Larotrectinib, selitrectinib, repotrectinib and zurletrectinib were obtained from InnoCare. Cabozantinib was purchased from MedChem Express. All drugs were dissolved in DMSO to produce 10 mM stocks and stored at −20 °C. Cabozantinib was added to our studies because it was previously reported to be specifically active against TRK xDFG mutant tumors [20, 22].

NTRK fusion-positive colorectal cancer cell lines IRC, Kor1, and KM12 were utilized. The IRC and the KM12 cell lines were obtained from Dr. Bardelli and cultured according to the described protocols [14, 19, 23]. The Kor1 cell line was obtained from Lee et al. [24] and cultured on laminin pre-treated plates in RPMI media supplemented with 10% Fetal Bovine Serum (FBS). All cell lines were authenticated by STR and periodically screened for the presence of mycoplasma. For the mutant models, two human-derived colorectal cancer cell lines (IRC) and two mouse-derived glioma models (NLS) were used. The single mutant IRC LMNA-NTRK1 TRKA G595R primary cell line was obtained from Dr. Bardelli and established from the tumor of a patient who had progressed on earlier generation TKI treatment (the specimen was used to generate a Patient-Derived-Xenograft from which the cell line was then established) [14]. The double mutant IRC LMNA-NTRK1 TRKA G595R/G667C cell line was established following chronic exposure of the single mutant patient-derived cell line to increasing concentrations of repotrectinib. Sequencing of the double mutant cell line was performed using MSK-IMPACT [25]. IRC primary cell lines were cultured with DMEM/F12 50:50 Mix (Corning) supplemented with 10% FBS. NLS Bcan-Ntrk1 Trka G598R and Bcan-Ntrk1 Trka G670C single mutants as well as double mutants NLS Bcan-Ntrk1 Trka G598R/G670A and NLS Bcan-Ntrk1 Trka G598R/G670C isogenic models were established using CRISPR/Cas9 to knock-in TRKA solvent-front and xDFG mutations into tumor cells derived from a Bcan-Ntrk1-driven glioma mouse model. Trka G598R is the mouse ortholog to human TRKA G595R while Trka G670A and Trka G670C are the mouse ortholog to human TRKA G667A and TRKA G667C, respectively. Mouse p53 −/−Bcan-Ntrk1 glioma cells were plated on laminin-coated plates and cultured with Neurocult Stem Cell Basal Media with Proliferation Supplements (Stem Cell Technologies) [26]. Ba/F3 cell lines stably expressing WT and mutant versions (i.e. G595R, G667C/A, V573M, F589L, F633L, F617L, G623R/E, G639R, G696A/C, G709C, V608D/M) of human TRKA, TRKB or TRKC were generated using either lentiviral transduction or standard transfection methods by a contract research organization (Kyinno, Beijing, China).

CellTiter-Glo Cell Viability Assays (Promega) and crystal violet clonogenic assays were performed on primary NTRK fusion-positive colorectal cancer cell lines, mouse Ba/F3 cell lines and isogenic mouse glioma cell lines. For the CellTiter-Glo assays, three biological replicates were performed, with each condition being assayed in triplicate determinations [22]. Cells were seeded in a 96-well plate in the afternoon at optimal density. The following morning, larotrectinib, selitrectinib, repotrectinib, zurletrectinib and cabozantinib (1:2, 1:3 or 1:4 dilutions with a maximum concentration of 10 µM) were added. Plates were removed from the incubator 72 h later and CellTiter-Glo reagent was added. Absorbance was read at 490 nm in accordance with Promega’s protocol. Data is presented as a survival percentage on the y-axis (mean ± STDEV) normalized to the control DMSO-treated cells deemed 100% viable. Drug concentrations on the x-axis are represented as a base 10 logarithm (LOG). For the crystal violet assays, cells were seeded in a 24-well plate at optimal densities in the afternoon. The following morning, larotrectinib, selitrectinib, repotrectinib and zurletrectinib were added at concentrations ranging from 1 nM to 5000 nM. Crystal violet plates were removed from the incubator 72 h later, washed with PBS, fixed with 4% paraformaldehyde for 15 min, and stained with crystal violet for 10 min. Crystal violet was washed off and plates were left to dry prior to imaging. Clonogenic assays were performed in three biological replicates.

Cell lines were seeded in six-well plates in the afternoon at 600,000 cells per well in full media. The following morning, cells were treated with 100 nM of each compound for 30 min. For isogenic mouse glioma cell lines Bcan-Ntrk1 Trka G598R and Bcan-Ntrk1 Trka G598R/G670A cells were seeded in six-well plates in the afternoon at 800,000 cells per well in full media. The following evening, cells were put into starvation with Neurocult Basal media containing only penicillin and streptomycin. The following morning, cells were treated with 100 nM of each compound for 30 min. Upon completion of the 30 min treatment total protein lysates were extracted in 75 uL of radioimmunoprecipitation assay (RIPA) buffer containing phosphatase and protease inhibitors. Lysates were quantified using bicinchoninic acid in accordance with the manufacturer’s protocol. Lysates were separated with sodium dodecyl sulfate-polyacrylamide gel electrophoresis gels in accordance with standard methods. All membranes were probed with the following antibodies: phospho-p44/42 MAPK (Erk1/2; T202/Y204) clone D13.14.4E (4370S, Cell Signaling Technology); total ERK1/2 (9102S, Cell Signaling Technology); pan TRK clone A7H6R (92991S, Cell Signaling Technology); phospho-TRKA (Y674/675) clone C50F3 (4621S, Cell Signaling Technology) and β-actin clone 13E5 (4970S, Cell Signaling Technology). Western blots were performed in three biological replicates and a representative western blot is shown.

Recombinant kinases were purchased from SignalChem Biotech Inc (Richmond, BC, Canada). In vitro kinase assays were performed following the optimization of the HTRF Kinase Assay (Cisbio) [27]. Compounds larotrectinib, selitrectinib, repotrectinib, and zurletrectinib were tested in 10-dose IC50 mode with four-fold serial dilutions starting at 100 nM (for TRKA, TRKB, and TRKC) or 1 or 10 µM (for TRKA mutant kinases). ATP concentrations ranged from 1 to 10 µM based on internal optimization. Substrate concentrations ranged from 1 to 10 µM at Km of each kinase.

Evaluation of zurletrectinib’s chemical structure revealed numerous similarities with the related macrocyclic, pyrazolopyrimidine, pan-TRK, DFG-Din, Alpha-Cin, and Activation-Loop-in ATP-competitive inhibitor, repotrectinib. We accessed the Protein Data Bank (PDB; https://www.rcsb.org/) and retrieved available crystal structures of the wild-type TRKA-Repotrectinib complex (PDB ID: 7VKO) and the solvent-front mutant TRKA G595R (PDB ID: 7VKN). Unfortunately, no TRKB or TRKC DFG-Din plus Alpha-Cin crystal structures were found on PDB suitable for zurletrectinib docking. We thus modeled these from FASTA sequences using SWISS-MODEL (http://swissmodel.expasy.org) [28] homology modeling starting from TRKA 7VKO as a template. To examine the influence of secondary mutations on the zurletrectinib-TRK interaction, we generated several solvent-front and xDFG mutant proteins using Pymol (http://pymol.org) [29]. Before docking, proteins were prepared using the Protein Preparation Wizard in Schrödinger’s Maestro. We obtained the chemical structure of zurletrectinib from PubChem and prepared it using Maestro’s LigPrep function (https://www.schrodinger.com/life-science/learn/white-papers/protein-preparation-wizard/). The receptor grid was centered on catalytic spine 6 (CS6) residue (TRKA L656, TRKB L699, TRKC L686), which is located in the center of the ATP binding pocket. A cubic grid box of dimensions 30 Å × 30 Å × 30 Å was constructed. Docking constraints were set to ensure the generation of a hydrogen bond with the peptide backbone of the hinge’s third (H3) residue (TRKA M593, TRKB M636, TRKC M620). We executed ligand docking simulations using the Glide module in standard precision mode [30], with the hydrogen-bond generation with H3 residue set as a constraint. The highest-ranking poses were visually inspected in Maestro and juxtaposed with the related macrocyclic inhibitor, repotrectinib. We analyzed interactions, such as hydrogen bonds and hydrophobic contacts for both WT and secondary mutant proteins.

KM12 and Ba/F3 LMNA-NTRK1, TRKA G595R xenografts were generated by subcutaneous injection of five and six million cells into the right flank of 6–8 weeks old BALB/c nude female mice, respectively. Treatment with zurletrectinib, selitrectinib and larotrectinib started at tumor reaching around 150 mm3. Xenografts were randomized into different groups (8 mice per group) based on tumor size and dosed orally with zurletrectinib (0.1, 0.3, 1 mg/kg BID for KM12 models and 0.3, 1, 3 mg/kg BID for Ba/F3 models), selitrectinib (30 mg kg−1 BID per day) and larotrectinib (30 mg kg−1 BID) for constant 10 days or 23 days for KM12 or Ba/F3 models, respectively. Eight mice per group were included in each experiment. Randomization was conducted without utilizing a blind method. Tumors were measured twice weekly using calipers and body weight was also assessed twice weekly. At the end of each treatment, animals were euthanized. The vehicles were formulated as follows: were first dissolved in Cremophor EL/ethanol (50:50) as a 4× stock solution, vortexed, sonicated and then diluted to a 1× solution with saline before use.

All pharmacodynamics (PK) studies in rats were conducted according to protocols approved by the Animal Care and Use Committee at InnoCare pharmacology company in China. Male Sprague Dawley (SD) rats (n = 3 per time point) received a single oral administration of 10 mg/kg of zurletrectinib, selitrectinib and repotrectinib. Zurletrectinib was suspended in 0.5% methylcellulose; repotrectinib and selitrectinib were reconstituted in 0.5% methylcellulose containing 10% Cremophor EL. For the collection of plasma, brain tissue and cerebrospinal fluid (CSF), animals were euthanized with CO2 at 0.5 and 2 h post dosing and blood samples were in EDTA-treated microtubes. Brain tissues were homogenized and immediately flash frozen (−80 °C). Blood samples and CSF were spun at 2000 × g for 15 min at 4 °C, and plasma was removed and stored at −80 °C until liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis. The agents in rat plasma, brain homogenate or CSF samples were extracted and quantified by LC-MS/MS.

In vivo brain orthotopic experiments were carried out at Memorial Sloan Kettering Cancer Center (MSKCC) [20]. Two mouse Bcan-Ntrk1 glioma models were used: single solvent front mutant Trka G598R (clone 1) and double xDFG mutant Trka G598R/G670A (clone 3). For each model, five arms were used with six mice per arm, totaling 60 mice. Athymic female mice (6–8 weeks old) were anesthetized with ketamine/xylazine and administered a preoperative dose of buprenorphine prior to stereotactic intracranial injection. Each mouse was injected with 200,000 cells in a volume of 2 uL 3 mm deep, 1 mm to the right of the sagittal suture (midline), and 1 mm posterior to the coronal suture behind the bregma with the aim being to target the right lateral ventricle. One week post cell implantation, the mice were randomized into different treatment groups and treatments began. The drugs utilized were larotrectinib, selitrectinib, repotrectinib and zurletrectinib. The vehicles were as follows: 100% labrafac for larotrectinib, 1% CMC, 0.5% Tween80 for selitrectinib, 0.5% methyl cellulose for repotrectinib, and zurletrectinib was first dissolved in Cremophor EL/ethanol (50:50) as a 4× stock solution, vortexed, sonicated and then diluted to a 1× solution with saline. For both the Trka G598R model and the Trka G598R/G670A model, mice were treated five days a week BID with the larotrectinib and selitrectinib groups receiving doses of 30 mg/kg and the repotrectinib and zurletrectinib groups receiving doses of 15 mg/kg. Mouse weights were monitored twice a week and survival was recorded and graphed on GraphPad Prism 9 [31, 32].

Statistical analyses were conducted using GraphPad Prism 9 (GraphPad Software Inc.). For CellTiter-Glo assays, raw data obtained were normalized to the DMSO control utilizing Microsoft Excel. GraphPad Prism 9 was used to determine the 50% growth inhibition concentration (IC50) using non-linear regression and curve fitting [22, 31]. Data is presented as the mean ± standard deviation (STDEV) of all replicates. Measurements were assessed for normal distribution and homogeneity of variance using appropriate tests and were analyzed using analysis of variance or T-Tests, and nonparametric tests were used for non-normal distributed data. Animal studies utilized five arms with six or eight mice per arm. A Kaplan–Meier test was used to assess lifespan characteristics and for comparison between groups [33]. Median survivals were obtained using GraphPad Prism 9. Sample size was based on statistical power analysis to ensure sufficient capability to detect a pre-specified effect size. We assumed a moderate effect size, with an alpha level set at 0.05, and a power (1-β) set at 0.80. P values were calculated using a Log-rank or Mantel–Cox test using GraphPad Prism 9. A P value < 0.05 was considered statistically significant.