ORIGINAL ARTICLE Gene Expression Signatures Associated with The

Home , ACSM3, ANTXR2, CCNB1IP1, Exoribonuclease, FBXO9, LMAN1

Gene expression signatures associated with the resistance to imatinib

Y-J Chung1,6, T-M Kim1,6, D-W Kim2, H Namkoong3, HK Kim3, S-A Ha3, S Kim3, SM Shin3, J-H Kim4, Y-J Lee4, H-M Kang1 and JW Kim3,5

1Department of Microbiology, College of Medicine, The Catholic University of Korea, Seoul, Korea; 2Department of Internal medicine, College of Medicine, The Catholic University of Korea, Seoul, Korea; 3Molecular Genetic Laboratory, College of Medicine, The Catholic University of Korea, Seoul, Korea; 4Seoul National University Biomedical Informatics (SNUBI) and Human Genome Research Institute, College of Medicine, Seoul National University, Seoul, Korea; and 5Department of Obstetrics and Gynecology, College of Medicine, The Catholic University of Korea, Seoul, Korea

Imatinib (imatinib mesylate, STI-571, Gleevec) is a selective Transcriptional amplification or point mutations in the BCR-ABL tyrosine kinase inhibitor that has been used as a tyrosine kinase domain of BCR-ABL have been frequently highly effective chemoagent for treating chronic myelogenous 3–5 leukemia. However, the initial response to imatinib is often observed in resistant clinical cases or in vitro cell models. followed by the recurrence of a resistant form of the disease, Although these changes are considered almost sufficient cause which is major obstacle to many therapeutic modalities. The of obtaining imatinib-resistance, previous studies have also aim of this study was to identify the gene expression signatures shown that resistance can occur without any apparent ampli- that confer resistance to imatinib. A series of four resistant fication or point mutations in BCR-ABL.6–8 In addition, the K562 sublines was established with different imatinib dosage overexpression of multidrug resistance genes has been suggested (200, 400, 600 and 800 nM) and analyzed using microarray 9,10 technology. The transcripts of the genes showing universal or to confer the resistance to imatinib. These reports suggest dose-dependent expression changes across the resistant that heterogeneous mechanisms might be responsible for sublines were identified. The gene sets associated with the imatinib-resistance. To identify key genetic elements responsi- imatinib-resistance were also identified using gene set enrich- ble for imatinib-resistance, various approaches including global ment analysis. In the resistant K562 sublines, the transcription- gene expression analysis using a microarray have been and apoptosis-related expression signatures were upregulated, used.7,11,12 whereas those related to the protein and energy metabolism were downregulated. Several genes identified in this study With the aim of deciphering gene expression signatures such as IGF1 and RAB11A have the potential to become associated with imatinib-resistance, this study focused on the surrogate markers useful in a clinical evaluation of imatinib- transcripts showing biologically relevant expression changes by resistant patients without BCR-ABL mutation. The expression examining the expression profiles across four imatinib-resistant signatures identified in this study provide insights into the sublines established with different imatinib dosage and primary mechanism of imatinib-resistance and are expected to facilitate CML cases. We present expression profiles and candidate signal the development of an effective diagnostic and therapeutic strategy. pathways associated with imatinib-resistance through improved Leukemia (2006) 20, 1542–1550. doi:10.1038/sj.leu.2404310; bioinformatical approaches including gene set enrichment published online 20 July 2006 analysis. Keywords: drug resistance; expression array; imatinib mesylate; STI-571 Materials and methods

Establishment of imatinib-resistant cell lines The resistance to imatinib was established by exposing the erythroid leukemic K562 cell lines to increasing imatinib Introduction concentrations by 50 nM every 2 weeks. During the culture with imatinib, a series of four resistant sublines exposed to Imatinib (imatinib mesylate, STI-571, Gleevec) is a selective various imatinib concentrations (200, 400, 600 and 800 nM) tyrosine kinase inhibitor that has been successfully used to treat were established. Any mutations of the BCR-ABL kinase domain chronic myloid leukemia (CML). Targeting to BCR-ABL fusion were screened as described previously.4 In addition, the kinase, the primary pathogenic molecule for disease, imatinib expression level of BCR-ABL kinase was examined using real- achieved remarkable remission rates up to B80%, especially for 1 time reverse transcriptase-polymerase chain reaction (PCR) with initial phase of CML. However, relapse after the initial the same primers used for mutation analysis and for human hematologic and cytologenetic response frequently occurred 2 glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as inter- in late-stage disease. This suggests that BCR-ABL targeted nal control. therapy can be hindered by the acquisition of resistance, which seriously reduces the clinical efficacy of imatinib. RNA preparation and hybridization Correspondence: Professor JW Kim, Molecular Genetic Laboratory, Total RNA of K562 and imatinib-resistant sublines was isolated College of Medicine, The Catholic University of Korea, Seoul 137- with Trizol (Invitrogen, Carlsbad, CA) according to the 040, Korea. manufacturer’s instruction. The quantity and quality of extracted E-mail: [email protected] 6These authors contributed equally to this work RNA was assessed by using a Bioanalyzer 2100 (Agilent Received 20 January 2006; revised 23 May 2006; accepted 6 June Technologies, Waldbronn, Germany). Applied Biosystems Human 2006; published online 20 July 2006 Genome Survey Microarray Version 2.0 representing a set Expression signatures associated with imatinib-resistant phenotype Y-J Chung et al 1543 of B30 000 human genes were used to analyze the expression rate methods.17 The transcripts significant (qo0.05) for up- and profiles of the five cell lines (parental sensitive K562 and four downprogression were further tested for the significance in 600 resistant sublines). Digoxigenin-UTP-labeled cRNA was gener- and 800 nM t-tests with Welch correction and examined to be ated and linearly amplified from 5 mg of total RNA using Applied coordinately changed across resistant sublines. Detailed de- Biosystems Chemiluminescent RT-IVT Labeling Kit V2.0. Array scription with used parameters is available in Supplementary hybridization, Chemiluminescence detection, image acquisition note 1. and analysis were performed using Applied Biosystems Chemi- luminescence Detection Kit and Applied Biosystems 1700 Chemiluminescent Microarray Analyzer according to the Gene set enrichment analysis manufacturer’s protocol. The hybridization reaction was tripli- Functionally related genes were retrieved from public gene cated per sample. database: Gene Ontology, BioCarta (http://www.biocarta.com), KEGG (Kyoto Encyclopedia of Genes and Genomes) and GenMAPP (Gene MicroArray Pathway Profiler).18–20 According Data processing to gene-versus-annotation matches, genes were categorized into Image of each array hybridization was collected using the 1700 annotation-specified gene sets. A total of 3774 gene sets were analyzer, which is equipped with high-resolution, large-format collected, whose members have a common functional annota- CCD camera, including two ‘short’ chemiluminescent images tion and are not necessarily exclusive. Enrichment analysis was with 5 s exposure length for gene expression analysis, two performed using parametrical analysis of gene enrichment that fluorescent images for feature finding and spot normalization directly calculates the significance of enrichment based on and 2QC images for spectrum cross-talk correction. Images Z-statistics.21 To perform enrichment analysis, we further were auto-gridded and the chemiluminescent signals were restricted initial gene sets into 721 sets containing at least 10 quantified. After the background subtraction, spot intensity data genes in minimal gene set.21 We used two kinds of parametrical were normalized by using variance stabilization and normal- 13 values, relative difference r.d. and Fk to calculate the signifi- ization method. cance level based on Z-statistics. Total six enrichment analyses were performed using four sets of r.d. representing comparisons between four resistant sublines and control K562 and two sets of Determination of differentially expressed transcripts One-way analysis of variance analysis was performed on the Fk representing up- and downprogression. We determined average expression of 29 098 genes across five cell lines activated gene sets when they are significantly enriched in top including those of parental K562 cell lines. The most variable r.d. for both comparisons of 600 and 800 nM and also in top Fk in 3000 transcripts were chosen and hierarchical clustering via upprogression and vice versa for repressed gene sets. average linking was performed using custom software, Cluster and Treeview (http://rana.lbl.gov/EisenSoftware.htm).14 To determine whether a transcript is up- or downregulated in resistant Real-time quantitative PCR assay sublines, the expression change compared to sensitive parental Total RNA from blood samples of six CML patients before and control was measured by calculating the relative difference, r.d. after acquiring imatinib-resistance (kindly donated from Korean as following: Leukemia Cell and Gene Bank) was used for real-time quantitative PCR (QPCR) analysis for six candidate genes; CASP4, TNFRF21, RAB11A, IGF1, LYN and RHOA. Three of xrsðiÞÀxcontðiÞ r:d:ðiÞ the patients have BCR-ABL mutation and the other three have s ðiÞþs ðiÞ rs cont no mutation. The first-strand cDNA was synthesized using Moloney-murine leukemia virus Reverse Transcriptase (Invitro- The xrsðiÞ and srsðiÞ stand for the mean and s.d. of (i) gene gen, USA) and used for PCR. Real-time QPCR was performed expression in resistant sublines and xcontðiÞ and scontðiÞ are those of sensitive parental cell lines, respectively. We used arbitrary using Mx3000P QPCR system with software MxPro Version 3.00 cutoff of 74 r.d. to select up- or downregulated transcripts in (Stratagene, La Jolla, CA, USA). The real-time QPCR mixture of resistant sublines. Array processing and calculations were 20 ml contains 10 ng of cDNA, 1 Â SYBR Green Tbr polymerase performed using Microsoft Excel and R-package 6.0 (http:// mixture (FINNZYMES, Finland), 0.5 Â ROX and primers of www.r-project.org/). 20 pmol. GAPDH was used as an internal control in each procedure. The thermal cycling was as follows: 10 min at 951C, followed by 40 cycles of 10 s at 941C, 30 s at 55–601C and 30 s Identification of transcripts with dose-dependent at 721C. To verify specific amplification, melting curve analysis expression change was performed (55–951C, 0.51C/s). Relative quantification was 22 To identify genes whose expression changes in dose-dependent performed by the DDCT method. The sequence information of manner in resistant sublines, we combined two statistical primers used for real-time QPCR is available in Supplementary approaches.15 First, five mean expression levels of control and note 2. four resistant sublines were considered as continuum corresponding to imatinib dose increment from 200 to 800 nM. The significance level of up- and downprogression was deter- Results mined under the alternative hypotheses, H1: mcontpm200 nM pm400 nMpm600 nMpm800 nM or the vice versa,H1: mcontX Expression profile of imatinib-resistant K562 sublines m200 nMXm400 nMXm600 nMXm800 nM, respectively. We used Four imatinib-resistant K562 sublines are denoted as KR200 nM, Bartholomew’s homogeneity law for ordered alternatives, and KR400 nM,KR600 nM and KR800 nM according to the imatinib doses modified F-statistics for comparison of more than two groups.16 used to establish the corresponding sublines. We first examined Multiple testing adjustment was performed using Q-value whether or not the imatinib-resistance of the sublines was owing package (http://faculty.washington.edu/~jstorey/qvalue/) that to the increased transcription or from a point mutation of the converts obtained P-value into q-value based on false discovery BCR-ABL kinase domain, which were commonly reported in

Leukemia Expression signatures associated with imatinib-resistant phenotype Y-J Chung et al 1544 previous studies.4 Neither a point mutation at the known hot Identification of transcripts universally changed across spots in the kinase domain nor transcriptional upregulation of resistant sublines the BCR-ABL gene was observed in the four resistant sublines The transcripts universally up- or downregulated in all four (data not shown). Then, the expression profiles of the K562 and resistant sublines relative to the K562 control were first imatinib-resistant sublines were examined using an oligo- investigated. Using the relative difference of 74 r.d. as cutoff, microarray encompassing B30 000 human genes. Unsuper- 16 and 11 universally changed transcripts were found to be vised clustering of the 3000 most variable transcripts discrimi- over- or underexpressed (Table 1). Apoptosis-related genes such nated the sublines by imatinib dosage used to induce resistance as CASP4, TNFRSF21 and TNFRSF9 were universally upregu- (Figure 1). According to the clustering, the expression profiles of lated in the resistant sublines. The transcription and signal-

the high-dose groups (KR600 nM and KR800 nM) were distinguished transduction-associated genes, NOTCH2, SQSTM1 and TGIF, from those of the low-dose groups (KR200 nM and KR400 nM) and were also upregulated. In case of NOTCH2, its product is parental K562. positive regulator of renin–angiotensin system (RAS) signal transduction in leukemogenesis and has been implicated in the molecular cross-talk with the mitogen-activated protein (MAP) kinase.23,24 Transforming growth factor-b induced factor (TGIF) is the transcriptional co-repressor forming a negative feedback loop in transforming growth factor (TGF)-b signaling. Among the downregulated genes in the resistant sublines, IGF1 and PLCD1 were found to be associated with signal transduction. IGF1 encodes a growth factor that regulates somatic growth and development as well as tumorigenesis.25 Phospho- liphase C, delta 1 (PLCD1) is involved in signal transduction by producing the second messenger molecules, diacylglycerol and inositol-1,4,5-trisphosphate.26 Cell-cycle-associated molecules such as CCNB1P1 were identified and found to function in the progression of the cell cycle through the G2/M phase.27 800nM 600nM 200nM 400nM K562 Expression changes correlated with imatinib doses in KR KR KR KR resistant sublines Secondly, genes that showed dose–response pattern between expression and imatinib dosage across the four resistant sublines were investigated (Table 2). Figure 2 illustrates the expression pattern of up- and downprogression across the sublines. Of the 22 genes showing a positive correlation with the increasing imatinib dosage, four transcription-related genes were identified, NMI, FOXO3A, ZNF226 and ZNF140. The NMI-encoded protein interacts with transcription factors such as N-MYC, C-MYC and the signal transducer and activator of transcription. In addition, a high expression of the NMI-encoded protein has been demonstrated in myeloid leukemia.28 As a signal- transduction-related transcript, latent TGF-b binding protein-1 is involved in the TGF-b signaling pathway by producing a protein that interacts with the TGF-b molecules. One putative ATP- dependent drug transporter, ABCA8, is also identified among the genes whose expression changes showed positive correlation with imatinib dosage.29 Cell proliferation-associated genes were identified in the transcripts that showed negative correlation with imatinib dosage. Of these, translocation of myeloid leukemia factor 1 (MLF1) has been frequently observed in myelodysplastic syndrome and acute myeloid leukemia.30 Among the metabolism and signal transduction-related genes, RAB11A is a RAS-related GTP-binding protein with putative roles in the tumorigenesis.31

Gene set enrichment analysis Gene set enrichment analysis was used to identify the imatinib- resistance-associated gene sets by measuring the statistical Figure 1 Hierarchical clustering of the imatinib-resistant K562 significance of the enrichment for B700 functionally annotated sublines according to the expression profiles. The top 3000 variable gene sets. Table 3 lists the activated or repressed gene sets transcripts in the five samples were selected and analyzed using the hierarchical clustering method. In the dendrogram, low-dose resistant identified in enrichment analysis. There are 15 activated gene sublines (KR200 nM and KR400 nM) along with the parental K562 were sets: four gene sets functionally related to transcription regula- distinguished from those of the high-dose group (KR600 nM and tion; three gene sets putatively related to transcription, such as KR800 nM). DNA binding, nucleic acid binding and zinc ion binding; two

Leukemia Table 1 Genes universally up- or downregulated across imatinib-resistant K562 sublines

Representative RNA Symbol Locus Description Relative difference (r.d.)a

200 nM 400 nM 600 nM 800 nM

Upregulatedb AA291203 NOTCH2 1p13–p11 Notch homolog 2 8 7.1 13 8.1 NM_016378 VCX3A Xp22 Variable charge, X-linked 3A 5.5 18 12 4.9 NM_013452 VCX Xp22 Variable charge, X-linked 10 11 8.9 7.1 U25804 CASP4 11q22.2–q22.3 Caspase-4, apoptosis-related cysteine peptidase 6 5.8 8.5 4.6 BE568134 TNFRSF21 6p21.1–12.2 Tumor necrosis factor receptor superfamily, member 21 5.5 5.4 5.8 5.5 NM_003059 SLC22A4 5q23.3 Solute carrier family 22 (organic cation transporter), member 4 13 16 8.4 8.6 NM_005462 MAGEC1 Xq26 Melanoma antigen family C, 1 7 9.7 9.4 6.8 NM_000610 CD44 11p13 CD44 antigen 4.5 6.4 7 8.2 NM_003192 TBCC 6pter-p12.1 Tubulin-speciﬁc chaperone c 5.8 4.2 6.8 5.7 NM_014821 KIAA0317 14q24.3 6.9 6.2 11 4.1 NM_001561 TNFRSF9 1p36 Tumor necrosis factor receptor superfamily, member 9 7.3 9.8 7.3 9.6 NM_003900 SQSTM1 5q35 Sequestosome 1 4.8 9.4 4.6 11 NM_002642 PIGC 1q23–q25 Phosphatidylinositol glycan, class C 4.4 9.1 16 16 BC001811 RRS1 8q13.1 RRS1 ribosome biogenesis regulator homolog 4.2 6.3 8.8 8 D86227 UMPS 3q13 Uridine monophosphate synthetase 4.9 4.4 6.9 8.4 phenotype Chung imatinib-resistant Y-J with associated signatures Expression NM_003244 TGIF 18p11.3 TGFB-induced factor 4.2 4.1 4.5 5.4

Downregulated NM_021178 CCNB1IP1 14q11.2 Cyclin B1-interacting protein 1 À7.6 À6.2 À6.3 À5.6 al et NM_005622 ACSM3 16p13.11 Acyl-CoA synthetase medium-chain family member 3 À9.1 À17 À7.7 À7.5 BE872563 KIF2 5q12–q13 Kinesin heavy-chain member 2 À4.5 À6.1 À4.4 À4.8 NM_173357 SSX6 Xp11.2 Synovial sarcoma, X breakpoint 6 À5.5 À5.5 À7.3 À4.3 NM_052965 C1orf19 1q25 Chromosome 1 open-reading frame 19 À4.4 À6.6 À8.1 À5 NM_012255 XRN2 20p11.2–p11.1 50–30 exoribonuclease 2 À4.2 À7.3 À6.4 À5.2 NM_005570 LMAN1 18q21.3–q22 Lectin, mannose binding, 1 À11 À10 À5.2 À7 AU144912 IGF1 12q22–q23 Insulin-like growth factor 1 (somatomedin C) À4.9 À12 À10 À12 NM_058172 ANTXR2 4q21.21 Anthrax toxin receptor 2 À5.8 À4.7 À7.5 À6.6 NM_006225 PLCD1 3p22–p21.3 Phospholipase C, delta 1 À5.1 À5.9 À4.6 À6.1 NM_004390 CTSH 15q24–q25 Cathepsin H À4.1 À6.2 À4.7 À4.4 aRelative difference, or r.d. represents the expression change of the gene for corresponding resistant subline relative to K562 control. bUp- and downregulated genes are determined when they passed the cutoff of +4 r.d. and À4 r.d. across four resistant sublines, respectively. Leukemia 1545 Expression signatures associated with imatinib-resistant phenotype Y-J Chung et al 1546 Table 2 Genes whose expression changes are correlated with imatinib doses across resistant sublines

Representative Symbol Cytoband Description RNA

Directly correlated U39573 LPO 17q23.1 Lactoperoxidase NM_015919 ZNF226 19q13.2 Zinc-ﬁnger protein 226 NM_002365 MAGEB3 Xp21.3 Melanoma antigen family B, 3 NM_014623 MEA1 6p21.3–p21.1 Male-enhanced antigen 1 AV757675 OPTN 10p13 Optineurin AF176704 FBXO9 chr6p12.3–p11.2 F-box protein 9 AF490510 NDNL2 15q13.1 necdin-like 2 NM_016378 VCX Xp22 Variable charge, X-linked AI986120 LTBP1 2p22–p21 Latent transforming growth factor b binding protein 1 BE568134 TNFRSF21 6p21.1–12.2 Tumor necrosis factor receptor superfamily, member 21 NM_002771 PRSS3 9p11.2 Protease, serine, 3 (mesotrypsin) NM_001561 TNFRSF9 1p36 Tumor necrosis factor receptor superfamily, member 9 NM_003440 ZNF140 12q24.32–q24.33 Zinc-ﬁnger protein 140 (clone pHZ-39) NM_022837 FLJ22833 2q32.3 Hypothetical protein FLJ22833 NM_002194 INPP1 2q32 Inositol polyphosphate-1–phosphatase NM_004267 CHST2 7q31 Carbohydrate (N-acetylglucosamine-6-O) sulfotransferase 2 NM_007168 ABCA8 17q24 ATP-binding cassette, subfamily A (ABC1), member 8 BG473130 KIF1A 2q37.3 Kinesin family member 1A NM_001455 FOXO3A 6q21 Forkhead box O3A NM_000944 PPP3CA 4q21–q24 Protein phosphatase 3 (formerly 2B), catalytic subunit, alpha isoform (calcineurin A alpha) AF399844 PRICKLE1 12q12 Prickle-like 1 (Drosophila) NM_004688 NMI 2p24.3–q21.3 N-myc (and STAT) interactor

Inversely correlated NM_006010 ARMET 3p21.1 Arginine-rich, mutated in early stage tumors NM_022443 MLF1 3q25.1 Myeloid leukemia factor 1 BK000686 SSX6 Xp11.2 Synovial sarcoma, X breakpoint 6 NM_013446 MKRN1 7q34 Makorin, ring ﬁnger protein, 1 NM_016542 MASK Xq26.2 Mst3 and SOK1-related kinase NM_018291 FLJ10986 1p32.1 AB209913 ANTXR2 4q21.21 Anthrax toxin receptor 2 XM_521278 MGC27005 X NM_001745 CAMLG 5q23 Calcium modulating ligand NM_004135 IDH3G Xq28 Isocitrate dehydrogenase 3 (NAD+) gamma NM_004226 STK17B 2q32.3 Serine/threonine kinase 17b (apoptosis-inducing) NM_019087 ARFRP2 5p15.2 ADP-ribosylation factor related protein 2 M29644 IGF1 12q22–q23 Insulin-like growth factor 1 (somatomedin C) NM_004663 RAB11A 15q21.3–q22.31 RAB11A, member RAS oncogene family NM_016021 UBE2J1 6q15 Ubiquitin-conjugating enzyme E2, J1 (UBC6 homolog, yeast) NM_002979 SCP2 1p32 Sterol carrier protein 2 NM_005622 ACSM3 16p13.11 Acyl-CoA synthetase medium-chain family member 3 NM_021178 CCNB1IP1 14q11.2 Cyclin B1-interacting protein 1 NM_052965 C1orf19 1q25 Chromosome 1 open-reading frame 19 NM_013979 BNIP1 5q33–q34 BCL2/adenovirus E1B 19 kDa-interacting protein 1

apoptosis-associated gene sets; and gene sets related to TGF-b sion, RAB11A and IGF1 were consistently downregulated in signaling pathway. Also, there are 17 repressed gene sets; gene BCR-ABL mutation-negative cases, whereas upregulated in two sets involved in biosynthesis, transport and folding, protein of the three BCR-ABL mutation-positive cases. LYN and RHOA, metabolism and three small GTPase signal transduction- whose upregulation was previously reported to be related with associated gene sets. the imatinb-resistance, were also investigated. Although they were up-regulated in a BCR-ABL mutation-negative cell line model,6,12 underexpression was observed in two of three BCR- Expression proﬁle in primary CML cases ABL mutation-negative cases in this study. Rather, the expres- To verify whether the expression changes accompanied by the sion of LYN and RHOA was signiﬁcantly upregulated in two of acquisition of imatinib-resistance in in vitro model are relevant three mutation-positive cases. However, the statistical power in primary CML cases, six CML patients showing imatinib- was not strong enough owing to the limited number of clinical resistance were analyzed. Log2 ratio of gene expression values samples. Further analysis with more patients will be required to before versus after acquisition of resistance was calculated for validate the result of this study. the six patients with (n ¼ 3) or without the BCR-ABL mutation

(n ¼ 3) (Figure 3). Expression ratio of K562 versus KR800 nM was also demonstrated as reference. CASP4 and TNFRSF21 were Discussion upregulated after the acquisition of imatinib-resistance in four and three of the six patients, respectively. However, the The aim of this study was to identify gene expression signatures expression ratio was not distinctively different between BCR- associated with imatinib-resistance by global gene expression ABL mutation-positive and -negative patients. For underexpres- analyses on a series of imatinib-resistant K562 sublines. By

Leukemia Expression signatures associated with imatinib-resistant phenotype Y-J Chung et al 1547 developed gene set enrichment analysis, which links the members of the genes to specific molecular pathways, was adopted to ensure the biological implication of our result.32 The essential signatures whose activation or repression is specific to the imatinib-resistance were identified using enrichment analysis combined with single gene approach. Most of the activated or repressed gene sets identified in enrichment analysis was associated with basic cellular physio- logy, reflecting the global metabolic changes in the imatinib- resistant sublines. Almost half of the 15 activated gene sets having transcription-related annotations showed that global upregulation of the genes involved in transcription is prevalent in the imatinib-resistant sublines. However, there is not enough evidence to pinpoint the increased global transcriptional activity as key determining factor.33 Similarly, two apoptosis-related gene sets activated in the resistant sublines are likely to represent the continuing cellular responses to imatinib. As shown in clinical samples, the frequency of upregulation for CASP4 and TNFRFS21 was not significantly different between BCR-ABL mutation-positive and -negative cases, indicating that upregulation of apoptosis is a general response to imatinib both in vitro and in vivo cases. Among the repressed gene sets, there were the protein and energy metabolism-associated gene sets, indicating the imatinib-resistant sublines have a lower protein and energy metabolism. The decreased protein metabolism and energy production might be owing to the dormant status of the resistant sublines, which is one of the putative mechanisms by which BCR-ABL primitive progenitor cells evade imatinib- induced apoptosis.33 The signal-transduction pathway-associated signature is worth consideration. For TGF-b signaling, TGIF and LTBP were found to be universally upregulated in a dose-dependent Figure 2 Expression changes correlated with the imatinib dosage by manner in the imatinib-resistant sublines examined in this the resistant sublines. (a) The expression values of the 22 genes, whose study. A previous study reported that TGF-b transcription was changes in expression were found to have a positive correlation with 11 imatinib dosage (y axis), are plotted against the imatinib dosage (x axis) upregulated in the imatinib-resistant KCL-22 cell lines. The by the imatinib-resistant sublines. The modified expression values, roles of the TGF-b pathway in tumorigenesis has been recently which are reduced from the original values to cope with Bartholo- highlighted along with its diverse roles such as cellular mew’s homogeneity law, were used. (b) The plot of the expression proliferation and possible tumor promoters.34,35 These findings values versus the imatinib dosage for 20 negatively correlated genes. of activation of TGF-b signaling indicate that it plays a key role in the acquisition of imatinib-resistance. The three gene sets associated with the small GTPase signaling pathway were fitting the serial expression profiles of four resistant sublines into repressed, which is indicative of the specific repression of the the analyses, it was possible to perform a more detailed pathway in the imatinib-resistant sublines. RAS-MAP kinase statistical analysis that is not feasible by simple dichotomous signaling is one of major downstream effectors and contributes comparison of sensitive versus resistant lines.11,12 Clear to the pathogenesis of BCR-ABL kinase.36 The specific inhibition discrimination of the expression profiles between the low- and of the downstream pathway such as RAS signaling, as shown in high-dose resistant sublines by unsupervised clustering supports this study, might be a putative mechanism for leukemic cells to the existence of heterogeneous genetic mechanisms in our become independent of growth inhibition directed by BCR-ABL imatinib-resistant K562 model. kinase. In single gene approach, the universally observed expression This study also identified several transcripts, which were changes shared by all resistant sublines might represent the significantly up- or downregulated, possibly associated with expression features closely associated with the survival of cells imatinib-resistance. They have potential to become surrogate in the presence of imatinib. And observed dose–response markers useful in a clinical evaluation of the imatinb-resistant relationship between expression and imatinib dosage suggests cases as demonstrated in our study of primary imatinib-resistant that transcriptional level of the genes is important in acquiring cases. IGF1 and RAB11A showed consistent downregulation imatinib-resistance. Among the transcripts identified, three after the acquisition of imatinib-resistance in BCR-ABL muta- upregulated (VCX, TNFRSF21 and TNFRSF9) and six down- tion-negative primary cases. This suggests that the measurement regulated (SSX6, ANTXR2, IGF1, ACSM3, CCNB1P1 and of expression level for these candidate genes could be used in C1orf19) genes were identified to have both universal and the prediction of developing of imatinib-resistance in BCR-ABL dose-dependent changes. This finding supports the biological mutation-negative patients. In case of LYN and RHOA whose potential of these nine genes to be putative surrogate markers up-regulation was reported in imatinib-resistance cell line indicating imatinib-resistance. However, in functional aspects, model, upregulation was not dominant in our BCR-ABL such genes have a wide variety of molecular functions among mutation-negative cases. This indicates that the heterogeneous which the essential ones such as signal transduction or expression patterns of in vivo cases might not be concordant proliferation are difficult to determine. Therefore, a recently with the observation of in vitro model.

Leukemia Leukemia 1548 xrsinsgaue soitdwt mtnbrssatphenotype imatinib-resistant with associated signatures Expression

Table 3 Gene sets activated or repressed in imatinib-resistant sublines

a b Annotated function Gene number 200 nM 400 nM 600 nM 800 nM Up- or down-progression

Activatedc Transcription (GO) 627 2.25E–05 1.07E–02 o10E–12 o10E–12 o10E–12 Zinc ion binding (GO) 613 6.40E–07 2.93E–04 o10E–12 o10E–12 o10E–12 DNA binding (GO) 519 2.18E–04 2.37E–04 1.13E–09 3.21E–10 o10E–12 DNA-dependent regulation of transcription, (GO) 813 3.85E–05 1.49E–03 o10E–12 o10E–12 2.50E–11 Proteasome degradation (G) 39 8.93E–04 8.08E–06 1.62E–02 3.56E–02 2.42E–09 Transcription factor activity (G) 511 1.30E–03 4.58E–03 9.53E–07 3.72E–10 2.25E–05 Apoptosis (G) 35 2.80E–02 3.87E–01 7.02E–03 1.53E–02 1.17E–04 Protein binding (GO) 743 4.02E–04 5.67E–01 2.25E–05 7.95E–03 2.57E–04 Structural constituent of ribosome (GO) 136 2.60E–02 3.32E–01 2.98E–02 4.86E–03 4.36E–04 Brain development (GO) 20 2.35E–02 2.60E–01 1.24E–02 2.13E–02 1.51E–02

Regulation of apoptosis (GO) 28 3.28E–02 6.43E–01 4.32E–05 1.33E–02 2.70E–02 Chung Y-J Nucleic acid binding (GO) 254 8.83E–03 4.52E–01 1.10E–03 3.62E–07 2.30E–05 Hemoglobin’s chaperone (B) 11 3.90E–01 o10E–12 2.21E–03 1.58E–08 1.29E–03

TGF b signaling pathway (G) 30 2.27E–01 3.51E–01 5.70E–05 1.66E–04 4.86E–03 al et Transcription corepressor activity (GO) 61 2.57E–01 8.75E–02 1.94E–02 2.50E–03 4.55E–02

Repressed Protein biosynthesis (GO) 186 1.73E–03 5.87E–05 1.46E–04 2.92E–02 1.16E–10 GTP binding (GO) 175 1.47E–02 8.77E–08 1.70E–04 6.58E–05 1.32E–03 Transferase activity (GO) 545 3.32E–04 1.63E–03 1.86E–03 3.37E–06 1.25E–08 Ribosomal proteins (G) 67 2.94E–05 1.48E–04 1.18E–07 2.25E–05 4.07E–10 Protein transport (GO) 136 8.83E–03 1.67E–04 2.04E–03 1.26E–03 9.04E–08 Metabolism (GO) 207 1.11E–02 3.01E–05 2.47E–02 2.04E–02 2.78E–06 Citrate cycle (TCA cycle; K) 13 6.86E–03 2.04E–02 4.32E–03 8.53E–03 2.72E–03 ER to Golgi transport (GO) 12 4.04E–03 5.51E–05 1.62E–02 4.62E–02 1.24E–02 GTPase activity (GO) 87 1.27E–01 3.68E–02 3.67E–02 2.46E–02 4.40E–02 Small GTPase-mediated signal transduction (GO) 110 2.10E–01 3.60E–03 7.38E–03 8.97E–04 1.53E–02 Krebs TCA cycle (G) 25 5.69E–03 7.35E–02 3.94E–02 4.33E–02 1.74E–02 Cholesterol biosynthesis (G) 12 1.99E–08 4.19E–01 8.77E–09 o10E–12 6.14E–04 Protein folding (GO) 133 1.65E–01 6.30E–01 7.40E–03 1.32E–05 2.69E–08 Unfolded protein binding (GO) 105 2.86E–01 6.72E–01 3.32E–02 8.80E–04 3.84E–04 Mitochondrial electron transport, NADH to ubiquinone (GO) 17 1.02E–01 3.52E–01 1.79E–03 1.96E–03 8.60E–09 Binding (GO) 198 2.00E–01 1.91E–01 1.76E–02 2.11E–03 4.85E–04 aAnnotated function is based on the functional categorization of genes in public gene database and symbols in the parentheses represent the origin of data set: GO – Gene Ontology; B – BioCarta; G – GenMAPP and K – KEGG. b,c Gene sets significantly enriched both for two comparisons of 600 and 800 nM and also significant as direct correlated with imatinib doses are considered as activated while repressed gene sets are determined vice versa. Expression signatures associated with imatinib-resistant phenotype Y-J Chung et al 1549 resistance. The resistance-associated expression signatures identified in this study would provide a deeper understanding of mechanism for the acquisition of resistance as well as for developing improved diagnostic and therapeutic strategies to deal with imatinib-resistant cases.

Acknowledgements

We thank the Korean Leukemia Cell and Gene Bank for providing the imatinib-resistant CML samples and S Yim for critical appraisal of this paper. This study was supported by a grant of the Korea Health 21 R&D Project, Ministry of Health & Welfare, Republic of Korea (0405-BC02-0604-0004), the Catholic Medical Center Research Foundation (2004) and Ministry of Science and Technology (2004-01303; M1-0416-22-0006), Republic of Korea.

References

1 O’Dwyer ME, Druker BJ. Chronic myelogenous leukaemia – new therapeutic principles. J Intern Med 2001; 250: 3–9. 2 Druker BJ, Sawyers CL, Kantarjian H, Resta DJ, Reese SF, Ford JM et al. Activity of a specific inhibitor of the BCR-ABL tyrosine kinase in the blast crisis of chronic myeloid leukemia and acute lymphoblastic leukemia with the Philadelphia chromosome. N Engl J Med 2001; 344: 1038–1042. 3 Hochhaus A, Kreil S, Corbin A, La RP, Lahaye T, Berger U et al. Roots of clinical resistance to STI-571 cancer therapy. Science 2001; 293: 876–880. 4 Shah NP, Nicoll JM, Nagar B, Gorre ME, Paquette RL, Kuriyan J et al. Multiple BCR-ABL kinase domain mutations confer polyclonal resistance to the tyrosine kinase inhibitor imatinib (STI571) in chronic phase and blast crisis chronic myeloid leukemia. Cancer Cell 2002; 2: 117–125. 5 Gorre ME, Mohammed M, Ellwood K, Hsu N, Paquette R, Rao PN et al. Clinical resistance to STI-571 cancer therapy caused by BCR- ABL gene mutation or amplification. Science 2001; 293: 876–880. 6 Donato NJ, Wu JY, Stapley J, Gallick G, Lin H, Arlinghaus R et al. BCR-ABL independence and LYN kinase overexpression in chronic myelogenous leukemia cells selected for resistance to STI571. Blood 2003; 101: 690–698. 7 Hofmann WK, de VS, Elashoff D, Gschaidmeier H, Hoelzer D, Koeffler HP et al. Relation between resistance of Philadelphia- chromosome-positive acute lymphoblastic leukaemia to the tyrosine kinase inhibitor STI571 and gene-expression profiles: a gene-expression study. Lancet 2002; 359: 481–486. 8 Mahon FX, Deininger MW, Schultheis B, Chabrol J, Reiffers J, Goldman JM et al. Selection and characterization of BCR-ABL positive cell lines with differential sensitivity to the tyrosine kinase inhibitor STI571: diverse mechanisms of resistance. Blood 2000; 96: 1070–1079. 9 Hegedus T, Orfi L, Seprodi A, Varadi A, Sarkadi B, Keri G. Figure 3 Expression changes of six candidate genes in imatinb- Interaction of tyrosine kinase inhibitors with the human multidrug resistant primary CML cases. (a) The expression changes after the transporter proteins, MDR1 and MRP1. Biochim Biophys Acta acquisition of imatinib-resistance are shown as log ratio (gene 2 2002; 1587: 318–325. expression values of before versus after acquisition of resistance) for 10 Widmer N, Colombo S, Buclin T, Decosterd LA. Functional two overexpression candidates, CASP4 and TNFRSF21. The positive consequence of MDR1 expression on imatinib intracellular log ratio indicates the expression of the corresponding gene increased 2 concentrations. Blood 2003; 102: 1142. after the imatinib-resistance was acquired and vice versa for negative 11 Tipping AJ, Deininger MW, Goldman JM, Melo JV. Comparative log ratio. The error bar depicts s.d. of expression values. Three BCR- 2 gene expression profile of chronic myeloid leukemia cells innately ABL mutation-negative CML cases (left) and the other three BCR-ABL resistant to imatinib mesylate. Exp Hematol 2003; 31: 1073–1080. mutation-positive cases (right) are separately displayed along with 12 Ohmine K, Nagai T, Tarumoto T, Miyoshi T, Muroi K, Mano H K562 cell line (BCR-ABL mutation negative, K562 versus KR )in 800 nM et al. Analysis of gene expression profiles in an imatinib-resistant the center. (b) The log expression ratio is shown for two under- 2 cell line, KCL22/SR. Stem Cells 2003; 21: 315–321. expression candidates RAB11A and IGF1 associated with imatinb- 13 Huber W, von HA, Sultmann H, Poustka A, Vingron M. Variance resistance. (c) LYN and RHOA are demonstrated for expression ratios. stabilization applied to microarray data calibration and to the quantification of differential expression. Bioinformatics 2002; 18: S96–S104. In conclusion, global expression analysis associated with 14 Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis imatinib-resistant in vitro and primary cases revealed number of and display of genome-wide expression patterns. Proc Natl Acad genes and gene sets possibly associated with the imatinib- Sci USA 1998; 95: 14863–14868.

Leukemia Expression signatures associated with imatinib-resistant phenotype Y-J Chung et al 1550 15 Tseng YH, Butte AJ, Kokkotou E, Yechoor VK, Taniguchi CM, progression through regulation of cyclin B levels. Mol Cell Biol Kriauciunas KM et al. Prediction of preadipocyte differentiation by 2003; 23: 2109–2122. gene expression reveals role of insulin receptor substrates and 28 Bao J, Zervos AS. Isolation and characterization of Nmi, a novel necdin. Nat Cell Biol 2005; 7: 601–611. partner of Myc proteins. Oncogene 1996; 12: 2171–2176. 16 Bartholomew DJ. A test of homogeneity for ordered alternatives. 29 Tsuruoka S, Ishibashi K, Yamamoto H, Wakaumi M, Suzuki M, Biometrika 1959; 46: 36–48. Schwartz GJ et al. Functional analysis of ABCA8, a new 17 Storey JD, Tibshirani R. Statistical significance for genomewide drug transporter. Biochem Biophys Res Commun 2002; 298: studies. Proc Natl Acad Sci USA 2003; 100: 9440–9445. 41–45. 18 Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, Foulger R 30 Matsumoto N, Yoneda-Kato N, Iguchi T, Kishimoto Y, Kyo T, et al. The Gene Ontology (GO) database and informatics resource. Sawada H et al. Elevated MLF1 expression correlates with Nucleic Acids Res 2004; 32: D258–D261. malignant progression from myelodysplastic syndrome. Leukemia 19 Kanehisa M, Goto S, Kawashima S, Nakaya A. The KEGG 2000; 14: 1757–1765. databases at GenomeNet. Nucleic Acids Res 2002; 30: 42–46. 31 Gebhardt C, Breitenbach U, Richter KH, Furstenberger G, Mauch 20 Dahlquist KD, Salomonis N, Vranizan K, Lawlor SC, Conklin BR. C, Angel P et al. c-Fos-dependent induction of the small ras-related GenMAPP, a new tool for viewing and analyzing microarray data GTPase Rab11a in skin carcinogenesis. Am J Pathol 2005; 167: on biological pathways. Nat Genet 2002; 31: 19–20. 243–253. 21 Kim SY, Volsky DJ. PAGE: parametric analysis of gene set 32 Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, enrichment. BMC Bioinform 2005; 6: 144. Lehar J et al. PGC-1alpha-responsive genes involved in oxidative 22 Livak KJ, Schmittgen TD. Analysis of relative gene expression data phosphorylation are coordinately downregulated in human dia- using real-time quantitative PCR and the 2(ÀDelta Delta C(T)) betes. Nat Genet 2003; 34: 267–273. Method. Methods 2001; 25: 402–408. 33 Holyoake TL, Jiang X, Jorgensen HG, Graham S, Alcorn MJ, Laird 23 Aster JC, Pear WS. Notch signaling in leukemia. Curr Opin C et al. Primitive quiescent leukemic cells from patients with Hematol 2001; 8: 237–244. chronic myeloid leukemia spontaneously initiate factor-indepen- 24 Zeng Q, Li S, Chepeha DB, Giordano TJ, Li J, Zhang H et al. dent growth in vitro in association with up-regulation of expression Crosstalk between tumor and endothelial cells promotes tumor of interleukin-3. Blood 2001; 97: 720–728. angiogenesis by MAPK activation of Notch signaling. Cancer Cell 34 Elliott RL, Blobe GC. Role of transforming growth factor Beta in 2005; 8: 13–23. human cancer. J Clin Oncol 2005; 23: 2078–2093. 25 Zumkeller W. IGFs and IGFBPs: surrogate markers for diagnosis and 35 Yan Z, Deng X, Friedman E. Oncogenic Ki-ras confers a more surveillance of tumour growth? Mol Pathol 2001; 54: 285–288. aggressive colon cancer phenotype through modification of 26 Stallings JD, Tall EG, Pentyala S, Rebecchi MJ. Nuclear transloca- transforming growth factor-beta receptor III. J Biol Chem 2001; tion of phospholipase C-delta1 is linked to the cell cycle and 276: 1555–1563. nuclear phosphatidylinositol 4, 5-bisphosphate. J Biol Chem 2005; 36 Mandanas RA, Leibowitz DS, Gharehbaghi K, Tauchi T, Burgess 280: 22060–22069. GS, Miyazawa K et al. Role of p21 RAS in p210 bcr-abl 27 Toby GG, Gherraby W, Coleman TR, Golemis EA. A novel RING transformation of murine myeloid cells. Blood 1993; 82: finger protein, human enhancer of invasion 10, alters mitotic 1838–1847.

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

Leukemia