ATP-Competitive Inhibitors Midostaurin and Avapritinib Have Distinct Resistance Profiles in Exon 17–Mutant KIT

KIT is a type-3 receptor tyrosine kinase (RTK); other type-3 RTKs are FLT3, PDGFR, and CSF1R. Physiologically, KIT is activated by stem cell factor and has multiple downstream effectors, including PI3K, RAS/MAPK, and JAK/STAT (1).

KIT is pathologically activated in a variety of cancers. The majority of oncogenic KIT mutations are in exon 11, which encodes the regulatory juxtamembrane (JM) domain, or exon 17, which encodes the activation loop of the kinase domain (2–5). Exon 11 mutations activate KIT by relieving the autoinhibition of the JM domain, while exon 17 mutations shift the conformational equilibrium of the kinase to the active state (6–8). For unclear reasons, exon 11 mutations predominate in gastrointestinal stromal tumor (GIST) and melanoma, whereas exon 17 mutations, exemplified by KIT D816V, predominate in systemic mastocytosis (SM), acute myeloid leukemia (AML), and germinomas (2–5).

Historically, exon 17–mutant KIT has been a challenging drug target, while exon 11–mutant KIT has been targetable with clinically available TKIs (1–5, 9–11). The first-generation of KIT inhibitors (imatinib, sunitinib, and regorafenib) transformed GIST driven by exon 11–mutant KIT from a lethal disease to a chronic condition (12). Nonetheless, over 50% of patients with GIST relapse with secondary resistance mutations in exon 13 or 14, which encode the drug/ATP-binding pocket, or exon 17, which encodes the activation loop (13). In addition, cancers with primary de novo exon 17 mutations, such as SM and AML, are insensitive to first-generation KIT TKIs because exon 17–mutant KIT is constitutively active and these drugs exclusively bind the inactive conformation (9–11, 14, 15).

The concept of conformational states affecting TKI binding led to classification of ATP-competitive TKIs as “type 1” or “type 2” (14, 16, 17). Type 1 TKIs bind the active kinase conformation, whereas type 2 TKIs, which include imatinib, sunitinib, and regorafenib, bind the inactive kinase conformation (6, 14, 15). Inactive conformations are referred to as “DFG-out” conformations because the Mg-binding Asp-Phe-Gly (“DFG”) motif, which is conserved at the N terminus of kinase activation loops and commonly makes conformation-specific molecular interactions with TKIs, is oriented out of the active site (6, 15–18).

Midostaurin (PKC412) and avapritinib (BLU-285) are the first type 1 TKIs to demonstrate clinical activity in malignancies harboring KIT exon 17 mutations. In April 2017, the FDA approved midostaurin for advanced systemic mastocytosis (ASM) based on a single-arm, open-label phase II trial of midostaurin in heavily pretreated patients with ASM, which showed a 60% overall response rate on the basis of modified Valent and Cheson criteria (19). Early phase I results of avapritinib in ASM are also encouraging, with a 72% overall response rate in heavily pretreated patients on the basis of modified IWG-MRT-ECNM response criteria (20). Although these trials are based on different response criteria, both strongly support the use of KIT-directed therapy in ASM.

Secondary kinase domain mutations are the best-characterized mechanism of acquired resistance to TKIs. These substitutions typically mediate resistance through three mechanisms: (i) directly interfering with TKI binding through steric hindrance or loss of molecular interactions (6, 14, 18, 21), (ii) increasing ATP affinity (22), and/or (iii) destabilizing the kinase conformation required for TKI binding (8, 23). One particularly problematic amino acid in kinases, termed the gatekeeper residue, resides in the back of the drug/ATP-binding site and controls access to a deep hydrophobic pocket accessed by many TKIs (14, 15). Gatekeeper mutations commonly cause TKI resistance and can act through all mechanisms described above (21–27).

Secondary kinase domain mutations capable of conferring resistance to type 1 KIT TKIs have not been described previously (26, 28, 29). We sought to identify secondary point mutations in KIT D816V that confer resistance to midostaurin and avapritinib with the hope that this knowledge will inform the next iteration of drug development efforts targeting KIT. We assessed candidate mutations for their ability to confer resistance to midostaurin and avapritinib, and determined these drugs have nonoverlapping resistance profiles: while T670I, a gatekeeper mutation, confers a high degree of resistance to avapritinib, it retains sensitivity to midostaurin. Computational studies, supported by experimental evidence, unexpectedly predict the KIT T670I gatekeeper mutation can induce distant conformational changes in the phosphate-binding loop (P-loop) that impair TKI binding, and support the development of next-generation KIT TKIs that minimally interact with the region surrounding the P-loop.

KIT was amplified from M230 melanoma cells and cloned into Gateway pENTR1A vector. The D816V mutation was generated by QuikChange (Agilent). MSCVpuro KIT D816V was generated via the LR clonase reaction (30) between pENTR1A- c-KIT D816V and MSCVpuroRFA. Secondary mutations were generated by QuikChange (Agilent), or by digestion and then ligation of purchased gene blocks (Integrated DNA Technologies) containing the desired secondary mutations. All plasmids were verified by diagnostic restriction digest and Sanger sequencing. See Supplemental Materials and Methods for details.

Parental Ba/F3 cells were purchased from DSMZ. Stable Ba/F3 lines were generated by retroviral spinfection with mutated plasmid as described previously (31). gDNA was extracted from each cell line, KIT was amplified by PCR, and sequenced to confirm incorporation of the correct KIT mutant.

PKC412/Midostaurin (Selleckchem), avapritinib/BLU-285 (ChemGood), and sunitinib (Sigma) were purchased. Stock solutions were prepared in DMSO and stored at −80°C (avapritinib and sunitinib) or −20°C (midostaurin).

Cells expressing KIT D816V primary mutations were plated at 2,000 cells per well in 96-well white opaque tissue culture plates (Corning) and treated with inhibitor or DMSO. Cells expressing primary V560D mutations were plated at 20,000 cells per well in 25 ng/mL of stem cell factor in 96-well plates and treated with inhibitor or DMSO. After 48 hours, cell proliferation was assessed with the CellTiter-GLO Luminescent Cell Viability Assay (Promega). IC₅₀s were calculated with GraphPad Prism 6 software.

Cells were starved for 2 hours, treated with inhibitor or DMSO for 2 hours, and then lysed. Lysates were resolved by SDS-PAGE, transferred to nitrocellulose, and blotted. See Supplementary Materials and Methods for more details.

An active-like conformation of KIT was built based on the ATP-bound structure (PDB ID: 1PKG). Missing domains were added using the SwissModel Server with PDB ID: 3G0E as a reference (8, 32, 33). Mutations were introduced using the rotamer search implemented in Chimera (34). To generate drug-bound models, ligand was docked into the D816V active site using Gold with midostaurin–DYRK1A complex as a reference (PDB ID: 4NCT; ref. 35). The apo-models were subjected to short-time MD simulations (∼11.5 ns) using the AMBER14 suite (36) and the equilibrated structure was used as a reference to maintain an “active-like” form. Models of the apo-double mutants were compared with the models of apo-D816V and drug-bound D816V. ChimeraX (37), a virtual reality tool, was used for three-dimensional investigation of the structures; PyMOL (Schrödinger) was used to generate figures. See Supplementary Materials and Methods for more details.

We previously adapted an XL1-Red E. coli saturation mutagenesis assay to identify problematic mutations for TKIs targeting BCR-ABL1 and FLT3, many of which were validated in clinical isolates (31, 38–40). However, KIT containing plasmids are highly unstable in this bacterial strain, rendering this technique unsuitable. Therefore, we generated a targeted panel of KIT alleles containing a primary activating exon 17 mutation (D816V) and candidate secondary resistance mutations (Fig. 1A). Two complementary methods were used to select candidate resistance mutations. First, we did a literature search for clinically observed KIT mutations associated with KIT TKI resistance. We identified two categories of mutations (Supplementary Table S1): activation loop substitutions that favor an active conformation, and alterations in the ATP/drug-binding pocket that sterically and/or chemically interfere with drug binding. Given that midostaurin and avapritinib were developed for activation loop-mutant KIT, and that the D816V activation loop mutation is the primary mutation for our studies, we reasoned that secondary activation loop mutations were unlikely to confer resistance to these TKIs. In contrast, we hypothesized resistance caused by steric and/or chemical changes within the active site were as likely to impact type 1 TKIs as type 2 TKIs because both are ATP competitive and access the active site. We identified two such secondary mutations discovered in patients with GIST who lost response to type 2 TKIs: V654A and T670I (Fig. 1A). V654A confers resistance to imatinib and other TKIs by eliminating van der Waals interactions important for drug binding (21). T670I, a gatekeeper mutation (14), confers imatinib resistance by abolishing a hydrogen bond between imatinib and KIT, and through steric hindrance conferred by the extra methyl of the Ile compared with Thr (18, 21).

The second method for identifying resistance mutations involved extrapolating from previous work describing resistance mutations in the type 3 RTK FLT3 (40, 41). The kinase domains of FLT3 and KIT share 64% sequence identity, and the root-mean-square deviation of the two kinase domains is 0.68 Å, indicating high structural homology (Fig. 1A and B). We hypothesized KIT mutations that confer resistance to type 1 KIT TKIs would be analogous to FLT3 mutations that confer resistance to type 1 FLT3 TKIs. We identified three FLT3 mutations that confer resistance to type 1 TKIs: N676K, Y693C, and D698N (Fig. 1A). N676K was discovered in a patient with FLT3 internal tandem duplication (ITD)-positive AML who relapsed on midostaurin (41). When transduced into 32D cells, FLT3 ITD/N676K confers resistance to midostaurin (41). The analogous mutation in KIT, N655K, has been described in patients with GIST, and confers resistance to the type 2 TKI nilotinib (42, 43). Two additional resistance mutations, Y693C and D698N, were identified in an in vitro saturation mutagenesis screen of FLT3 ITD, and shown to confer resistance to crenolanib and midostaurin, two type 1 FLT3 TKIs (40). The analogous mutations in KIT, Y672C and D677N, have not been reported. Notably, each mutation results from a single nucleotide change, which facilitates the genesis of most clinically identified resistance mutations. KIT D816V readily transforms Ba/F3 cells to IL3 independence (Supplementary Fig. S1), and KIT D816V harboring each of these secondary mutations retained transformation potential.

We first determined the sensitivity of our allelic series to midostaurin. Cell proliferation assays confirmed growth of Ba/F3 KIT D816V cells is inhibited by midostaurin in a dose-dependent manner, with an IC₅₀ of 36 nmol/L (Fig. 2A; Supplementary Fig. S2A and S2B). Addition of V654A, T670I, N655K, Y672C, or D677N secondary mutations conferred varying degrees of midostaurin resistance relative to D816V alone, with the greatest relative resistance associated with V654A, N655K, and D677N (Fig. 2A; Supplementary Fig. S2A and S2B). Consistent with a previous report that demonstrated midostaurin is effective against KIT with a JM domain primary mutation and a T670I secondary mutation (26), KIT D816V/T670I conferred only a 5-fold increase in IC₅₀ compared with D816V alone, similar to the degree of relative resistance conferred by Y672C, but less than any of the other mutants tested. Western blot analysis of phospho-KIT and global phosphotyrosine confirmed KIT D816V/Y672C and D816V/T670I retain biochemical sensitivity to midostaurin, but KIT D816V/V654A, D816V/N655K, and D816V/D677N are resistant (Fig. 2C–E; Supplementary Fig. S3A–S3C).

Avapritinib has a strikingly different chemotype than midostaurin (Supplementary Fig. S4). The chemical differences suggest the inhibitors might interact with distinct residues, and display disparate activity against secondary kinase domain mutants. To test this hypothesis, we treated our Ba/F3 KIT mutants with avapritinib. We found the IC₅₀ of avapritinib was 3.8 nmol/L in Ba/F3 KIT D816V cells, which is 10-fold lower than the IC₅₀ of midostaurin (Fig. 2B; Supplementary Fig. S2B). Although addition of secondary V654A, N655K, Y672C, or D677N mutations resulted in a small increase in IC₅₀ relative to KIT D816V alone, avapritinib generally retained potency against these secondary mutations (IC₅₀s ranging from 1.1 to 16 nmol/L; Fig. 2B; Supplementary Fig. S2B). In contrast, the IC₅₀ of avapritinib toward Ba/F3 KIT D816V cells expressing the secondary gatekeeper mutant, T670I, was approximately 70-fold higher (270 nmol/L) than KIT D816V alone (Fig. 2B). Western blots examining phosphorylated KIT and global phospho-tyrosine confirmed biochemical resistance (Fig. 2E).

Early clinical trial experience with avapritinib in heavily pretreated GIST, which commonly harbors primary KIT JM domain mutations (2), shows V654A and T670I mutations are associated with clinical resistance to avapritinib (44). The overall response rate (ORR) in patients with GIST with V654A or T670I mutations was 0% (n = 25), compared with a 26% ORR (n = 84) in patients who lacked these mutations; the stable disease rate was also lower in patients with these mutations (28% vs. 51%). Furthermore, the rate of progressive disease was substantially higher in patients with preexisting V654A or T670I mutations compared with patients who lacked these mutations (72% vs. 23%; ref. 44). However, as stated above, our experiments showed avapritinib potently inhibited proliferation of Ba/F3 D816V/V654A cells (IC₅₀ 16 nmol/L; Figs. 2E; Supplementary Fig. S2B). We therefore assessed whether the nature of the primary activating mutation in KIT (a JM vs. a D816V mutation) influences the potency of avapritinib against KIT mutants with secondary V654A or T670I mutations. We generated KIT alleles with an activating primary V560D mutation, which is a common JM domain substitution in GIST (8), and a secondary V654A or T670I mutation. KIT V560D and all KIT JM mutants we have tested are insufficient to transform Ba/F3 cells to IL3 independence (Supplementary Fig. S5). Therefore, experiments were performed in the presence of KIT ligand (stem cell factor; SCF). The IC₅₀ of avapritinib in Ba/F3 KIT V560D cells was over 10-fold greater than the IC₅₀ of avapritinib in Ba/F3 V560D/D816V cells (Fig. 3). Notably, the IC₅₀ of avapritinib in Ba/F3 KIT V560D/D816V cells supplemented with SCF was nearly identical to that for Ba/F3 KIT D816V cells without SCF, indicating that the D816V mutation renders KIT highly sensitivity to avapritinib regardless of the presence of SCF (Figs. 2E and 3). Ba/F3 KIT V560D cells harboring secondary V654A or T670I substitutions were considerably less sensitive to avapritinib than their counterparts with primary D816V mutations (⁠|${\rm{I}}{{\rm{C}}_{{\rm{5}}0}}_{_{{\rm{V56}}0{\rm{D}}/{\rm{V654A}}}}$| 245 nmol/L vs. |${\rm{I}}{{\rm{C}}_{{\rm{5}}0}}_{_{{\rm{D816V}}/{\rm{V654A}}}}$| 16 nmol/L; |${\rm{I}}{{\rm{C}}_{{\rm{5}}0}}_{_{{\rm{V56}}0{\rm{D}}/{\rm{T67}}0{\rm{I}}}}$| 610 nmol/L vs. |${\rm{I}}{{\rm{C}}_{{\rm{5}}0}}_{_{{\rm{D816V}}/{\rm{T67}}0{\rm{I}}}}$| 270 nmol/L; Figs. 2E and 3).

The nonoverlapping resistance profiles of midostaurin and avapritinib support the hypothesis that these compounds interact with different residues within the KIT D816V active site, and suggest that avapritinib may interact directly with T670. To investigate potential drug–protein binding interactions, we performed molecular docking studies. We developed a model of apo-KIT D816V using a crystal structure of KIT in the active conformation (PDB: 1PKG; ref. 6), then separately docked midostaurin and avapritinib into the active site (Fig. 4A). Both midostaurin and avapritinib are predicted to bind KIT in the interdomain cleft between the N- and C-terminal lobes, consistent with their known ATP-competitive activity. However, because the compounds have distinct shapes, the pockets they are predicted to occupy differ (Fig. 4B). Midostaurin, a large bulky compound, projects its 3-pyrroline-2-one head group toward the hinge region, where it is able to make two hydrogen bonding interactions with the backbone amides of E671 and C673 (Fig. 4C; Supplementary Fig. S6A). The phenyl ring of midostaurin's tail extends down from the adenosine-binding pocket and accesses a pocket close to D677. In addition, residues T670 and V654 are positioned to make hydrophobic interactions with midostaurin (Fig. 4C; Supplementary Fig. S6A). In contrast, avapritinib occupies a longer, thinner region within the active site (Fig. 4B and D), and is predicted to make only one hydrogen bond with the backbone of the hinge region, at residue C673 (Fig. 4D; Supplementary Fig. S6B). The docking studies predict an additional hydrogen bond between the primary amine of avapritinib and the side chain carboxylic acid of D810, which is part of the DFG motif (Fig. 4D, inset). The fluorophenyl group of avapritinib is adjacent to the P-loop, a flexible loop in the active site that helps coordinate the phosphates of ATP during phosphoryl transfer (Fig. 4D; ref. 45). The predicted interactions that the DFG and P-loop make with avapritinib are unique compared with midostaurin, which binds far from both these motifs (Fig. 4C and D). Despite the observation that the T670I mutation is highly resistant to avapritinib, this TKI is not predicted to bind close to the T670 gatekeeper (Fig. 4D).

Molecular docking studies suggest clear differences in how midostaurin and avapritinib bind KIT D816V, providing a rationale for their distinct resistance profiles. However, analysis of residues within 5 Å of the docked TKIs, a range that encompasses hydrogen bonding and hydrophobic interactions, fails to explain how N655K and D677N confer resistance to midostaurin, or how T670I confers resistance to avapritinib (Fig. 4C and D; Supplementary Fig. S6).

To elucidate possible structural mechanisms by which these mutations confer resistance, we performed molecular dynamics (MD) simulations. We built models of apo forms of KIT single (D816V) and double (D816V/V654A, D816V/T670I, D816V/N655K, and D816V/D677N) mutants, and compared these with the docked poses of midostaurin and avapritinib in D816V. Consistent with previous modeling (21), the V654A mutation is predicted to reduce hydrophobic interactions between residue 654 and the 3-pyrroline-2-one head group of midostaurin (Supplementary Fig. S7A and S7B). This effect is not observed for avapritinib, which binds more than 5 Å from V654 (Supplementary Figs. S6B and S7). The predicted resistance mechanism conferred by mutation of neighboring N655 to lysine appears similar, also reducing hydrophobic interactions between residue 654 (in this case, the native valine) and midostaurin. In the MD simulation of apo D816V/N655K, the K655 side chain can hydrogen bond to the side chain of neighboring N649, pulling the loop that holds V654 away from the midostaurin-binding pocket, thus reducing the ability of V654 to form hydrophobic interactions with midostaurin and mimicking the effect of the V654A mutation (Supplementary Fig. S7C).

Because gatekeeper residues often interact directly with drugs (e.g., KIT T670 forms a hydrogen bond with imatinib; ref. 6), we initially hypothesized that avapritinib directly interacts with T670, but midostaurin does not. However, our docking studies strongly argue against this hypothesis. In fact, in our model, T670 is more than 5 Å from avapritinib, while the side chain methyl of T670 has favorable close hydrophobic contacts to midostaurin (Fig. 4C and D). These hydrophobic interactions between T670 and midostaurin are likely retained upon mutation to the hydrophobic amino acid Ile, and the increased steric bulk of Ile compared with Thr likely explains the small increase in IC₅₀ of midostaurin against the D816V/T670I mutant compared with D816V alone. Because T670I confers a high degree of relative resistance to avapritinib despite being far from avapritinib in our model, and because it appears unlikely that this substitution increases ATP affinity based upon its retention of sensitivity to midostaurin compared with avapritinib, we hypothesized that remote structural changes induced by the gatekeeper mutation might impair avapritinib binding. Structural changes have been ascribed to gatekeeper mutations in kinases, such as ABL and SRC, where gatekeeper mutations push the kinase toward an active conformation through stabilization of a hydrophobic spine (23). By forcing the conformational equilibrium toward the active state, these structural changes contribute to type 2 TKI resistance (23). In KIT, our MD simulations support the hypothesis that T670I alters the conformation of the active site. However, our studies suggest a novel change that involves rigidification of the N-terminal lobe rather than stabilization of a hydrophobic spine. Comparison of the last MD frames of the apo-models of D816V/T670I and D816V predicts the larger aliphatic side chain of I670, compared with native T670, results in increased hydrophobic contacts between residue 670 and residues in proximity of the P-loop (Fig. 5A and B). The additional interactions provided by the I670 side chain are predicted to cause a global rigidification of the N-terminal lobe of D816V/T670I compared with D816V, as demonstrated by lower b-factor values for D816V/T670I in the last nanosecond of the simulation compared with D816V alone (Supplementary Fig. S8A and S8B). The rigidification is predicted to have a profound effect on the conformation of the P-loop, positioning the P-loop closer to the binding pocket of the fluorophenyl moiety of avapritinib (Fig. 5C). Changing this pocket should impair avapritinib binding, but not midostaurin binding, because midostaurin does not extend into this pocket, providing a mechanistic hypothesis for why T670I selectively confers resistance to avapritinib.

To test this hypothesis, we generated gatekeeper mutants with varying hydrophobicity. We expected less hydrophobic gatekeeper mutants, such as D816V/T670A, would retain sensitivity to avapritinib, while more hydrophobic gatekeeper mutants, such as D816V/T670V, which has a similar degree of hydrophobicity to Ile, would be resistant. Consistent with these predictions, avapritinib potently inhibited D816V/T670A with an IC₅₀ of 20 nmol/L, but was relatively resistant to the Val gatekeeper substitution (IC₅₀ 360 nmol/L; Fig. 6A). The T670V mutation requires a double amino acid change, which makes it less likely that this mutation will arise clinically. Also consistent with our predictions, D816V/T670A and D816V/T670V both retain sensitivity to midostaurin, demonstrating increased hydrophobicity of the gatekeeper has no effect on midostaurin as long as the side chain is small enough to avoid steric clash (Fig. 6B). Overall, these data support our model that the increased hydrophobicity of Ile causes increased hydrophobic packing that alters the position of the P-loop to selectively impair avapritinib binding.

Midostaurin and avapritinib are the first clinically active type 1 KIT TKIs developed to target exon 17–mutant KIT (19, 46). The early clinical efficacy of midostaurin and avapritinib in SM suggests exon 17–mutant KIT represents a driver mutation in this disease. However, identification of secondary resistance mutations to midostaurin and avapritinib in patient isolates is essential to provide definitive proof that exon 17–mutant KIT is a valid drug target, and to guide development of future KIT TKIs. Therefore, we sought to prospectively identify point mutations within the KIT kinase domain that confer resistance to midostaurin and/or avapritinib. We show that in the setting of a primary KIT D816V mutation, midostaurin and avapritinib are susceptible to secondary resistance mutations in vitro, but their resistance profiles are distinct. Secondary V654A, N655K, and D677N mutations confer resistance to midostaurin, whereas T670I confers selective resistance to avapritinib. Mechanistically, our docking and MD studies predict the V654A mutation acts as described previously (21), conferring resistance by decreasing hydrophobic interactions between V654 and midostaurin. Interestingly, the N655K mutation is also predicted to decrease hydrophobic interactions between residue 654 and midostaurin, via a mechanism that includes formation of a novel hydrogen bond and concerted conformational changes. The mechanism(s) underlying the selective resistance of D677N to midostaurin is unclear and undergoing further study. Of the mutations assessed, only the T670I gatekeeper mutation confers significant resistance to avapritinib in the setting of a primary KIT D816V mutation. MD simulations suggest a unique resistance mechanism in which the increased hydrophobicity of the Ile compared with Thr leads to rigidification of the N-terminal lobe and movement of the P-loop into the avapritinib binding pocket. In support of this hypothesis, we found substituting T670 for a less hydrophobic amino acid (alanine) results in retention of sensitivity to avapritinib.

Currently, there is limited data on clinical isolates from midostaurin-treated relapsed or refractory patients, and no studies of samples from avapritinib-treated relapsed or refractory patients. A recent clinical study evaluating genetic predictors of midostaurin response demonstrated that midostaurin reduces the KIT D816V allele burden in SM, which represents the strongest on-treatment predictor for improved survival (29). However, mutations in SRSF2, ASXL1, and RUNX1 also have a major impact on midostaurin response, and lack of response was not associated with on-target resistance mutations in KIT (29). This suggests the mutational landscape of SM is complex, an assertion that is further supported by whole-exome sequencing (WES) studies that show an average of 35 nonsynonymous mutations per patient with ASM (47). On the basis of these data, it seems likely that midostaurin's effect may depend, in part, upon oncogenic pathways not involved in canonical KIT signaling. Midostaurin is a derivative of the pan-kinase inhibitor staurosporine, and lacks potency and specificity toward KIT when compared with avapritinib (46, 48). Other targets of midostaurin include FLT3, PDGFRA, PKC, and other kinases important in myeloid development (46, 49, 50). Notably, pharmacokinetic studies show serum concentrations of midostaurin decrease sharply after the first administration in patients, suggesting that midostaurin induces its own metabolism and may not maintain high enough concentrations in the blood to sustain KIT inhibition beyond the initial treatment period (51). Given data from over a decade of experience with midostaurin, it is likely that its moderate efficacy and lack of on-target resistance mutations in SM result from the combination of poor pharmacokinetic properties, low potency, and general lack of specificity for KIT in a disease with significant genetic complexity (29, 52).

In contrast, avapritinib is a highly selective and potent KIT inhibitor (46). Although the genetic complexity of SM may pose a challenge for the success of all KIT-targeted therapies in SM, the early success of avapritinib in SM strongly suggests that avapritinib is exerting its effect through KIT inhibition, and that avapritinib may apply sufficient pressure on the disease to select for resistance conferring KIT mutants. In addition, comparison of WES in GIST with WES in SM shows GIST has a higher mutational burden than SM (35–60 mutations/sample in GIST vs. 25 mutations/sample in SM; refs. 47, 53, 54). Given that GIST is at least as genetically complex as SM, and that resistance to KIT TKIs in GIST often involves on-target KIT mutations, it seems plausible that on-target mutations will confer avapritinib resistance in SM. In addition, AML has been shown to harbor relatively few mutations in coding regions (55), and the potency of avapritinib may enable the first assessment of exon 17–mutant KIT as an oncogenic driver mutation in this disease. An analogous scenario was observed in the validation of FLT3 as a drug target in FLT3-ITD–positive AML. In that disease, the first several FLT3 inhibitors, including midostaurin, failed to achieve deep responses, raising the possibility that pathologically activated FLT3 was not a disease driver. Subsequently, quizartinib, the first potent and selective inhibitor of FLT3-ITD, was shown to induce deep remissions in a substantial proportion of patients (56). Moreover, acquired resistance to quizartinib was highly associated with secondary resistance mutations in FLT3-ITD, thus validating FLT3-ITD as a therapeutic target and driver of AML (31, 57).

The ability of secondary mutations to confer clinical resistance to targeted therapeutics is highly dependent upon the concentration of drug safely achievable in patients. Only translational studies of clinical isolates obtained from patients with acquired resistance to targeted therapy can provide definitive evidence for the clinical importance of candidate resistance mutations. We found the high nanomolar concentration of avapritinib required to inhibit the proliferation of Ba/F3 KIT D816V/T670I cells is similar to the concentration of avapritinib required to inhibit the proliferation of Ba/F3 KIT V560D/V654A cells. Because the V654A mutation in patients with GIST with JM-mutant KIT is associated with clinical resistance to avapritinib (44), our studies point to the T670I mutation in KIT D816V as a candidate mediator of acquired clinical resistance to avapritinib. Avapritinib is the most potent and selective type 1 KIT TKI described to date, and our data suggest that the KIT D816V/T670I mutant is a high-value target for efforts to rationally design the next generation of potent type 1 KIT TKIs. Our MD simulations predict the T670I mutation induces resistance to avapritinib through a novel mechanism involving neither steric clash nor increased ATP affinity, two previously implicated resistance mechanisms for gatekeeper mutations. Rather, MD simulations predict distant conformational changes in the P-loop that contract the drug-accessible area adjacent to this region are primarily responsible for avapritinib resistance. These data support the development of potent type 1 KIT inhibitors that not only avoid interactions with T670, but can tolerate significant flexibility in the P-loop, or perhaps avoid interaction with this region altogether. Moreover, it is possible that gatekeeper mutations may impact the P-loop region in other kinases, and as efforts are undertaken to develop TKIs that retain activity against gatekeeper mutants in a broad range of kinases, it may be important to consider the potential impacts of gatekeeper substitutions on P-loop architecture. In addition, we provide rationale for clinically assessing midostaurin and avapritinib as second-line therapeutics for select secondary KIT mutations that arise upon initial treatment with the other agent. In light of the nonoverlapping resistance profiles of avapritinib and midostaurin, and the challenges of finding a single drug that can overcome the complexity of KIT TKI resistance, strategies that combine avapritinib with either midostaurin or a more potent type 1 TKI that retains activity against T670I, may forestall the development of clinical resistance and warrant clinical investigation in patients with malignancies harboring exon 17–mutant KIT.

M.P. Jacobson has ownership interest (including stock, patents, etc.) in and is a consultant/advisory board member for Schrodinger. No potential conflicts of interest were disclosed by the other authors.

Conception and design: B. Apsel Winger, D. Garrido Ruiz, N.P. Shah

Development of methodology: B. Apsel Winger, W.A. Cortopassi, D. Garrido Ruiz, L. Ding, R. Esteve-Puig, M.P. Jacobson, N.P. Shah

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Apsel Winger, L. Ding, K. Jang, A. Leyte-Vidal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B. Apsel Winger, W.A. Cortopassi, D. Garrido Ruiz, L. Ding, K. Jang, N. Zhang, R. Esteve-Puig, M.P. Jacobson, N.P. Shah

Writing, review, and/or revision of the manuscript: B. Apsel Winger, W.A. Cortopassi, D. Garrido Ruiz, L. Ding, K. Jang, R. Esteve-Puig, M.P. Jacobson, N.P. Shah

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Apsel Winger, R. Esteve-Puig

Study supervision: N.P. Shah

This study was supported by grant from: NCI CA176091 (to N.P. Shah), NCI 2T32CA108462-11, St. Baldrick's Foundation Award ID: 527421, and the Frank A. Campini Foundation (to B. Apsel Winger), and The China Scholarship Council Award grant no. 201706545010 (to N. Zhang); this work used the Extreme Science and Engineering Discovery Environment, which is supported by National Science Foundation grant no. ACI-1548562 (to M.P. Jacobson). N.P. Shah acknowledges the generous support of the Yen family. We gratefully thank Kevin Shannon, Kevan Shokat, Jonathan Winger, and Debajyoti Datta for helpful discussions.

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.