Characterizing and Overriding the Structural Mechanism of the Quizartinib-Resistant FLT3 “Gatekeeper” F691L Mutation with PLX3397

Secondary mutations in the tyrosine kinase domain (TKD) remain the most commonly encountered cause of acquired clinical resistance to small-molecule tyrosine kinase inhibitors (TKI) in human cancer (1–4). Recent pharmaceutical efforts have focused on the development of “type II” kinase inhibitors, which bind to a relatively nonconserved inactive kinase conformation and exploit an allosteric site adjacent to the ATP-binding pocket as a potential means to increase kinase selectivity, although recent data suggest such efforts may be misguided (5). Commonly, kinase domain (KD) mutations effect resistance to type II inhibitors via two mechanisms: (i) substitution of amino acid positions directly involved in binding inhibitor, or (ii) mutation of residues that stabilize the inactive kinase conformation required for binding (6). Type I inhibitors, which bind to the more conserved active kinase conformation, are typically vulnerable only to mutations at inhibitor contact residues and therefore may be generally affected by a reduced spectrum of resistance-causing KD mutations (7, 8).

FMS-like tyrosine kinase 3 (FLT3) is a class III receptor tyrosine kinase that is expressed in hematopoietic cells. Mutations in FLT3 represent the most common genetic alteration in patients with acute myeloid leukemia (AML; ref. 9) and are comprised primarily of constitutively activating internal tandem duplication (ITD) mutations (of 1–100 amino acids) in the juxtamembrane domain and, less commonly, point mutations, typically within the kinase activation loop. We recently identified secondary KD mutations in FLT3-ITD that can cause preclinical and acquired clinical resistance to the highly potent type II FLT3 inhibitor quizartinib (10), which achieved a composite complete remission (CRc) rate of approximately 50% in relapsed or chemotherapy-refractory FLT3-ITD⁺ AML patients in monotherapy studies (11). An in vitro saturation mutagenesis screen of FLT3-ITD identified five quizartinib-resistant KD mutations at three residues: the “gatekeeper” F691 residue and two amino acid positions within the kinase activation loop (D835 and Y842), a surprisingly limited spectrum of mutations for a type II inhibitor. Mutations at two of these residues (F691L and D835V/Y/F) were subsequently identified in each of eight samples analyzed at the time of acquired clinical resistance to quizartinib (10), a finding that definitively validated FLT3 as a therapeutic target in AML. The type II multikinase inhibitor sorafenib, which also has some clinical activity in FLT3-ITD⁺ AML, is ineffective against all identified quizartinib resistance–causing mutants, in addition to other mutant isoforms (10, 12). Although we (8) and others (13, 14) recently identified the type I inhibitor crenolanib as an equipotent inhibitor of quizartinib-resistant D835 mutants, no FLT3 inhibitor has demonstrated equipotent inhibition of the F691L mutant, including the ABL/FLT3 inhibitor ponatinib, which was rationally designed to retain activity against the problematic gatekeeper T315I mutant in BCR–ABL (15).

Mutation of the conserved kinase gatekeeper residue in the ATP-binding pocket, which controls access to an allosteric site adjacent to this pocket, represents a well-documented cause of clinical resistance to inhibitors of a number of pathologically activated kinases in addition to FLT3-ITD and BCR–ABL, including FIPL1–PDGFRα (3), KIT (16), EGFR (2), and EML4–ALK (4). Cocrystal structure analyses of TKI-bound kinases have suggested that the molecular mechanism of TKI resistance due to gatekeeper mutations is largely attributable to steric hindrance resulting from substitution of a bulkier amino acid residue for a smaller (commonly threonine) gatekeeper residue (6, 17). In the case of the EGFR gatekeeper T790M mutation, increased affinity for ATP has been reported to also play a role (18). Notably, FLT3 is one of a limited number of kinases to harbor a bulky phenylalanine at the gatekeeper position. Although the non–ligand-bound structure of intracellular FLT3 has been described (19), no cocrystal structure of FLT3 in complex with ATP or any other ligand has been reported, greatly limiting our structural understanding of this pathologically important kinase. To elucidate the structural mechanism of resistance mediated by F691L and other quizartinib-resistant mutants, we solved the cocrystal structure of quizartinib bound to the FLT3 KD, with the ultimate aim of informing the rational development of novel FLT3 inhibitors with the potential to retain activity against these mutants.

In the costructure with FLT3 (Fig. 1A; Supplementary Fig. S1; Supplementary Table S1), quizartinib adopts the canonical type II kinase inhibitor–binding mode with the imidazobenzothiazole “head” occupying the adenine binding pocket and the t-butyl-isoxazole “tail” residing in the allosteric back pocket of the kinase. The terminal t-butyl group is surrounded by hydrophobic side chains, including L802, from the αE helix that marks the distal end of the back pocket (Fig. 1B). The urea linker between the phenyl and isoxazole rings is hydrogen-bonded to both E661 of the αC helix and the backbone NH of D829 of the Asp-Phe-Gly (DFG) motif, a pattern shared by other urea- and amide-containing type II kinase inhibitors, including imatinib (20), sorafenib (21), and ponatinib (22). The middle phenyl ring forms edge-to-face interactions with both gatekeeper F691 and F830 of the conserved DFG motif. Edge-to-face interaction is the preferred interface between nonsequential aromatic residues inside proteins (23). Perturbation of any component of this compact unit (either F691 or F830) is expected to cause substantial loss of binding affinity. This explains why mutations in FLT3 that confer clinical resistance to quizartinib have been thus far confined to two residues: the gatekeeper F691 and D835, a key residue of the activation loop that determines the conformational state of the DFG motif and thus the orientation of F830. This binding mode places the imidazobenzothiazole of quizartinib inside the narrow cleft between the N- and C-terminal lobes of the kinase abutting the hinge region (Fig. 1A). Quizartinib is an unusual kinase inhibitor in that it lacks a readily identifiable adenine-like moiety that can simultaneously fit the interlobe cleft shape and the hinge hydrogen bonding requirements. Forcing quizartinib to conform to the adenine-like hinge interactions led to binding mode predictions that are not supported by crystallographic studies (ref. 10; Supplementary Fig. S2A and S2B). Unexpectedly, the cocrystal structure revealed a workaround—instead of engaging the hinge directly, the imidazobenzothiazole group recruits a water molecule to form bidentate hydrogen bonds with two polar backbone atoms of the hinge (Fig. 1B; Supplementary Fig. S3). This orientation of the imadazobenzothiazole confers site specificity to the attached morpholinoethoxy aqueous solubilizing group as observed previously (24).

The conjugation between the imidazobenzothiazole and the middle phenyl ring restricts the orientation of the phenyl ring and thus contributes to the edge-to-face interactions. Interestingly, sorafenib—another diarylurea—also exhibited approximately 1,000-fold loss in potency against the gatekeeper mutant F691L compared with native FLT3 (10). The phenyl ring in sorafenib is linked to a conventional hinge-binding scaffold through a flexible ether linker. The optimal fit of sorafenib in FLT3 entails a tilted middle ring orientation relative to quizartinib (Fig. 1C). The cause for F691L-induced resistance to diarylureas is not the orientation of the middle ring, but rather the urea NH proximal to the phenyl ring, which clashes with the C^δ atom of the mutated L691 (Fig. 1D, E, and F). Supporting the conjecture that the urea linker rather than the geometry of the phenyl ring is responsible for F691L resistance, ponatinib, which contains a shorter amide linker (i.e., lacking the corresponding NH of a urea linker; Fig. 1G and H), is less affected (17-fold) by F691L despite having additional substitution on the middle ring (15).

To identify FLT3 inhibitors with activity against gatekeeper mutants, we thus turned to novel chemical structures that possess abridged connectors between the middle and tail rings. PLX3397, a triple kinase inhibitor of CSF1R (enzymatic IC₅₀ = 13 nmol/L), KIT (enzymatic IC₅₀ = 27 nmol/L), and FLT3-ITD (enzymatic IC₅₀ = 11 nmol/L), represents a new chemotype containing a pyridine (rather than phenyl) middle ring and a methyl amine (rather than amide or urea as in other type II kinase inhibitors) linker (W. Tap et al.; submitted for publication; Fig. 2A). Based on in vitro phosphorylation assays, PLX3397 inhibits FLT3 signaling in cells harboring FLT3-ITD (e.g., biochemical IC₅₀ = 18 nmol/L in MV4;11 cells) but is significantly less effective in cells containing only native FLT3 (e.g., biochemical IC₅₀ = 1.8 μmol/L in RS4;11 cells), suggesting bias toward binding the ITD-mutant form of FLT3. This is corroborated by in vitro assays using recombinant proteins that show preferential binding of PLX3397 to FLT3-ITD (IC₅₀ = 10 nmol/L) compared with native, autoinhibited FLT3 (0.4 μmol/L). The anti–FLT3-ITD activity was confirmed in vivo in a MV4;11 xenograft model (Fig. 2B). Based on its chemical structure and inferred binding pose (Fig. 2C and D; Supplementary Fig. S4), we speculated that PLX3397 would retain activity against the F691L mutant. We performed proliferation studies of PLX3397 and quizartinib for Ba/F3 cells expressing FLT3-ITD and FLT3-ITD/F691L and confirmed that PLX3397, in contrast with quizartinib, has the ability to inhibit the proliferation of both isoforms (Supplementary Table S2). PLX3397 equally inhibited the proliferation of MOLM14 F691L cells, which have an acquired F691L mutation in FLT3-ITD following selection in quizartinib, and parental MOLM14 cells (Fig. 2E and Supplementary Table S3). Importantly, PLX3397 achieved equipotent inhibition of FLT3 phosphorylation and downstream signaling in MOLM14 F691L and parental MOLM14 cells in the presence of human plasma, albeit at higher concentrations than those required to inhibit FLT3 signaling in media (Fig. 2F; Supplementary Fig. S5). This suggests that if sufficient plasma drug levels can be obtained, PLX3397 would be expected to be clinically active in both patients with native FLT3-ITD and patients who acquired the FLT3-ITD/F691L as a consequence of prior quizartinib or sorafenib therapy. In contrast, quizartinib demonstrated decreased activity against MOLM14 F691L compared with parental cells in similar assays (Supplementary Fig. S6A and S6B).

We next sought to assess the activity of PLX3397 against other quizartinib-resistant mutants. Although PLX3397 potently inhibited growth and signaling of Ba/F3 cells expressing the FLT3-ITD/F691L mutant, cells expressing quizartinib-resistant activation loop mutants (at residues D835 and Y842) demonstrated substantial relative resistance to PLX3397 (Supplementary Fig. S7A and S7B; Supplementary Table S4), suggesting that PLX3397 may be particularly vulnerable to activation loop substitutions.

To more broadly assess the vulnerability of PLX3397 to resistance-causing KD mutations within FLT3-ITD and to predict mutations that might confer clinical resistance to PLX3397, we utilized a well-validated in vitro saturation mutagenesis screen (10). Given the significant promise for clinical activity of PLX3397 in patients who develop resistance to quizartinib or sorafenib due to the F691L mutation, we subjected both FLT3-ITD and FLT3-ITD/F691L to mutagenesis. In contrast with the small number of resistance mutations identified in a similar screen for quizartinib resistance (5 substitutions at 3 residues; ref. 10), we identified 18 distinct PLX3397 resistance–conferring substitutions at 10 different amino acid residues after selection in 2 μmol/L PLX3397 (∼20× IC₅₀ for FLT3-ITD). Twelve mutations were identified on the background of FLT3-ITD (in 78 resistant clones analyzed); these substitutions, and an additional six mutations, were also identified on the background of FLT3-ITD/F691L (in 117 resistant clones analyzed; Fig. 3A and B). We recreated all identified mutations (in both the ITD and ITD/F691L backgrounds) and introduced them into Ba/F3 cells to confirm their ability to confer resistance to PLX3397. All mutations conferred relative resistance to PLX3397 in proliferation and biochemical studies when compared with FLT3-ITD alone (Fig. 3C and D; Supplementary Table S4). The majority of identified mutations (15 of 18) occurred in the kinase activation loop (Fig. 3A and B), suggesting that perturbing the conformation of the activation loop is a primary means to disrupt PLX3397 binding (Fig. 4A and B). In addition, some mutations at these sites may favor or stabilize the active state of the kinase that is inaccessible to type II inhibitors. Although the remaining three mutations were found in the N-terminal lobe and hinge, none makes direct contact with the inhibitor, and their effects are mediated by long-range interactions (Fig. 4B, C, and D). In fact, M664I and N676S can both bias the conformation of the activation loop in favor of the active state either directly or indirectly (Fig. 4B and C). Importantly, no resistance-conferring mutation at the F691 gatekeeper position was identified, whereas this residue was the most commonly mutated residue in our previous screen for quizartinib-resistant mutants. All mutations (with the exception of D835N) caused a higher degree of relative resistance in the FLT3-ITD/F691L background compared with the FLT3-ITD alone (at least ∼2–5×), suggesting that in general, the aggregate resistance for any compound mutation is higher than for each component mutation alone.

Based upon its encouraging preclinical activity, a phase I/II clinical trial of PLX3397 in patients with relapsed or chemotherapy-refractory FLT3-ITD⁺ AML (NCT01349049) was initiated, testing doses ranging from 800 mg to 5,000 mg daily. Although the full results of this clinical trial will be reported elsewhere, we undertook translational studies of clinical samples to test our hypothesis that acquired resistance to PLX3397 would be due to mutations at residues other than the F691L gatekeeper position. Using a deep sequencing approach we previously successfully utilized to identify KD mutations occurring on the ITD-containing allele in patient samples (10), we analyzed pretreatment and relapse blood or bone marrow samples from 9 patients who achieved a partial or complete response on PLX3397 at any dose level (with or without recovery of peripheral blood counts) and who subsequently relapsed on the drug (Supplementary Table S5). All assessed responders were treated with PLX3397 ≥1,400 mg daily, the minimally effective dose. In six of nine patient samples, secondary FLT3 KD mutations were acquired at the time of disease relapse (Table 1), thereby demonstrating that PLX3397 is a clinically active FLT3 inhibitor that achieves clinical responses via suppression of FLT3 signaling. The majority of FLT3 KD mutations detected in relapse samples occurred at amino acid residues identified in our in vitro mutagenesis screen. Three patients (1.14, 5.03, and 7.08) also harbored low-frequency mutations of uncertain significance at other residues in addition to those implicated in PLX3397 resistance (Table 1; Supplementary Table S6). However, the F691L substitution was not observed in any of the samples. In addition, a striking degree of polyclonality was present in some samples, including one (5.03) that exhibited seven distinct mutations at three amino acid residues in FLT3-ITD at the time of relapse, illustrating the critical reliance of this leukemia on FLT3-ITD kinase activity, as well as confirming the vulnerability of PLX3397 to a larger spectrum of FLT3 KD mutations compared with quizartinib. One patient (1.14) was noted to have a triple-compound M837G/S838R/D839H mutation acquired at the time of relapse. Recreation of the D839H mutation alone and the compound M837G/S838R/D839H mutations in Ba/F3 cells confirmed ability to cause PLX3397 resistance (Supplementary Table S4). None of the identified mutations in any relapse sample were detectable on ITD-containing alleles in pretreatment samples (Table 1).

In four cases, PLX3397-resistant KD mutations were found in both FLT3-ITD-containing (ITD⁺) and non–FLT3-ITD-containing (ITD⁻) alleles, consistent with prior observations made by our group (unpublished data). In one sample obtained from a patient (8.02) treated with a relatively low dose of PLX3397 (1,400 mg), a D835Y mutation was identified only in ITD⁻ alleles at relapse, which comprised the majority of the relapse sample (6,497 ITD⁻ reads vs. 19 ITD⁺ reads). The pretreatment sample from this patient was also comprised of a majority of ITD⁻ alleles (3,844 ITD⁻ vs. 11 ITD⁺ reads), and no activating mutations in other PLX3397 targets (CSF1R or KIT) were observed (data not shown). These findings suggest that this patient may have responded primarily through inhibition of native FLT3, a phenomenon described with other FLT3 inhibitors (11). Interestingly, patient 5.02, who had a detectable D835Y mutation in ITD⁻ alleles pretreatment, achieved clearance of bone marrow blasts on 2,000 mg of PLX3397, but subsequently relapsed with a highly resistant FLT3-ITD/D835Y clone. These observations are consistent with our finding that D835V/Y mutations cause a lesser degree of resistance to PLX3397 in the absence of an ITD (Supplementary Fig. S7A and S7B) and suggest that these mutations may retain clinical sensitivity to higher doses of PLX3397.

In three of the nine cases we analyzed, no secondary FLT3 KD mutations were identified at the time of PLX3397 resistance, suggesting that in these cases, non–FLT3-dependent resistance mechanisms may drive therapeutic resistance. However, in these patients, examination of the KD of the other PLX3397 targets KIT or CSF1R revealed no mutations (data not shown). Efforts to define off-target, non–FLT3-dependent mechanisms of resistance in these samples are currently ongoing.

The lack of secondary F691L mutations as a mechanism of acquired clinical resistance to PLX3397 in all 9 patients available for analysis strongly suggests that this mutation can be effectively suppressed by PLX3397. In further support of this concept, steady-state plasma from 2 patients treated with PLX3397 at the phase II dose of 3,000 mg contained sufficient drug to completely inhibit FLT3 phosphorylation in MOLM14 F691L cells (Fig. 5).

The development of therapeutic resistance to clinically effective cancer therapies is a pervasive problem that has challenged cancer biologists since effective treatments were first identified. In the era of targeted therapy, mutations that prevent drug binding, particularly at kinase gatekeeper residues, have negatively affected the durability of response achieved with numerous clinically active TKIs. The development of kinase inhibitors that retain activity against resistance-conferring KD mutations therefore represents a critical challenge.

Overriding clinical resistance through rational compound design necessitates a detailed structural understanding of the target kinase and its interaction with inhibitors. From a structural perspective, FLT3 is one of the least understood of all clinically relevant kinase targets, with only one reported crystal structure obtained in the inactive apo, non–ligand-bound form. Here, we report the cocrystal structure of FLT3 complexed with quizartinib, the first highly clinically active FLT3 TKI. This structure reveals features of quizartinib binding to FLT3 that differ from how the prototypic type II inhibitor imatinib interacts with BCR–ABL, including water-mediated interactions with the kinase hinge and remarkably optimized interactions with the ATP-binding pocket and the allosteric pocket. Unexpectedly, the orientation of quizartinib bound to FLT3 is completely opposite to that predicted by previous molecular docking studies. Most surprisingly, the interaction of quizartinib with FLT3 appears to depend upon two edge-to-face interactions with the aromatic side chains of the catalytically crucial F830 residue in the kinase activation loop and the F691 gatekeeper residue. Only strong disruptions of this critical interaction through substitutions at F691 and at the D835 residue, which is key to stabilizing the inactive conformation of the activation loop, are sufficient to induce clinical resistance to quizartinib, whereas substitutions in other less critical regions of interaction appear insufficient to cause clinical resistance. The clinically active FLT3 inhibitor PLX3397, which relies upon a larger number of weak interactions for effective binding, nonetheless tolerates clinically problematic substitutions at the F691 residue. Although this feature enables PLX3397 to retain activity against mutations at F691, its vulnerability to KD mutations at other residues is increased as a consequence.

Ultimately, the prevention of therapeutic resistance due to KD mutations may require combination therapy with clinically active inhibitors that exhibit nonoverlapping resistance profiles. By virtue of its ability to retain activity against gatekeeper F691 substitutions, PLX3397 holds promise as a critical cornerstone therapy, which if used in conjunction with an inhibitor that retains activity against FLT3 activation loop mutants may minimize KD mutation–mediated resistance and thereby improve response durability and therapeutic outcomes. To that end, our studies promise to further facilitate the rational development of the next generation of potent FLT3 inhibitors.

Quizartinib/AC220, 1- (5- (tert-Butyl)isoxazol-3-yl)-3- (4- (7- (2-morpholinoethoxy)benzo[d]imidazo[2,1-b]thiazol-2-yl)phenyl)urea, was purchased from ChemShuttle (Wuxi), and PLX3397, 5-[(5-chloro-1H-pyrrolo[2,3-b]pyridin-3-yl)methyl]-N-[[6- (trifluoromethyl)-3-pyridyl]methyl]pyridin-2-amine, was synthesized following the procedures used previously (W. Tap et al.; submitted for publication).

To enable crystallization, FLT3 KD containing residues H564-V958 with internal deletion H711–H76l, an N-terminal 6xHis-tag (19), and a TEV cleavage site was expressed in baculovirus-infected Sf9 insect cells. FMS proteins were purified through a combination of IMAC and further purified by ion-exchange and size-exclusion chromatography (SEC). The FLT3 SEC pool was autophosphorylated by the addition of 1 mmol/L ATP and 5 mmol/L MgCl2 (1-hour incubation at 4°C). Following kinase activation, the FLT3 protein was incubated with 100 μmol/L quizartinib for 1 hour at room temperature. The protein was then concentrated to a final concentration of 10 mg/mL. Crystals were grown with vapor diffusion sitting drop method and incubated at 293K. Crystallization drops contained 200 nl FLT3-native protein (10 mg/mL) coconcentrated with quizartinib and mixed with 200 nl of reservoir solution containing 100 mmol/L Na-Citrate, pH 5.5, and 20% PEG3000. Needles (0.10–0.15 mm long) were observed after 12 weeks. Crystals were cryoprotected with 25% glycerol before freezing. Diffraction data were collected at the Advanced Light Source beamline 8.3.1. FLT3 quizartinib costructure was solved by molecular replacement with the program MOLREP and using the structure of apo-FLT3 (Protein Data Bank ID code 1RJB) as a starting model. The structure was refined with the program PHENIX. Cycles of manual rebuilding with COOT and structure refinement were continued until there was no further improvement, indicated by the convergence of the Rfree factor. The crystallographic data and refinement statistics are provided in Supplementary Table S1.

All three compounds have been previously cocrystallized with other class III receptor tyrosine kinases (PLX3397 in CSF1R, sorfenib in KDR, and ponatinib in KIT). Given the high degree of homology between these kinases and FLT3, we expected these inhibitors would exhibit the same overall binding modes. For each inhibitor, we performed structural alignment between the known costructure containing the inhibitor and the costructure of FLT3 and quizartinib with an emphasis on optimal registration of ATP-binding apparatus. The pose of the inhibitor was then “copied” from its costructure with another kinase and “pasted” into FLT3 to replace quizartinib. One round of energy minimization using the default parameter of MOE (Chemical Computing Group) was used to refine the structural model.

Random mutagenesis was performed as previously described (10). Cells were selected in 2 μmol/L PLX3397 in soft agar. After 10 to 21 days, visible colonies were plucked and expanded in 2 μmol/L PLX3397.

Sequencing was performed from amplified genomic DNA from colonies expanded from soft agar as previously described (10).

Mutations isolated in the screen were engineered into pMSCVpuroFLT3-ITD by QuikChange mutagenesis (Agilent Technologies) according to the manufacturer's recommendations.

Stable Ba/F3 lines were generated by retroviral spinfection with the appropriate mutated plasmid as previously described (10). Ba/F3 cell lines were originally obtained from the laboratory of Charles Sawyers in 2006 and have not been authenticated. MV4;11 and MOLM14 cells for in vitro proliferation and Western Blot assays were obtained from the laboratory of Scott Kogan in 2008 and authenticated by Promega STR analysis in May 2014. All cell lines were Mycoplasma-free.

MV4;11 cells were purchased from the ATCC in April 2008 and were not authenticated. Cells were expanded to a density of 0.5 × 10⁶/mL. Cells were washed three times with PBS, after which the cells were resuspended in to the final density of 50 × 10⁶ cells/mL in PBS before inoculation. There were 8 nu/nu outbred, 6-week-old male mice per group (24 mice total for vehicle and two dose groups). A sample size of 8 mice per group provided at least 80% power to detect a 2 standard deviation difference of mean tumor volume between two groups with a t-sided type I error of 1%. Mice were fed Laboratory Rodent Diet. Municipal tap water was used ad libitum. Mice were housed at 18°C to 26°C and 50% ± 20% relative humidity in rooms with at least ten room air changes per hour. Photoperiod was diurnal: 12 hours light and 12 hours dark. Mice were injected with a MV4;11 cell suspension in 100 μL PBS + 100 μL Matrigel (5 × 10⁶ cells per mouse) in the lower left abdominal flank. Treatments (vehicle and PLX3397 at two dose levels, all given daily via oral gavage) started when the average tumor size reached 125 mm³. Animals in the 3 groups of 8 were randomly assigned to match mean tumor volume measurements. Dosing duration was 21 days. Tumor measurements were taken with electronic microcaliper 3 times weekly (Monday, Wednesday, and Friday). Investigators were not blinded to animal group assignment. All animal studies were conducted in accordance with the Institute for Laboratory Animal Research Guide for the Care and Use of Laboratory Animals and the U.S. Department of Agriculture Animal Welfare Act and local animal ethics committees.

Exponentially growing cells (5 × 10³ cells/well for Ba/F3, 1 × 10⁴ cells/well for others) were plated in each well of a 96-well plate with 0.1 mL of RPMI 1640 + 10% FCS containing the appropriate concentration of drug in triplicate. Cells were allowed to expand for 2 or 3 days (as indicated), and cellular proliferation was assessed using ATP content as a measure of cell viability using either CellTiter-Glo reagent (Promega) or ATPlite 1step Luminescence Assay reagent (Perkin-Elmer) according to the manufacturer's recommendation on a SpectraMax M3 microplate reader using SpectraMax Pro Software (Molecular Devices). The value at varying concentrations of drug was normalized to the median value in the no-drug sample for each mutant, and a mean value was calculated. Numerical IC₅₀ values were generated using nonlinear best-fit regression analysis using Prism 5 software (GraphPad). IC₅₀ values reported are the calculated mean result of 3 or more independent experiments. All cell viability data shown are representative of experiments replicated 3 or more times.

Exponentially growing Ba/F3 cells stably expressing FLT3-mutant isoforms were plated in RPMI medium 1640 + 10% FCS supplemented with kinase inhibitor at the indicated concentration. After a 90-minute incubation, the cells were washed in PBS and lysed in buffer (50 mmol/L HEPES, pH 7.4, 10% glycerol, 150 mmol/L NaCl, 1% Triton X-100, 1 mmol/L EDTA, 1 mmol/L EGTA, 1.5 mmol/L MgCl₂) supplemented with protease and phosphatase inhibitors. The lysate was clarified by centrifugation and quantitated by bicinchoninic acid assay (Thermo Scientific). Protein was subjected to sodium dodecyl sulfate polyacrylamide electrophoresis and transferred to nitrocellulose membranes. Immunoblotting was performed using anti–phospho-FLT3 (3464; 1:1,000), anti–phospho-STAT5 (9351; 1:1,000), anti-STAT5 (9363; 1:1,000), anti–phospho-S6 (2211; 1:1,000), anti-S6 (2317; 1:1,000), anti–phospho-ERK (9101; 1:1,000), anti-ERK (9102; 1:1,000; Cell Signaling), and anti-FLT3 S18 antibody (1:1,000; Santa Cruz Biotechnology). Data shown in figures are representative of 3 technical replicates.

Plasma inhibitory assay was performed as previously described (25) with some modifications. Briefly, 7 × 10⁶ parental MOLM14 cells and MOLM14 cells harboring the indicated mutations were resuspended in 1 mL of plasma from either a healthy control containing the indicated concentration of the drug or a patient treated with the indicated dose of the inhibitor at steady state. All samples were collected under the University of California, San Francisco, Institutional Review Board–approved cell banking protocol (CC#112514). Informed consent was obtained in accordance with the Declaration of Helsinki. Cells were incubated at 37°C for 2 hours. FLT3 was immunoprecipitated from 400 μg of total protein using anti-FLT3 S18 antibody (1:50) and used for Western immunoblot with anti-FLT3 (1:1,000) and anti–phospho-tyrosine antibody 4G10 (1:1,000; EMD Millipore). The remaining lysate was normalized and used for Western immunoblot as above.

Nine cases of acquired resistance to PLX3397 were analyzed, including all known patients with bone marrow responses who subsequently relapsed while still on drug treatment. Patients were enrolled on the multicenter phase I/II clinical trial of PLX3397 in relapsed or refractory AML (NCT01349049). The clinical trial protocol, including analysis of resistance samples, was approved by the Institutional Review Board of all participating institutions. Details of the clinical trials and results will be reported elsewhere. All patients were positive for FLT3-ITD mutation and negative for D835 mutation by standard clinical testing at the time of enrollment. Eligibility for the study required that patients be unsuitable for, relapsed on, or refractory to standard induction chemotherapy. Samples were collected before treatment and at the time of disease progression. Only patients who had achieved bone marrow partial remission (PR) or composite complete remission (CRc = CR + Cri + CRp) using modified International Working Group Criteria were analyzed. CR was defined as patients who had bone marrow regenerating normal hematopoietic cells and a morphologic leukemia-free state with an absolute neutrophil count (ANC) >1 × 10⁹/L, platelet count ≥100 × 10⁹/L, and normal marrow differential with < 5% blasts, and who were red blood cell (RBC) and platelet transfusion independent (defined as 4 weeks without RBC transfusion and 1 week without platelet transfusion) with no evidence of extramedullary leukemia. CRp was defined as CR except for incomplete platelet recovery (<100 × 10⁹/L). CRi was defined as CR except for incomplete hematologic recovery with residual neutropenia (ANC ≤1 × 10⁹/L) with or without residual thrombocytopenia (platelet count <100 × 10⁹/L). PR was defined as bone marrow regenerating normal hematopoietic cells with evidence of peripheral recovery with no (or only a few regenerating) circulating blasts and with a decrease of at least 50% in the percentage of blasts in the bone marrow aspirate, with total marrow blasts between 5% and 25%. For CRp, CRi, and PR, patients did not need to be RBC or platelet transfusion independent. All patients gave informed consent according to the Declaration of Helsinki both to participate in the clinical trials and for collection of samples.

For sequencing, frozen Ficoll-purified mononuclear cells obtained from blood or bone marrow were lysed in TRIzol (Invitrogen), and RNA was isolated according to the manufacturer's protocol. cDNA was synthesized using Superscript II (Invitrogen) as per the manufacturer's protocol. The FLT3 KD and adjacent juxtamembrane domain were PCR amplified from cDNA as previously described (10). The CSF1R and KIT KDs were amplified from cDNA and sequenced by direct sequencing. Primer sequences are available upon request.

PCR product containing the FLT3 KD was generated from patient cDNA as previously described above, using high-fidelity DNA polymerase (10). We prepared PCR products for Pacific Biosciences single-molecule real-time sequencing (26) using standard commercial kits and reagents and sequenced on the Pacific Biosciences RSII instrument as previously described, but using P4-C3 enzyme and chemistry alongside 3-hour collection times for circular consensus sequencing (CCS; 10).

Data from single molecule, real-time (SMRT) sequencing were analyzed using an extension to the methods previously described in Smith and colleagues (10). CCS reads were separated by sample using barcoded primer sequences, and all reads with full-length alignment to the targeted FLT3 interval were oriented and passed for downstream ITD analysis. We employed a variant of “dot plotting” (27) which is commonly used to visually compare DNA sequences; a match (dot) between two sequences was defined as a shared substring of size 10, but extensions to exact matches were permitted for 10 mers with up to 2 mismatches. This enabled us to detect perfectly tandem duplications of a single codon and nontandem duplications of 10 bp or more. In short, for a target sequence (t) and a query sequence (q), t was dot plotted against itself and q. The two resulting matrices were then summed column-wise, to generate a pair of distinct vectors, T and Q, of length /t/. T and Q represented the relative abundance of particular substrings in t and q, respectively, projected onto the target interval. Next, we normalized Q by element-wise division of T, yielding Q_Norm. A two-state hidden Markov model was then applied on Q_Norm in order to identify putative intervals of duplication. Once all reads had been partitioned by sample into ITD⁺ and ITD⁻ sets, variant analysis was performed using Quiver (28) on each read subset to identify candidate mutations.

Pacific Biosciences sequencing data were processed, mapped, and variant called using SMRT Analysis version 2.0.1. Custom scripts for detecting ITDs were written using Python version 2.7 and Cython version 0.21.1 and are available upon request.

C.C Smith reports receiving clinical trial research support from Plexxikon. A.E. Perl is a consultant/advisory board member for Actinium Pharmaceuticals, Ambit Biosciences, Arog Pharmaceuticals, Asana Biosciences, and Astellas Pharmaceuticals. N.P. Shah reports receiving commercial research grants from Ambit, Ambit Biosciences, ARIAD, and Plexxikon. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C.C. Smith, C. Zhang, A. Kasarskis, E.A. Burton, J. Zhang, M.H. Le, B.L. West, N.P. Shah

Development of methodology: C.C. Smith, A. Bashir, R. Sebra, R. Shellooe, B. Powell, E.A. Burton, J. Zhang, G. Habets

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.C. Smith, K.C. Lin, E.A. Lasater, E. Massi, L.E. Damon, R. Sebra, A. Perl, R. Shellooe, H. Carias, B. Matusow, W. Spevak, H.H. Hsu, G. Habets, B.L. West, N.P. Shah

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.C. Smith, C. Zhang, K.C. Lin, E.A. Lasater, E. Massi, L.E. Damon, M. Pendleton, A. Bashir, A. Perl, H. Carias, B. Matusow, J. Zhang, G. Habets, G. Bollag, N.P. Shah

Writing, review, and/or revision of the manuscript: C.C. Smith, C. Zhang, M. Pendleton, A. Bashir, R. Sebra, A. Perl, A. Kasarskis, G. Tsang, H. Carias, E.A. Burton, W. Spevak, P.N. Ibrahim, H.H. Hsu, G. Habets, B.L. West, G. Bollag, N.P. Shah

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.C. Smith, K.C. Lin, E. Massi, M. Pendleton, P.N. Ibrahim, M.H. Le, H.H. Hsu, G. Bollag

Study supervision: C. Zhang, B. Powell, M.H. Le, N.P. Shah

Other (crystal structure determination and analysis): Y. Zhang

Other (designed and synthesized PLX3397): J. Zhang

N.P. Shah acknowledges the generous support of Arthur Kern, Mark Maymar, and the Edward S. Ageno family. This work is dedicated to Ilana Massi.

This work was supported by grants from the NCI (1R01 CA166616-01, to N.P. Shah), the NIH T-32 Molecular Mechanisms of Cancer (to E.A. Lasater), and the Leukemia and Lymphoma Society (to C.C. Smith and N.P. Shah). C.C. Smith is an American Society of Hematology Faculty Scholar and recipient of a Hellman Family Foundation Early Career Faculty Award.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.