of Ile 1246 , thereby sterically blocking the position of the short ␤9-strand formed in an active A-loop conformation. A comparison of ALK R1275Q with the triply phosphorylated IRK in complex with an ATP analog and a substrate peptide (PDB 1IR3) highlights these differences (Fig.  1E) (42).
Other features aside from the A-loop that characterized the wild-type and F1174L ALK structures as inactive are also present in the R1275Q structure. These include the position of the ␣C-helix, the relative degree of closure between the N-and C-terminal lobes of the kinase domain, as well as the position of the C-terminal portion of the activation loop that blocks the substrate binding site. Even the ␤-turn N-terminal to the kinase domain is preserved, despite the loss of hydrogen bonding interactions between it and the A-loop. These structural similarities are highlighted by the relatively small r.m.s. deviation of 0.6 Å for 271 C␣ atoms when comparing the R1275Q and wildtype ALK structures, but excluding the A-loop from the calculation. As in the F1174L structure, the majority of the polypeptide sequence from Asn 1093 to Leu 1404 is visible in the electron density, although portions of the N terminus (Arg 1084 -Tyr 1092 ), C terminus (Leu 1404 -Lys1410), P-loop (Gly 1123 -Gly 1128 ), A-loop (Arg 1279 -Gly 1287 ), and ␤2-␤3 loop (Gly 1137 -Asp 1141 ) are disordered. Enough of the A-loop is ordered to observe Tyr 1278 , one of the residues phosphorylated upon ALK activation, but the electron density shows an unphosphorylated side chain, which is consistent with our characterization of the  OCTOBER 26, 2012 • VOLUME 287 • NUMBER 44 protein as unphosphorylated coming out of Sf9 cells. The fact that Tyr 1278 makes no specific hydrogen bonding interactions in this structure and is adjacent to the beginning of the disordered region of the A-loop is clearly suggestive of more facile autophosphorylation of this residue. As previous studies have shown that Tyr 1278 , the first residue in the activation loop YXXXYY motif, is a key driver of ALK activation, the R1275Q structure provides a structural rationale of the activating nature of this mutant (40,41).

Alternative Activation Loop Conformations of ALK
Models of Crizotinib Binding to the F1174L and R1275Q Neuroblastoma Mutants-The F1174L and R1275Q ALK neuroblastoma mutants can be inhibited by ATP-competitive inhibitors including crizotinib (33,35). Literature IC 50 values for crizotinib inhibition of the F1174L and R1275Q ALK mutants range from 89 -130 and 67-85 nM in in vitro enzyme assays, respectively (35). In cellular assays, it is generally accepted that the F1174L mutant displays reduced sensitivity to crizotinib relative to the R1275Q mutant or to wild-type enzyme, although the reported level of reduced sensitivity varies (33,35,36,54). In the clinical setting, recent reports also identify the F1174L variant and the related F1174C variant as secondary mutations conferring resistance to crizotinib therapy in patients harboring an oncogenic ALK fusion protein (36,37). The reduced sensitivity of F1174L ALK to crizotinib is reported to be due, at least in part, to a reduced K m,ATP and an increased catalytic efficiency in this mutant (35). Having determined the structures of the apo forms of the F1174L and R1275Q ALK kinase domain, we sought to generate models of these mutants in complex with crizotinib, an agent that is currently being tested in clinical trials in neuroblastoma (trial No. NCT00939770, www.clinicaltrials.gov).
Model generation of crizotinib binding to the F1174L and R1275Q neuroblastoma mutants used the published co-crystal structure of crizotinib in complex with wild-type ALK (19) and took advantage of the similarities between the mutant and wildtype structures. Both secondary structure matching and simple least squares matching algorithms gave excellent superposition of the F1174L-ALK and R1275Q-ALK structures onto the wildtype ALK-crizotinib structure (PDB ID 2XP2). Superposition of the F1174L-ALK and wild-type ALK-crizotinib structures gave a r.m.s. deviation of 0.5 Å for 283 Ca atoms and alignment of the R1275Q-ALK structure gave a r.m.s. deviation of 0.4 Å for 271 Ca atoms (A-loop residues were excluded). As described above, the F1174L and R1275Q mutations do not disrupt the conformation of Asp 1270 or Phe 1271 at the beginning of the activation loop that define one face of the ATP-binding site of ALK. Consequently, superposition of the mutant structures onto the ALK-crizotinib co-crystal structure generated models that showed crizotinib fitting into the ATP-binding site of both the F1174L and R1275Q mutants with no steric clashes and no need for invoking alternate side chain conformations of amino acids lining the active site pocket (Fig. 2). These findings are consistent with the reported in vitro inhibition of each mutant by crizotinib and with the in vivo inhibition of R1275Q-ALKdriven neuroblastoma xenografts (35,55). The models fail to explain the reduced sensitivity of the F1174L-ALK mutant to crizotinib, but a higher catalytic efficiency of the F1174L ALK mutant is consistent with the same binding mode for the inhib-itor despite weaker activity in vivo. Indeed, similar results have been reported for the L1196M-ALK mutant, a clinically observed kinase gatekeeper mutation that confers resistance to crizotinib treatment (56). Much like the F1174L mutant, the L1196M mutant is reported to have a higher catalytic efficiency than the wild-type enzyme, but it is still inhibited by crizotinib at concentrations 10-fold higher than the wild-type enzyme in vitro assays (57). The crystal structure of crizotinib bound to the L1196M-ALK mutant shows the same binding conformation as in the wild-type enzyme (PDB 2YFX). These results provide precedent for the model of crizotinib-inhibited F1174L-ALK.
Crystal Structure of the ALK Kinase Domain with a Type II Inhibitor-During the course of our medicinal chemistry work targeting ALK, we identified a number of compounds from other kinase programs that were potent ALK inhibitors (data not shown). From these compounds, one chemotype based on a benzoxazole core was intriguing because it was known to be a type II inhibitor of VEGFR-2 and other closely related receptor tyrosine kinases (44,58). Compound 1 (shown in Fig. 3A) has an IC 50 ϭ 0.256 M for wild-type ALK in an in vitro enzyme assay. It inhibits F1174L ALK somewhat more weakly (IC 50 ϭ 0.734 M), but it is quite potent against the R1275Q ALK mutant (IC 50 ϭ 0.016 M). If compound 1 was a Type II inhibitor of ALK, one could rationalize its increased potency against the R1275Q mutant based on the altered conformation of the ALK A-loop in the R1275Q variant as detailed above. To confirm the binding mode of compound 1 in ALK and investigate the molecular details of a DFG-out ALK structure, we sought to co-crystallize compound 1 with the ALK kinase domain. Although we were unsuccessful with the wild-type protein, we successfully crystallized it with the R1275Q ALK kinase domain. The structure was solved to 2.45-Å resolution and the electron density for the bound inhibitor was very clear (Fig. 3B). As predicted, the activation loop had undergone a dramatic rearrangement into a DFG-out conformation (44). As part of this rearrangement, the carbonyl group of Gly 1269 , the residue immediately preceding the DFG sequence, flipped 180 degrees and formed a hydrogen bond to His 1247 of the HRD sequence. To our knowledge, this is the first ALK crystal structure to show Gly 1269 in this more standard orientation. Phe 1271 of the DFG sequence now occupies parts of the ribose and phosphate region of that ATP-binding site where it sits adjacent to Val 1180 . The pocket formerly occupied by Phe 1271 is now largely filled by the 4-chlorophenyl group of the inhibitor. The central benzoxazole ring fills a largely hydrophobic pocket between catalytic Lys 1150 and gatekeeper residue Leu 1196 and the dimethoxyquinoline ring is hydrogen bonded to Met 1199 from the kinase hinge region. The nitrogen atom of the benzoxazole ring is within hydrogen bonding distance of the backbone NH of Asp 1270 from the DFG sequence but the hydrogen bond geometry is poor, probably due to the effect of the larger leucine gatekeeper residue in orienting the plane of the central benzoxazole ring. The secondary amine linking the benzoxazole and 4-chloroquinoline rings is also involved in a specific hydrogen bond with Glu 1167 from the ␣C-helix, which is in turn bonded to catalytic Lys 1150 . Although this Lys 1150 -Glu 1167 hydrogen bond is also present in the apo-ALK R1275Q structure, the position of the N-terminal portion of the ␣C-helix in the cocrystal structure with compound 1 is shifted outwards from the body of the kinase by nearly 3 Å. Concurrent with this shift is disorder in the ␤-turn segment preceding the kinase domain. Although our protein construct began with Arg 1084 , the first visible residue was Ser 1103 . That compound 1 can bind to this altered ALK structure, trapping a DFG-out conformation, highlights the inherent ability of the protein to undergo various structural perturbations. Such perturbations are a necessary part of shifting from an inactive to an active kinase conformation. From a drug discovery perspective, such varied inactive conformations can also present new pockets to exploit during inhibitor design. The structure reported here of R1275Q ALK in complex with compound 1 does just that in presenting a previously unobserved active site topology in ALK that may be useful for the design of novel type II inhibitors of this kinase.
Crystal Structure of the ALK Kinase Domain with a Non-Type II Extended Binding Inhibitor-In addition to identifying a type II inhibitor of ALK, we also identified a series of molecules that bound to the ALK kinase domain in an extended conformation, trapping the protein somewhere between a type I and type II configuration. A report detailing our work on this series of compounds has appeared recently, and it includes a crystal structure of the wild-type ALK kinase domain in complex with one such compound (PDB 4DCE) (45). The crystal structure of another, more potent compound (Fig. 4A) from this series (IC 50 ϭ 0.016 M) was also determined with wild-type ALK and the salient features of this structure are summarized here. Sitting in an extended conformation, the aminopyrimidine ring of compound 2 hydrogen bonds to Met 1199 from the kinase hinge region, the piperidine ring provides the proper trajectory for the amide functionality to traverse the pocket adjacent to gatekeeper residue Leu 1196 , and the 3-trifluoromethoxybenzyl group occupies part of the pocket normally filled by Phe 1271 from the DFG sequence (Fig. 4B). The amide makes specific hydrogen bonds to catalytic Lys 1150 and to the backbone carbonyl oxygen atom of Gly 1269 , the residue preceding the DFG sequence. The Glu 1269 carbonyl moiety pivots by about 40 degrees relative to its position in the wild-type, apo-ALK structure to optimize hydrogen bond formation with the inhibitor. Asp 1270 and Phe 1271 shift to allow formation of the pocket that accommodates the 3-trifluoromethoxybenzyl group of the inhibitor. The side chain of Phe 1271 caps this pocket and effectively blocks access to bulk solvent. The shifting of these resi-   OCTOBER 26, 2012 • VOLUME 287 • NUMBER 44 dues in the DFG sequence disrupts ␣-helix formation in the A-loop and residues from Met 1273 -Gly 1287 are disordered in the structure. Despite the absence of the A-loop ␣-helix, the conformation of the ␤-turn N-terminal to the kinase domain remains intact and the gross features of the kinase domain structure are maintained. These include the position of the ␣C-helix and the relative degree of closure between the N-and C-terminal domains of the kinase. The same maintenance of overall structural features despite a conformational change in the A-loop was also observed in the R1275Q ALK structures and it emphasizes the plasticity of the activation loop in the unphosphorylated ALK protein. Such observations are not surprising because the activation loop of kinases has long been known to be particularly amenable to altered conformations, especially in the inactivated form (59). These varied conformations give rise to uniquely shaped pockets that are exploitable for drug design, and the "DFG-shifted" form of ALK in complex with compound 2 provides another such view.

DISCUSSION
Successful determination of the ALK crystal structures reported here relied upon the identification of an appropriate protein construct that was amenable to crystallization. We utilized an in situ proteolysis protocol to generate initial crystals and then analyzed the resulting species to obtain two protein constructs that were well suited for structural studies. Others have arrived at suitable constructs by independent means. It is interesting to note, however, that of the five different protein constructs used in the ALK kinase domain structures reported to date, including the one in this report, all have crystallized in the same space group. The intermolecular packing arrangement that is obviously favored by the ALK kinase domain is mediated in part by interactions between the N-terminal and C-terminal portions of the crystallizable sequences (see supplemental Fig. S2). These interactions would be absent in smaller protein constructs and may explain why all of the ALK protein constructs from which structures have been reported contain at least the minimal sequence 1094 -1407.
The alternative, inactive conformations of the ALK kinase domain presented here demonstrate that structural perturbations occur within the ALK kinase domain and that they can be stabilized and characterized by protein crystallography with the use of the appropriate mutant or inhibitor. In the case of the R1275Q ALK neuroblastoma mutant, the structure showed that disruption of the hydrogen bonding interactions between the ␣C-helix and the ␣-helical A-loop observed in the wild-type structures was enough to shift the equilibrium to a new A-loop conformation. In this altered A-loop conformation, Tyr 1278 is no longer engaged in a hydrogen bond with Cys 1097 from the N-terminal ␤-turn of the kinase and therefore its phosphorylation, and hence the activation of ALK, may be more facile. The other reported Arg 1275 neuroblastoma mutation, R1275L, is likely to work through the same mechanism (29). The structure of the F1174L ALK neuroblastoma mutant showed no such changes, although the structural results do not rule out the hypothesis that disruption of the packing around Phe 1174 in the hydrophobic cluster involving residues from the A-loop, C-helix, C-terminal kinase lobe, and N-terminal ␤-turn shifts a conformational equilibrium toward an active form of the kinase, as others have suggested (35,39). Mutations in neuroblastoma of Phe 1174 to residues other than leucine, as well as mutations of Phe 1245 to other smaller hydrophobic residues (Cys, Val, Leu, Ile), likely work the same way (23,37).
The models of crizotinib bound to the F1174L and R1275Q neuroblastoma mutants strongly suggest a binding mode for the inhibitor identical to its structure in the wildtype protein. Because the apo structures of F1174L and R1275Q ALK showed no disruption to the residues lining the ATP-binding site of the ALK kinase domain, this result was consistent with our expectations. In other ALK neuroblastoma mutants that are expected to function similarly to the F1174L and R1275Q mutants, as outlined above, the binding mode of crizotinib is expected to be similarly unaffected. Although some mutants may display a reduced sensitivity to crizotinib, as shown for the F1174L and L1196M mutations, diminished potencies resulting from competition with tighter binding of ATP can be overcome by administering higher concentrations of inhibitor (35,57). Inhibitors with higher potencies on these ALK mutants will also be useful and reports of selected examples of such molecules have already appeared (33,52,54).
Of the ATP competitive inhibitors that have been structurally characterized with ALK to date, all bind to a similar protein conformation, namely the inactive, unphosphorylated wildtype conformation with the short ␣-helix at the start of the A-loop. Several of these inhibitors, NVP-TAE684, PHA-E429, as well as crizotinib, take advantage of binding to a hydrophobic "shelf" on top of Leu 1256 and adjacent to Gly 1269 in addition to their more traditional interactions with the hinge region of the kinase (Fig. 5). The exception is CH5424802, which binds in a more linear fashion to the hinge region of ALK and extends further back into the pocket adjacent to gatekeeper residue Leu 1196 . Both types of inhibitors display exquisite potency on ALK and a third class of molecules combining elements of both of the first two classes produces another type of potent, selective ALK inhibitor (60). Importantly, the contour of the portion of the ATP site into which these inhibitors bind is not affected in the F1174L or R1275Q neuroblastoma mutants, as described above, nor is it expected to be perturbed upon ALK kinase domain phosphorylation. ALK is somewhat unique in this sense, because other kinases with defined inactive A-loop conformations display a greater change in the contour of the ATPbinding site. IRK and IGF-1R fall into this category, despite their similarity to ALK, because their unphosphorylated A-loops shift to a DFG-out conformation (43,61). Even kinases with inactive conformations displaying ␣-helical segments in their A-loops, such as Fak and Nek2, differ from ALK because the position of the DFG segment is perturbed as a result of the A-loop helix (62,63). Of course, inactive kinase conformations can be targeted by potent, small molecule inhibitors and there are multiple clinical examples of such compounds. The most notable is Gleevec (imatinib), which is a type II kinase inhibitor targeting the DFG-out conformation of its targets c-Abl and c-kit (64). Crizotinib itself does this in its complex with c-Met, the kinase toward which it was initially targeted, by interacting with Tyr 1230 on a unique, inactive conformation of the A-loop (19). Perturbations to the wild-type, inactive ALK kinase domain structure do occur and alternative conformations can be trapped by small molecule inhibitors as we have shown. The co-crystal structures of ALK with compounds 1 and 2 provide examples of two such alternative, inactive conformations. As shown in the complex with compound 1, ALK can be inhibited by type II inhibitors. The complex with compound 2 shows another A-loop geometry that can be exploited by inhibitors that bind in an extended conformation. Interestingly, the extended hydrophobic pocket in both the DFG-out and DFGshifted ALK structures are lined on one side with Phe 1174 and Phe 1245 , two residues that are mutated in neuroblastoma. This observation raises the intriguing question of whether compounds could be designed to target mutants of these residues with increased potency and selectivity over the wild-type enzyme.
In conclusion, we have presented crystal structures of the ALK kinase domain containing the two most common activating mutations in neuroblastoma. A novel A-loop conformation in the R1275Q mutant structure helps explain its status as an activating mutation. In addition, we used these structures to generate models of their complexes with crizotinib, the first ALK-targeted therapy to receive FDA approval. These models strongly suggest that crizotinib binds to these mutants in the same manner as it does to the wild-type enzyme. Also presented were structures of the ALK kinase domain in complex with small molecule, ATP-competitive inhibitors that bound to two alternative, inactive conformations of the ALK A-loop. One small molecule was a classic type II inhibitor and the other binds in a somewhat unique, DFG-shifted conformation of the A-loop. Collectively, these structures present novel ALK conformations that may prove useful in the structure-based design of a new generation of inhibitors of ALK, a kinase that is already a clinically validated target in oncology.