The amount of genomic information about leukemia cells currently far exceeds our overall understanding of the precise genetic events that ultimately drive disease development and progression. Effective implementation of personalized medicine will require tools to distinguish actionable genetic alterations within the complex genetic landscape of leukemia. In this study, we performed kinase inhibitor screens to predict functional gene targets in primary specimens from patients with acute myeloid leukemia and chronic myelomonocytic leukemia. Deep sequencing of the same patient specimens identified genetic alterations that were then integrated with the functionally important targets using the HitWalker algorithm to prioritize the mutant genes that most likely explain the observed drug sensitivity patterns. Through this process, we identified tyrosine kinase nonreceptor 2 (TNK2) point mutations that exhibited oncogenic capacity. Importantly, the integration of functional and genomic data using HitWalker allowed for prioritization of rare oncogenic mutations that may have been missed through genomic analysis alone. These mutations were sensitive to the multikinase inhibitor dasatinib, which antagonizes TNK2 kinase activity, as well as novel TNK2 inhibitors, XMD8-87 and XMD16-5, with greater target specificity. We also identified activating truncation mutations in other tumor types that were sensitive to XMD8-87 and XMD16-5, exemplifying the potential utility of these compounds across tumor types dependent on TNK2. Collectively, our findings highlight a more sensitive approach for identifying actionable genomic lesions that may be infrequently mutated or overlooked and provide a new method for the prioritization of candidate genetic mutations. Cancer Res; 76(1); 127–38. ©2015 AACR.
A tremendous amount of information now exists detailing the genetic alterations in leukemia cells. Despite this wealth of genomic data, our understanding of the functional significance of these genetic events lags far behind. One major challenge is how to sort through all of the known mutations to find novel therapeutic targets. To this end, we have developed an algorithm called HitWalker (1), which can prioritize gene mutations based on functional data, detailing the underlying vulnerabilities of the leukemia cells. These functional data are derived from a kinase inhibitor screening platform (2). These kinase inhibitor screens are run on primary patient samples, and then the kinase targets of the effective drugs are calculated based on the known efficacy of the drugs against various targets (1). This study is the first example of the HitWalker algorithm being used to identify a novel therapeutic target, TNK2.
Tyrosine Kinase Nonreceptor 2 (TNK2) is a cytoplasmic kinase also known as ACK1 (activated CDC42-associated kinase; ref. 3). TNK2 is part of a family of cytoplasmic tyrosine kinases that also includes TNK1 (3). TNK2 was originally identified based on its binding to the cell-cycle regulator, CDC42 (4). Together, TNK2 and CDC42 regulate cellular attachment and migration (5).
TNK2 is composed of several functional domains, including a sterile alpha motif (SAM) domain, a tyrosine kinase domain, a SH3 protein–protein interaction domain, a CDC42/RAC-interactive (CRIB) domain, and a region that is homologous to the EGFR-binding domain (EBD) of Mig6 and a ubiquitin association domain. In addition to phosphorylation in response to EGFR signaling, TNK2 can also be activated by other receptor tyrosine kinases (6), and phosphorylation of the TNK2 activation loop by SRC is required for its kinase activity (7).
Multiple mechanisms by which TNK2 contributes to solid tumors have been documented. TNK2 mutations have been found in renal cancer cells and also in lung, ovarian, and gastric cancers (8, 9). TNK2 genomic amplification has been associated with late-stage or metastatic lung and prostate cancers (10). Overexpression of TNK2 promoted metastasis in a mouse model of breast cancer (10). Finally, TNK2 signaling is disrupted in prostate (11), breast (12), and gastrointestinal (13) tumors.
TNK2 can activate several protumorigenic signaling pathways, including modulation of the prosurvival AKT signaling pathway (14), phosphorylation of androgen receptor leading to androgen-independent prostate cancer growth (11, 15), and negatively regulate the tumor suppressor Wwox (16). In myeloid malignancies, specifically chronic neutrophilic leukemia and atypical chronic myeloid leukemia, nonmutated TNK2 has been shown to be a functional target in patients with CSF3R mutations (17). TNK2 levels increase in BaF3 cells harboring oncogenic CSF3R mutations, which have upregulated JAK kinase and SRC signaling promoting IL3-independent growth (17).
In this report, we find a novel link between TNK2 mutations and leukemia using a combination of drug screening and deep sequencing of primary patient samples to identify novel mutations of TNK2 that are therapeutically targetable in leukemia. These mutations are oncogenic and sensitive to previously developed TNK2 inhibitors. In addition, we developed novel, potent TNK2 inhibitors that exhibit greater specificity for TNK2.
Materials and Methods
Sequencing of patient samples
Genomic DNA was extracted from pellets of 5 × 106 purified primary patient sample mononuclear cells using the DNeasy Blood and Tissue Kit (Qiagen). Deep sequencing was performed on a panel of 1,862 kinase-associated genes as previously described (17). The mutated regions of TNK2 were verified via Sanger sequencing (GeneWiz) and analyzed using Sequencher and DNASTAR software. All clinical samples were obtained with informed consent by the patients. The study was approved by the Institutional Review Boards at Oregon Health and Science University.
Genomic variants were prioritized based on the HitWalker variant prioritization for personalized functional cancer genomics algorithm as described previously (1). Functional screening of kinase inhibitors was performed using primary patient samples, and dependence on specific kinases was predicted based on an inhibitor sensitivity algorithm, as previously described (2). The gene mutations found within each patient sample were prioritized by HitWalker integration of mutant genes with the functionally predicted targets.
TNK2 transcript variant 1 in pDONR was purchased from Genecopoeia (GC-Y4392). TNK2 mutations were made using the QuikChange II XL site-directed mutagenesis Kit (Agilent Technologies) and the following primers: TNK2-D163E-F gcctgaagcccgaggtcctgagccagc, TNK2-D163E-R gctggctcaggacctcgggcttcaggc, TNK2-R806Q-F cgctgccaccccagctctcaagctc, TNK2-R806Q-R gagcttgagagctggggtggcagcg, TNK2-P494fs-F gctcaggaggtcggggggtccatg, TNK2-P494fs-R catggaccccccgacctcctgagc, TNK2-P632fs-F ctgcccccccccgcccgcctatg, TNK2-P632fs-R cataggcgggcgggggggggcag, TNK2-S808*-F cccaggtgagctttagagccggggtgg, TNK2-S808*-R ccaccccggctctaaagctcacctggg, TNK2-Q831fs-F gcctggatcacctgggggtggcgtac, and TNK2-Q831fs-R gtacgccacccccaggtgatccaggc. Wild-type and mutated TNK2 were transferred into a Gateway-converted version of pMXs-IRES-Puro (Cell Biolabs, Inc.) or MSCV-IRES-GFP using a Gateway LR Clonase kit (Invitrogen). Plasmid sequences were confirmed via Sanger sequencing (Eurofins).
Cell culture and transfection
The human embryonic kidney, 293T17 cell line (ATCC), was grown in DMEM (Invitrogen) medium with 10% FBS (Atlanta Biologicals), l-glutamine, penicillin/streptomycin (Invitrogen), and amphotericin B (HyClone). Murine retrovirus was created by cotransfecting TNK2 plasmids and the EcoPac plasmid (provided by Dr. Rick Van Etten, Tufts University, Medford, MA) into 293T17 using FuGENE 6 (ProMega). Supernatants were harvested 72 hours later. Ba/F3 cells were maintained in RPMI 1640 medium with 10% FBS, 15% conditioned medium from the WEHI cell line (a source of IL3), l-glutamine, penicillin/streptomycin, and amphotericin B.
Ba/F3 cytokine-independent growth assays
Ba/F3 cells were infected with murine retrovirus obtained from 293T17 cells via addition of polybrene and spinoculation at 30°C for 90 minutes at 2,500 rpm. Infected cells were placed under puromycin selection for 72 hours. Cells were expanded after being washed three times and finally plated at 5 × 105 cells/mL in RPMI 1640 medium with 10% FBS, penicillin/streptomycin, l-glutamine, and amphotericin B. Viable cell counts were measured using propidium iodide exclusion with a Guava personal cell analysis system (Millipore).
pTNK2 inhibition assay
293T cells expressing TNK2 were plated in 6-well format at a density of 250,000 cells per well 48 hours prior to treatment. Cells were then treated with a 100 μL of XMD8-87 or XMD16-5 at 5 μmol/L and with 9 1:1 serial dilutions down to ≈10 nmol/L. Two additional samples were treated with DMSO only. Cells were then incubated for 6 hours at 37°C. Protein extraction was accomplished by adding 300 μL of lysis buffer (complete Lysis-M EDTA-free: Roche + complete protease inhibitors Roche + PhosSTOP: Roche) to cells after removing media. Plates were gently shaken for 5 minutes at room temperature. Lysates were collected and cleared of incompletely solubilized material by spinning for 10 minutes at maximum speed in a microcentrifuge. Samples were prepared for SDS-PAGE using the EPage loading buffer (Invitrogen) by diluting 10 μL of lysate with 0.3 μL β-mercaptoethanol and 3.5 μL 4× loading buffer. Samples were heated at 70°C for 10 minutes and then run on EPage 8% gels (Invitrogen). For Western blot detection, EPage gels were transferred to nitrocellulose using the Invitrogen iBlot transfer instrument. After transfer, blots were blocked for 45 minutes in 5% milk diluted in TBS-T (TRIS-buffered saline containing 0.1% tween), then were washed 3 times for 5 minutes with TBS-T. Primary antibody (Millipore ACK1-PO-Y284 #09-142) was diluted into blocking buffer and incubated for 2 hours at room temperature. Horseradish peroxidase (HRP)–labeled anti-rabbit secondary was diluted into blocking buffer and incubated for 30 minutes. After 3 × 5 minute washes in TBS-T, Pierce SuperSignal chemiluminescent substrate was added and imaged using a Fugi chemiluminescence Imager.
TNK2 ELISA assay
293T cells expressing TNK2 were plated in 96-well format at a density of 50,000 cells per well 1 day prior to inhibitor treatments. Cells were treated with TNK2 inhibitors for 6 hours, washed in Dulbecco's Phosphate-Buffered Saline, and lysed in ice-cold RIPA buffer (Boston Bioproducts) containing Halt protease and phosphatase inhibitor cocktail (Thermo Scientific). Plates were incubated on ice for 10 to 20 minutes. Ninety-six–well ELISA plates were precoated with anti-ACK chicken antibody overnight, then washed, and blocked in PBS with 1% BSA. Sample from lysed cells was incubated for 2 hours at room temperature, washed, and detected with ACK1-PO-Y284 antibody or total ACK antibody (A11; Santa Cruz Biotechnology) for 1 hour. Plates were then washed and incubated with anti-Rabbit HRP (Cell Signaling Technology) and then detected using Glo Substrate (R&D) and a Pherastar Microplate Reader (BMG Labtech).
KiNativ is based on biotinylated acyl phosphates of ATP and ADP that irreversibly react with protein kinases on conserved lysine residues in the ATP-binding pocket. The compounds were screened following the detailed procedures described in KiNativ methodology (18).
Enzymatic kinase assay
Kinase targets were tested with biochemical enzymatic kinase assays using the SelectScreen Kinase Profiling Service (Life Technologies Corporation) to determine IC50 values. The compounds were assayed at 10 concentrations (3-fold serial dilutions starting from 1 μmol/L) at an ATP concentration equal to the ATP Km,app for the assay following the detailed procedures described in the SelectScreen Customer Protocol and assay conditions documents located at www.invitrogen.com/kinaseprofiling.
Infected Ba/F3 cell lines were distributed in 96-well plates with increasing concentrations of TNK2 inhibitors. Control lines (parental Ba/F3s, pMX empty vector, and TNK2 wild-type–expressing) were plated in RPMI 1640 medium with 10% FBS, 15% WEHI-conditioned medium, l-glutamine, penicillin/streptomycin, and amphotericin B, whereas posttransformation lines were plated in WEHI-deficient media. Cells were plated at a density of 6 × 104 cells/mL and treated with the following inhibitors for 72 hours: dasatinib (SellekChem), AIM-100 (CalBiochem; ref. 19), XMD8-87, and XMD16-5. All conditions were plated with three replicates. Cell viability was measured using a methanethiosulfonate (MTS)-based assay (CellTiter96 Aqueous One Solution; Promega), and absorbance (490 nm) was read at 1 and 3 hours after adding reagent, using a BioTek Synergy 2 plate reader. MTS absorbances of inhibitor-treated wells were normalized to those of untreated cells. Regression curve fit analysis of the normalized data to determine IC50 values was performed using GraphPad Prism software.
Transfected 293T17 cells were treated with cell lysis buffer (Cell Signaling Technologies) with added complete mini protease inhibitor mixture tablets (Roche), phenylmethylsulfonyl fluoride, and Phosphatase Inhibitor Cocktail 2 (Sigma-Aldrich). Lysates were spun at 8,000 rpm for 10 minutes at 4°C to pellet cell debris, mixed 2:1 with 3× GS sample buffer with β-ME [75 mmol/L Tris (pH 6.8), 3% sodium dodecyl sulfate, 15% glycerol, 8% β-mercaptoethanol, and 0.1% bromphenol blue], and heated at 95°C for 5 minutes. Lysates were run on criterion 4% to 15% Tris-HCl gradient gels (Bio-Rad), transferred to a polyvinylidene difluoride membrane, and blocked for 30 minutes in TBS-T with 5% BSA. Blots were probed overnight at 4°C with anti-ACK1 (TNK2) rabbit antibody (Abcam; catalog no. ab65108), anti-ACK1 Phospho Y284 rabbit antibody (Abcam; ab74091), anti-ACK N-terminal antibody (Abcam; ab137506), anti-actin mouse antibody (Millipore), followed by anti-rabbit or anti-mouse IgG HRP conjugate secondary antibodies. Blots were developed using Clarity chemiluminescent substrate (Bio-Rad) or SuperSignal West Dura Extended Duration Substrate (Life Technologies) and imaged using a Bio-Rad ChemiDoc MP Imaging System or exposed to X-ray film.
Functional genomic prioritization identifies TNK2 as an important target in a patient with acute myelogenous leukemia
In order to identify the actionable genetic drivers in a patient with acute myelogenous leukemia (AML; 10-514), we utilized functional inhibitor screen data to prioritize genomic mutations found by sequencing of 1,862 kinase-associated genes. The functional data were obtained by screening the primary patient sample against a panel of small-molecules composed largely of kinase inhibitors. An IC50 value was then calculated for each inhibitor and compared with the median of the IC50s for the full patient sample cohort that has been screened to date. Inhibitors for which the IC50 value was less than 20% of the cohort median were considered effective. As an example, the cohort median IC50 observed for dasatinib to date has been 109 nmol/L, and in comparison, the two specimens analyzed for this study exhibited significantly lower IC50s of 6.8 and 22.6 nmol/L (Fig. 1A) The full list of effective drugs was used to predict specific kinase targets that were important for mediating the observed drug sensitivity profile, using a previously described algorithm (2). These functional targets were then integrated with mutant genes observed within this same specimen using the previously described HitWalker algorithm, which annotates functional and genomic targets onto the STRING database of protein–protein interaction networks to calculate the shortest path between mutated genes to functional gene targets (1). In this way, we prioritized mutant genes that were likely to explain the drug sensitivity pattern observed for AML patient specimen 10-514 (Fig. 1B). In this specimen, two genes were both highly ranked functional targets and were themselves mutated genes—the tyrosine kinases EPHA8 and TNK2 (Figs. 1B and 2A and B).
Prioritized TNK2 mutations are highly oncogenic
To determine whether or not the prioritized mutations were oncogenic, we tested them in a cytokine-independent growth assay. The murine pro-B Ba/F3 cell line normally requires the cytokine IL3 for growth, but can grow in the absence of IL3 if a transforming oncogene is expressed. To test the oncogenic capacity of the EPHA8 and TNK2 mutations, we introduced wild-type and mutant versions of EPHA8 and TNK2 into Ba/F3 cells and withdrew IL3. The EPHA8 mutation exhibited no transforming capacity in this assay (data not shown), whereas the TNK2 D163E mutation allowed the cells to grow in the absence of IL3 (Fig. 2C). Based on the oncogenic capacity of this novel TNK2 mutation, we looked for other patient samples in which a TNK2 mutation had been highly ranked by HitWalker. In a chronic myelomonocytic leukemia (CMML) patient sample (10-210), a TNK2 R806Q mutation was also highly prioritized (Supplementary Fig. S1 and Fig. 2A and B). This mutation also transformed Ba/F3 cells to cytokine-independent growth. The D163E mutation occurs in the tyrosine kinase domain of TNK2, while that R806Q mutation is in the EBD of TNK2 (Fig. 2A). Analysis of these TNK2 mutations by Western blot revealed that the TNK2 D163E mutation results in increased phosphorylation of TNK2, whereas the R806Q mutation does not. This suggests that the D163E mutation may increase kinase activity of TNK2, whereas the R806Q mutation may lead to increased signaling through alternative mechanisms.
Structural consequences of the D163E and R806Q mutations
From a structural perspective, the two activating mutations of TNK2, D163E and R806Q, are rather conservative and do not drastically change the electrostatic properties of the TNK2 domain in which they are located. Based on available structural data, we tried to infer a possible mechanism as to how the mutants could influence kinase activity. For TNK2, structural data are available for the kinase domain in its active conformation (in the presence and absence of ATP-pocket ligands) as well as in its inactive conformation as part of a kinase–SH3 domain construct. Based on these structural studies, homodimerization has been proposed to allosterically activate TNK2 kinase activity similar to EGFR (20). The latter forms an asymmetric head-to-tail dimer, with the N-lobe of one molecule interacting with the C-lobe of the second molecule (21). Another example of kinase domain dimerization is BRAF that forms a symmetric side-to-side dimer involving primarily residues of the N-lobe (22). In contrast with EGFR and BRAF, the structures of the TNK2-active state show a head-to-head symmetric dimer involving residues in proximity to the catalytically active αC-helix in the N-lobe (20).
One of the two activating mutations, the D163E, resides in the N-lobe of the kinase domain, and therefore existing structural data could be used to build a hypothesis explaining the activating effect of the D to E mutation. The D163 residue is located in the loop connecting the β3 strand with the αC-helix (Fig. 3A). This loop is not present in some of the structures deposited in the PDB database, which might reflect a certain degree of flexibility. Nonetheless, Protein Data Bank entry 4ewh presents the active state of the kinase, including this loop region (23), in which the side chain of D163 is solvent exposed and may be involved in the dimerization interface (Fig. 3A). In the interface, however, no potentially interacting residue, which would engage the aspartate carboxyl group in either an ionic interaction (Lys/Arg) or a hydrogen bond (Ser/Thr), could be found in the near proximity (<5Å).
In a different approach, comparison of the active and inactive conformation of the TNK2 kinase domain reveals the classical rearrangements within the N-lobe (Fig. 3B; refs. 20, 23). Specifically, in the inactive state, the αC-helix undergoes an outward movement, and also the N-terminal β-sheet is repositioned. These conformational changes also influence the position and orientation of the β3-αC loop that hosts the D163 residue. While in the active conformation, the side chain of D163 points to the outside, it is buried between the C-terminal β-sheet and the αC-helix of the N-lobe in the inactive conformation (Fig. 3B). Similar to the active state situation, there is no functional group in proximity of 5Å of D163 that would lock the residue in this position in the inactive state (Fig. 3C). However, one could well imagine that upon a D163E mutation, the longer and thus bulkier glutamate side chain causes steric clashes within the core of the N-lobe (Fig. 3D). As a consequence, the β3-αC loop would undergo a conformational switch to a more outward-oriented position of the D163E residue, thereby stabilizing the αC-helix in the active conformation.
A closer look at the sequence of this region of TNK1 and TNK2 across different species reveals additional interesting observations strengthening our above made hypothesis. While in TNK2, the KPD(161-163) segment in the β3-αC loop is only conserved in mouse and human, TNK1 possesses a Glu at position of D163. This glutamate, however, is either accommodated in a longer β3-αC loop or preceded by less bulky residues, such as glycine. Furthermore, in other cytoplasmic tyrosine kinases, this motif in the β3-αC loop is often composed of a KP[S/G] sequence that is also less bulky as compared with KP[D/E].
The importance of this β3-αC loop for the regulation of kinase activity is further corroborated by reports about other tyrosine kinases whose activity was influenced upon mutations in this region of the kinase domain. For example, targeted mutations in this region of SRC have shown to be activating (24, 25), whereas alanine scanning in this loop of CSK has led to a loss of kinase activity due to disruption of the kinase–SH2 interface (25, 26). Also for ABL kinase mutation of residue, E275 has been shown to influence kinase activity (27).
In contrast with D163, the second mutation R806Q lies in a region of TNK2 that is structurally undefined. Based on the primary sequence, this region has been identified as the EBD also known as MIG6 homology region (MHR). Secondary structure prediction by PsiPred (http://bioinf.cs.ucl.ac.uk/psipred/) indicates that this region is very likely to be unfolded. As the name implies and further shown by sequence alignments, this stretch of TNK2 possesses high sequence identity to be a part of the MIG6 protein that regulates EGFR kinase activity upon interaction with its kinase domain (Supplementary Fig. S2). Structural studies on a complex between EGFR and a MIG6 peptide have been published by Kuriyan and colleagues (28). MIG6 binds to the C-lobe of the kinase domain with one segment and likely extends toward the kinase-active site with a segment C-terminal to the first one leading to kinase inhibition. Within the same publication, the authors also speculate on a possible function of the EBD of TNK2 (28). Not only the EBD is highly homologous to MIG6 but also the C-lobes of the kinase domains of EGFR and TNK2 are very similar. This led to the hypothesis that the EBD of TNK2 could play a role in an autoinhibition mechanism of TNK2. The R806Q mutation lies in the segment of the EBD that would by homology bind to the C-lobe of the kinase (Fig. 2), and this interaction is required as a docking interaction so that the second segment can be engaged in further interactions with either the activation loop or residues from the N-lobe of the kinase. If now this mutation weakens the binding to the kinase C-lobe, the autoinhibition of TNK2 would be lost, resulting in an active kinase.
TNK2 mutations confer dasatinib sensitivity
In the TNK2-mutant leukemia samples, TNK2 was prioritized partly due to sensitivity of these samples to the multikinase inhibitor dasatinib (Fig. 1A). Dasatinib has been shown to exhibit binding to TNK2 in the low nanomolar range (29), suggesting that dasatinib sensitivity in these cases could be mediated through targeting of TNK2. To test whether the TNK2 mutations directly confer sensitivity to dasatinib, we treated the Ba/F3 cell lines expressing the TNK2 mutants with dasatinib and evaluated effects on cell growth and viability. Control or mutant cell lines grown in IL3 were completely insensitive to dasatinib, because the cytokine support renders their survival independent of TNK2. In contrast, both TNK2-mutant cell lines growing in the absence of IL3 were highly sensitive to dasatinib with IC50s of less than 10 nmol/L.
Novel TNK2 inhibitors
Despite dasatinib's ability to potently inhibit TNK2, its promiscuity as a kinase inhibitor (30) makes it difficult to conclude that its cellular potency is derived primarily from TNK2 inhibition. Therefore, we sought to identify structurally distinct TNK2 inhibitors with improved kinase selectivity that could serve as more precise mechanistic probes. We searched our in-house database for inhibitors that exhibited potent and selective binding to TNK2, which resulted in the identification of XMD8-87 (31) and XMD16-5 (Fig. 4A; ref. 32). The kinase selectivities of these two compounds were evaluated by screening against a diverse panel of 241 kinases using a chemical proteomic approach, KiNativ (Fig. 4B; ref. 33). The selectivity of XMD8-87 was also assessed using KinomeScan (Ambit Biosciences; ref. 29) across a panel of 402 human kinases at concentration of 10 μmol/L (see supporting information for complete profiling data). The top hits from both screening tests (kinases exhibit binding inhibition of 80% or more in Kinativ or 90% or more in KinomeScan) were further investigated by an enzymatic kinase assay (SelectScreen Kinase Profiling Services; Supplementary Table S1). This analysis revealed that both of these compounds are much more selective inhibitors of TNK2 than dasatinib. Activity of XMD8-87 and XMD16-5 against TNK2 was confirmed by ELISA (Fig. 4C). To confirm that these compounds could inhibit TNK2 in a cellular context, we evaluated their ability to inhibit TNK2 autophosphorylation by immunoblot analysis in 293T cells (Fig. 4D). Both compounds were found to be potent in the inhibition of TNK2 phosphorylation.
Selective TNK2 inhibitors potently block proliferation of TNK2-mutant cell lines
Dasatinib is a potent multikinase inhibitor and has many targets other than TNK2, such as ABL and SRC kinases. As such, we next wanted to test whether inhibitors with greater specificity to TNK2 are also effective against Ba/F3 cells that are dependent on mutant TNK2. In addition to XMD8-87 and XMD16-5, we also evaluated a structurally distinct third previously reported TNK2 inhibitor, AIM-100 (15). All three of the inhibitors exhibited potent inhibition of growth of the TNK2 mutant expressing cell lines while having little or no effect on the control cells out to the highest tested concentrations (1,000 nmol/L; Fig. 4B–D). The compound XMD8-87 had IC50s of 38 nmol/L and 113 nmol/L for the D163E and R806Q mutations (Fig. 4B), and the compound XMD16-5 had IC50s of 16 nmol/L and 77 nmol/L for the D163E and R806Q mutations (Fig. 4C). Both compounds show lower IC50 compared with the previously reported TNK2 inhibitor AIM-10021, which had IC50s of 91 nmol/L for the D163E cells and 320 nmol/L for the R806Q cells (Fig. 4D).
Mutations of gatekeeper residue inhibit drug sensitivity
In order to confirm that the death of cells harboring TNK2 mutations was specifically due to inhibition of TNK2, we designed potentially inhibitor-resistant alleles by mutating the “gatekeeper” threonine (T205) to a methionine (T205M) or an isoleucine (T205I). This residue was selected for mutation, as it is well characterized as a hot-spot that can confer resistance in many of the therapeutically important kinases, such as BCR-ABL, c-KIT, PDGFR, and EGFR (34–37). In addition, from crystallography and molecular modeling, dasatinib, XMD8-87, and XMD16-5 are all predicted to engage in hydrogen bonding interaction with the hydroxyl of T205. We engineered these gatekeeper mutations in the context of the most potent oncogene of the TNK2 mutations, D163E. We then generated IL3-independent Ba/F3 lines and tested the cell lines for sensitivity to dasatinib, AIM-100, XMD8-87, and XMD16-5 (Fig. 5). The gatekeeper mutations conferred near-complete resistance of the D163E-mutant cell lines to all of the compounds tested, indicating that the effects of these inhibitors on TNK2 cell lines are largely due to on-target effects on TNK2. It is important to note, however, that the gatekeeper mutations are in and of themselves activating, and thus increased growth rate of the cells could affect drug sensitivity.
TNK2 mutations in other tumor types are sensitive to the novel TNK2 inhibitors
Analysis of The Cancer Genome Atlas data using cBioPortal (38) revealed that many tumor types harbor TNK2 mutations. Of particular interest were several recurrent truncation mutations found in TNK2. Because these truncation mutations resulted in complete or partial loss of the portion of TNK2 that is homologous to MIG6, which in addition to binding EGFR also participates in negative regulation of the kinase domain, we hypothesized that they could be activating. We generated four of these truncation mutations: P494fs [found in several cell lines in the Cancer Cell Line Encyclopedia (CCLE); ref. 39], P632fs (CCLE, glioblastoma, pancreatic, and stomach cancers), S808* (bladder cancer), and Q831fs (stomach cancer; Fig. 7A). Of particular interest, the S808* and Q831fs mutations were markedly overexpressed when expressed in 293T17 cells (Fig. 7B), with a corresponding increase in phosphorylation (Fig. 7C). To determine whether autophosphorylation of these overexpressed TNK2 mutants could be blocked with TNK2 inhibitors, we treated the cells with dasatinib, XMD8-87, or XMD16-5 and indeed found that the vast majority of phosphorylation was blocked. These data suggest that these novel TNK2 inhibitors may be useful therapeutically in other tumor types harboring activating mutations in TNK2.
In this study, we have identified two oncogenic mutations in TNK2 from leukemia patient samples using the HitWalker algorithm, which uses functional data to prioritize mutations that may be functionally relevant in the patient sample. This method allowed us to rapidly prioritize TNK2 as a novel therapeutic target in leukemia. It also has the advantage of allowing for the identification of mutations that are of functional importance, but may be overlooked due to their rarity in genomic data alone. The first mutation, D163E, is located in the tyrosine kinase domain of TNK2. A previous report also identified a mutation in the kinase domain (E346K) from a patient with ovarian endometroid carcinoma (9). The E346K mutation had the highest levels of phosphorylation at tyrosine 284, of all the mutations screened in that study (9). This is consistent with our finding that the D163E mutant, also in the kinase domain, results in high levels of TNK2 phosphorylation (Fig. 2D). Interestingly, although the D163E and E346K mutations are in different lobes of the kinase domain, they result in similar phenotypes. Molecular modeling suggests that although a fairly conservative substitution, the D163E mutation may stabilize the αC-helix of the N-lobe in an active conformation (Fig. 3).
The second mutation is present in the Mig6 homology domain (MHC, aka EBD) of TNK2. This mutation, R806Q, is very close to a synthetic mutation F820A designed to block interaction of the MHD with the kinase domain (9). This interaction of the kinase domain with the MHD is thought to inhibit the kinase activity of TNK2 (9). Interestingly, the F820A mutation was hyperphosphorylated (9), whereas the R806Q mutation is not (Fig. 2D), suggesting that different regions of the MHD may have different functional importance. The R806Q mutation resides within a segment that is highly homologous to the C-lobe–binding peptide of Mig6, and therefore may be involved in kinase autoinhibition. In summary, based on TNK2 kinase domain structures as well as sequence alignments and homology, it is possible to derive hypotheses giving a possible explanation for the two activating mutations in this kinase. To further elucidate the molecular mechanism of kinase activation and a possible autoinhibition mechanism of the MHD domain, further experimental evidence is needed.
Mutated TNK2 may also represent a therapeutic target in other malignancies. In this study, we identified two truncation mutations found in solid tumors that led to receptor overexpression and a correspondingly high level of phosphorylation (Fig. 7). Interestingly, these two mutations both lead to a loss of a portion of the MHD, which may negatively regulate kinase activity, but also a loss of the ubiquitin association domain (UBD; Fig. 7A). The UBD is the region that interacts with ubiquitin ligases, leading to protein degradation and turnover, and therefore the loss of the UBD could contribute to protein overexpression (40).
Dasatinib is a multitargeted kinase inhibitor initially developed against ABL and SRC kinases (30). Dasatinib sensitivity in samples from patients with TNK2 mutations (Fig. 1B) could result from either direct inhibition of TNK2 kinase activity itself or indirectly through inhibition of SRC kinases by dasatinib, because SRC can activate TNK2 by directly phosphorylating the TNK2 activation loop (7). Alternatively, any number of other targets of dasatinib could be relevant. Dasatinib has been shown to exhibit low nanomolar binding affinity for both TNK2 and SRC with Kds of 5.6 and 0.5 nmol/L, respectively (30). Other TNK2 inhibitors, such as AIM-100, have been shown to be effective in cell culture models of prostate cancer (15).
In this study, we newly describe two kinase inhibitors that exhibit greater selectivity for TNK2 compared with dasatinib. XMD8-87 in particular demonstrated a high degree of selectivity for TNK2 (Fig. 4B) and is also more potent than the previously reported TNK2 inhibitor, AIM-100 (Fig. 4). Furthermore, XMD8-87 and XMD16-5 potently inhibit phosphorylation of TNK2 truncation mutations found in solid tumor types (Fig. 7C). These compounds represent promising new lead chemical candidates for further development of clinically applicable, selective TNK2 inhibitors across multiple tumor types.
Here, we prioritized TNK2 mutations as important functional targets using the HitWalker algorithm (1). This highlights the utility of the HitWalker algorithm in identifying actionable genomic lesions, including those that are infrequently mutated and may otherwise be overlooked in the analysis of genomic data alone. It is important to note that other genes not sequenced as part of the panel of kinase-associated genes used in this study could also contribute to the kinase sensitivity observed in these samples, and it will therefore be important in the future to perform a more comprehensive genomic analysis through whole-exome sequencing. Given the large number of mutations present in a wide variety of tumors, the ability to prioritize genetic lesions greatly reduces the time and resources necessary to validate candidate mutations. Furthermore, this study highlights the utility of drug screening data for understanding the underlying vulnerabilities of leukemia cells and their accompanying gene mutations.
Disclosure of Potential Conflicts of Interest
N.S. Gray has ownership interest (including patents) in Dana Farber. B.J. Druker reports receiving other commercial research support from NIH STTR via subcontract from Oncotide Pharmaceuticals, Novartis Pharmaceuticals clinical trial funding to institution, Bristol-Myers Squibb clinical trial funding to institution, and ARIAD clinical trial funding to institution, and is a consultant/advisory board member for MolecularMD. J.W. Tyner reports receiving other commercial research support from Aptose Biosciences, Array Biopharma, Constellation Pharmaceuticals, Incyte, and Janssen Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.
Conception and design: J.E. Maxson, J. Wang, X. Deng, N.S. Gray, J.W. Tyner
Development of methodology: J.E. Maxson, S.K. McWeeney, R.E. Middleton, J.W. Tyner
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.E. Maxson, M.L. Abel, S.B. Luty, H. Sun, J. Gorenstein, S.B. Hughes, R.E. Middleton, B.J. Druker, J.W. Tyner
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.E. Maxson, M.L. Abel, X. Deng, S. Reckel, S.B. Luty, H. Sun, S.B. Hughes, B. Wilmot, S.K. McWeeney, O. Hantschel, R.E. Middleton, J.W. Tyner
Writing, review, and/or revision of the manuscript: M.L. Abel, X. Deng, S. Reckel, B. Wilmot, O. Hantschel, N.S. Gray, B.J. Druker, J.W. Tyner
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.L. Abel, J. Wang, D. Bottomly, B. Wilmot, S.K. McWeeney, J.W. Tyner
Study supervision: N.S. Gray, B.J. Druker, J.W. Tyner
Other (developed inhibitor used in the article): N.S. Gray
This work was supported by Howard Hughes Medical Institute and The Leukemia & Lymphoma Society. J.E. Maxson is supported by a Leukemia & Lymphoma Society Fellow Award and a Medical Research Foundation Early Clinical Investigator Award, and NCI K99 CA190605-01. J.W. Tyner is supported by grants from the V Foundation for Cancer Research, the Gabrielle's Angel Foundation for Cancer Research, and the NCI (5R00CA151457-04; 1R01CA183947-01). N.S. Gray, X. Deng, and J. Wang are supported by NIH grant (5 R01 CA173469-02).
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.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received March 27, 2015.
- Revision received August 24, 2015.
- Accepted September 7, 2015.
- ©2015 American Association for Cancer Research.