KIT-dependent and -independent genomic heterogeneity of resistance in gastrointestinal stromal tumors – TORC1/2 inhibition as salvage strategy

Thomas Mühlenberg1,2, Julia Ketzer1,2, Michael C. Heinrich3, Susanne Grunewald1,2, Adrian Marino-

Enriquez4, Marcel Trautmann5, Wolfgang Hartmann5, Eva Wardelmann5, Jürgen Treckmann6, Karl
Worm7, Stefanie Bertram7, Thomas Herold2,7, Hans-Ulrich Schildhaus8, Hanno Glimm2,9, Albrecht
Stenzinger2,10, Benedikt Brors2,11, Peter Horak2,12, Peter Hohenberger13, Stefan Fröhling2,12, Jonathan
A. Fletcher4, Sebastian Bauer1,2

Affiliations:
9 1 Dept. of Medical Oncology, Sarcoma Center, West German Cancer Center, University Duisburg-
10 Essen, Medical School, Essen, Germany
11 2 German Cancer Consortium (DKTK), Heidelberg, Germany
12 3 Portland VA Health Care System, Knight Cancer Institute, Oregon Health and Science University,
13 Portland, Oregon, USA
14 4 Dept. of Pathology, Brigham and Women’s Hospital, Harvard Medical School, Boston,
15 Massachusetts, USA
16 5 Gerhard Domagk Institute of Pathology, University Hospital Münster, Münster, Germany
17 6 Dept. of Visceral and Transplant Surgery, Sarcoma Center, West German Cancer Center,
18 University Duisburg-Essen, Medical School, Essen, Germany
19 7 Institute of Pathology, University Hospital of Essen, University of Duisburg-Essen, Germany
20 8 Institute of Pathology, Universitätsmedizin Göttingen, Göttingen, Germany
21 9 Department of Translational Oncology, National Center for Tumor Diseases (NCT) Dresden,
22 Dresden University Hospital, Dresden, Germany
23 10 Institute of Pathology, Heidelberg University Hospital, Heidelberg, Germany
24 11 Dept of Applied Bioinformatics, German Cancer Research Center (DKFZ), Heidelberg University,
25 Heidelberg, Germany
26 12Department of Translational Oncology, National Center for Tumor Diseases (NCT) Heidelberg,
27 German Cancer Research Center (DKFZ), Heidelberg University Hospital, Heidelberg, Germany
28 13 Mannheim, University Medical Center, Mannheim, Germany

Running title: Sapanisertib in genetically heterogeneous GIST
31 Keywords: Soft-tissue sarcoma; GIST; imatinib resistance; sapanisertib; Clinical drug resistance;
32 Novel mechanisms of drug resistance
33 Financial support: This work was supported by funds from the fundraising event “Sarkomtour”
34 (www.sarkomtour.de; S. Bauer). Whole-exome/genome and RNA sequencing were funded by the
35 DKTK Joint Funding Program (S. Fröhling). Further support was received from the GIST Cancer
36 Research Fund (M.C. Heinrich), and VA Merit Review Grant (M.C. Heinrich, 2I01BX000338-05).
37 Corresponding authors:
38 Dr. Thomas Mühlenberg Dr. Sebastian Bauer
39 Universitätsklinikum Essen Universitätsklinikum Essen
40 Innere Klinik (Tumorforschung) Innere Klinik (Tumorforschung)
41 Hufelandstraße 55 Hufelandstraße 55
42 D- 45147 Essen D- 45147 Essen
43 Phone: +49 201 / 723-85097 Phone: +49 201 / 723-2014
44 Fax: +49 201 / 723-3112 Fax: +49 201 / 723-3112
45 [email protected] [email protected]
47 The authors declare no potential conflicts of interest.

1 Abstract:
2 Sporadic gastrointestinal stromal tumors (GIST), characterized by activating mutations of KIT or
3 PDGFRA, favorably respond to KIT inhibitory treatment but eventually become resistant. The
4 development of effective salvage treatments is complicated by the heterogeneity of KIT
5 secondary resistance mutations. Recently, additional mutations that independently activate KIT-
6 downstream signaling have been found in pretreated patients – adding further complexity to the
7 scope of resistance. We collected genotyping data for KIT from tumor samples of pretreated
8 GIST, providing a representative overview on the distribution and incidence of secondary KIT
9 mutations (n=80). Analyzing next generation sequencing data of 109 GIST, we found that 18%
10 carried mutations in KIT-downstream signaling intermediates (NF1/2, PTEN, RAS, PIK3CA,
11 TSC1/2, AKT, BRAF) potentially mediating resistance to KIT inhibitors. Notably, we found no
12 apparent other driver mutations in refractory cases that were analyzed by whole exome/genome
13 sequencing (13/109). Employing CRISPR/Cas9 methods, we generated a panel of GIST cell
14 lines harboring mutations in KIT, PTEN, KRAS, NF1, and TSC2. We utilized this panel to
15 evaluate sapanisertib, a novel mTOR kinase inhibitor, as a salvage strategy. Sapanisertib had
16 potent antiproliferative effects in all cell lines, including those with KIT-downstream mutations.
17 Combinations with KIT- or MEK- inhibitors completely abrogated GIST-survival signaling and
18 displayed synergistic effects. Our isogenic cell line panel closely approximates the genetic
19 heterogeneity of resistance observed in heavily pretreated GIST patients. With the clinical
20 development of novel, broad spectrum KIT inhibitors, emergence of non-KIT-related resistance
21 may require combination treatments with inhibitors of KIT-downstream signaling such as mTOR
22 or MEK.

1 Introduction
2 Gastrointestinal stromal tumors (GIST) are the most common sarcomas of the GI tract and are
3 characterized by activating mutations of the KIT or PDGFRA receptor tyrosine kinases (1,2).
4 Most patients respond to the KIT/PDGFRA inhibitor imatinib (IM) but eventually progress due to
5 secondary resistance mutations in KIT (3,4). Second and third line KIT-inhibitors have limited
6 clinical benefit and only for a subset of patients (5-7). The development of effective salvage
7 treatments is hampered by the heterogeneity of resistance mutations in KIT often observed
8 within a single patient (8-10).
9 The various KIT mutations in GIST activate the PI3K/AKT/mTOR and RAS/RAF/MAPK signaling
10 pathways, and these same pathways are also activated in other GIST subtypes, even in so-
11 called wild type GIST (11). Recently, activating mutations in these signaling cascades, including
12 PI3K, KRAS, PTEN, and NF1, have been shown to emerge in later treatment lines, representing
13 resistance mechanisms that cannot likely be addressed using direct KIT-inhibition approaches
14 (11,12). Therefore, novel treatment strategies beyond the direct inhibition of KIT may become a
15 crucial factor in GIST treatment in the near future. However, preclinical models recapitulating this
16 heterogeneity of resistance, which would alleviate research towards this goal, do not exist yet.
17 We used the CRISPR/Cas9 (Clustered Regularly Interspaced Short Palindromic Repeats/
18 CRISPR associated protein 9) system as a novel powerful tool to generate new GIST cell line
19 models (13).
20 Inhibitors of KIT-downstream signaling intermediates (PI3K, mTOR, MEK) have been evaluated
21 in GIST for several years but have not transitioned to advanced clinical development.
22 Rapamycin-analogues targeting mTOR have only been examined in refractory GIST in
23 combination with imatinib – which could not reverse imatinib-resistance. However, next-
24 generation mTOR kinase inhibitors have not yet been tested. Sapanisertib (INK0128, MLN0128,
25 TAK-228) represents a novel class of ATP-competitive mTOR inhibitors, specifically inhibiting
26 mTOR kinase in both mTOR complexes (mTORC) 1 and 2 (14). The compound has been shown
27 to be a more potent inhibitor of mTOR signaling than rapamycin, one that can overcome intrinsic
28 and acquired resistance to rapamycin, and is well tolerated in vivo and in early clinical trials (15-
29 18).
30 Here, we sought to generate a GIST cell line panel comprising KIT-dependent and –independent
31 mechanisms of resistance to current KIT-inhibitors, approximating the clinical situation in vitro.
32 We then utilized this panel to evaluate the efficacy of sapanisertib alone and in combination with
33 KIT- or MEK-inhibition.
1 Methods
2 Patients
3 All patients were previously diagnosed as GIST patients by routine pathological review. In a
4 retrospective study we gathered all consecutive sequencing data available (Sanger + NGS) for
5 GIST patients who underwent routine molecular-pathology review in Essen. Furthermore we
6 compiled panel NGS data of routine molecular-pathology review from sarcoma centers of
7 Göttingen and Münster, Germany and Portland, OR, USA, as well as WES/WGS data from GIST
8 patients who participated in the DKTK MASTER program (19) in Heidelberg. Due to the nature
9 of these data (i.e. anonymized molecular pathology review only, except for the DKTK MASTER
10 cohort), no clinical data, or sequencing raw data of these patients are available and no written
11 informed consent could be requested or given. In this study exclusively anonymized sequencing
12 data were analyzed, allowing no inference to patient identity and medical history except for
13 diagnosis. The study was approved by the institutional review board (IRB; ethics committee) of
14 the Medical School of the University of Duisburg-Essen was conducted in accordance with the
15 Declaration of Helsinki. For the DKTK MASTER cohort, all patients provided written informed
16 consent under a protocol approved by the ethics committee of Heidelberg University, and the
17 study was conducted in accordance with the Declaration of Helsinki.
Illumina – panel sequencing of tumor samples (63 patients)
20 Multiplex PCR and purification was performed with the GeneRead DNAseq Custom Panel V2,
21 GeneRead DNAseq Panel PCR Kit V2 (QIAgen) and Agencourt® AMPure® XP Beads
22 (Beckmann). A total amount of 44ng DNA was used to perform multiplex PCR (four primer pools
23 with 11ng each). Library preparation was performed using NEBNext Ultra DNA Library Prep Set
24 for Illumina (New England Biolabs; NEB), according to the manufacturer’s recommendations
25 applying 24 different indices per run. The pooled library was sequenced on MiSeq (Illumina;
26 2×150 bases paired-end run) and analyzed by the Biomedical Genomics Workbench (CLC Bio,
27 QIAgen). Within the CLC Cancer Research Workbench demultiplexed paired-end sequencing
28 data was mapped to human genome (version hg19). A local realignment was performed to reach
29 better alignment quality, especially for regions with small insertions or deletions. All reads which
30 were mapped outside of targeted-regions were deleted after the mapping process. In a filtering-
31 step all reference-variants and variants found in dbSNP common, 1000 genome project and
32 HapMap were deleted. An allele-frequency of minimum 2% and coverage of at least 100
33 mapped-reads were applied. Samples with less than 50% of mapped bases against hg19 were
34 categorized as not analyzable. For deviations from this protocol, as performed for patients from

1 Münster and Göttingen, and detailed information on sequencing panels see Supplemental
2 Material.
Ion Torrent – panel sequencing of tumor samples (33 patients)
5 Targeted sequence analysis was performed with a custom AmpliSeq panel (Life Technologies,
6 Grand Island, NY) that includes 24 genes (AKT1, AKT2, AKT3, ATM, BRAF, CDKN2A, HRAS,
7 KIT, KRAS, MAP2K1, NF1, NRAS, PDGFRA, PIK3CA, PTEN, PTPN11, RB1, SDHA, SDHAF1,
8 SDHAF2, SDHB, SDHC, SDHD, TP53). Sequencing was carried out on an Ion Torrent PGM
9 instrument, and Torrent Suite Software v3.2 was used for sequence alignment and variant
10 calling (20).
Whole exome/genome and RNA sequencing (13 patients)
14 DNA was extracted from 13 GIST samples from patients participating in the DKTK MASTER
15 program (19) along with corresponding peripheral blood or adjacent normal tissue and subjected
16 to whole exome/genome and RNA sequencing as described previously (21,22). Reads were
17 mapped to the 1000 Genomes Phase II assembly of the human reference genome (NCBI build
18 37.1). Genome sequencing data were aligned using Burrows-Wheeler Aligner, BWA (version
19 0.6.2 or 0.7.15). BAM files were sorted with SAMtools (version 0.1.19)(23), and duplicates were
20 marked with Picard tools (version 1.125). Single-nucleotide variants (SNVs) and small
21 insertions/deletions (indels) were analyzed using a previously reported bioinformatics workflow
22 (24). Copy number variants (CNVs) were extracted from the whole exome sequencing samples
23 with the help of CNVkit (version 0.8.3.dev0)(25) and from the whole genome sequencing (WGS)
24 data using our in-house CNV calling pipeline ACEseq (26). Structural variants were detected in
25 whole exome sequencing (WES) data using CREST (27). All events were annotated with
26 RefSeq genes using BEDTools (28). RNA sequencing (RNA-Seq) data generated on the HiSeq
27 2500 platform were processed as described before (24), RNA-Seq data generated on the HiSeq
28 4000 platform were aligned using STAR 2.5.1b (29). Relative RNA expression of 467 predefined
29 cancer relevant genes, compared to median RPKM (Reads Per Kilobase Million) in a cohort of
30 149 diverse cancer patients, is reported. Overexpressed genes with RPKM fold change > 10 and
31 Z-score > 1 and underexpressed genes with RPKM fold change < 0.25 and Z-score <-1 were
32 evaluated. Sequencing data were deposited in the European Genome-phenome Archive under
33 accession EGAS00001003405.
CRISPR/Cas9 mediated gene editing

1 To generate GIST-T1/V654A, GIST-T1/D816A, and GIST-T1/G12R suitable guide sequences
2 targeting KIT exon 13 and 17, and KRAS exon 2, respectively, were identified using the online
3 tool CRISPOR (30) (www.crispor.org). For both KIT exons two adjacent guides were selected. A
4 forward oligo containing the t7 RNA polymerase promoter sequence and the respective guide
5 sequence, as well as a reverse oligo containing the generic single guide RNA sequence were
6 purchased at MWG Eurofins. For guide RNA design we employed “FE-modified” sequences
7 described by Chen et al. (31). The t7 RNA polymerase DNA template was generated by a fill-in
8 PCR using Q5 high fidelity DNA polymerase (NEB), running 10 cycles of 62°C / 20sec and 72°C
9 / 2min. sgRNA was then transcribed with t7 RNA polymerase (NEB), according to
10 manufacturer’s instructions and precipitated by phenol/chlorophorm extraction, using standard
11 protocols.
12 GIST-T1 cells were seeded in a T25 flask at low density and after 72h of growth cells were
13 trypsinized, washed and resuspended in electroporation buffer (Buffer SF, Lonza). 105 cells
14 were mixed with 0.5-0.7µl Recombinant Cas9 (20µM; NEB), 0.5-0.7µl in vitro transcribed sgRNA
15 (20µM) per guide, and 0.5-0.7µl single stranded DNA template (ssODN (MWG Eurofins);
16 500µM), carrying the desired mutation to be introduced via the homology directed repair (HDR)
17 pathway. Cells were electroporated using the program DN100 on the Amaxa Nucleofector 4D
18 (Lonza). On the next day the bulk cells were selected with IM 100nM until outgrowth of a
19 resistant population was achieved. Sanger sequencing confirmed heterozygous mutations KIT
20 V654A and D816A, and KRAS G12R, as well as silent mutations induced by the HDR templates.
21 In the attempt to generate GIST-T1/G12R we furthermore generated a subline harboring a
22 heterozygous 23bp deletion in KRAS (GIST-T1-KRAS) starting in exon 2, codon 9, causing a
23 frameshift and premature stop codon within the exon. Interestingly, these cells showed strong
24 activation of RAS downstream signaling (see results). For a detailed list of sequences see
25 Supplemental Material.
CRISPR/Cas9 mediated gene knock out
28 For the generation of T1-PTEN and T1-TSC2, suitable guide sequences against PTEN, TSC2,
29 with sticky-end overhangs were ordered from MWG Eurofins. Oligos were annealed, and cloned
30 into the “lentiCRISPR v2” vector (Addgene Plasmid #52961), according to the published protocol
31 (13). For PTEN knockout the resulting plasmid was transfected by nucleofection as described
32 above. 24h after transfection cells were selected with puromycin 2µM for 96h. For TSC2
33 knockout the plasmid was transduced after lentiviral Packaging, according to standard protocols.
34 5d after transfection cells were selected with puromycin 2µM for 96h. Cells were then further
35 selected with IM 100nM until outgrowth of a resistant population. For NF1 knockout cell (T1-

1 NF1), lentiviral constructs encoding SpCas9 (pXPR_BRD001, Broad Institute) and an anti-NF1
2 guide RNA (pXPR_BRD003) were transduced by two consecutive lentiviral infections, followed
3 by one week of 2µM of puromycin selection, and further selection with IM 100nM until outgrowth
4 of a resistant population. Knock out was confirmed by western blot and next generation
5 sequencing. For a detailed list sequences see Supplemental Material.
Cell lines
8 Apart from the cell lines described above, further IM-sensitive (GIST-T1, GIST882, GIST430)
9 and IM-resistant (GIST430/654, GIST-T1-D816E, GIST48B) cell lines were studied. GIST-T1
10 and GIST882 were established from human, untreated, metastatic GISTs and carry primary
11 activating mutations in exons 11(V560_Y578del) and 13 (K642E), respectively. GIST430/654
12 was established from a GIST that had progressed, after initial clinical response during IM
13 therapy and harbors a primary activating mutation in exon 11 (51bp del V560-Y578) and
14 secondary resistance mutation in exon 13 (V654A). GIST430 is an IM-sensitive subline, with
15 only the primary KIT exon 11 (51bp del V560-Y578) mutation but no secondary resistance
16 mutation, derived from the same GIST culture as the GIST430/654 line. GIST48B, despite
17 retaining the activating KIT mutation in all cells, expresses KIT transcript and protein at
18 essentially undetectable levels. GIST-T1 was established by Takahiro Taguchi (Kochi University,
19 Kochi, Japan). GIST882 were cultured in RPMI1640 containing 15% FBS and 1% Pen/Strep
20 (Gibco). All other cell lines were cultured in IMDM containing 10% FBS and 1% Pen/Strep. Cell
21 lines are regularly authenticated by sequencing of endogenous mutations in KIT, confirmation of
22 KIT-expression, and response to KIT inhibitor treatment. In the course of this study all cell lines
23 were regularly tested for mycoplasma contamination by PCR and by MycoAlert Mycoplasma
24 Detection Kit (Lonza).
Reagents and Antibodies
27 Sapanisertib, imatinib, sunitinib, ponatinib, trametinib and everolimus (RAD001) were purchased
28 form Selleck chemicals. A primary polyclonal rabbit antibody against KIT was purchased from
29 Dako. A monoclonal mouse antibody against beta-actin was purchased from Sigma. All other
30 primary and secondary antibodies used in this study were purchased from Cell Signaling
31 Technologies.
Western Blot
34 Cells were plated in six-well plates and on the next day treated with different inhibitors or vehicle
35 control. After 24h of treatment lysis buffer (1% NP-40, 50mM Tris-HCl pH 8.0, 100mM sodium
fluoride, 30mM sodium pyrophosphate, 2mM sodium molybdate, 5mM EDTA and 2mM sodium
2 vanadate; freshly adding 0.1% 10mg/mL aprotinin and leupeptin as well as 1% 100mM PMSF
3 and 200mM sodium vanadate) was added, and cells were scraped off and then lysed while
4 rotating for 1h at 4°C. Lysates were centrifuged at 4°C for 30min at 18,000rcf and protein
5 concentration was determined using the Bio-Rad Protein Assay (Bio-Rad Laboratories). Protein
6 concentration was adjusted to 2µg/µl (if not otherwise specified), SDS-loading buffer (0.5M Tris-
7 HCl pH 6.7, 10% SDS, 2.5% DTT, 50% Glycerol, and 0.05% bromophenol blue) was added and
8 lysates were incubated for 5min at 95°C. Equal amounts of Protein (30µg per lane, if not
9 otherwise specified) were separated on SDS-PAGE Gels (NuPAGE 4-12%, Life Technologies)
10 and blotted onto nitrocellulose-membranes (GE Healthcare/Amersham-Biosciences). After
11 blocking with Net-G buffer (1.5M NaCl, 50mM EDTA, 500mM Tris, 0.5% Tween 20 and 0.4%
12 gelatine) membranes were incubated at 4°C overnight with the respective primary antibody.
13 After washing (Net-G), membranes were incubated for 2h at room temperature with a secondary
14 antibody (in Net-G) and washed again. Changes in protein expression and phosphorylation as
15 visualized by chemiluminescence were captured and quantified using a FUJI LAS3000 system
16 with Science Lab 2001 ImageGauge 4.0 software (Fujifilm Medial Systems). Usually 2–4
17 gels/membranes were prepared from the same experiment/lysates, to enable clean stains of
18 proteins with similar or nearby molecular weight as well as stains of total proteins and their
19 phosphorylated counterparts. Membranes were consecutively stained with different antibodies of
20 different molecular weights. Beta-Actin served as loading control for each membrane and a
21 representative stain is shown.

23 Sulforhodamin B assay
24 Cell viability was evaluated by Sulforhodamin B (SRB; Sigma-Aldrich) assay after 72h of
25 treatment, as previously described (32). Cells were treated with increasing concentrations of
26 DMSO-dissolved compounds, sapanisertib, trametinib, imatinib, sunitinib, regorafenib, ponatinib.
27 Mean values were normalized to DMSO-solvent control and the mean standard error was
28 calculated. All experiments were carried out in triplicate/quadruplicate cultures at least twice and
29 a representative example is shown.
For dose-finding experiments 1000 cells/well were seeded in white 384-well plates (Greiner)
33 using the Multidrop (Thermo Scientific) and allowed to attach overnight. Respective compounds
34 were added in duplicates or triplicates using the digital dispenser Tecan D300e, and normalized
35 to identical solvent volumes. After 72h Cell Titer Glo (Promega) reagent was added according to

1 manufacturer’s instructions. Luminescence was measured on the Tecan Spark M10. The
2 combinatorial index (CI) was calculated according to the method by Chou-Talalay (33), using
3 CalcuSyn Software (BioSoft). For confirmation the most effective combinations were then
4 evaluated in triplicates in 96-well plates using the SRB assay.

1 Results
2
3 GIST patients display heterogeneity of resistance to KIT-inhibition
4 First we reviewed the in-house sequencing data bases for mutations in GIST patients found
5 during routine clinical testing in our centers. We thus compiled a cohort of 80 patients with
6 secondary resistance mutations in KIT, 11 of which displayed more than one such mutation in
7 the particular examined biopsies (Figure 1). Most of these were point mutations in exon 13
8 (V654A) or exon 17 (involving codons 820, 822, or 823). However, we also found less common
9 mutations in amino acids D677, C809, and S840.
10 Next, we queried our next generation sequencing patient data (n=109) for mutations in KIT
11 downstream effectors, such as NF1, N/H/KRAS, BRAF, PTEN, PIK3CA, AKT, and TSC1/2. We
12 found that a subset of patients (18%; 20/109) displayed such downstream mutations, potentially
13 causing resistance to KIT/PDGFRA-inhibitors (Tables 1 and 2). Notably, 10% (3/31) of patients
14 with primary PDGFRA mutations displayed mutations in downstream signaling intermediates,
15 indicating a similar incidence of these events as in KIT mutated GIST. Except for the BRAF-
16 mutated cases, all patients with downstream effector mutations also displayed activating
17 mutations in KIT or PDGFRA, most of which were accompanied by secondary resistance
18 mutations (13/18; Tables 1 and 2).
19 Thirteen mostly heavily pretreated (median treatment lines = 4) patients with TKI-resistant GIST
20 were recruited into the DKTK MASTER molecular stratification program (19) in which tumors are
21 analyzed by whole exome/genome and RNA sequencing to identify clinically actionable
22 aberrations. In these datasets, we identified primarily alterations in genes that are involved in the
23 PI3K or MAPK signaling pathways (Table 2). Notably, no driver mutations apart from KIT and its
24 downstream pathways were detected (Supplemental Figure S1). Furthermore, RNA-Seq data
25 revealed mostly lineage-specific markers among the highest ranking transcripts (KIT, PDGFRA,
26 ETV1; Table 2).

28 Novel cell line panel approximates genetic heterogeneity of GIST patients
29 To expand our panel of models recapitulating the heterogeneity of resistance in GIST we applied
30 CRISPR/Cas9 mediated gene editing and knockout. We thus induced heterozygous point
31 mutations of KIT in exon 13 (V654A) and exon 17 (D816A), respectively, in addition to the
32 endogenous primary exon 11 mutation in GIST-T1. As expected, these cells displayed a much
33 higher resistance to IM as well as an inhibitory profile towards the approved second- and third
34 line KIT-inhibitors sunitinib (SU) and regorafenib (RE), matching their respective secondary
35 mutations (Table 3).
10

1 To investigate the effect of oncogenic mutations in KIT-downstream signaling intermediates on
2 KIT-inhibitor sensitivity, we generated further GIST-T1 sublines: T1/G12R-HOM/HET harboring
3 homozygous and heterozygous G12R mutations in KRAS, respectively, as well as T1-KRAS
4 with a heterozygous KRAS deletion (see methods; Figure 2). The cell lines T1-PTEN, T1-TSC2,
5 and T1-NF1 carry homozygous knockout of PTEN, TSC2, and NF1, respectively. While TSC2
6 and NF1 deficient cells were completely resistant to KIT inhibition, cells with loss of PTEN
7 displayed some sensitivity to IM treatment, which still efficiently inhibits KIT-dependent MEK
8 signaling (Table 3). However, these cells would still continue to grow, albeit slower, at high IM
9 concentrations (10µM; Suppl. Figure S2). Interestingly the heterozygous mutation of KRASG12R
10 did not cause a notable increase in tolerance to KIT inhibition, while cells carrying homozygous
11 KRASG12R were, similar to T1-KRAS, partly resistant (Suppl. Figure S3). As depicted in Table 3,
12 cell lines harboring mutations in KIT or its downstream signaling intermediates are highly
13 resistant to all currently approved GIST treatments.
14
15 Sapanisertib has antiproliferative effects in IM-sensitive and IM-resistant GIST cell lines
16 We then sought to evaluate the therapeutic potential of KIT-downstream inhibition using the
17 mTOR kinase inhibitor sapanisertib. In cell viability assays after 3 days of treatment, sapanisertib
18 displayed IC50 values between 20nM (GIST430/654) and 70nM (T1-G12R) (Figure 3A).
19 Sensitivity towards sapanisertib was independent of secondary mutations, sensitivity to IM, and
20 KIT expression (GIST48B). Strikingly, cell lines harboring KIT-independent resistance mutations
21 in KRAS, PTEN, TSC2 and NF1 were similarly sensitive to sapanisertib treatment.

23 Sapanisertib efficiently abrogates mTORC1/2 signaling
24 To elucidate the effects of sapanisertib on intracellular signaling we conducted western blot
25 experiments in GIST cell lines of different origins. We observed a dose-dependent inhibition of
26 ribosomal protein S6 phosphorylation (pS6; as marker for mTOR activation) starting at 1 - 5nM
27 with complete inhibition at 50-100nM in all cell lines (Figure 2b). Notably, in this concentration
28 range sapanisertib mediated inhibition of mTORC1 led to a strong inhibition of AKT
29 phosphorylation (pAKT), followed by loss of 4E-BP1 phosphorylation (p4E-BP1). In contrast,
30 20nM everolimus (RAD001), while also potently inhibiting pS6, did not inhibit pAKT or p4e-BP1,
31 but instead increased their phosphorylation levels. Interestingly, sapanisertib treatment,
32 especially at higher concentrations, led to a dose dependent increase of ERK1/2
33 phosphorylation (pERK; Figure 2b).

1 Co-treatment inhibits feedback induction of MEK/ERK-signaling

2 mTOR inhibition alone appeared to leave the cells with an escape route via MEK/ERK driven
3 survival, as indicated by induction of pERK1/2 (Figure 2b). Therefore, we combined sapanisertib
4 either with KIT-inhibition or with the clinical MEK inhibitor trametinib (Figure 3). Combinations
5 with KIT-inhibitors displayed the strongest effects in cell lines with KIT-downstream mutations,
6 which was to be expected as GIST-T1-PTEN and KRAS-mutated sublines displayed residual
7 sensitivity to IM alone (Table 3, Suppl. figures S2 and S3). Strikingly, in GIST-T1-TSC2,
8 combination of sapanisertib and IM completely abrogated S6 and 4E-BP1 phosphorylation, even
9 at sapanisertib doses as low as 10nM, whereas IM alone did not inhibit S6 and 4E-BP1
1 phosphorylation. These cells displayed reduced baseline KIT-expression which increased after
2 IM-treatment (Figures 2 and 3). Combinations of sapanisertib with trametinib completely
3 abrogated both KIT-downstream signaling axes in all investigated cell lines.

14 Combinational treatment with KIT- or MEK-inhibitors displays synergistic effects
15 To further elucidate the potential of combinational treatment, we next conducted multi-dose
16 combination proliferation experiments with sapanisertib, combined with either KIT-inhibition or
17 trametinib. Cells were treated with 5-100nM of each inhibitor and with each possible combination
18 of two drugs. We then calculated the combinatorial Index (CI) for each combination according to
19 the Chou-Talalay method, which describes synergy at CI < 1, additivity at CI = 1, and
20 antagonism at CI > 1. We found that concentrations of sapanisertib between 25nM – 100nM
21 yielded the strongest combinatorial effects in all cell lines (Table 4). In most cell lines
22 combinations with IM yielded at best moderate additive effects (CI: 0.5 – 0.58) at low
23 concentrations of IM (50-200nM). However, in TSC2 deficient cells the combination displayed a
24 strong synergy signified by the lowest CI-value of 0.26 (Table 4). In the sublines harboring
25 secondary KIT mutations (V654A and D816A) cells were treated with sunitinib and ponatinib,
26 respectively. These combinations also had moderate synergistic effects.
27 Combinations of sapanisertib and trametinib showed strong synergistic effects in all examined
28 cell lines (lowest CI values: 0.126 – 0.37; Table 3; Suppl. Figure S3). Notably, the lowest CI-
29 value overall of 0.126 was obtained in TSC2 deficient cells when sapanisertib 25nM was
30 combined with trametinib 200nM. However, even at lower concentrations of trametinib 25nM and
31 50nM, the combination with sapanisertib 25nM had strong synergistic effects (CI-values 0.33
32 and 0.19, respectively; Suppl. Figure S3). All calculated CI-values, as well as the results of the
33 cytotoxicity experiments they were generated from are depicted in Supplemental Figure S3.

1 Discussion

2 Activating mutations of KIT or PDGFRA are the oncogenic hallmarks of GIST that lead to ligand-
3 independent downstream activation of the PI3K and RAS/RAF/MAPK signaling cascades (2,34).
4 Inhibition of KIT by imatinib abrogates KIT phosphorylation and consequently signaling through
5 these pathways. While most GISTs exhibit long-lasting responses to KIT-inhibitor therapy the
6 majority of patients eventually progress. Secondary mutations in KIT have been identified as the
7 main mechanism of resistance in resection specimens of patients failing imatinib (35). This is
8 accompanied by re-activation of PI3K and RAS/RAF/MAPK signaling. Notably, these two
9 pathways seem to be required for GIST homeostasis and proliferation regardless of the
1 presence of KIT or PDGFRA mutations – thus defining crucial tissue-specific pathways.
2 Therefore, in a broader sense, GIST could be defined not by KIT activation itself, but rather by
3 conjoined activation of its downstream signaling intermediates (11).
4 Therapeutic success in GIST is therefore not defined only by successful inhibition of KIT but, in
5 extension, by abrogation of KIT downstream signaling. Sustained therapeutic KIT inhibition is
6 confounded by the heterogeneity of secondary KIT IM-resistance mutations, while activity of
7 therapies targeting KIT downstream signaling might depend on the types of mutations activating
8 these downstream pathways. Very recently, genomic activation of KIT-downstream pathways
9 has even been observed in treatment-naïve GIST, further underscoring the clinical relevance of
10 KIT-independent mutations within these pathways (12,36).
11 In an international collaborative effort we interrogated the pathology databases of several GIST
12 centers, and thus compiled a representative spectrum and incidence of secondary KIT mutations
13 in a large cohort of TKI-refractory GIST. Interestingly, we found a lower incidence of exon 14
14 mutations compared to reports from the early 2000s (7,35). This may be reflective of the
15 availability of additional KIT inhibitors in recent years which effectively inhibit the exon 14
16 gatekeeper mutations. Also, there could be technical bias, as KIT exon 14 was not sequenced in
17 the early years of routine pathological diagnosis in all centers, these patients may have been
18 falsely classified as not carrying a secondary KIT mutation. Our data indicate the need for more
19 potent KIT-inhibitors, with activity against the full spectrum of resistance mutations. Of note,
20 latest generation KIT-inhibitors show broader and more specific inhibitory profiles (37,38)
21 although no drug has yet been shown to be equally and universally potent against all KIT
22 mutations.
23 Recently, next generation panel sequencing was introduced into routine pathological analysis in
24 many centers, covering genes whose activation or loss could confer resistance apart from
25 secondary KIT mutations. We were thus able show the incidence of non-KIT mutations in
26 random samples of TKI-resistant GIST specimens sent for genotyping. In a fraction larger than

1 expected, these comprise KIT-downstream signaling intermediates in the PI3K/AKT/mTOR and
2 RAS/RAF/MAPK pathways. Of note, most mutations of tumor suppressor genes appeared
3 heterozygous in this cohort, which might result from homozygous mutations present in only a
4 subset of the neoplastic cells, which would not be surprising given that these were generally
5 secondary KIT/PDGFRA-inhibitor resistance mutations. Another possibility is that inactivating
6 mutations in the remaining tumor suppressor allele were often present but undetected, which
7 can occur in the instance of large indels or promoter-region mutations. We also cannot exclude
8 the possibility that some inactivations were truly heterozygous, and that haploinsufficiency in this
9 context is biologically meaningful.
10 Although, this “tertiary resistance” has rarely been reported yet our results underscore the
11 importance and functional relevance of mutations in these pathways (11,12). We speculate that
12 these mutations are frequently not tested for or even not added to routine pathology reports – ,
13 despite the availiability of the data, e.g. when only KIT sequencing is requested. Given the
14 advent of broader KIT inhibitors, we expect that these novel mechanisms represent a clinically
15 relevant cause of treatment resistance, as GIST cells are crucially depending on the PI3K and
16 RAS/RAF-MAPK pathways. Future studies, employing plasma sequencing with sequencing
17 panels optimized for GIST will most likely reveal these mutations to be more common than
18 previously expected.
19 To our surprise, comprehensive DNA and RNA sequencing in a subset of patients revealed no
20 other apparent driver mutations that may have replaced KIT signaling as the dominant
21 oncogenic pathway. This finding underscores the requirement for concomitant activation of PI3K
22 and MAPK signaling GIST cell homeostasis. Inhibiting KIT downstream signaling may therefore
23 prove to be a necessary, effective, and actionable strategy.
24 To help validate our hypotheses, we expanded our panel of GIST cell lines, to model
25 mechanisms of the complex heterogeneity observed in the clinic. Employing CRISPR/Cas9
26 methods we induced specific point mutations in KIT, and KRAS and thus show for the first time
27 that precise genomic editing in GIST cell lines is possible and is a valuable tool to generate
28 clinically relevant models. Up to now, KIT-inhibitor studies for GIST were often conducted in
29 Ba/F3 cells, lacking the GIST specific cellular background (39). In fact, perturbations of GIST-
30 specific KIT-downstream signaling are probably not optimally modelled in those systems. To
31 date, virtually no in vitro studies have been conducted in GIST harboring mutations in signaling
32 intermediates downstream of KIT (40).
33 Furthermore, our newly generated cell lines will not only enable improved inhibitor research but
34 may also yield relevant insights into GIST biology. Of note, we observed that KIT expression
35 decreased in GIST-T1 sublines with mutations in KRAS, NF1, and TSC2, but not PTEN.

1 Inhibition of mTOR upon sapanisertib treatment subsequently increased the levels of total KIT
2 (Figures 2, 3). Especially in cells lacking TSC2, this reactivation also reinstated KIT-dependence
3 and thus sensitivity to IM treatment (Figure 3, Table 4). These findings indicate that KIT-
4 independent mutations may supplant the role of KIT and may impact KIT expression levels. Loss
5 of KIT has been occasionally observed in clinical specimens and also in GIST cell lines grown in
6 vitro (41). In this context our modified sublines may serve as informative models to better
7 understand the feedback-regulation of KIT by its main intracellular signaling pathways. We were
8 furthermore surprised to find that, at least in our GIST-T1 derived cell line model, the
9 heterozygous mutation of KRASG12R was not able to confer KIT-inhibitor resistance. We assume
10 that GIST-T1 is particularly dependent on the RAS/RAF/MEK signal so that the activation of a
11 single allele does not compensate for the complete block of the upstream KIT-dependent signal.
12 In other cancers with KRAS-mediated mechanisms of TKI-resistance very little data is published
13 on the zygosity of secondary mutations. Notably, Serrano et al. recently reported a KRASG12R
14 mutated resistant GIST clone bearing a hemizygous mutation (12).
15 Targeting mTOR has been a strategy in clinical trials based on the observation that PI3K-
16 activation is a signaling hallmark in GIST regardless of the presence of secondary resistance
17 mutations (42). However, clinical success may have been hampered by pharmacological
18 interactions as well as the selection of a refractory treatment setting, in which imatinib was
19 unlikely to inhibit clones with secondary resistance mutations (43). For imatinib-resistant clones,
20 mTOR inhibition (everolimus) alone is most likely not sufficient to fully control tumor growth and
21 it may require a combination with a broader KIT-inhibitor or with inhibitors of MEK. Based on
22 preclinical findings in GIST (44), imatinib in combination with MEK is currently being tested in the
23 clinic (NCT01991379). Another approach based on the in vivo studies by van Looy et. al (45)
24 recently looked at the combination of imatinib and the PI3K-inhibitor Alpelisib (BYL719).
25 Unfortunately, the results of this study are not yet available.
26 In contrast to everolimus, sapanisertib inhibits not only mTORC1, but also mTORC2 (46). This
27 distinct inhibitory profile, causing a strong decrease of AKT- as well as 4E-BP1-phosphorylation,
28 may yield superior clinical efficacy. Notably, compared to other rapalogs, which also inhibit
29 mTORC2, sapanisertib has been shown to only cause grade 1 and 2 hyperglycemia and only in
30 a subset of patients (17,47). We now report that sapanisertib has strong anti-proliferative effects
31 in IM-sensitive and IM-resistant cell lines, including KIT-negative GIST. As Slotkin et al. have
32 shown sapanisertib has antitumor effects in a panel of bone and soft tissue sarcoma cell lines
33 and xenograft models (48). In their study, cell lines displaying in vitro IC50 values similar to the
34 ones described herein were also inhibited in vivo at a dose of 1mg/kg/day. Furthermore,
35 effective concentrations for this drug against the IM-resistant GIST models are well within the
1 range of clinically achievable plasma levels (17,18). Currently sapanisertib is investigated in
2 several clinical phase 2 trials, in entities including lung cancer, acute lymphoblastic leukemia and
3 soft tissue sarcomas (clinicaltrials.gov).
4 However, RAS/RAF/MAPK signaling, unperturbed by sapanisertib, is similarly important for
5 GIST cell proliferation and survival (42). Combinations of sapanisertib with approved KIT
6 inhibitors display moderate synergistic effects and may represent a feasible clinical strategy,
7 which warrants further investigation. To date, MEK inhibitors show clinical toxicity profiles
8 requiring careful management in combination therapies (49,50). We found strong synergistic
9 effects when combining sapanisertib with trametinib, which was to be expected as this
10 combination inhibits the two major routes of GIST proliferation and survival. These strong
11 combinational effects might allow for dose reduction of one or both drugs which may reduce side
12 effects and thus become more attractive to patients, especially to those with non-KIT resistance
13 mutations. Although it is possible that the particular prototypic compounds or their combinations
14 selected for our in vitro studies may display in vivo toxicity, we are convinced that inhibition of
15 the two major oncogenic signaling axes in GIST will prove to be a clinically feasible treatment
16 option. We speculate that in a disease exceptionally dependent on these pathways, such as
17 GIST, the therapeutic window of such drug combinations may be even more favorable than in
18 other cancers.
19 In summary, our data strongly underscore the need for comprehensive sequencing of KIT as
20 well as of KIT-related signaling molecules that may contribute to KIT-inhibitor resistance in GIST.
21 With the advent of more potent KIT-inhibitory molecules we hypothesize that mutations of genes
22 coding for KIT-downstream signaling intermediates will become more prevalent. Future
23 treatment strategies, both in untreated and pretreated GIST may benefit from integrating potent
24 inhibitors of these pathways. The novel cell lines presented herein may provide meaningful
25 models for the validation of such new drug combinations.26

Acknowledgments
The authors sincerely thank Miriam Christoff for the expert technical assistance.

1 References
2
3 1. Hirota S, Isozaki K, Moriyama Y, Hashimoto K, Nishida T, Ishiguro S, et al. Gain-of-function
4 mutations of c-kit in human gastrointestinal stromal tumors. Science 1998;279(5350):577-80.
5 2. Heinrich MC, Corless CL, Duensing A, McGreevey L, Chen CJ, Joseph N, et al. PDGFRA activating
6 mutations in gastrointestinal stromal tumors. Science 2003;299(5607):708-10.
7 3. Verweij J, Casali PG, Zalcberg J, LeCesne A, Reichardt P, Blay JY, et al. Progression-free survival in
8 gastrointestinal stromal tumours with high-dose imatinib: randomised trial. Lancet 9 2004;364(9440):1127-34.
10 4. Blanke CD, Demetri GD, Von Mehren M, Heinrich MC, Eisenberg B, Fletcher JA, et al. Long-term
11 results from a randomized phase II trial of standard- versus higher-dose imatinib mesylate for
12 patients with unresectable or metastatic gastrointestinal stromal tumors expressing KIT. J Clin 13 Oncol 2008;26(4):620-25.
14 5. Demetri GD, Reichardt P, Kang YK, Blay JY, Rutkowski P, Gelderblom H, et al. Efficacy and safety
15 of regorafenib for advanced gastrointestinal stromal tumours after failure of imatinib and
16 sunitinib (GRID): an international, multicentre, randomised, placebo-controlled, phase 3 trial. 17 Lancet 2012;381(9863):295-302.
18 6. Heinrich MC, Corless CL, Blanke CD, Demetri GD, Joensuu H, Roberts PJ, et al. Molecular
19 Correlates of Imatinib Resistance in Gastrointestinal Stromal Tumors. Journal of Clinical Oncology 20 2006;24(29):4764-74.
21 7. Heinrich MC, Maki RG, Corless CL, Antonescu CR, Harlow A, Griffith D, et al. Primary and
22 Secondary Kinase Genotypes Correlate With the Biological and Clinical Activity of Sunitinib in
23 Imatinib-Resistant Gastrointestinal Stromal Tumor. Journal of Clinical Oncology 24 2008;26(33):5352-59.
25 8. Wardelmann E, Merkelbach-Bruse S, Pauls K, Thomas N, Schildhaus HU, Heinicke T, et al.
26 Polyclonal evolution of multiple secondary KIT mutations in gastrointestinal stromal tumors
27 under treatment with imatinib mesylate. Clin Cancer Res 2006;12(6):1743-49.
28 9. Agaimy A, Wunsch PH, Hofstaedter F, Blaszyk H, Rummele P, Gaumann A, et al. Minute gastric
29 sclerosing stromal tumors (GIST tumorlets) are common in adults and Sapanisertib frequently show c-KIT
30 mutations. Am J Surg Pathol 2007;31(1):113-20.
31 10. Kawanowa K, Sakuma Y, Sakurai S, Hishima T, Iwasaki Y, Saito K, et al. High incidence of
32 microscopic gastrointestinal stromal tumors in the stomach. Hum Pathol 2006;37(12):1527-35.
33 11. Li B, Garcia CS, Marino-Enriquez A, Grunewald S, Wang Y, Bahri N, et al. Conjoined
34 hyperactivation of the RAS and PI3K pathways in advanced GIST. Journal of Clinical Oncology 35 2016;34(15_suppl):e22520-e20.
36 12. Serrano C, Wang Y, Marino-Enriquez A, Lee JC, Ravegnini G, Morgan JA, et al. KRAS and KIT
37 Gatekeeper Mutations Confer Polyclonal Primary Imatinib Resistance in GI Stromal Tumors:
38 Relevance of Concomitant Phosphatidylinositol 3-Kinase/AKT Dysregulation. J Clin Oncol 2014.
39 13. Ran FA, Hsu PD, Wright J, Agarwala V, Scott DA, Zhang F. Genome engineering using the CRISPR-
40 Cas9 system. Nature protocols 2013;8(11):2281-308.
41 14. Hsieh AC, Liu Y, Edlind MP, Ingolia NT, Janes MR, Sher A, et al. The translational landscape of
42 mTOR signalling steers cancer initiation and metastasis. Nature 2012;485(7396):55-61.
43 15. Hassan B, Akcakanat A, Sangai T, Evans KW, Adkins F, Eterovic AK, et al. Catalytic mTOR inhibitors
44 can overcome intrinsic and acquired resistance to allosteric mTOR inhibitors. Oncotarget 45 2014;5(18):8544-57.
46 16. Sun SY. mTOR kinase inhibitors as potential cancer therapeutic drugs. Cancer Lett 2013;340(1):1- 47 8.
48 17. Ghobrial IM, Siegel DS, Vij R, Berdeja JG, Richardson PG, Neuwirth R, et al. TAK-228 (formerly
49 MLN0128), an investigational oral dual TORC1/2 inhibitor: A phase I dose escalation study in
17

1 patients with relapsed or refractory multiple myeloma, non-Hodgkin lymphoma, or
2 Waldenstrom’s macroglobulinemia. American journal of hematology 2016;91(4):400-5.
3 18. Burris HA, 3rd, Kurkjian CD, Hart L, Pant S, Murphy PB, Jones SF, et al. TAK-228 (formerly
4 MLN0128), an investigational dual TORC1/2 inhibitor plus paclitaxel, with/without trastuzumab,
5 in patients with advanced solid malignancies. Cancer chemotherapy and pharmacology 6 2017;80(2):261-73.
7 19. Horak P, Klink B, Heining C, Groschel S, Hutter B, Frohlich M, et al. Precision oncology based on
8 omics data: The NCT Heidelberg experience. Int J Cancer 2017;141(5):877-86.
9 20. Beadling C, Neff TL, Heinrich MC, Rhodes K, Thornton M, Leamon J, et al. Combining highly
10 multiplexed PCR with semiconductor-based sequencing for rapid cancer genotyping. J Mol Diagn 11 2013;15(2):171-76.
12 21. Heining C, Horak P, Uhrig S, Codo PL, Klink B, Hutter B, et al. NRG1 Fusions in KRAS Wild-Type
13 Pancreatic Cancer. Cancer Discov 2018.
14 22. Chudasama P, Mughal SS, Sanders MA, Hubschmann D, Chung I, Deeg KI, et al. Integrative
15 genomic and transcriptomic analysis of leiomyosarcoma. Nat Commun 2018;9(1):144.
16 23. Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, et al. The Sequence Alignment/Map
17 format and SAMtools. Bioinformatics 2009;25(16):2078-9.
18 24. Kordes M, Roring M, Heining C, Braun S, Hutter B, Richter D, et al. Cooperation of BRAF(F595L)
19 and mutant HRAS in histiocytic sarcoma provides new insights into oncogenic BRAF signaling. 20 Leukemia 2016;30(4):937-46.
21 25. Talevich E, Shain AH, Botton T, Bastian BC. CNVkit: Genome-Wide Copy Number Detection and
22 Visualization from Targeted DNA Sequencing. PLoS Comput Biol 2016;12(4):e1004873.
23 26. Kleinheinz K, Bludau I, Huebschmann D, Heinold M, Kensche P, Gu Z, et al. ACEseq – allele specific
24 copy number estimation from whole genome sequencing. bioRxiv 2017.
25 27. Wang J, Mullighan CG, Easton J, Roberts S, Heatley SL, Ma J, et al. CREST maps somatic structural
26 variation in cancer genomes with base-pair resolution. Nat Methods 2011;8(8):652-4.
27 28. Quinlan AR, Hall IM. BEDTools: a flexible suite of utilities for comparing genomic features. 28 Bioinformatics 2010;26(6):841-2.
29 29. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA-
30 seq aligner. Bioinformatics 2013;29(1):15-21.
31 30. Haeussler M, Schonig K, Eckert H, Eschstruth A, Mianne J, Renaud JB, et al. Evaluation of off-
32 target and on-target scoring algorithms and integration into the guide RNA selection tool
33 CRISPOR. Genome biology 2016;17(1):148.
34 31. Chen B, Gilbert LA, Cimini BA, Schnitzbauer J, Zhang W, Li GW, et al. Dynamic imaging of genomic
35 loci in living human cells by an optimized CRISPR/Cas system. Cell 2013;155(7):1479-91.
36 32. Muhlenberg T, Grunewald S, Treckmann J, Podleska L, Schuler M, Fletcher JA, et al. Inhibition of
37 KIT-glycosylation by 2-deoxyglucose abrogates KIT-signaling and combination with ABT-263
38 synergistically induces apoptosis in gastrointestinal stromal tumor. Plos One 39 2015;10(3):e0120531.
40 33. Chou TC, Talalay P. Quantitative analysis of dose-effect relationships: the combined effects of
41 multiple drugs or enzyme inhibitors. Advances in enzyme regulation 1984;22:27-55.
42 34. Hirota S, Nishida T, Isozaki K, Taniguchi M, Nakamura J, Okazaki T, et al. Gain-of-function
43 mutation at the extracellular domain of KIT in gastrointestinal stromal tumours. JPathol 44 2001;193(4):505-10.
45 35. Heinrich MC, Corless CL, Blanke CD, Demetri GD, Joensuu H, Roberts PJ, et al. Molecular
46 correlates of imatinib resistance in gastrointestinal stromal tumors. J Clin Oncol 47 2006;24(29):4764-74.