ORIGINAL ARTICLE

KIT oncoprotein interactions in gastrointestinal stromal tumors:

therapeutic relevance

M-J Zhu1,2, W-B Ou1,2, CDM Fletcher1,2, PS Cohen3, GD Demetri2,4 and JA Fletcher1,2

1Department of Pathology, Brigham & Women’s Hospital, Boston, MA, USA; 2Departments of Pathology, Pediatrics and Medicine,Harvard Medical School, Boston, MA, USA; 3Novartis Oncology, Florham Park, NJ, USA and 4Ludwig Center at Dana-FarberCancer Institute, Boston, MA, USA

Most gastrointestinal stromal tumors (GISTs) expressoncogenic and constitutively active forms of the KIT orplatelet-derived growth factor receptor alpha (PDGFRA)receptor tyrosine kinase proteins, and these kinaseoncoproteins serve as targets for effective therapies. Giventhat mutant KIT oncoproteins serve crucial transformingroles in GISTs, we evaluated interactions with the KIToncoproteins and determined signaling pathways thatare dependent on KIT oncogenic activation in GISTs.Tyrosine-phosphorylated KIT oncoproteins interacted withPDGFRA, PDGFRB, phosphatidylinositol 3-kinase (PI3-K)and PKCh in GIST cells, and these interactions wereabolished by KIT inhibition with imatinib or PKC412 orKIT RNAi. Notably, tyrosine-phosphorylated PDGFRAwas prominent in frozen GIST tumors expressing KIToncoproteins, suggesting that KIT-mediated PDGFRAphosphorylation is an efficient and biologically conse-quential mechanism in GISTs. Activated signaling inter-mediates were identified by immunoaffinity purification oftyrosine-phosphorylated proteins in GIST cells before andafter treatment with KIT inhibitors, and these analysesshow that GRB2, SHC, CBL and MAPK activation arelargely KIT dependent in GISTs, whereas PI3-K, STAT1and STAT3 activation are partially KIT dependent. Inaddition, we found that phosphorylation of severaltyrosine kinase proteins – including JAK1 and EPHA4– did not depend on KIT activation. Likewise, paxillinactivation was independent of the KIT oncogenic signal.These studies identify signaling pathways that can provideboth KIT-dependent and KIT-independent therapeuticsynergies in GIST, and thereby highlight clinical strate-gies that might consolidate GIST therapeutic response toKIT/PDGFRA inhibition.Oncogene (2007) 26, 6386–6395; doi:10.1038/sj.onc.1210464;published online 23 April 2007

Keywords: KIT; PDGFRA; kinase; interaction; signaltransduction

Introduction

Gastrointestinal stromal tumors (GISTs) are the mostcommon mesenchymal tumors of the digestive tract.GIST cells share morphological and immunophenotypicfeatures – including expression of the KIT receptortyrosine kinase (RTK) – with interstitial cells of Cajal(ICC), which are the pacemaker cells responsible forcoordinating gastrointestinal peristaltic activity (Hein-rich et al., 2002). Most GISTs (85%) contain gain-of-function KIT oncogenic mutations, which are associatedwith constitutive KIT tyrosine phosphorylation, andactivation of cell proliferation and survival signalingpathways (Duensing et al., 2004b). Alternately, a smallsubset of GISTs has oncogenic mutations in the geneencoding platelet-derived growth factor receptor alpha(PDGFRA) that confer constitutive activation (Hein-rich et al., 2003; Hirota et al., 2003). Although KIT andPDGFRA oncogenic mutations are mutually exclusivein GISTs, approximately 30% of KIT-mutant GISTsexpress PDGFRA strongly (Heinrich et al., 2003).Notably, the oncogenic KIT and PDGFRA mutationscan be targeted effectively with imatinib mesylate(Gleevec) in GIST cell culture models and in more than80% of GIST patients (Tuveson et al., 2001; Demetriet al., 2002; Verweij et al., 2004).KIT is a member of the PDGFR subfamily of RTK

proteins, which feature extracellular Ig domains and asplit kinase domain (Heinrich et al., 2002). KIT under-goes homodimerization and autophosphorylation whenbound by the ligand, stem cell factor, or when activatedby oncogenic mutations (Hirota et al., 1998; Rubinet al., 2001). In addition, KIT can heterodimerize withthe PDGFR family members FLT3 (Otto et al., 2001)and PDGFRA (Hirota et al., 2003), and presumablywith other RTKs. Therefore, the clinical success of KITinhibitors in patients with GIST might result frominactivation of KIT downstream signaling pathways andKIT heterodimerization partners.Although most GIST patients have dramatic thera-

peutic responses to imatinib, some patients are resistantto imatinib, and very few patients exhibit a completeresponse to this drug (Demetri et al., 2002). The aim ofthe studies reported herein was to evaluate cell signalingpathways transmitting proliferation and survival signalsfrom KIT, and to determine which signaling intermedi-

Received 1 June 2006; revised 23 October 2006; accepted 9 March 2007;published online 23 April 2007

Correspondence: Dr M-J Zhu or Dr JA Fletcher, Department ofPathology, Brigham andWomen’s Hospital, 75 Francis Street, Boston,MA 02115, USA.E-mails: mzhu2@partners.org or jetcher@partners.org

Oncogene (2007) 26, 6386–6395& 2007 Nature Publishing Group All rights reserved 0950-9232/07 $30.00

www.nature.com/onc

ates are dependent on KIT activation in GIST. Thesestudies, performed in human GIST cell lines before andafter treatment with KIT inhibitors, interrogatedsignaling proteins that play key roles in tyrosinekinase-mediated cell proliferation, survival and adhe-sion. Mechanisms of oncogenic signaling pathwayactivation were evaluated further by determining KITinteractions with various kinase and adaptor proteins.These studies show that certain signaling intermediatesare largely KIT dependent in GISTs, whereas others arenot. These insights will be helpful in designing rationalcombination-targeted therapy strategies in GIST.

Results

KIT interactions in primary GISTsKinase protein interactions with oncogenic KIT wereevaluated in two untreated GIST specimens with hetero-zygous KIT exon 11 (GIST3) or exon 9 (GIST4) mutations(Figure 1). GIST3 and GIST4 lacked PDGFRA genemutations (data not shown), but expressed phosphorylatedPDGFRA in addition to constitutively phosphorylatedKIT (Figure 1). Phosphatidylinositol 3-kinase (PI3-K)p85 immunoprecipitates revealed a dominant 170/160 kD tyrosine-phosphorylated doublet in both GISTs(Figure 1), consistent in size with the transmembraneforms of PDGFRA (170 kD) and KIT (160 kD). PI3-Kp85 interactions with KIT and PDGFRA wereconfirmed by immunostaining (Figure 1). These findingssuggest that KIT and PDGFRA were the predominantactivated RTKs in GIST3 and GIST4. In addition,interaction between PDGFRA and the mature form ofKIT was evident in the PDGFRA immunoprecipita-tions (Figure 1). Because KIT and PDGFRA are

structurally similar RTKs, the possibility of a spuriouscoimmunoprecipitation, resulting from direct bindingof KIT by the PDGFRA antiserum, was evaluated. Thispossibility was excluded using a KIT-mutant GISTthat expressed KIT strongly but lacked PDGFRAexpression: KIT was not detected in the PDGFRAimmunoprecipitations from this GIST (data not shown).ABL was not demonstrably phosphorylated, and wasnot complexed with KIT, in GIST3 or GIST4 (Figure 1).In sum, these findings demonstrate interactions betweenKIT, PDGFRA and PI3-K p85, and suggest thatABL – despite the precedent for interaction with KITin human leukemias (Hallek et al., 1996) – neithercomplexes substantially with KIT nor is demonstrablyactivated in GIST.

Kinase expression varies in GIST cell linesVariability of key kinase expression was evaluated byimmunoblotting in the GIST882, GIST48 and GIST430cell lines. KIT, PI3-K p85 and PKCy were expressed atsimilar levels in each of these lines, whereas PDGFRA,PDGFRB and ABL expression varied considerably,being strongest in GIST882, GIST48 and GIST430,respectively (Figure 2).

Kinase interactions in imatinib-sensitive and imatinib-resistant GIST cell linesKinase activation and interactions were evaluated in theGIST882 and GIST48 cell lines. Kinase interactions inthese cell lines were evaluated before and afterpharmacologic inhibition of KIT/PDGFR by imatinibin GIST882 (Figure 3) and PKC412 in the imatinib-resistant GIST48 (Figure 4). Wild-type PDGFRA andPDGFRB were phosphorylated and coprecipitated withoncogenic KIT (Figures 3a and 4a), suggesting that

Figure 1 KIT-dependent kinase interactions in primary, frozen, GISTs. Internal standards are frozen GISTs, GIST1 (KIT mutant)and GIST2 (PDGFRA mutant), which express high levels of KIT and PDGFRA, respectively. GIST3 (a) and GIST4 (b) each containKIT mutations: the first three lanes in each panel are total cell lysates, whereas the remaining four lanes are immunoprecipitates of KIT(Santa Cruz, sc-168), PDGFRA (Santa Cruz, sc-338), ABL (Santa Cruz, sc-131) and PI3-K p85 (Upstate, Lake Placid, NY, USA, 06-497), evaluating kinase activation and interactions. The PY99 immunostain shows KIT and PDGFRA activation, and suggests thatactivated KIT and PDGFRA are complexed with PI3-K p85. KIT complexing with PDGFRA and PI3-K p85 is confirmed by the KITimmunostains of PDGFRA and PI3-K p85 immunoprecipitates, respectively. PI3-K p85 complexing with KIT and PDGFRA, but notABL, is also shown by the PI3-K immunostains of KIT, PDGFRA and ABL immunoprecipitates, respectively.

KIT oncoprotein interactions in GISTM-J Zhu et al

6387

Oncogene

PDGFRs heterodimerize with, and are cross-phos-phorylated by, the constitutively activated KIT mutantsin GIST882 and GIST48. KIT inactivation inhibited thecomplexing between KIT and PDGFR (Figures 3b and4b). PI3-K p85 and p110a, and PKCy also complexedwith oncogenic KIT, as shown best in the KIT andPKCy immunoprecipitates, respectively, and as forPDGFR, these interactions were partially inhibited byimatinib (Figures 3 and 4). Likewise, PI3-K p85interactions with PKCy were inhibited by imatinib(Figures 3b and 4b). The interaction between KIT andPKCy appeared to be relatively weak, being mostconvincing in the GIST882 cells. Interestingly, ABLphosphorylation and interactions between KIT andABL were not detectable in either GIST882 or GIST48(Figures 3 and 4).

KIT-dependent signaling in GIST cell linesExpression of activated signaling intermediates wasevaluated by immunoblotting in phosphotyrosineimmunopurifications from confluent GIST882 andGIST48 cell lines. The rationale for evaluating signalingintermediates in confluent cells was to highlight activa-tion events that might be strictly dependent on the KIToncogenic signal, irrespective of whether the cells wereactively proliferating. KIT dependence was determinedby comparing expression of each protein before andafter KIT/PDGFR inhibition by imatinib (in GIST882cells) and PKC412 (in GIST48 cells). Phosphorylation

patterns differed between GIST882 and GIST48, as canbe seen by comparing protein expression ratios in theinput (total protein) and phosphotyrosine eluate (acti-vated protein) lanes (Figure 5). For example, GRB2,FAK and PDGFRA were more strongly phosphory-lated – on a per molecule basis – in GIST882 than inGIST48 (Figure 5). However, KIT, PI3-K p85, PI3-Kp110a and SHC were strongly tyrosine phosphorylatedin both cell lines (Figure 5). In addition, there weresubstantial similarities between GIST882 and GIST48with respect to which of the protein activation eventswere KIT/PDGFR dependent. PI3-K p85, PI3-K p110a,GRB2 and SHC tyrosine phosphorylation were sub-stantially KIT/PDGFR dependent in both cell lines,whereas SRC-family kinases and FAK were minimallyKIT/PDGFR dependent in GIST882, and KIT/PDGFR independent in GIST48 (Figure 5). Weak,partially KIT dependent, tyrosine phosphorylation wasdemonstrated for STAT1 (GIST48 only), STAT3,JAK2, ABL (GIST48 only), MAPK (GIST882 only),PKCy, CBL (GIST882 only) and HSP90. Weak, KITindependent, tyrosine phosphorylation was demon-strated for EPHA4, JAK1 and paxillin. However, theSRC-family kinase data do not exclude the possibilityof functional alterations, in that shifts from SFKc-terminal inhibitory tyrosine phosphorylation to inter-nal activating tyrosine phosphorylation can occur with-out substantially changing the net phosphorylationlevels. The weakly activated wild-type PDGFRB inGIST48 was inhibited by PKC412, and the stronglyactivated PDGFRA in GIST882 was inhibited byimatinib (Figure 5). These findings are in keeping withthe immunoprecipitation studies (Figure 3), and suggestthat PKC412 or imatinib-mediated PDGFR inhibitionresults both from direct effects and inactivation ofheterodimerized KIT oncoproteins. It is particularlynotable that PKCy tyrosine phosphorylation was inhibi-ted by imatinib (GIST882) and PKC412 (GIST48),which is in keeping with KIT-dependent activationof this characteristic GIST kinase (Duensing et al.,2004a).

Density-dependent effects on signaling protein activationThe density (% confluence) of cancer cell lines, whengrown as monolayers in vitro, has been shown to impactphosphorylation of STATs and other signaling proteins(Vultur et al., 2004). In order to determine whetherGIST cell density impacts KIT-dependent signaling, wecompared phosphoprotein expression, with and withoutKIT inhibitor treatment, in 100% confluent (‘confluent’)vs 75% confluent (‘subconfluent’) GIST48 cells. Thesestudies were performed in phosphotyrosine immunoaf-finity-purified fractions from the GIST cells, and showedthat PI3-K p85, FAK and paxillin phosphorylation weresubstantially increased in 75% confluent cells, comparedto 100% confluent cells (Figure 6a and b). Indeed, FAKphosphorylation was tenfold higher in the 75% con-fluent cells than in 100% confluent cells (Figure 6b). TheFAK cell density effects were independent of KITactivation, in that FAK remained strongly phosphorylated

Figure 2 Kinase expression in GIST cell lines with KIT oncogenicmutations. KIT, PI3-K p85 and PKCy were expressed at similarlevels in each of these lines, whereas PDGFRA, PDGFRB andABL expression varied considerably, being strongest in GIST882,GIST48 and GIST430, respectively.

KIT oncoprotein interactions in GISTM-J Zhu et al

6388

Oncogene

in subconfluent GIST48 after PKC412 treatment. Bycontrast, the accentuated PI3-K p85 and paxillinphosphorylation in 75% confluent GIST48 was KITdependent, as shown after PKC412-mediated KITinactivation, which reduced the phosphoprotein expres-sion to levels found in the 100% confluent GIST48(Figure 6a and b). These studies were corroborated, intotal cell lysates of GIST48 at four different celldensities, showing stronger FAK Y397 and paxillinY118 phosphorylation at lower cell density (Figure 6c).CBL and PKCy were also more strongly phosphory-lated – and in a KIT-dependent manner – at 75%confluence, whereas PI3-K p110a and JAK1 were morestrongly phosphorylated in 100% confluent cells (Figure6a and b). STAT, JAK2 and HSP90 tyrosine phosphory-lation were unaffected by GIST48 cell density (Figure 6aand b).

Preferential interactions with the mature form of KITInteractions with the two major KIT forms (migrating,in relationship to the molecular size markers in thisstudy, at 160 and 145 kD, but also described tradition-ally as migrating at 145 and 125 kD) were evaluated inimmunoprecipitations with KIT antibody recognizing

only the smaller, immature, cytoplasmic, KIT form(KIT N-terminal polyclonal rabbit antibody, SantaCruz, Santa Cruz, CA, USA, sc-5535) or both KITforms (mouse monoclonal antibody, Santa Cruz, sc-13508). These studies show that PDGFRA, PI3-K andGRB2 complex preferentially withmatureKIT (Figure 7).Interestingly, both immature and mature KIT coimmu-noprecipitated with PI3-K and PKCy, whereas only themature forms of KIT and PDGFRA coimmunoprecipi-tated (Figures 1, 3a and 7). These findings suggest thatthe cytoplasmic PI3-K and PKCy proteins mightmodulate all forms of KIT signaling, whereas PDGFRAprimarily modulates mature KIT signaling.

KIT RNAi corroboration of small molecule KIT-inhibitorassaysBecause imatinib and PKC412 are multikinase inhibi-tors, and therefore not specific for KIT, we used KITRNAi methods to confirm key findings from the studies(described above) on KIT-dependent GIST cell signal-ing mechanisms. KIT knockdown was accomplished byinfecting GIST882 cells with lentiviral shRNA vectors,and corroborated that PI3-K p85 and CBL activa-tion were KIT dependent (Figure 8a). KIT-mediated

Figure 3 Kinase activation and interactions in untreated (a) and imatinib-treated (b) GIST882 cells. Lanes 1 and 2 contain GIST882total cell lysates, with and without imatinib treatment, respectively. Lanes 3–10 contain GIST882 lysates immunoprecipitated withnormal rabbit serum (lane 3) or antibodies to KIT (Santa Cruz, sc-168; lane 4), PDGFRA (Santa Cruz, sc-338; lane 5), ABL (SantaCruz, sc-131; lane 6), PI3-K p85 (Upstate, 06-497; lane 7), PDGFRB (Santa Cruz, sc-432; lane 8), PKCy (Santa Cruz, sc-1875; lane 9)and PI3-K p110a (Cell Signaling #4254; lane 10). The PY99 and KIT immunostains show activation-dependent complexing of KITwith PDGFRA, PDGFRB, PKCy and PI3-K. Complexing between PI3-K p85 and p110a, as expected, is not KIT dependent.

KIT oncoprotein interactions in GISTM-J Zhu et al

6389

Oncogene

activation of PDGFRA was evaluated by PDGFRAimmunostaining of anti-phosphotyrosine immunopreci-pitates (Figure 8a) and by phosphotyrosine immunos-taining of PDGFRA immunoprecipitates (Figure 8b),and both of these studies showed decreased PDGFRAactivation after shRNA KIT knockdown. Thus, thesefindings are consistent with oncogenic KIT-mediatedPDGFRA cross-activation in GIST cells.

Discussion

KIT and PDGFRA mutational activation events aremutually exclusive in GISTs, but these alternateoncogenic tyrosine kinase mechanisms have similarbiologic and transforming consequences (Heinrichet al., 2003). Although KIT expression is a diagnostichallmark of the majority of GIST cells, the small subsetof GISTs with PDGFRA oncoproteins feature uncha-racteristically low – and even undetectable – KITexpression (Heinrich et al., 2003; Medeiros et al.,2004). On the other hand, approximately 30% of KIT-mutant GISTs express wild-type PDGFRA at high

levels, comparable to those in PDGFRA-mutant GISTs(Heinrich et al., 2003). Previous studies demonstratethat PDGFRA oncoproteins expressed by transfectedPDGFRA mutants in vitro can bind with and cross-activate wild-type KIT (Hirota et al., 2003). Herein, wedemonstrate interactions and evidence for cross-activa-tion between oncogenic KIT and wild-type PDGFRAproteins in human GIST tissues. Complexing betweenKIT and PDGFRA was shown in frozen samples ofprimary human GISTs as well as in the GIST882cell line (Figures 1 and 3). The KIT and PDGFRAcomplexes were dependent on activation of the KIToncoproteins, as evidenced by disruption of the com-plexes after GIST882 exposure to imatinib (Figure 3). Inaddition, the studies in frozen primary GISTs shed lighton the nature of KIT oncogenic signaling complexes inuntreated GISTs, and reveal that another knownimatinib target – PDGFRA – complexes with and islikely coactivated by KIT oncoproteins in GIST. Thesefindings were confirmed by KIT RNAi knockdownassays, in which KIT shRNA-mediated inhibitionresulted in decreased PDGFRA activation (Figure 8),underscoring that PDGFRA cross-activation can bemediated by KIT oncoproteins in GISTs. These studies

Figure 4 Kinase activation and interactions in untreated (a) and PKC412-treated (b) GIST48 cells. Lanes 1 and 2 contain GIST48total cell lysates, with and without PKC412 treatment, respectively. Lanes 3–10 contain GIST48 lysates immunoprecipitated withnormal rabbit serum (lane 3) or antibodies to KIT (Santa Cruz, sc-168; lane 4), PDGFRA (Santa Cruz, sc-338; lane 5), ABL (SantaCruz, sc-131; lane 6), PI3-K p85 (Upstate, 06-497; lane 7), PDGFRB (Santa Cruz, sc-432; lane 8), PKCy (Santa Cruz, sc-1875; lane 9)and PI3-K p110a (Cell Signaling, Danvers, MA, USA, #4254; lane 10). The PY99 and KIT immunostains show activation-dependentcomplexing of KIT with PKCy and PI3-K. Complexing between PI3-K p85 and p110a, as expected, is not KIT dependent.

KIT oncoprotein interactions in GISTM-J Zhu et al

6390

Oncogene

show how the KIT oncogenic signal can be amplified toinvolve other RTKs in GIST, and suggest that imatinibtherapeutic efficacy might be based, in part, oninhibition of both the KIT oncoprotein and hetero-dimerization (PDGFRA) partners.Interestingly, the interactions between KIT and

PDGFRA involved predominantly the mature, trans-membrane forms of these kinases (Figures 1 and 7).Therefore, KIT:PDGFRA interactions, in GIST, likelyresult from heterodimerization – or other complexing –at the cell surface, with resultant activation of canonicalRTK signaling cascades. The GIST882 kinase interactionstudies, performed in confluent cultures, demonstrateKIT:PDGFRA interactions under basal unstimulatedconditions in GIST. Similarly, our studies revealed weak

associations between KIT and PDGFRB in GIST(Figures 3 and 4). However, our studies do not showinteractions between ABL and KIT, and nor was ABLdemonstrably activated in the GIST frozen tumors andcell lines (Figures 1, 3 and 5). These findings suggest thatABL, although a classical imatinib target anda KIT-interacting protein in hematological models(Hallek et al., 1996), does not modulate or enhancethe KIT oncogenic signal in GISTs.PI3-K pathways play crucial roles in transmitting

survival and proliferation signals from various RTKoncoproteins (Herbst et al., 1995; Sordella et al., 2004).In addition, multiprotein complexes involving KIT,PI3-K and PDGFRB have been demonstrated aftertransfection of KIT and PDGFRB constructs in 293

Figure 5 (a–d) KIT/PDGFR-dependent activation of kinase signaling proteins in GIST882 (with and without KIT/PDGFRinhibition by imatinib) and GIST48 (with and without KIT/PDGFR inhibition by PKC412). Total cell lysates (lanes 1 and 4 in eachblot) and tyrosine-phosphorylated proteins immunoaffinity purified from the cell lysates (lanes 2 and 3 in each blot) wereimmunostained with antibodies (specified in Materials and methods) to evaluate KIT-dependent activation of various signalingproteins. KIT activation (large arrows in PY99 stains) was substantially suppressed after inhibitor treatment in both cell lines, as wasPDGFR activation. Treatment-mediated KIT/PDGFR inhibition was accompanied by dephosphorylation of PI3-K p85, PI3-Kp110a, GRB2, SHC, PKCy, STATs, MAPK, CBL and HSP90.

KIT oncoprotein interactions in GISTM-J Zhu et al

6391

Oncogene

cells (Herbst et al., 1995). In the present studies, the p85regulatory and p110 catalytic PI3-K subunits wereconstitutively associated with activated KIT and PDG-FRA in GIST frozen tumors and cell lines (Figures 1, 3and 7). These associations were largely abrogated afterKIT inhibition by imatinib or PKC412 (Figures 3b and4b). PI3-K activation was partially KIT dependent(Figures 5 and 6) and was regulated by cell density(Figure 6).To characterize KIT/PDGFR-dependent signaling

pathways in GISTs, we used an efficient phosphoty-rosine eluate approach combining phosphotyrosineaffinity purification with KIT kinase inhibition in celllines GIST882 and GIST48. This approach enabledefficient immunoblot evaluations of KIT/PDGFR-dependent tyrosine phosphorylation in a panel ofsignaling intermediates, in a single immunoblottingexperiment. KIT was inhibited by imatinib treatmentin GIST882, and by PKC412 treatment in the imatinib-resistant GIST48 cell line. The staurosporine derivativePKC412 was originally identified as an inhibitor of

protein kinase C (PKC) and subsequently shown toinhibit other kinases including VEGFR2, PDGFR andKIT (Fabbro et al., 2000), with activity in someimatinib-resistant GIST mutations (Debiec-Rychteret al., 2005). The phosphotyrosine eluate studiesconfirmed the immunoprecipitation assays (Figures 1and 3), showing imatinib/PKC412 inhibition of KIT,PDGFRA and PI3-K (Figure 5). Comparisons betweentotal cell lysates and phosphotyrosine purifications(Figure 5) showed that KIT, PI3-K (regulatory andcatalytic subunits), SHC and GRB2 were stronglytyrosine phosphorylated in a KIT-dependent mannerin both GIST882 and GIST48 cell lines. These findingsare consistent with KIT oncogenic activation of theRAS/RAF/MAPK and PI3-K/AKT pathways, whichare regulated by activation of GRB2/SHC and PI3-K,respectively (Tauchi et al., 1994; Blume-Jensen et al.,1998). Recruitment of GRB2 and SHC adaptor proteinsto RTK-derived oncogenes has been shown to besufficient for morphological cell transformation andexperimental metastases (Saucier et al., 2002), whereas

Figure 6 (a–c) Cell density effects on KIT/PDGFR-dependent activation of kinase signaling proteins in GIST48, with and withoutKIT/PDGFR inhibition by PKC412. As in Figure 5, total cell lysates (lanes 1 and 4 in a and b) and immunoaffinity-purified tyrosine-phosphorylated proteins (lanes 2 and 3 in a and b) were immunostained to evaluate KIT-dependent activation of various signalingproteins. Tyrosine phosphorylation was evaluated in cells harvested from 100% confluent (images from Figure 5) vs 75% confluentcultures (a and b). PI3-K p85, FAK and Paxillin were more strongly phosphorylated in 75% confluent cultures, PI3-K p110a andJAK1 were more strongly phosphorylated in 100% confluent cultures. In the 75% confluent cultures, the accentuated phosphorylationof PI3-K p85 and paxillin, but not FAK, is KIT dependent. STAT, JAK2 and HSP90 phosphorylation were unaffected by cultureconfluence. Increased FAK and paxillin phosphorylation, in subconfluent cultures, was confirmed in GIST48 total cell lysatesimmunostained with antibodies to phospho-FAK and phospho-paxillin (c). These studies showed that FAK and paxillin tyrosinephosphorylation were reduced in 100% confluent cultures, compared to the subconfluent cultures.

KIT oncoprotein interactions in GISTM-J Zhu et al

6392

Oncogene

PI3-K appears to play a particularly crucial role intyrosine kinase oncoprotein signaling (Sordella et al.,2004). Therefore, our findings suggest that PI3-K,

GRB2 and SHC might serve as alternate therapeutictargets, whose inhibition could be beneficial in GISTpatients resistant to imatinib.Our present studies corroborate previous reports that

STAT activation, in GISTs, is relatively weak and onlypartially dependent on the activated KIT oncoprotein(Duensing et al., 2004b). This situation differs fromthe KIT kinase-domain mutants in mast cell disease,where STAT is activated strongly and crucial to KITtransforming activity (Ning et al., 2001). In addition,whereas STAT activation is regulated by cell density in acarcinoma model (Vultur et al., 2004), being morestrongly activated in subconfluent cultures than inconfluent cultures, we did not observe marked increasein STAT activation in subconfluent GIST cultures.The involvement of cell adhesion proteins has not

been evaluated previously in GISTs, and our studiesshow that the FAK and paxillin adhesion-relatedproteins are activated in a cell density-dependentmanner. That is, activation of both FAK and paxillinactivation was strongest in subconfluent, actively pro-liferating, cultures. FAK activation was KIT indepen-dent in both confluent and subconfluent cultures(Figures 5 and 6), whereas paxillin activation was KITindependent in confluent cultures and partially KITdependent in subconfluent cultures. These findingssuggest that imatinib impact on adhesion proteins, inclinical GISTs, might vary in different parts of the sametumor, depending on the local architecture and celldensity. Notably, the KIT-independent nature of FAKactivation, in these GIST models, suggests that FAKtherapeutic inhibition might provide clinical synergiesfor KIT inhibition by imatinib.The novel PKC family member, PKCy, is an

intriguing biomarker in GIST, given its narrow rangeof expression in normal cells, and preliminary evidence

Figure 7 PDGFRA, PI3-K p85 and GRB2 interactions withmature vs immature KIT oncoproteins. GIST882 lysates wereimmunoprecipitated with antibodies recognizing immature KITonly (Santa Cruz, sc-5535) vs both immature and mature KIT(Santa Cruz, sc-13508). The immunostains show that PDGFRA,PI3-K p85 and GRB2 interact predominantly with the mature KIToncoprotein.

Figure 8 Kinase activation in GIST882 cells after infection with lentiviral empty vector (lanes 1 and 3) vs KIT shRNA (lanes 2 and 4).In each panel, lanes 1 and 2 are total cell lysates, whereas lanes 3 and 4 are immunoprecipitates evaluating kinase activation.Immunostains of the PY99 (Santa Cruz, sc-7020) immunoprecipitates (a) showed that KIT shRNA infections reduced the expressionof total and phospho-KIT to 50% of control levels, resulting in 35, 51 and 52% reductions in tyrosine-phosphorylated PI3-K p85, CBLand PDGFRA, respectively. Protein expression quantitation was performed using a FUJI LAS1000þ digital capture system.Similarly, phosphotyrosine immunostaining of PDGFRA immunoprecipitates (b) demonstrated 60% reduction in PDGFRAphosphorylation, after KIT shRNA knockdown.

KIT oncoprotein interactions in GISTM-J Zhu et al

6393

Oncogene

for a positive regulatory role in KIT oncogenic signalingin GIST (Duensing et al., 2004a). We show, for the firsttime, that PKCy associates with KIT oncoproteins inGIST cell lines (Figure 3), and is tyrosine phosphory-lated in a KIT-dependent manner, particularly insubconfluent cells (Figures 5 and 6). Strong PKCyexpression has been reported – in adult tissues – only inT cells and ICC, and in the leukemias and GISTs thatderive, respectively, from these cell lineages. Therefore,it is possible that PKCy inhibition might be a minimallytoxic and highly effective therapy for GIST. Ourpreliminary studies with PKCy inactivation or RNAiknockdown (Hubert, Ou and Fletcher, unpublished) doindeed suggest that PKCy expression and activation arecrucial to GIST cell survival and proliferation.

Materials and methods

ReagentsAntibodies for immunoprecipitation and immunoblotting aredescribed in the Supplementary methods.

Tissue specimens and cell linesGIST tissue specimens and cell lines are described in theSupplementary methods.

Functional studiesBiochemical correlates for KIT inhibition were evaluated inGIST882 and GIST48 cells after incubation for 4 h in serum-free media with 2mM imatinib and 1mM PKC412 (provided byNovartis Pharma, Basel, Switzerland), respectively.

KIT shRNA studiesA KIT lentiviral shRNA (short hairpin RNA) was obtainedfrom Dr William Hahn (Dana-Farber Cancer Institute andBroad Institute RNAi consortium). This shRNA was as-sembled by ligating KIT forward [50-CCGGCCATAAGGTTTCGTTTCTGTACTCGAGTACAGAAACGAAACCTTATGGTTTTTG-30] and reverse [50-AATTCAAAAACCATAAGGTTTCGTTTCTGTACTCGAGTACAGAAACGAAACCTTATGG-30] oligomers into the AgeI and EcoRI sites of a pLKO.1purolentiviral vector. Lentivirus was produced by cotransfectingpLKO.1puro empty vector or pLKO.1puro-KIT (shRNA),

pCMVDR8.91 and pMD.G helper virus packaging plasmids(at a 10:10:1 ratio) into 293T cells. These transfections wereperformed using lipofectamine and PLUS reagent, andlentivirus supernatants were harvested at 24, 36, 48 and 60 h.Viral titers were determined in GIST882 cells, according to aprotocol from Invitrogen (Carlsbad, CA, USA). GIST882 cellswere infected, in the presence of 8 mg/ml of polybrene, andwere then lysed for immunoprecipitation and western blotanalysis after 20 days of selection with 2.5 mg/ml puromycin.

Immunoprecipitation and western blot analysisImmunoprecipitation and immunoblotting methods wereessentially as described, and are detailed in the Supplementarymethods. The chemiluminescence signals were captured andquantified using a FUJI LAS1000plus system with Science Lab2001 ImageGauge 4.0 software (Fujifilm Medical Systems,Stamford, CT, USA).

Phosphotyrosine affinity purificationThe phosphotyrosine fraction was isolated from GIST cellsusing a minicolumn loaded with 1ml of anti-phosphotyrosine-sepharose beads (Zymed, South San Francisco, CA, USA).The anti-phosphotyrosine affinity columns were first washedten times with phosphate-buffered saline containing 0.02%sodium azide followed by ten washes with cell lysis buffer. Thecolumns were then incubated at 41C overnight with 10mg ofGIST protein lysate, washed seven times with lysis buffer, andthe bound tyrosine-phosphorylated proteins were then elutedby loading the column ten times with 1ml of 100mMphenylphosphate. The phosphotyrosine protein eluates weredialysed to remove the phenylphosphate, then evaluated byimmunoblotting, using 10% of the eluates per gel lane.

Acknowledgements

We thank Yongchun Zhou for critical review of the article andvaluable discussions, Chang-Jie Chen, Anette Duensing andMaureen Thyne for useful discussions and technical assistance.This work was supported by grants from an anonymousdonor, the Life Raft Group, Cesarini Team for the Pan-Massachusetts Challenge, the Virginia and Daniel K LudwigTrust for Cancer Research, the Ronald O Perelman Fund forCancer Research, the Stutman GIST Cancer Research Fund,the Rubenstein Foundation and Leslie’s Links.

References

Blume-Jensen P, Janknecht R, Hunter T. (1998). The kitreceptor promotes cell survival via activation of PI 3-kinaseand subsequent Akt-mediated phosphorylation of Bad onSer136. Curr Biol 8: 779–782.

Debiec-Rychter M, Cools J, Dumez H, Sciot R, Stul M,Mentens N et al. (2005). Mechanisms of resistance toimatinib mesylate in gastrointestinal stromal tumors andactivity of the PKC412 inhibitor against imatinib-resistantmutants. Gastroenterology 128: 270–279.

Demetri GD, von Mehren M, Blanke CD, Van den AbbeeleAD, Eisenberg B, Roberts PJ et al. (2002). Efficacy andsafety of imatinib mesylate in advanced gastrointestinalstromal tumors. N Engl J Med 347: 472–480.

Duensing A, Joseph NE, Medeiros F, Smith F, Hornick JL,Heinrich MC et al. (2004a). Protein Kinase C theta(PKCtheta) expression and constitutive activation in

gastrointestinal stromal tumors (GISTs). Cancer Res 64:5127–5131.

Duensing A, Medeiros F, McConarty B, Joseph NE,Panigrahy D, Singer S et al. (2004b). Mechanisms ofoncogenic KIT signal transduction in primary gastrointes-tinal stromal tumors (GISTs). Oncogene 23: 3999–4006.

Fabbro D, Ruetz S, Bodis S, Pruschy M, Csermak K, Man Aet al. (2000). PKC412 – a protein kinase inhibitor with abroad therapeutic potential. Anticancer Drug Des 15: 17–28.

Hallek M, nhauser-Riedl S, Herbst R, Warmuth M, WinklerA, Kolb HJ et al. (1996). Interaction of the receptor tyrosinekinase p145c-kit with the p210bcr/abl kinase in myeloidcells. Br J Haematol 94: 5–16.

Heinrich MC, Corless CL, Duensing A, McGreevey L, Chen CJ,Joseph N et al. (2003). PDGFRA activating mutationsin gastrointestinal stromal tumors. Science 299: 708–710.

KIT oncoprotein interactions in GISTM-J Zhu et al

6394

Oncogene

Heinrich MC, Rubin BP, Longley BJ, Fletcher JA. (2002).Biology and genetic aspects of gastrointestinal stromaltumors: KIT activation and cytogenetic alterations. HumPathol 33: 484–495.

Herbst R, Shearman MS, Jallal B, Schlessinger J, Ullrich A.(1995). Formation of signal transfer complexes between stemcell and platelet-derived growth factor receptors and SH2domain proteins in vitro. Biochemistry 34: 5971–5979.

Hirota S, Isozaki K, Moriyama Y, Hashimoto K, Nishida T,Ishiguro S et al. (1998). Gain-of-function mutations of c-kit inhuman gastrointestinal stromal tumors. Science 279: 577–580.

Hirota S, Ohashi A, Nishida T, Isozaki K, Kinoshita K,Shinomura Y et al. (2003). Gain-of-function mutations ofplatelet-derived growth factor receptor alpha gene in gastro-intestinal stromal tumors. Gastroenterology 125: 660–667.

Medeiros F, Corless CL, Duensing A, Hornick JL, OliveiraAM, Heinrich MC et al. (2004). KIT-negative gastrointest-inal stromal tumors: proof of concept and therapeuticimplications. Am J Surg Pathol 28: 889–894.

Ning ZQ, Li J, Arceci RJ. (2001). Signal transducer andactivator of transcription 3 activation is required forAsp(816) mutant c-Kit-mediated cytokine-independent sur-vival and proliferation in human leukemia cells. Blood 97:3559–3567.

Otto KG, Jin L, Spencer DM, Blau CA. (2001). Cellproliferation through forced engagement of c-Kit andFlt-3. Blood 97: 3662–3664.

Rubin BP, Singer S, Tsao C, Duensing A, Lux ML, Ruiz Ret al. (2001). KIT activation is a ubiquitous feature ofgastrointestinal stromal tumors. Cancer Res 61: 8118–8121.

Saucier C, Papavasiliou V, Palazzo A, Naujokas MA, KremerR, Park M. (2002). Use of signal specific receptor tyrosinekinase oncoproteins reveals that pathways downstream fromGrb2 or Shc are sufficient for cell transformation andmetastasis. Oncogene 21: 1800–1811.

Sordella R, Bell DW, Haber DA, Settleman J. (2004).Gefitinib-sensitizing EGFR mutations in lung cancer acti-vate anti-apoptotic pathways. Science 305: 1163–1167.

Tauchi T, Feng GS, Marshall MS, Shen R, Mantel C, PawsonT et al. (1994). The ubiquitously expressed Syp phosphataseinteracts with c-kit and Grb2 in hematopoietic cells. J BiolChem 269: 25206–25211.

Tuveson DA, Willis NA, Jacks T, Griffin JD, Singer S,Fletcher CD et al. (2001). STI571 inactivation of thegastrointestinal stromal tumor c-KIT oncoprotein: biologi-cal and clinical implications. Oncogene 20: 5054–5058.

Verweij J, Casali PG, Zalcberg J, LeCesne A, Reichardt P,Blay JY et al. (2004). Progression-free survival in gastro-intestinal stromal tumours with high-dose imatinib: rando-mised trial. Lancet 364: 1127–1134.

Vultur A, Cao J, Arulanandam R, Turkson J, Jove R,Greer P et al. (2004). Cell-to-cell adhesion modulates Stat3activity in normal and breast carcinoma cells. Oncogene 23:2600–2616.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).

KIT oncoprotein interactions in GISTM-J Zhu et al

6395

Oncogene