Myosin Light-Chain Kinase Inhibition Potentiates the Antitumor Effects of Avapritinib in PDGFRA D842V-Mutant Gastrointestinal Stromal Tumor

Gastrointestinal stromal tumors (GISTs) are the most common type of sarcoma. They usually occur in the gastrointestinal tract and are driven by a mutation in either the KIT or PDGFRA receptor tyrosine kinase. KIT mutations are responsible for 75% of GISTs. Meanwhile, PDGFRA mutations drive approximately 10% of GISTs and most commonly reside in the juxtamembrane domain (exon 12) or the activation loop (exon 18) of the receptor (1, 2). Over the past 20 years, the advent of tyrosine kinase inhibitors (TKIs) has revolutionized the treatment of unresectable or metastatic GISTs. Imatinib is used as first-line therapy followed by sunitinib and then regorafenib for most KIT-mutated GISTs. Ripretinib also provided clinical benefit for advanced GISTs after regorafenib (3). However, the PDGFRA D842V mutation is resistant to imatinib and sunitinib (1, 2, 4–6). In one study, there were no responses in 31 patients with advanced GISTs with a PDGFRA D824V mutation treated with imatinib, and the median progression-free survival was only 2.8 months versus 28.5 months for patients with other PDGFRA mutations (5).

Avapritinib is a selective and potent inhibitor of activated KIT and PDGFRA mutant receptor tyrosine kinases (7). Avapritinib has been approved by the FDA for the treatment of adults with unresectable or metastatic GISTs that harbor a PDGFRA exon 18 mutation, including D842V (8). Avapritinib has shown significant clinical activity and durable responses in patients with unresectable/metastatic PDGFRA D842V-mutant GISTs (9–11). Although avapritinib is safe to use, cognitive effects such as memory impairment, confusion, and encephalopathy were observed in 40% to 57% of patients (9, 10). Avapritinib-related adverse events generally depended on the dose and exposure of treatment (9). Therefore, other strategies are needed to treat PDGFRA D842V-mutant GISTs.

We have previously developed genetically engineered mouse models (GEMMs) of Kit-mutated GISTs under the control of the endogenous Kit promoter to investigate oncogenic KIT signaling, the antitumor immune response, and new therapies including immunotherapy (12–16). GEMMs with a common GIST PDGFRA activating mutation expressed under the control of the endogenous promoter develop hyperplasia of stromal fibroblasts in embryos and progressive fibrosis in adults (17). However, these mice do not develop GISTs, even after increasing PDGFRA signaling with Ink4a/Arf deletion. Patient-derived xenografts (PDXs) are an alternative model to investigate tumorigenesis and drug efficacy in human cancer. Although PDX models of GISTs with primary and secondary KIT mutations are available and have been used to study avapritinib and other TKIs (7, 18, 19), a PDX of PDGFRA-mutant GIST has not been reported. The lack of an in vivo model of PDGFRA-mutant GIST limits investigation and understanding of this tumor.

Myosin light-chain kinase (MYLK) was first studied in muscle contraction. The kinase activity of MYLK is activated after binding the Ca²⁺–calmodulin complex and results in the phosphorylation of the myosin light-chains. Phosphorylation enables myosin to bind to actin, which results in contraction (20–23). Phosphatases then decrease myosin light-chain phosphorylation allowing muscle relaxation (21). In smooth muscle, when not bound to calmodulin, MYLK is phosphorylated at two sites by a cyclic AMP-dependent kinase that decreases both MYLK kinase activity and myosin light-chain phosphorylation (21, 24–26). Several kinases negatively regulate MYLK, including protein kinase A (PKA), protein kinase C (PKC), p21-activated kinase (PAK), and Ca²/calmodulin-dependent protein kinase II (CaMK II; ref. 26). In mammals, MYLK is encoded by the MYLK1 and MYLK2 genes. Although MYLK2 expression is restricted to skeletal muscles, MYLK1 is expressed in diverse cell types and tissues, including brain, platelets, secretory cells, and muscle cells (27, 28). MYLK is a major component of the actin cytoskeleton (29, 30). Through cytoskeletal organization and rearrangement, MYLK regulates many cell functions, such as proliferation, adhesion, migration, and epithelial barrier formation (31–33). Although MYLK mostly functions via phosphorylation of muscle and non muscle myosin light-chains, it can control cell migration by maintaining membrane tension rather than myosin light-chain phosphorylation (32). Recent evidence shows that MYLK is not limited to regulation of myosins, but is part of a multicomponent signaling pathway involving several other protein kinases (34).

We have now developed a GIST PDX model of the PDGFRA D842V mutation and studied the effects of imatinib and avapritinib. We discovered that avapritinib was highly effective but increased tumor cell expression of MYLK. MYLK inhibition allowed lower dose avapritinib to be effective. Our results support the idea of combination MYLK/avapritinib therapy, which may reduce the therapeutic dose of avapritinib and minimize avapritinib-related adverse events, such as altered cognition.

An 80-year-old male developed a peritoneal metastasis after removal of a primary gastric GIST. The tumor was resistant to imatinib and sunitinib. After surgical resection of the patient's tumor, fresh pieces measuring a few millimeters were implanted subcutaneously in the flank of female NOD SCID gamma (NSG, institutional animal facility) mice. These initial tumors were harvested and processed as single cells as described below and stored at −80°C. Then, 10⁶ cells in 200-μL solution PBS/Matrigel 50/50 were inoculated in the flanks of NSG mice. Tumors from passages 3 and 4 were used for the current experiments.

GIST T1 cells were cultured in RPMI containing l-glutamine and 25 mmol/L HEPES (Gibco), supplemented with 10% FBS, penicillin and streptavidin, and 0.1% 2-mercaptoethanol. Cells were pelleted, resuspended in PBS, and inoculated in the flank of NSG mice as described previously.

For in vivo experiments, avapritinib (SelleckChem and Blueprint Medicines) was dissolved in 0.5% carboxymethyl cellulose and 1% Tween 80 and administered by daily gavage. Imatinib was dissolved in water and administered in the drinking water as before (35). ML-7 (SelleckChem) was diluted in DMSO at 50 mg/mL and then diluted to 1 mg/mL in 30% PEG 400, 1% Tween 80. For cell culture experiments, avapritinib and ML7 were dissolved in DMSO; imatinib was dissolved in water. Experiments were performed in complete media (10% FBS) in a final concentration of 0.4% DMSO.

Tumor volumes were measured every 2 to 4 days with calipers and calculated via the formula V = (L × L × W)/2, where L is the longest dimension and W is the perpendicular length. Statistical analysis was performed with GraphPad Prism (version 7.0, RRID:SCR_002798). Comparison between two groups was done by unpaired t test analysis assuming unequal variances. Comparison between three or more groups was done by one-way ANOVA. Statistical significance was achieved when P < 0.05.

To prepare single cells from the GIST PDGFRA D842V PDXs, tumors were minced in 2.5 mL of 5 mg/mL collagenase IV (Sigma-Aldrich) and DNAse I (0.5 mg/mL, Roche Diagnostics) in HBSS as described previously (36), incubated at 37°C for 15 minutes, and then quenched with the addition of 2-mL FBS. The solution was filtered over a 100-μmol/L filter and washed with 20-mL PBS. The cells were centrifuged and resuspended in 20-mL PBS and passed through a 40-μmol/L filter. After centrifugation, the cell pellet was resuspended in RPMI containing 10% FBS and plated. For Annexin V and Phalloidin staining, cells were treated 24 hours after plating with ML-7 and avapritinib for 24 hours in complete RPMI media containing 10% FBS. For Phalloidin and DOG1 staining, cells were fixed in 4% PFA for 30 minutes before staining with Phalloidin (Abcam), following the manufacturer's protocol, or co-staining with Phalloidin and DOG1 (Cell Signaling Technologies). For co-staining, after cell permeabilization, cells were incubated with CAS Block (Thermo Fisher Scientific) for 30 minutes then with DOG1 in CAS block solution, overnight at 4°C. After three washes for 5 minutes in PBS, the secondary antibody against DOG1, Alexa Fluor plus 647 (Thermo Fisher Scientific), was added for 1 hour at room temperature. After three washes for 5 minutes in PBS, staining with Phalloidin followed the manufacturer's protocol. For Annexin V staining, cells were processed for flow cytometry as indicated below. Live, apoptotic, and dead cells were defined as 7-AAD⁻Annexin⁻, 7-AAD⁻Annexin⁺, and 7-AAD⁺Annexin⁺, respectively.

Viability assays of 5 × 10⁴ GIST T1 cells in RPMI1640 with 10% FCS were performed in 96-well flat bottom plates using Cell Counting Kit-8 (Dojindo) as instructed. There were five to 10 replicates per experiment.

Flow cytometry was performed using a LSRFortessa (BD Biosciences). Single cells from PDGFRA D842V PDXs tumors were prepared as above and processed as described previously (36). Single-cell suspensions were stained using an antibody cocktail in 100 μL of PBS plus 5% FBS in the dark at 4°C, washed, and analyzed immediately by flow cytometry. Human-specific antibodies conjugated to various fluorochromes were purchased from BioLegend - CD117 (clone 104D2), PDGFRA (clone 16A1). 7-AAD and Annexin V staining were performed using the BioLegend Apoptosis Detection Kit, as directed.

Next-generation RNA-seq of tumors from PDGFRA D842V PDXs treated with vehicle, imatinib, or avapritinib was performed by the institutional Next Generation Sequencing Core facility using an Illumina HiSeq 2500 platform. Raw counts of the indicated genes were normalized and analyzed using the R software package DESeq2 (RRID:SCR_000154). RNA-seq data are deposited in GEO. First, the RNA sequencing results of all the samples were normalized using the regularized-logarithm transformation found in the DESeq2 package. The Euclidean distance between each sample was then calculated and produced in a heatmap with 0 meaning no similarity and 1 meaning perfect similarity. The plotPCA function was also used to produce a PCA plot of all the samples. Finally, a heat map of select important genes form the normalized data was generated of all the samples using the R package ComplexHeatmap.

For histologic and IHC analyses, treated tumors were immediately fixed in 4% paraformaldehyde at 4°C overnight. After paraffin embedding, 5 μmol/L sections were stained for H&E, trichrome, Ki67 (Thermo Fisher Scientific), or cleaved caspase-3 (Cell Signaling Technology). Stained slides were scanned with an Aperio VERSA 200 (Leica). Proliferation index was measured by counting Ki67-positive cells in at least four 250-μmol/L × 250-μmol/L nonoverlapping fields per tumor, in at least four tumors per treatment group. Quantification of the surface stained with trichrome dye was measured with Fiji (RRID:SCR_002285; ImageJ2, RRID:SCR_003070).

To prepare tumor lysates, snap-frozen tumors were homogenized in a 850 homogenizer (Thermo Fisher Scientific) in lysis buffer and then incubated on ice for 30 minutes. Lysates were cleared by 15 minutes centrifugation at 4°C and then fractionated by SDS-PAGE. Protein extracts were prepared from tumors of several PDXs to assess variability in drug response. Phospho-PDGFRA (Y754), phospho-PDGFRA (Y762), phospho-PDGFRA (Y1018), phospho-PDGFR (Y849/Y857), PDGFRA, phospho-KIT (Y703), phospho-KIT (Y719), KIT, phospho-p44/42 ERK (Y202/Y204), p44/42 ERK, phospho-S6 Protein (Ser235/236), S6 protein, phospho-AKT (Ser473), AKT, phospho-STAT3 (Y705), STAT3, and GAPDH were obtained from Cell Signaling Technologies. MYLK antibody was obtained from Sigma.

Fifty micrograms of tumor lysate was incubated overnight with rotation at 4°C, with the antibody at 1/10 dilution in 60 μL total volume. Twenty microliters of Protein A/G PLUS-Agarose beads (Santa Cruz Biotechnology) was added for 2 hours with rotation at 4°C. Samples were then washed three times with 500-μL lysis buffer. The immunoprecipitate was then resuspended in 20 μL of 4× loading buffer containing 20% 2-mercaptoethanol and heated at 95°C for 10 minutes, then vortexed and centrifuged, and the supernatant was loaded on an SDS-PAGE gel. Anti-phospho-tyrosine PY-20 (BD Biosciences) and anti-PDGFRA (Santa Cruz Biotechnology) were used.

The PKA Colorimetric Activity Kit (Invitrogen) was used according to the manufacturer's instructions. PKA standards were serially diluted in the kinase reaction buffer provided to generate a standard curve. To prepare the samples, tumor lysates from the PDGFRA D842V PDXs and GIST T1 cell lysates were also serially diluted in the kinase reaction buffer. Several dilutions were tested, and the optimal dilution that fell within the standard curve was used to determine the PKA activity. The PKA colorimetric activity was read with a microtiter plate reader at 450 nm.

For transient MYLK knockdown, GIST T1 cells were transfected with 30 nmol/L of ON-TARGETplus SMARTpool siRNA for human MYLK (L-005351–00–0005) or nontarget control siRNA (D-001810–10–05; Thermo Fisher Scientific) using Lipofectamine RNAiMAX (Invitrogen) for 24, 48, or 72 hours.

Total RNA was extracted from tumor tissues or cells using RNAeasy Plus Mini Kit (Qiagen). One microgram RNA was reverse transcribed with the Taqman Reverse Transcription Kit using oligo (dT)₁₆ (Applied Biosystems) and amplified with the following PCR Taqman assays (Applied Biosystems): Hs01019589_m1 PDGFRB; Hs00998018_m1 PDGFRA; Hs00174029_m1 KIT; Hs00364926_m1 MYLK, repeated in triplicates. qPCR was performed using the QuantStudio 3 PCR system. Data were calculated as described (37).

Informed written consent was obtained from the patient. All studies were conducted in accordance with the Declaration of Helsinki, after approval by an institutional review board, and in accordance with an assurance filed with and approved by the U.S. Department of Health and Human Services. Animal experiments were approved by the University of Pennsylvania Institutional Animal Care and Use Committee (IACUC).

The RNA-seq data presented in this study are publicly available through the Sequencing Read Archive under accession No. PRJNA945239. Other data generated in this study are available within the article and its supplementary data files.

NSG mice developed a tumor 8 to 12 weeks after subcutaneous inoculation of a single-cell suspension of a human PDGFRA D842V mutant GIST in the right flank (Fig. 1A and B). The original human tumor and the PDX had epithelioid cells with high proliferation based on Ki67 staining, diffuse DOG1 (more so in the patient) and PDGFRA staining, and patchy KIT staining. The MSK IMPACT assay (38), which profiles more than 400 commonly mutated genes in cancer, was performed on the original patient tumor as well as passage 1 of the PDX. Both samples had a heterozygous PDGFRA D842V mutation. Furthermore, both samples had deletion of CDKN2A and CDKN2B, and accordingly their RNA expression was not detectable by RNA-seq.

To further validate the model, RNA-seq samples from untreated PDGFRA D842V PDX passage 4 were compared with RNA-seq from the original tumor and with untreated human KIT-driven and PDGFRA-driven GIST samples from our published RNA-seq data (39). The principal component analysis showed that the four PDX samples clustered with the patient specimen and with the PDGFRA-driven GISTs, whereas they were distinct from KIT-driven GISTs (Fig. 1C). The Euclidean distance between each sample also showed that the PDX samples were closer to the original tumor than to the other PDGFRA GIST specimens, whereas a heat map from selected important genes showed similarities between the PDX samples and the original tumor (Supplementary Fig. S1A and S1B). Altogether, these data indicate that the PDGFRA D842V PDX is representative of the original tumor.

PDGFRA-mutant GISTs are known to express activated PDGFRA and KIT (2). Therefore, we performed Western blots to assay for phosphorylation of these oncogenic receptor tyrosine kinase drivers. Several commercial antibodies were inefficient at detecting PDGFRA phosphorylation at sites Y754, Y762, and Y1018 (not shown). Therefore, we performed immunoprecipitation of total PDGFRA followed by staining for the phospho-tyrosine PY-20, which showed phosphorylated PDGFRA in all samples (Fig. 1D). We confirmed the results with the converse experiment by immunoprecipitating the phospho-tyrosines with anti-PY-20 followed by Western blot analysis using an anti-PDGFRA antibody (Fig. 1D). The dual phospho-PDGFRA Y849/phospho-PDGFRB Y857 antibody yielded similar results (Fig. 1E), so we used this antibody thereafter as an indicator of P-PDGFRA because there was negligible PDGFRB expression (Supplementary Fig. S1C). KIT was also expressed, as we had reported previously in PDGFRA-mutant GISTs (39) and activated as determined by the phosphorylation of KIT-Y703 (Fig. 1E). We also found that KIT ligand was expressed in our model, possibly accounting for some of the KIT activation (Fig. 1E).

GISTs with PDGFRA D842V mutations are known to be resistant to imatinib (2, 6). To determine the effect of imatinib in our PDX model, we treated NSG mice with established tumors for 2 weeks. There was a slight reduction in tumor growth (Supplementary Fig. S2A). By flow cytometry, there was a decrease in cell surface PDGFRA and KIT expression (Supplementary Fig. S2B). The decrease appeared to be either posttranscriptional or due to internalization because bulk tumor PDGFRA and KIT mRNA expression were unchanged (Supplementary Fig. S2B). There was no apparent difference by histologic analysis using H&E or trichrome staining and no apparent increase in apoptosis with imatinib using cleaved caspase-3 immunostaining (Supplementary Fig. S2C and S2D). However, imatinib reduced Ki67 staining by 26% (Supplementary Fig. S2C and S2D). Oncogenic signaling at 2 weeks was not affected by imatinib. Phosphorylation of PDGFRA and KIT and the downstream mediators of PI3K (i.e., AKT and S6), ERK, and STAT3 pathways were intact (Supplementary Fig. S2E). Altogether, these data demonstrate that imatinib slightly reduces tumor proliferation.

Crenolanib is a potent type I inhibitor of PDGFRA/PDGFRB and FLT3. In CHO and BaF3 biochemical assays, crenolanib inhibited PDGFRA D842V mutations (40). To determine the effect of crenolanib in our PDX model, mice with established tumors were treated for 6 hours to investigate oncogenic PDGFRA signaling. P-PDGFRA was not inhibited, and the downstream signaling was not affected, except for P-STAT3 (Supplementary Fig. S3A). Accordingly, the tumor volume of PDGFRA D842V PDX treated for 2 weeks was unchanged compared with vehicle-treated mice (Supplementary Fig. S3B). These data indicate that crenolanib had little effect on the PDGFRA D842V PDX.

Avapritinib was designed specifically for PDGFRA D842V-mutant GISTs (7). To determine the immediate effect of avapritinib on PDGFRA oncogenic signaling, mice with established tumors were treated for 6 hours with 10 or 30 mg/kg avapritinib. At 10 mg/kg, avapritinib inhibited P-PDGFRA and the downstream PDGFRA/KIT signaling effectors AKT, S6, STAT3, and ERK (Fig. 2A). Notably, P-KIT was not inhibited. Increasing the dose to 30 mg/kg efficiently inhibited P-PDGFRA and its downstream signaling effectors as well as P-KIT (Fig. 2A). Avapritinib at a dose of 30 mg/kg for 6 or 24 hours did not alter tumor proliferation by Ki67 staining (Fig. 2B).

To determine the effect of longer therapy, mice with established tumors were treated for 2 weeks. Avapritinib at either 10 or 30 mg/kg achieved about a 60% reduction in tumor size (Fig. 2C) and a 90% reduction in tumor weight (Fig. 2D). There was a substantial decrease in tumor cellularity by H&E staining, with a more pronounced effect at the higher dose. However, the two doses achieved a similar decrease in proliferation (Ki67) and increase in apoptosis (cleaved caspase-3; Fig. 2E). Flow cytometry gated on PDGFRA-positive tumor cells showed that surface PDGFRA and KIT expression were reduced by avapritinib 10 mg/kg (Fig. 2F). Analysis of quantitative mRNA expression of mice treated with 10 and 30 mg/kg for 2 weeks revealed a decrease in PDGFRA and KIT mRNA expression, suggesting that the decrease in cell surface expression resulted from decreased transcription (Supplementary Fig. S4A and S4B). Two weeks of avapritinib 10 or 30 mg/kg inhibited the phosphorylation of PDGFRA, KIT, AKT, and S6, but not STAT3 and ERK (Fig. 2G; Supplementary Fig. S4C). Overall, avapritinib therapy in the PDGFRA D842V PDX leads to a durable tumor response via inhibition of PDGFRA and downstream targets.

To identify additional targets to exploit during avapritinib therapy, we treated cohorts of mice with established PDX tumors with avapritinib, imatinib, or vehicle for 2 weeks and performed bulk tumor RNA-seq. Principal component analysis showed that vehicle- and imatinib-treated tumors clustered together, highlighting the limited efficacy of imatinib (Fig. 3A). In contrast, avapritinib-treated tumors had a distinct molecular signature. Gene set enrichment analysis (GSEA) showed that several pathways were enriched by avapritinib compared with vehicle or imatinib treatment. Among the top hits were the p53 pathway, hypoxia, NFkB signaling, and surprisingly myogenesis (Fig. 3B). In fact, there was upregulated expression of several heavy and light-chain myosin genes in avapritinib-treated tumors (Fig. 3C). In particular, MYLK was upregulated by avapritinib treatment (Fig. 3C and D). Interestingly, MYL9, one of the downstream targets of MYLK, was also upregulated by avapritinib treatment (Fig. 3C). MYLK upregulation was validated by qPCR (Fig. 3E) and Western blot analysis (Fig. 3F).

To confirm the relevance of MYLK in human GISTs, we interrogated our published RNA-seq data on 75 human GISTs (39). We discovered that MYLK was expressed in KIT-mutant GISTs and to even a greater magnitude in PDGFRA-mutant GISTs (Fig. 3G). These results demonstrated that MYLK is expressed in GISTs and that avapritinib upregulates MYLK expression.

Because avapritinib increased MYLK expression, we sought to determine whether MYLK inhibition would improve the antitumor effects of low-dose avapritinib. Cohorts of mice with established tumors were treated with avapritinib (3 mg/kg) alone or in combination with the MYLK inhibitor ML-7 (10 mg/kg; ref. 41). Tumors were examined at 6 hours to determine the immediate effects of both drugs on PDGFRA signaling before any long-term compensatory mechanism could occur. Although low-dose avapritinib reduced P-PDGFRA, P-AKT, P-S6, and P-STAT3, it did not inhibit P-KIT or P-ERK (Fig. 4A). As expected, ML-7 alone did not inhibit any of the signaling components. The combination of avapritinib and ML-7 was more effective in immediately inhibiting P-AKT (Fig. 4A).

Two weeks of treatment with avapritinib 3 mg/kg resulted in stable tumor volume (Fig. 4B). ML-7 at 10 mg/kg slowed tumor growth compared with control. However, the combination of both drugs had an additive effect, as tumor volume was less than avapritinib therapy alone (Fig. 4B). Although ML-7 alone did not affect the expression of its target MYLK, avapritinib or combination therapy increased MYLK RNA expression similarly (Fig. 4C). The increase in MYLK mRNA expression was associated with an increase in MYLK protein (Fig. 4D).

Low-dose avapritinib reduced cellularity and increased the stromal response by H&E, but to a lesser extent than higher doses (Figs. 2E and 4E). Although avapritinib at 10 and 30 mg/kg abrogated tumor proliferation, avapritinib at 3 mg/kg inhibited cell proliferation by 77% (Figs. 2E and 4E and F). In combination with ML-7, avapritinib reduced tumor proliferation by 86% (Fig. 4E and F). Staining for cleaved caspase-3 indicated that ML-7 increased apoptosis, whereas avapritinib or combination therapy did not, at least at 2 weeks (Fig. 4E). ML-7 did not affect PDGFRA phosphorylation or PDGFRA downstream signaling at 2 weeks (Fig. 4G). Avapritinib at 3 mg/kg reduced P-PDGFRA, P-AKT, P-S6, and P-STAT3, but ERK and KIT were not inhibited. KIT might have been upregulated as a compensatory mechanism for PDGFRA inhibition. However, the combination of drugs did inhibit P-KIT (Fig. 4G). Taken together, these results suggest that low-dose avapritinib and ML-7 have additive effects in inhibiting oncogenic PDGFRA signaling and in reducing tumor volume.

Although we have not been able to generate a cell line from the PDX, we were able to grow cells in short-term culture. We confirmed that the cells were GIST cells by checking DOG 1 expression by Western blot analysis and immunofluorescence (Supplementary Fig. S5). Inhibition of MYLK by ML-7 has been shown to induce apoptosis by affecting the cytoskeleton (41). Accordingly, freshly isolated PDGFRA D842V GIST cells treated with ML-7 had increased apoptosis in a dose-dependent manner (Fig. 5A and B). To determine the effect of ML-7 on the cytoskeleton, short-term cultured PDGFRA D842V GIST cells were treated with ML-7 alone or in combination with 100-nmol/L avapritinib and stained with phalloidin to reveal actin fibers. ML-7 at 4 and 20 μmol/L did not substantially modify actin fibers. However, at 30 or 40 μmol/L ML-7, actin filaments were disrupted, changing the cell shape from elongated to spherical (Fig. 5C). Although avapritinib alone did not affect actin fibers, in combination with 20-μmol/L ML-7, actin fibers were disrupted to the same extent as high-dose ML-7.

To determine if the additive effects of avapritinib and ML-7 were specific to PDGFRA D842V GISTs, we treated GIST T1 cells, which harbor a KIT exon 11 mutation. Avapritinib 100 nmol/L, which effectively killed GIST T1 cells (Supplementary Fig. S6), doubled MYLK expression after 24 hours and increased it more than 15 times after 48 hours (Fig. 5D). Although avapritinib reduced GIST T1 viability after 24 hours, there was an additive effect in combination with ML-7 20 μmol/L (Fig. 5E).

Overall, these results suggest that ML-7 induces apoptosis and that with avapritinib, it has additive effects in disrupting the actin fibers and reducing cell viability in PDGFRA D842V GIST cells and GIST T1 cells.

ML-7 has been used routinely in cell culture at doses of 5 to 40 μmol/L (41–43). To determine if the effects observed with ML-7 were specific to MYLK inhibition, MYLK was knocked down in GIST T1 cells with siRNA (Supplementary Fig. S7A and S7B). Although MYLK knockdown after 24 hours had a mild effect on cell survival, possibly due to residual MYLK protein, after 48 hours, there was a net decrease in cell viability confirming the role of MYLK in cell survival (Fig. 5F; Supplementary Fig. S7A). Although ML-7 20 μmol/L reduced cell viability after 24 hours, the knockdown of MYLK abolished ML-7 effects indicating that ML-7 acted through MYLK (Fig. 5G). These data suggest that ML-7 targeted MYLK at the doses used.

ML-7 is known to inhibit PKA and PKC, albeit with an affinity of 70- and 140-fold lower than MYLK, respectively (Selleckchem specifications and ref. 44). To rule out that inhibition of PKA was responsible for the observed ML-7 effects, PKA kinase activity was measured in GIST T1 and PDX tumor samples lysates. ML-7 treatment up to 40 μmol/L for 24 hours did not inhibit GIST T1 PKA activity (Supplementary Fig. S7C). Similarly, PKA activity was unchanged in tumors from PDXs treated with ML-7 10 mg/kg, avapritinib 3 mg/kg, or a combination of both drugs (Supplementary Fig. S7D). Phosphorylation of PKA and PKC is associated with their kinase activity. Because there are several subunits of PKA and PKC, we looked at the phosphorylation of the most expressed isoforms according to our RNA-seq data. Phosphorylation of PKA catalytic subunit alpha (threonine 197) and PKC theta (threonine 538) were unchanged in the tumor lysates from the PDX treated with ML-7, avapritinib, or a combination of both drugs (Supplementary Fig. S7E). However, the phosphorylation of myosin light-chains, canonical targets of MYLK, was reduced in lysates of tumors that had been treated with ML-7 (Supplementary Fig. S7F). Altogether, these results confirm that at the doses used, ML-7 does not inhibit PKA and PKC activity but inhibits MYLK in GIST T1 and in the PDGFRA D842V PDX.

To determine if avapritinib-induced MYLK expression is specific to avapritinib therapy, we first treated GIST T1 cells with imatinib in culture. Imatinib also induced MYLK expression in a dose-dependent manner (Fig. 6A) and had additive effects with ML-7 in reducing cell viability (Fig. 6B). To test the relevance of these findings in vivo, we treated GIST T1 xenografts for 2 weeks with avapritinib or imatinib. Avapritinib increased MYLK expression, whereas imatinib did not (Fig. 6C and D). The oncogenic signaling inhibited by imatinib and avapritinib at 2 weeks was similar with mostly inhibition of only P-S6 and P-ERK. P-KIT Y719 was only slightly reduced by both drugs (Fig. 6E). Although imatinib only achieved stable tumor growth, consistent with our previous results (35) and those of others (45), avapritinib significantly reduced tumor size (Fig. 6F). Accordingly, Ki67 and trichrome staining of tumor sections showed that avapritinib was more potent in inhibiting tumor proliferation, reducing tumor cellularity, and inducing a stromal response than imatinib (Fig. 6G). Taken together, these data indicate that MYLK expression is induced when there is tumor cell cytotoxicity, regardless of whether imatinib or avapritinib is used. In agreement with this, MYLK was not overexpressed in the PDGFRA D842V PDX treated with imatinib (Supplementary Fig. S8), which had little effect and did not induce cell death (Supplementary Fig. S2A and S2C).

We have created the first PDX model of PDGFRA D842V-mutant GIST. The PDX tumors retained similar morphology, proliferation, gene expression profile, PDGFRA and KIT staining, and PDGFRA D842V mutation and CDKN2A/CDKN2B deletion. Although PDGFRA D842V mutations are well known to be clinically resistant to imatinib (5, 46), whether imatinib provides even a slight benefit is unknown, as there was no comparison to placebo in clinical trials. In our PDX, we found that tumor size and proliferation were slightly reduced by imatinib, while oncogenic signaling was not affected. This result may be particular to mice or the subcutaneous location of the tumor, and it is conceivable that imatinib altered nontumor cells, such as endothelial cells. Although in biochemical assays crenolanib inhibited PDGFRA D842V mutations (40), it did not inhibit the PDGFRA D842V PDX. In a dose-escalation study of crenolanib of 16 evaluable patients with advanced GIST who had showed no improvement with at least one tyrosine kinase inhibitor, there were only two patients of 16 who had partial responses and three had stable disease (47). Thus, crenolanib is not very effective against PDGFRA D842V-mutated GIST.

Avapritinib is a type I kinase inhibitor that binds to the active form of KIT and PDGFRA and is particularly suited to inhibit activation loop mutations (7). We confirmed that avapritinib is very efficient in reducing tumor size of both the PDGFRA D842V PDX as well as GIST T1 xenografts (7). Although avapritinib at 10 mg/kg or 30 mg/kg reduced tumor size by 60% in our PDX, the stromal response was more pronounced at 30 mg/kg. Avapritinib is highly effective in PDGFRA-mutant GISTs, but its use has been somewhat limited by the dose-dependent cognitive impairment in 40% of patients (9). Lowering the dose may reduce the secondary effects and perhaps even avapritinib resistance. However, it has been reported that in xenografts, avapritinib at 3 mg/mL achieved only stable tumor volume in KIT exon 11/17 mutation (7). Here, we show that combination avapritinib/MYLK therapy may allow a lower dose of avapritinib, potentially reducing the cognitive side effects of patients treated with avapritinib. Although we used simultaneous avapritinib and ML-7 therapy in vivo, there may be benefit to sequential therapy. Avapritinib resistance has been reported due to secondary resistance mutations in PDGFRA-mutant GISTs (48). Here, we show that MYLK expression can contribute to tumor persistence after avapritinib response and that MYLK therapy may enable a lower dose of avapritinib to be used.

Surprisingly, avapritinib upregulated several genes involved in myogenesis. Among them was MYLK, a kinase ubiquitously expressed and a component of the actin cytoskeleton (26). Selective inhibition of MYLK with ML-7 has been used for many purposes, including to sensitize MYLK-expressing cancer cells to apoptosis and delay tumor growth (41, 44, 49). The targets of MYLK responsible for protecting against apoptosis or promoting tumor growth are not known. Although the canonical targets of MYLK kinase activity are the myosin light-chains, which contribute to cell shape and motility, CRISPR knockdown experiments of MYLK in a kidney cell line revealed that MYLK phosphorylates a broad network of other proteins, including transcription regulators and signaling molecules involved in the MAP kinase signaling pathway (34). In the PDGFRA D842V PDX, ML-7 alone slowed tumor growth. It also increased the efficacy of low-dose avapritinib. Induction of MYLK was not specific to avapritinib or PDGRA mutation because we found that imatinib induced MYLK expression in GIST T1 cells. In contrast, imatinib did not induce MYLK expression in GIST T1 xenografts, which did not respond to imatinib as we (35) and others (45) have reported. Thus, MYLK is increased when there is GIST cell death, and it may be a general compensatory pathway in GIST after KIT or PDGFRA inhibition.

We cannot explain the apparent discrepancy between the high dose of ML-7 required for antitumor effects in vitro versus in vivo. Many investigators have used ML-7 doses in vitro at micromolar doses as stated previously. Similarly, micromolar doses were needed in our in vitro experiments. By knocking down MYLK in GIST T1 cells, we showed that the effect of 20 μmol/L ML-7 required MYLK to cause cell death. In contrast, we found that ML-7 at a dose of 10 mg/kg was effective in the PDX when used in combination with avapritinib 3 mg/kg. The typical dose of avapritinib in mice is 10 mg/kg (7, 50) and we used up to 30 mg/kg. Meanwhile, we routinely use imatinib at a dose of 45 mg/kg in mice. One possibility is that GIST cells are less dependent on MYLK in vitro. As with any kinase inhibitor, however, it is possible that ML-7 is working through off-target effects, at least in vitro. It is known that ML-7 can inhibit PKA and PKC at high doses, although we could not find evidence of that in our experiments using 40 μmol/L in GIST T1 cells.

The canonical targets of MYLK are Myl9, Myl12a, and Myl12b, which contribute to cell shape and motility. However, MYLK has been shown recently to regulate a broad network of other proteins. Isobe and colleagues removed MYLK from a kidney cell line using CRISPR and then assayed changes in 1,743 phosphopeptides (34). There were 29 phosphopeptides that were substantially decreased including P-PKC theta, while 78 were increased. The phosphopeptide screen also revealed alterations in 14 phosphoproteins that function in transcriptional initiation and elongation. Remarkably, the overall pattern of changes overlapped with the authors’ prior report of deleting PKA in the same cell line (51). Nevertheless, in PDGFRA D842V PDX tumors, ML-7 did not change P-PKA, PKA kinase activity, or P-PKC theta. Notably, MYLK and PKA deletion each increased P-ERK1, which we also observed after ML-7 treatment of the PDGFRA D842V PDX. Thus, MYLK can exert effects well beyond myosin that include nuclear proteins and P-PKC theta, and MYLK shares some of the downstream effects of PKA.

Abnormal expression of MYLK has been observed in cell lines, including prostate and breast cancer cell lines where it plays a proliferative and an anti apoptotic role (44, 49, 52). Activation of the MAPK pathway regulates cell motility through phosphorylation and activation of MYLK (53). The MAPK pathway was active in our PDX model and ERK1/2 were still active after 2 weeks of avapritinib. It is possible that MAPK signaling contributes to MYLK activation in GISTs. In a three-dimensional COS-7 culture model, ERK activation of MYLK coupled with the CAS-CRK complex played a role in cell invasion and in protecting cells from apoptosis (54). It is possible that tumors respond to avapritinib by increasing MYLK expression as a survival mechanism to escape apoptosis. Consistent with this, we found that MYLK inhibition increased apoptosis in cells derived from the PDX. Others have shown that MYLK protected against apoptosis in normal and Ras-transformed MCF-10A cells, where expression of dominant-negative MYLK led to apoptosis. Anti apoptotic signals generated by MYLK are upstream of integrin-mediated focal adhesion kinase (FAK) signaling (41). We previously reported in a mouse model of KIT-mutated GIST that imatinib activated FAK through integrin signaling (13). It will be important to determine if FAK signaling is active in PDGFRA-mutated GISTs and if avapritinib-induced expression of MYLK influences integrin/FAK signaling.

By creating a PDX model of a PDGFRA D842V-mutated GIST, we were able to investigate the effect of avapritinib on oncogenic PDGFRA signaling. While avapritinib was very effective, the tumor responded by increasing MYLK expression to resist cell death. Combining inhibition of MYLK with avapritinib had additive anti tumor effects. Our results suggest that simultaneous inhibition of avapritinib with MYLK will allow for a lower dose of avapritinib, possibly reducing its side effects while maintaining efficacy. Although we are not aware of clinical trials of MYLK inhibitors, they have been considered for a variety of clinical diseases (55).

R.P. DeMatteo reports grants from NIH and grants from David Foundation during the conduct of the study. No disclosures were reported by the other authors.

F. Rossi: Conceptualization, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. M. Liu: Conceptualization, formal analysis, validation, investigation, methodology, writing–review and editing. A. Tieniber: Data curation, formal analysis, validation, investigation, methodology, writing–review and editing. M.S. Etherington: Conceptualization, formal analysis, validation, investigation, methodology, writing–review and editing. A. Hanna: Data curation, formal analysis, validation, investigation, methodology, writing–review and editing. G.A. Vitiello: Conceptualization, data curation, validation, investigation, methodology, writing–review and editing. N.J. Param: Formal analysis, validation, investigation, writing–review and editing. K. Do: Formal analysis, validation, investigation, methodology, writing–review and editing. L. Wang: Formal analysis, investigation, methodology, writing–review and editing. C.R. Antonescu: Formal analysis, validation, investigation, writing–review and editing. S. Zeng: Conceptualization, formal analysis, validation, investigation, methodology, writing–review and editing. J.Q. Zhang: Conceptualization, formal analysis, validation, investigation, writing–review and editing. R.P. DeMatteo: Conceptualization, resources, supervision, funding acquisition, validation, writing–original draft, project administration, writing–review and editing.

The investigators were supported by NIH grants R01 CA102613 and T32 CA251063, the David Foundation, Betsy Levine-Brown and Marc Brown, and the GIST Cancer Research Fund (RPD).

The authors are grateful to Elena Rossi for assistance with experiments and Andrea L. Stout for assistance with confocal microscopy.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.