Response Markers and the Molecular Mechanisms of Action of Gleevec in Gastrointestinal Stromal Tumors¹

GISTs³ are the most common mesenchymal malignancies of the gastrointestinal tract. GIST neoplastic cells appear to belong to the same lineage as the interstitial cells of Cajal, which are the pacemaker cells of the gut (1). GISTs share many immunohistochemical, morphological, and ultrastructural features with interstitial cells of Cajal, supporting this hypothesis (2). GISTs are characterized by the presence of constitutively activated KIT (CD117), the RTK encoded by the c-KIT proto-oncogene, also known as stem cell factor receptor (3, 4). c-KIT is a member of the RTK subclass III family and has structural homology to the receptors for FLT3, platelet-derived growth factor, and macrophage colony-stimulating factor (5). The proposed mechanism of constitutive KIT phosphorylation in the majority of GISTs is a gain of function mutation most commonly involving exon 11, which encodes a portion of the cytoplasmic juxtamembrane domain. Other mutations involve exon 9 (extracellular domain), exon 13 (first part of the split tyrosine kinase domain), or exon 17 (phosphotransferase domain; Ref. 6). In addition, a recent study has found activating mutations in the PDGFRA oncoprotein in GISTs lacking c-KIT mutations, indicating an alternative oncogenic mechanism in GISTs (7).

GISTs are usually diagnosed in middle-aged to older adults and occur anywhere in the gastrointestinal tract, most commonly in the stomach (8). Malignant GISTs often present with synchronous metastatic disease, and even those amenable to complete surgical resection recur locally, with a 5-year actuarial survival rate of approximately 50% (9). Patients with metastatic disease uniformly have a poor prognosis with a median survival of 6–8 months because GISTs have been historically resistant to conventional and investigational therapy.

Imatinib (imatinib mesylate, Gleevec; Novartis, Basel, Switzerland), formerly known as STI-571, is an oral 2-phenylaminopyrimidine derivative. In preclinical studies, imatinib has been shown to inhibit the activity of c-ABL, BCR-ABL, and PDGFRA/B (10, 11). Several Phase I/II clinical trials have demonstrated the efficacy of imatinib in the treatment of chronic myelogenous leukemia patients with the BCR/ABL translocation (12–14). This drug has also been shown to be a specific in vitro inhibitor of c-KIT phosphorylation in several tumor cell lines (10, 15). A recent report indicated that imatinib rapidly inhibits c-KIT phosphorylation and tumor cell proliferation while inducing apoptosis in a human GIST cell line (16). The c-KIT target has been exploited in two completed Phase I/II clinical trials of GISTs, with over 50% of patients with unresectable or metastatic GISTs demonstrating a partial response, and only 13% of patients manifesting disease progression while receiving imatinib (17, 18).

Despite its clinical success, there are limited preclinical and clinical data on the molecular targets of imatinib downstream from c-KIT. Therefore, our strategy was to characterize genetic markers of response to imatinib with a gene expression profiling approach using an in vitro model of GIST. Results of these studies were then used to further elucidate the molecular mechanisms of imatinib and to assess the panel of genetic markers in a limited number of sequential biopsy specimens from GIST patients before and after Gleevec treatment.

GIST882 is an immortal GIST cell line that possesses a homozygous exon 13 missense mutation (i.e., K642E) in c-KIT (16). GIST882 cells were grown as reported previously (16).

For growth analysis, GIST882 cells were seeded at 6.5 × 10⁵ cells/60-mm dish. Imatinib (provided by Novartis Oncology and dissolved in water) was added directly to the media to achieve a final concentration of 1 or 10 μm. Cells were refed with conditioned media containing imatinib or with conditioned media without drug every 12 h. Cells were harvested and stained for the cell number, cell viability, and induction of apoptosis using Guava ViaCount and Guava Nexin reagents (Guava Technology Inc., Burlingame, CA). The cells were counted using a Guava Personal Cytometer, and the data were analyzed using the Guava CytoSoft software package.

Cells were trypsinized, centrifuged, and fixed in 70% ethanol at 4°C. Cell pellets were resuspended in 50 μg/ml propidium iodide in PBS for 30 min at 4°C. The stained cells were analyzed by flow cytometry performed on a FACScan, and the data were analyzed with Cell Quest software (Becton Dickinson).

A total of 10,367 human cDNA fragments corresponding to known genes and ESTs from the integrated molecular analysis of genomes and their expression (I.M.A.G.E.) consortium library (Research Genetics) were used to produce high-density cDNA microarrays in a manner described previously (19).

Imatinib (10 μm)-treated and untreated GIST882 cells were harvested simultaneously at 60–70% confluence. Total RNA was isolated using the guanidinium/isothiocyanate/phenol/chloroform method as described previously (20). Total RNA was DNase treated using “DNA free” kit (Ambion, Austin, TX) according to the manufacturer’s specifications. Fifteen μg of this DNase-treated RNA were reverse transcribed, and amino-allyl-dUTP was incorporated in a reaction containing 500 ng of oligo(dT) primers; 1× first strand buffer (Invitrogen, Carlsbad, CA); 0.01 m DTT; 500 μm each of dATP, dCTP, dGTP, and dTTP/amino-allyl-dUTP (2:3 ratio); 40 units of rRNasin (Promega, Madison, WI); and 200 units of SuperScript II reverse transcriptase (Invitrogen). After brief denaturation and annealing of the primers at 70°C for 8 min, the reaction was incubated at 42°C for 2 h, followed by alkali hydrolysis of RNA and cDNA purification using Microcon-30 columns (Millipore, Bedford, MA) according to the manufacturer’s instructions. The cDNA was then labeled with either Cy3 or Cy5 dye by a coupling reaction using FluoroLink monofunctional dyes (Amersham Pharmacia Biotech, Piscataway, NJ) according to the manufacturer’s specifications. Probes were then purified using StrataPrep PCR Purification Kit (Stratagene, La Jolla, CA). Two of the samples (one labeled with Cy3 and one labeled with Cy5) were combined, denatured, and preannealed in the presence of 10 μg of Cot-1 DNA (Invitrogen) and 10 μg of poly(dA) DNA. Hybridization and washings were performed as described previously (19). The above was repeated for each sample, except that the dyes used to label the RNAs were reversed. Furthermore, for each time point, the hybridizations were repeated using an independent source of RNA.

Intensity extraction and spot quality characteristics were assessed as described previously (19). The data were then analyzed using GeneSight 3 software (BioDiscovery, Inc., Marina Del Rey, CA) and with in-house software. The data were (a) corrected for background by subtraction of the local group median, (b) normalized using a piecewise linear normalization with 5 bins and typically >1000 data points/bin, and (c) limited to have a minimum expression level equal to an estimate of the minimum background. The data were then converted to log₂ values, and the mean and SD were determined for each intensity ratio by combining “dye-flip” replicates. At least two dye-flip experiments were performed for each time point. The mean and the coefficient of variance were calculated for these values and used for statistical analysis and clustering. Differential expression of individual genes was determined by confidence analysis (21) and maximum likelihood analysis (22) to obtain a final list of candidates at a >95% confidence level. The data were also clustered into 11 clusters using hierarchical clustering (23) with a Euclidean distance metric.

cDNA fragments of interest were obtained from original stock plates used for microarray fabrication by the plasmid isolation from the bacteria. Clones were resequenced, and the correct annotation and homology were identified using the BLAST (National Center for Biotechnology Information) against the GenBank/EMBL database. DNA and amino acid sequence analyses were performed with the GCG/Wisconsin Package version 9.1 (Genetics Computer Group).

Fifteen μg of total RNA were used for standard Northern blot analysis, as described previously (20). Blots were hybridized with ³²P-labeled cDNA probes corresponding to each of the genes.

Two μg of DNase-treated RNA were reverse transcribed using Moloney murine leukemia virus reverse transcriptase (New England Biolabs, Beverly, MA) in a 20-μl reaction containing 1× reverse transcription buffer (NEB), 0.5 mm each dNTP, 4 μm oligo(dT)_16–18 primer, 10 units of RNase inhibitor (Promega), and 200 units of reverse transcriptase. Primers were preannealed for 10 min at 70°C, and the reaction was incubated for 1 h at 42°C, followed by enzyme inactivation for 10 min at 90°C.

DNA fragments of SPRY4A, MAFbx, FZD8, PDE2A, RTP801, FLJ20898, ARHGEF2, and β-actin were amplified by PCR using cDNAs obtained by reverse transcription of mRNA from GIST cells and tumor biopsies as a template and the following sets of primers: Sprouty 4A (SPRY4A), 5′-CCGTTCTGTGGAGAGTCGATTTAC-3′ and 5′-GTCCCTCAGTGGCTCTCGACT-3′; Frizzled 8 (FZD8), 5′-ACAGTGTTGATTGCTATTAGCATG-3′ and 5′-GTGAAATCTGTGTATCTGACTGC-3′; ARHGEF2, 5′-AAGGACGGAGAAAGGGAGAA-3′ and 5′-CAAGACAGCAGTGACCCTGA-3′; PDE2A, 5′-CCGCGATCTTTCTCGTAGTC-3′ and 5′-CCCACTTCTGCTACCTGCTC-3′; MAFbx, 5′-GTCCTGGGGTGAAAGTGAAA-3′ and 5′-TCACAGCTCACATCCCTGAG-3′; FLJ20898, 5′-CCCGAGTGACTCTGTTTTCC-3′ and 5′-ACACCCAGTTGGAGGTGAAG-3′; RTP801, 5′-AGACACGGCTTACCTGGATG-3′ and 5′-TTGATGACTCGGAAGCCAGT-3′; and β-actin, 5′-CTCACCATGGATGATGATATCGC-3′ and 5′- CATGATGGAGTTGAAGGTAGTTTCGT-3′. PCR was performed in a reaction volume of 30 μl containing cDNA from 1 μl of the reverse transcription reaction described above as a template, 10 mm Tris-HCl (pH 8.3), 50 mm KCl, 1.5 mm MgCl₂, 0.001% gelatin, 0.5 μm of both the forward and reverse primer, 60 μm of each deoxyribonucleotide, 5% DMSO, and 0.5 unit of platinum Taq DNA polymerase (Invitrogen). After an initial denaturation step at 95°C for 5 min, DNA was amplified through 19, 21, 22, 23, 24, 25, and 27 cycles consisting of 5-s denaturing at 95°C, 60-s annealing at 55°C, and 90-s extension at 72°C. The products were resolved on a 1.5% agarose gel and visualized by UV light after staining with ethidium bromide. Images of the gels were obtained using an Alpha Imager 2200 documentation and analysis system (Alpha Innotech, San Leandro, CA). Images were captured within linear dynamic range and controlled for white color saturation. DNA bands were quantified using the Alpha Imager version 5.5 software package and the Fuji Image Gauge version 3.11 software package (Fuji Photo Film Co., Ltd.).

The MEK1/2 inhibitor U0126 (Promega) was dissolved in DMSO. GIST cells were cultured to 60–70% confluence. MEK inhibitor was added directly to the media to achieve a final concentration of 1, 10, and 30 μm. An equal amount of DMSO was added to control untreated cells as a vehicle control. The PI3K inhibitor LY294002 (CalBiochem, La Jolla, CA) was dissolved in DMSO and used at a concentration of 30 μm. Cells were treated for 1, 3, 6, and 24 h before RNA isolation.

Anti-β-actin monoclonal antibodies (Sigma, St. Louis, MO) were used at a dilution of 1:5000 in 5% dried milk. Anti-phospho-c-KIT (Tyr⁷¹⁹) polyclonal antibodies (Cell Signaling, Beverly, MA) were used at a dilution of 1:500 in 5% BSA. Anti-c-KIT monoclonal antibodies (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) were used at a dilution of 1:100 in 5% dried milk. Anti-phospho-ERK1/2 monoclonal antibodies (Cell Signaling), anti-ERK1 and anti-ERK2 polyclonal antibodies (Santa Cruz Biotechnology, Inc.), anti-AKT, anti-AKT/phospho-Thr³⁰⁸, and anti-AKT/phospho-Ser⁴⁷³ polyclonal antibodies (Cell Signaling) were each used at a dilution of 1:1000 dilution in 5% BSA. Cell lysate preparation and Western blot analysis were performed as discribed in Ref. 24.

With an institutional review board-approved study for human subjects and tissue collection and with informed consent, tumor specimens were obtained from seven patients enrolled in the CSTI571-B2222 clinical trial sponsored by Novartis Oncology. Eligibility criteria included histological confirmation of GIST with documentation of c-KIT expression, as well as evidence of unresectable recurrent and/or metastatic disease. Patients were randomized to receive 400 or 600 mg p.o. once daily dose of imatinib (17). Ultrasound-guided 14G core biopsies of nonnecrotic tumor were obtained before initiation of therapy as well as while on drug or shortly after drug was discontinued. The biopsies were typically obtained from patients after 7–31 days of treatment with imatinib (Table 1). Standard histological evaluation was used to assure adequacy of nonnecrotic tumor tissue in these specimens. Specimens were flash frozen and kept in liquid N₂. Patients were monitored for response to imatinib mesylate at 4–12-week intervals by one of the authors (M. v. M.). Patients who were subsequently found to have disease recurrence underwent another biopsy or surgical resection if clinically appropriate.

Patients 1, 2, and 3 had GIST of either the stomach or the small bowel and had a partial response to imatinib treatment defined per standard criteria. Patient 7 had two measurable abdominal tumors, one of which responded to the therapy, whereas the other progressed. This patient, however, did not have a pretreatment biopsy. Patient 4 had a partial response to imatinib initially but developed disease progression 3 months after initiation of the treatment. The patient was taken off drug, and a core tumor biopsy was obtained. Patients 5 and 6, with large intestine and gastric GIST, respectively, had no response to imatinib treatment; one of them (patient 5) remained with stable disease, and the other one (patient 6) developed disease progression. Additional tumor was obtained from patient 6 during palliative surgical resection ∼31 days after drug therapy. One to 2 μg of total RNA is typically obtained from a single biopsy and was used to evaluate SPRY4 and MAFbx expression by RT-PCR analysis.

We first assessed the response of GIST882 cells to imatinib in vitro. We found that both 1 and 10 μm imatinib dramatically inhibited cell growth within 24–48 h after exposure (Fig. 1). However, GIST cells treated with imatinib show no significant increase (<3% of total cell number) in either early or late apoptosis during 96-h treatment period as determined by annexin-V and 7 amino-actinomycin D staining. The fraction of viable cells was 93 ± 5% for both treated and untreated cells during entire course of treatment. We next performed cell cycle analysis of drug-treated cells and found a 57% (15.5% versus 6.6% of total cell amount) decrease in the amount of cells in S phase with addition of 1 or 10 μm of imatinib (Fig. 1). The variation of the total number of cells in S phase was <5% for independently repeated experiments.

To identify potential imatinib-specific genetic targets, we treated GIST882 with 10 μm imatinib for 6, 12, 24, and 48 h and isolated RNA. These RNA samples were labeled with Cy3 or Cy5 dyes and hybridized to cDNA microarrays containing known genes and ESTs. The significance and confidence analysis produced a list of 148 genes or ESTS that were differentially regulated in at least one time point at 95% confidence. The complete list of these genes can be found at online.⁴ In the hierarchical cluster analysis, one cluster showed significant down-regulation relative to the control at all time points in the cell culture experiment and also contained a number of genes identified as confidently differentially regulated (Fig. 2). This cluster contained 62 of the 10,368 genes in the experiment as a whole, with 34 genes in the cluster being down-regulated at 95% confidence in at least one condition (i.e., the intersection of the results of the confidence analysis with the down-regulated cluster). The analysis revealed that this cluster is quite small and very close to the root of the tree, indicating that these genes cluster well away from the other genes in the experiment. Because our interest was to identify potential surrogate genetic markers of response to imatinib, we further refined this list to only those genes that showed a durable response for all time points (at least 2.4-fold). The reason for this approach is that if genes fluctuate up and down after treatment, then they would be unlikely to be reliable markers when evaluating tumor samples from patients on clinical trials. Based on this criterion, we identified a total of seven known genes that displayed this pattern of expression after treatment. One gene was strongly up-regulated, and six genes were strongly down-regulated at all time points (Table 2). To confirm the microarray results, we first revalidated the annotations of our gene list (as describe in “Materials and Methods”) and verified that the seven genes were SPRY4A, ARHGEF2, RTP801, FLJ20898, FZD8, PDE2A, and MAFbx.

To validate the array data, we isolated a panel of cDNA probes and designed a set of PCR primers for each of the named genes, and performed Northern and RT-PCR analyses, respectively, using an aliquot of RNA reserved from the array experiments (Fig. 3; data not shown). We controlled for saturation of the PCR reaction by sampling each reaction together with a β-actin control reaction at different numbers of PCR cycles. All seven genes were differently expressed, as expected from our microarray screen (Fig. 3). SPRY4A, ARHGEF2, RTP801, FLJ20898, and FZD8 mRNA levels were diminished significantly within 6 h after addition of imatinib, and the level of PDE2A mRNA decreased measurably 12 h after drug treatment. In contrast, MAFbx levels were dramatically induced within 6 h of imatinib exposure and remained elevated throughout the treatment (Fig. 3).

We next examined GIST882 cells for constitutively activated c-KIT, as well as downstream mediators ERK1/2 and AKT, using anti-phosphorylated antibodies. c-KIT, ERK1/2, and AKT were found to be constitutively activated in exponentially growing GIST882 cells (Fig. 4A). The inhibitory efficacy of the drug on the activity of c-KIT, ERK1/2, and AKT was then evaluated using either 1 or 10 μm imatinib for 0.5–6 h. Both drug concentrations resulted in loss of phosphorylated c-KIT within 30 min and were specific only for the phosphorylated protein because the total levels of KIT protein were unaffected (Fig. 4A). Imatinib also potently inhibited the constitutive activation of AKT, as assessed using antibodies specific for phosphorylated threonine 308 or serine 473. There was a similar finding for phosphorylated ERK1/2. As with c-KIT, the total levels of AKT and ERK1/2 were unaffected by imatinib treatment (Fig. 4A). Furthermore, decreased phosphorylation of AKT and ERK1/2 was rapid and paralleled inhibition of c-KIT (Fig. 4A). We next evaluated the effects of imatinib on SPRY4A and MAFbx expression in conjunction with inhibition of c-KIT. As shown in Fig. 4B, the expression of both genes was altered within 3 h of inhibition of c-KIT. SPRY4A levels were noticeably decreased by 3 h and virtually undetectable by 6 h, and MAFbx mRNA levels were dramatically increased by 3 h and maintained throughout the course of treatment. Overall, we were able to verify that imatinib treatment resulted in decreased autophosphorylation of a K642E mutant of c-KIT by inhibiting c-KIT kinase activity rather than by down-regulating expression of the c-KIT protein in GIST882 cells. Furthermore, this inhibition leads to down-regulation of the activation of ERK1/2 and AKT.

Previous studies have shown that the ERK pathway positively regulates the expression of the Sprouty genes in mouse cells and that, in a limited number of tumor cell lines that exhibit constitutive activation of ERKs, SPRY1 and/or SPRY2 mRNA is elevated (25). To determine whether ERK1/2 pathways also regulated SPRY4A in GIST cells, we treated the GIST882 line with U0126 (30 μm), a MEK1 inhibitor, for 30 min, 1, 3, and 6 h. Complete suppression of activated ERK1/2 was seen by 30 min (Fig. 5A). Interestingly, U0126 inhibits phosphorylation of AKT shortly after treatment, but this effect was transient because phospho-AKT returned to initial levels within 3 h of treatment (Fig. 5A) and did not affect the levels of total AKT. We next sought to determine whether the panel of imatinib-responsive genes was dependent on ERK1/2 activation by MEK1. Expression patterns of SPRY4A, MAFbx, and ARHGEF2 in U0126 cells were similar to those after imatinib treatment (Fig. 5, B and C). PDE2A and RTP801 mRNA levels were decreased transiently, and the levels of FLJ20898 and FZD8 did not appear to change noticeably (Fig. 5B). The level of expression after addition of U0126 was comparable with that seen with imatinib, except that the duration of suppression with a single dose of imatinib was more prolonged (data not shown). We also evaluated the effect of LY294002, a PI3K inhibitor (26), and we observed little or no effect on expression levels of any of the imatinib-responsive genes in our restricted panel (Fig. 5D; data not shown). These results suggest that several of these genes are likely to be dependent on activation of the MAPK signaling pathway by c-KIT, whereas others may not be directly downstream of either MAPK or AKT signaling pathways.

Ultrasound-guided needle core biopsies obtained from GIST patients both before and after initiation of imatinib therapy were evaluated for expression of a down-regulated transcript (SPRY4A) and an up-regulated transcript (MAFbx) by RT-PCR analysis. Each biopsy was evaluated by a pathologist for evidence of viable GIST cells with c-KIT expression by routine histology and immunohistochemistry, respectively, before molecular analysis. Fig. 6 depicts the baseline and day 15 biopsy for patient 1 (a responder). There is persistence of malignant tumor cells with KIT expression in the biopsy taken while the patient was on imatinib. As evident in Fig. 7A, the expression levels of SPRY4 were dramatically decreased after imatinib therapy in tumors from patients (patients 1, 2, 3, and 7) who showed favorable clinical responses to the drug. The two nonresponsive patients (patients 5 and 6) expressed high levels of SPRY4A in both the pre- and posttreatment biopsies. In the patient (patient 4) who initially responded to the drug treatment but subsequently relapsed, the SPRY4A levels in the tumor decreased dramatically in the biopsy taken during clinical response but returned to the pretreatment levels during disease progression (Fig. 7A). We also assessed the expression level of MAFbx in clinical samples (patients 1, 4, 6, and 7). In patients 1 and 7, both of whom showed favorable clinical responses, we detected high levels MAFbx transcripts in the treated tumor. In the tumor sample from patient 4, the level of MAFbx increased during the response but returned to pretreatment level during the disease progression. In patient 6, who never responded to the drug, the level of MAFbx was not changed (i.e., not detected) in the paired samples (Fig. 7B).

For patient 6, a larger biopsy sample (31 days after imatinib treatment) was available, which allowed us to evaluate the c-KIT, AKT, and ERK1/2 protein levels after imatinib exposure in vivo. A Western blot of this nonresponding GIST indicated that c-KIT was constitutively activated even in the presence of imatinib therapy (Fig. 8A). For comparison, we included the GIST882 cell line and a sample of a myxoid liposarcoma, a sarcoma histotype that is generally not responsive to imatinib. As expected, the liposarcoma failed to express detectable levels of c-KIT (both phosphorylated and unphosphorylated; Fig. 8A). We also evaluated AKT and ERK1/2 in these samples. As one would predict, phosphorylated forms of AKT and ERK1/2 were present in the nonresponding GIST (Fig. 8, B and C). Interestingly, in the liposarcoma specimen, both ERK1/2 and AKT were constitutively activated (Fig. 8B), and SPRY4A was expressed at a level comparable with a typical GIST specimen (data not shown). These results confirm that c-KIT was not down-regulated in response to imatinib in this patient and that the expression levels of both SPRY4A and MAFbx were predictive of this lack of response.

To assess further why this patient failed imatinib therapy, we evaluated c-KIT for a somatic mutation. RNA was isolated from this tumor biopsy and reversed transcribed, and the cDNA was amplified by PCR. The entire transcript was sequenced, and a mutation was found in exon 9 of the c-KIT cDNA. The mutation identified was a 6-base insertion starting at nucleotide 1530 (1530ins6) that would result in the addition of two extra amino acids (Fig. 8D). Multiple alleles were sequenced, and it was determined that the tumor was heterozygous for the mutation (Fig. 8D; data not shown). These results suggest that this mutation in c-KIT may be associated with the inability of imatinib to down-regulate c-KIT activity.

The CSTI571-B2222 clinical trial was one of the first to use imatinib for the treatment of a solid tumor (17). Although this Phase II trial reported a high number of durable partial responses, 13% of patients never responded, and some responding patients have subsequently progressed. Questions remain as to both the basis of this response and the basis of primary and acquired resistance. We hypothesized that by evaluating gene expression profiles from GIST cell lines and then using these data to evaluate specimens from GIST patients taken before and after imatinib therapy, we would identify novel genetic biomarkers of this therapy and subsequently define additional downstream mediators of response. Using cDNA microarray approaches, we identified a panel of genes that were responsive to imatinib inhibition of c-KIT and its subsequent downstream signaling pathways. Our results showed that ∼1.4% (148 of 10,367) of the gene sequences evaluated were significantly (P < 0.05) differentially expressed in a GIST cell line after KIT kinase inhibition. Hierarchical clustering identified a single cluster of 62 genes showing down-regulation at all time points. Because hierarchical clustering cannot assess significance, we further reduced the focus from these 62 genes to only those 34 that showed statistically significant down-regulation. We narrowed this list of candidate genetic response markers to the genes with the strongest fold changes across all treatment time points, seven of which were well annotated and showed significant changes in expression at each time point. We first observed that GIST cells treated with imatinib undergo rapid growth arrest, without significant induction of apoptosis. A previous study of GIST882 cells found that a portion of the cells undergo programmed cell death in response to imatinib, albeit at later time points (16). Other studies have shown that imatinib can lead to inhibition of cell growth in the absence of measurable induction of apoptosis (27). Flow cytometry further confirmed that the imatinib-treated cells accumulate in G_0–1 and G₂, almost exhibiting a static growth state. Furthermore, there was no detectable sub-G₀ fraction, which would have been indicative of apoptotic cells. This parallels the clinical observation of the absence of initial rapid decrease in tumor volume on computerized tomography scan, whereas positron emission tomography scan indicates a rapid abrogation of metabolic activity suggestive of tumor dormancy (28). The genocentric approach used in this study identified potential gene products and signaling pathways that may be important in the therapeutic effects of imatinib. We found that the mRNA levels of Sprouty4A (SPRY4A) were the most significantly down-regulated and that the mRNA levels of MAFbx were dramatically up-regulated for the duration of drug treatment. We also observed and validated down-regulation of several other known genes (i.e., FZD8, PDE2A, RTP801, FLJ20898, and ARHGEF2) in the cell culture model. We then attempted to validate the gene transcript most significantly down-regulated (SPRY4A) and the only gene transcript up-regulated (MAFbx) in vitro by examining sequential GIST biopsy specimens. SPRY4A expression and MAFbx expression were found to be highly reliable predictors of immediate response to drug, even in small core tumor biopsies (Fig. 7). The rationale behind evaluating these two genes is that MAFbx has been reported to be expressed exclusively in heart and skeletal muscle (29). We have found for the first time that this gene is also expressed in smooth muscle, albeit at much lower levels than seen after imatinib treatment (data not shown). Therefore, by combining the expression of both an up-regulated gene and a down-regulated gene, we believe that this lends further support that tumor cells are present in the biopsies.

Our study also uncovered an activating mutation in c-KIT that was associated with a failure to respond to imatinib. This in-frame mutation (1530ins6) in exon 9 of c-KIT was the one of the first reported outside the exon 11 juxtamembrane domain (30). Henrich et al. (31) have reported that GISTs with exon 9 mutations have a lower response rate than GISTs with exon 11 mutations. It is possible that this exon 9 mutation activates the receptor via ligand-independent oligomerization. It was shown that imatinib binds to the Thr⁶⁷⁰ of the ATP-binding site of the kinase (corresponds to exon 14). Importantly, imatinib can only bind to an inactive conformation of the kinase, in which the NH₂-terminal part of the activation loop is folded into the ATP-binding site (32). Therefore, it is possible that the 1530ins6 mutation could lead to activation of this site, thereby preventing imatinib from binding and inhibiting the kinase activity. However, additional studies to evaluate a range of GIST-associated c-KIT mutations and their response to imatinib are needed.

The panel of genes uncovered in our study is also of potential interest from a functional aspect to further elucidate the molecular mechanisms of imatinib. There are little data regarding the therapeutic effect of imatinib on GISTs, other than that the drug inhibits activation of c-KIT and several downstream mediators. Our studies confirmed that both AKT and ERK1/2 signaling pathways are rapidly inhibited after exposure to imatinib. Interestingly, our studies with inhibitors of these pathways, LY294002, a PI3K inhibitor, and U0126, a MEK1 inhibitor, suggest that other signaling pathways may also be affected by imatinib treatment, and additional studies should elucidate this issue. Specifically, we found that mRNA levels of SPRY4A and ARHGEF2 are down-regulated, whereas MAFbx levels are up-regulated after treatment with U0126 or imatinib. Interestingly, the kinetics and pattern by which the two drugs inhibit these markers are quite distinct. Even though we have demonstrated that all seven markers respond to imatinib, albeit at different rates, inhibition of two major pathways downstream of c-KIT do not appear to account for all of the differences we observe in expression patterns (compare Fig. 5, B, C, and D). These results suggest that several of the markers identified are not directly downstream of any of these signaling pathways and warrant further investigation regarding the action of imatinib.

We have also used hierarchical clustering to identify a subset of genes that includes SPRY4A, which showed the highest degree of down-regulation. Expression levels of the transcripts for SPRY4A and ARHGEF2, for example, are consistently down-regulated by imatinib treatment in the GIST cell line. Interestingly the proteins encoded by these two genes appeared to be involved in the control of feedback loops for RAS signaling associated with coadapter protein binding and RAS activation (25). The coordinated down-regulation of these two genes and presumably their proteins suggests a potential mechanism for immediate downstream imatinib-induced inhibition of KIT signaling.

The down-regulation of the PDE2 transcript may directly relate to its role in the transduction of cell survival signals. c-GMP-stimulated PDE2 was shown to be up-regulated in certain cancers and functions to control the cAMP level in the cell (33). It was also shown that some RTKs may transduct signal not only through adapter proteins but also through secondary messengers. In this case, PDEs play a crucial role in transmitting the signal from the receptor and the response to various stimuli. It was recently shown that PDE3 plays a very important role in insulin-like growth factor I receptor signaling in pancreatic β cells (34), and PDE2A was shown to be down-regulated in exisulind-induced apoptosis (35). It was also shown that insulin-like growth factor I-dependent regulation of PDE3B might be linked to cell survival through PI3K and not p42/p44 MAPK (36). Recent studies demonstrated that GISTs overexpress genes encoding G-protein-coupled receptor 20 and protein kinase C θ (37). Both of these genes were not only highly expressed in GIST but also clustered very tight with it. PDEs were shown to be activated by protein kinase C and enhance the degradation of cyclic monophosphates (38), thus playing a critical role in signal transduction. This evidence further supports the suggestion of possible horizontal interactions between RTK and G-protein-coupled receptor signaling.

The discovery of the up-regulation of the muscle atrophy F-box (MAFbx) is of particular interest. MAFbx, also known as Atrogin-1, encodes a ubiquitin ligase, a protein that binds and mediates ubiquitination of specific substrates (29, 39). It has been shown that overexpression of MAFbx in the myotubules of mice results in muscular atrophy when compared with mice deficient in this gene (29). The presence of the up-regulated MAFbx transcript in imatinib-sensitive GIST suggests a potentially important role in promoting tumor necrosis, which is of interest, given that the GIST progenitor cells have myogenic differentiation properties. Furthermore, we found that treatment of GIST882 cells with LY294002 resulted in a modest but detectable increase in MAFbx mRNA levels (Fig. 5D). This is interesting, given that several proteins, including AKT in the PI3K/AKT cascade, have been shown to have a positive effect on muscle hypertrophy (40). Therefore, down-regulation of AKT, as observed in GISTs after treatment with imatinib, may have the opposite effect. Furthermore, it is intriguing to speculate that up-regulation of proteins, such as MAFbx, may also contribute to the generalized muscle cramping that is a common side effect in patients taking imatinib.

Clearly further work will need to be done in exploring the functional consequences of these genomic GIST profiles comparing molecular events before and after imatinib administration. This initial study with unique patient samples has yielded important observations regarding molecular profiling of GIST patients on imatinib therapy. The correlation of SPRY4A and MAFbx expression in relation to clinical response is quite uniform and compelling, particularly in patient 4, with initial response (SPRY4A down-regulated and MAFbx up-regulated) followed by disease progression (SPRY4A up-regulated and MAFbx down-regulated). SPRY4A expression in these patients is interesting, particularly with reference to a recent publication supportive of additional evidence of SPRY4A involvement in GISTs (41). This report indicates that SPRY1, SPRY4, and c-KIT are among the genes that distinguish GISTs from other sarcomas. Therapeutic implications are speculative; however, proteins such as those encoded by SPRY4A and RTP801, whose expression is elevated in GISTs, provide potential targets for drug discovery purposes, especially in tumors refractory to imatinib.

In summary, we have used a genomic-based approach to identify several surrogate response markers of imatinib therapy in a limited number of GIST cell lines and in GIST clinical specimens. Functional studies, which are under way, will hopefully resolve which of them are important in the therapeutic effects of this small molecule targeted agent. This study provides a platform for continuing investigations into the molecular basis for imatinib response in GISTs, which may result in the development of new therapeutic targets and reliable markers of clinical response.

We acknowledge Drs. Alfred Knudson and Phillip A. Godwin for insightful comments regarding the manuscript, Sharon Howard and the Tissue Culture Facility for technical assistance, and the DNA microarray facility, the DNA sequencing facility, the bioinformatics facility, and the Biosample Repository⁵ at Fox Chase Cancer Center for services and specimens. We also thank Dr. Rosaleen Parsons and Dr. Barton Milestone for expert assistance obtaining the core biopsies, Dr. Harry Cooper for pathologic assessment of the tumor biopsies, and Monica Davey for excellent care of the study patients and data management. We acknowledge the support of Dr. Sasa Dimitrijevic (Novartis, Basel, Switzerland).