Gene expression changes associated with progressionand response in chronic myeloid leukemiaJerald P. Radich*†‡, Hongyue Dai§, Mao Mao§, Vivian Oehler*, Jan Schelter§, Brian Druker†¶, Charles Sawyers�,Neil Shah�, Wendy Stock**, Cheryl L. Willman†,††, Stephen Friend§, and Peter S. Linsley§

*Divisions of Clinical Research, Fred Hutchinson Cancer Research Center, Seattle, WA 98109; §Rosetta Inpharmatics, Seattle, WA 98109; ¶OregonHealth & Science University, University Center, Portland, OR 97239; �University of California, Los Angeles, CA 90095; **University of ChicagoSchool of Medicine, Chicago, IL 60637; ††University of New Mexico Cancer Research and Treatment Center, Albuquerque, NM 87131; and†Southwest Oncology Group, Ann Arbor, MI 48106

Communicated by E. Donnall Thomas, Fred Hutchinson Cancer Research Center, Seattle, WA, December 13, 2005 (received for review June 24, 2005)

Chronic myeloid leukemia (CML) is a hematopoietic stem celldisease with distinct biological and clinical features. The biologicbasis of the stereotypical progression from chronic phase throughaccelerated phase to blast crisis is poorly understood. We used DNAmicroarrays to compare gene expression in 91 cases of CML inchronic (42 cases), accelerated (17 cases), and blast phases (32cases). Three thousand genes were found to be significantly (P <10�10) associated with phase of disease. A comparison of the genesignatures of chronic, accelerated, and blast phases suggest thatthe progression of chronic phase CML to advanced phase (accel-erated and blast crisis) CML is a two-step rather than a three-stepprocess, with new gene expression changes occurring early inaccelerated phase before the accumulation of increased numbersof leukemia blast cells. Especially noteworthy and potentiallysignificant in the progression program were the deregulation ofthe WNT��-catenin pathway, the decreased expression of Jun Band Fos, alternative kinase deregulation, such as Arg (Abl2), and anincreased expression of PRAME. Studies of CML patients whorelapsed after initially successful treatment with imatinib demon-strated a gene expression pattern closely related to advancedphase disease. These studies point to specific gene pathways thatmight be exploited for both prognostic indicators as well as newtargets for therapy.

genetics � blast crisis � PRAME � Wnt

Chronic myeloid leukemia (CML) is a hematopoietic stem celldisease with distinct biological and clinical features. CML

usually presents in chronic phase, in which the clonal expansion ofmature myeloid cells leads to an elevated white blood cell count.Without curative intervention, chronic-phase CML will invariablytransform through a phase of ‘‘acceleration,’’ often heralded by theappearance of increased immature myeloid cells in the bonemarrow and peripheral blood, as well as new cytogenetic changesin addition to the Philadelphia chromosome (Ph). Progression thenproceeds to blast crisis, with immature blast cells overwhelming theproduction of normal hematopoetic elements. Blast crisis is highlyresistant to treatment, with death generally occurring from infec-tion and bleeding complications secondary to the absence ofnormal granulocytes and platelets. The median time from diagnosisof chronic phase CML to progression to accelerated phase is �3–4years, but the range of timing is quite broad, encompassing from 0.5to 15 years (1).

There are several treatment options for CML. All treatments aremore successful when administered during the chronic phasedisease than in accelerated or blast phase. The only known curativetherapy for CML is stem cell transplantation, a complex andpotentially toxic modality that carries a high potential for morbidityand mortality (2). Nontransplant therapy includes IFN-�, whichproduces a major reduction in the proportion of Ph-positive cellsand extends the natural history of the disease in �10–20% ofpatients cases, with some alive and in remission for �10 years (3).IFN-� has largely been replaced by the tyrosine kinase inhibitor,

Imatinib Mesylate, which suppresses the Ph to the point where it isundetectable by cytogenetic evaluation (a complete cytogeneticremission or CCR) in �70% of newly diagnosed patients withchronic-phase CML disease (4). The long-term durability durationof such responses is unknown, as is potential for cure with imatinib.Complete molecular response, defined by the absence of theBcr-Abl fusion mRNA by RT-PCR, is unusual in patients with CCR(5), and patients who achieve a molecular response, but haveimatinib discontinued, show a relatively rapid reemergence ofdisease (6). Resistance to imatinib occurs (especially in advancedphase disease) often accompanied by point mutations in thetyrosine kinase domain of the abl gene (7). The natural history ofsuch relapses is unknown, although some appear to progress rapidlyinto advanced stage disease (8).

The genetic events that cause the progression of chronic phase toblast crisis CML are largely unknown. Numerous genetic abnor-malities have been demonstrated, including the acquisition ofadditional chromosomal abnormalities including such as a dupli-cation of the Ph�, isochromosome 17(p) resulting in the disruptionof TP53, and less commonly, the deletion of the p15�p16 tumorsuppressor genes (especially in lymphoid blast crisis) and theRUNX1-EV11 fusion (9, 10). Genetic instability is apparent, asevidenced by the in the additional chromosomal changes that occurwith progression, but standard assays of instability, such as alter-ations in minisatellite repeats, are detected infrequently (11, 12).

Unfortunately, clinical and molecular tests cannot predict whereon the ‘‘clock’’ of disease progression an individual lies at the timeof the diagnosis, and this makes it impossible to adapt therapy to thedegree of risk that faces an individual CML patient. We also cannotyet identify the subset of patients who are most likely to benefit froma specific treatment option. This study is aimed at determiningchanges in gene expression that occur in the evolution of the chronicphase to blast crisis, with the hope of identifying genes and pathwaysthat may be useful as prognostic markers or targets for therapeutics.

ResultsPhase-Specific Gene Expression in CML. We first examined the genesassociated with the phase of disease. We used as a reference a poolof 200 chronic phase bone marrows, and studied 91 individual casesof CML in chronic phase (n � 42), accelerated phase by blast countcriteria (n � 9) or by the occurrence of additional clonal cytogeneticchanges (n � 8), blast crisis (n � 28), and four cases of blast crisisin remission after chemotherapy. An ANOVA analysis revealed�3,500 genes (from a total of �24,000 genes) differentially ex-pressed across the different phases of disease by using a minimum

Conflict of interest statement: No conflicts declared.

Freely available online through the PNAS open access option.

Abbreviations: CML, chronic myeloid leukemia; CCR, complete cytogenetic remission.

Data deposition: The sequence reported in this paper has been deposited in the GeneOntology (GO) database (accession no. GSE4170).

‡To whom correspondence should be addressed. E-mail: jradich@fhcrc.org.

© 2006 by The National Academy of Sciences of the USA

2794–2799 � PNAS � February 21, 2006 � vol. 103 � no. 8 www.pnas.org�cgi�doi�10.1073�pnas.0510423103

statistical significance cutoff of P � 10�11 (Fig. 1 and Table 4, whichis published as supporting information on the PNAS web site). Thisset of genes identified in the progression from chronic to blast phaseis referred to here as the ‘‘phase reporter’’ gene set.

We examined the proposed biologic function of the genesassociated with CML phases by applying a biological annotationprogram based on the gene ontology (GO) and KEGG annotations.The major functional groups of genes in the phase reporter gene setare shown in Table 5, which is published as supporting informationon the PNAS web site. The functional groups most highly correlatedwith disease phase (accelerated�blast phase relative to chronicphase) included increased expression of nuclear genes, mitochon-

drial genes, RNA-binding genes, and protein biosynthesis genes,reflecting the increased proliferation and metabolism of progres-sive disease. Advanced-phase CML, compared to chronic phase,exhibited a decreased expression of genes involved in structuralintegrity and adhesion, as well as decreases in expression of genesinvolved in inflammatory and immune response. In addition,several protooncogenes and tumor suppressor genes are differen-tially expressed in advanced phase CML, including N- and H-ras,FLT3, yes, AF1q, CBFB, WT1, ORALOV1, Bcl-2, and PTPN11.

CML Appears to Be a ‘‘Two-Step’’ Disease. We examined whether theprogression of CML best fit a three-step model as often described

Fig. 2. Comparison of CML blasts to normal CD34� cells and correction for the pattern of gene expression. (A) Phase genes were corrected for normal CD34�

gene expression (ANOVA, P � 1 � 10�8). The gene expression of normal CD34� cells was subtracted from each disease sample. The resulting pattern reflects genesassociated with progression independent of normal blast biology. (B) Relative gene expression of the ‘‘top ten’’ genes found to be up-regulated in advancedphase disease compared to chronic phase.

Fig. 1. Genes associated with CML progression. Sam-ples from patients with of CML cases in chronic phase,accelerated by cytogenetic criteria only, acceleratedphase, blast crisis, and blast crisis ‘‘in remission’’ werecompared to a pool of chronic phase RNA (See Mate-rials and Methods for details). Approximately 3,500genes were significantly associated with progressivedisease at a significance level of P � 10�11. Each rowrepresents one sample, and each column representsone gene. Red color indicates overexpression relativeto the control pool, and green color indicates lowexpression.

Radich et al. PNAS � February 21, 2006 � vol. 103 � no. 8 � 2795

MED

ICA

LSC

IEN

CES

in the literature (chronic phase3 accelerated phase3 blast crisis),or whether the expression data were best explained by a two-stepmodel (chronic phase3 advanced phase; i.e., accelerated or blastphase). We compared the gene expression of accelerated phasecases versus blast crisis cases (Fig. 4, which is published as support-ing information on the PNAS web site). The correlation of expres-sion level between accelerated and blast phase gene expression wasstrong (r � 0.81). In addition, the correlation of gene expressionbetween accelerated phase and cases of accelerated phase definedby new cytogenetic abnormalities alone (i.e., chronic phase mor-phology but additional chromosomal changes besides the Ph) wasalso high (r � 0.61). These observations suggest that the differenceof gene expression between accelerated and blast phase is aquantitative change in expression rather a qualitative change.

Identification of ‘‘Progression-Specific’’ Gene Expression. We nextinvestigated how the phase reporter gene set was influenced by thegene expression signature of leukemia blasts and how these CMLblasts compared to normal immature CD34� cells. First, we madea direct comparison of the gene expression signature of six samplesof normal CD34� cells with CML blast samples containing �70%blasts (Fig. 5, which is published as supporting information on thePNAS web site). We found that the gene expression pattern of CMLblast cells was very similar to normal CD34� cells. To uncover geneexpression unique to CML progression (that is, associated withphase of disease, but not normal CD34� expression), we used twoapproaches. First, we found genes differentially expressed betweenCML blast crisis and normal immature CD34� cells (Fig. 6, whichis published as supporting information on the PNAS web site, 368genes with ANOVA P � 0.1%). Secondly, we mathematicallysubtracted the normal CD34� signature from each of the CMLsample and examined the resulting effect on the phase reportersignals (Fig. 2A, 386 genes with P � 10�8, and Table 6, which ispublished as supporting information on the PNAS web site). Thesetwo approaches identified 103 overlapping genes found in both genesets (Table 7, which is published as supporting information on thePNAS web site, hypergeometric P value: 1.3 � 10�99). Table 8,which is published as supporting information on the PNAS web site(identified by the second approach), shows the ‘‘top ten’’ genes up-and down-regulated that are associated with progression, indepen-dent of normal CD34� expression, based on log 10 ratio ofexpression compared to the chronic phase pool. These genessuggest a deregulation of genes of transcriptional regulation (GLI2,WT1, FOS, FOSB), signal transduction (SOCS2, Rras2, IL8), andapoptosis (GAS2). Also significantly associated with progressionwas PRAME (Preferentially Expressed Antigen of Melanoma), agene with unknown function, but associated with myeloid malig-nancy (13, 14). The expression levels of these genes are shown inFig. 2B.

Gene functions of the progression gene set were organized byGO and KEGG functional annotation (Table 1; for the full list, seeTables 9–11, which are published as supporting information on thePNAS web site). Of special relevance was aberrance in genesinvolved in the WNT��-catenin signaling pathway, alterations innucleosome genes, and genes involved in differentiation and apo-ptosis. In all, 16 genes associated with the �-catenin pathway weresignificantly deregulated in progression (e.g., cadherin, protocad-hedrin, PRICKLE1, CSNK1E, PLCB1, FZD2, LRP6, SMAD3,MDF1, etc.; see Fig. 7, which is published as supporting informationon the PNAS web site).

Progression appears to be associated with expression changes ofseveral cytoskeletal and adhesion molecules. Thus, there is anincrease in CD47, Crebl1, and ITGA5 expression, and a decreasein proto-cadhedrin, cadhedrin, actin, and �-actin. In addition, theabl family member abl2 (ARG) was significantly up-regulated indisease progression; by contrast, abl1, which is both expressed fromthe normal chromosome 9 and the t(9, 22) translocation, was notdifferentially expressed with progression.

MZF- and EF1�-Controlled Genes Are Aberrantly Expressed in Progres-sion. Gene expression networks defined by specific promoter reg-ulation were analyzed in the progression and phase reporter genesets (Table 2). The most significantly network of genes showingaberrant control in progression with those that contained a MZFpromoter or a EF1� promoter sequence (P � 10�15 and �10�11,respectively; Figs. 8 and 9, which are published as supportinginformation on the PNAS web site). Also significantly associatedwith progression were genes bearing SPI-B, Yin Yang, and AHR-ARNT promoter sequences (all with P values �10�9).

Gene Expression Profiles of Cases Treated with Imatinib. We studiedthe gene expression of CML patients for whom imatinib therapywas ineffective. We examined 15 cases of CML treated withimatinib, including nine patients who initially achieved a CCR onimatinib therapy, but then relapsed back into an apparent chronicphase by morphologic examination; three cases who had achieveda complete hematologic but no cytogenetic response; and three latechronic phase patients. All but one of these the chronic phasepatients who relapsed after a CCR had a point mutation in abl1,presumably abrogating imatinib activity (Table 3). In addition, onlyone of the relapsing cases had additional cytogenetic lesions.

Fig. 3 shows the expression pattern of these 15 CML cases. Wefound that the cases that appeared in chronic phase after relapsingafter an achievement of CCR had expression patterns similar toadvanced disease. This can be demonstrated by segregating allCML cases by the correlation of gene expression signature betweenthe boundaries of ‘‘most chronic’’ cases (bottom of the heat map)and ‘‘most advanced’’ gene expression (top of the heat map) for all3,000 genes in the phase reporter gene set. The majority of the poorresponse patients have gene expression profiles more consistentwith advanced disease rather than chronic phase. Both cases withT315I mutations, which have been shown to have especially poorprognosis (8, 15), have expression signatures more similar toadvanced disease than chronic phase (red arrows).

Table 1. Functional annotation of ‘‘progression’’ genes

Keyword Number (%)* Examples P value

Ribosome 18 (21) ROK13A 9 � 10�11

Wnt signaling 16 (11) Cadherin, MDI1,Prickle 1, FZD2

2 � 10�5

Nucleosome 22 (22) BSZ1A, HIST1H2AE 3 � 10�11

Sugarmetabolism

45 (26) RP1A, ALDOC,G6PD

4 � 10�9

Myeloiddifferentiation

27 (14) CEBPA, CEBPE,FOXO3A

3 � 10�8

Apoptosis 42 (10) GADD45G, BCL2,FOXO3A, MCL1

2 � 10�7

DNA damageresponse

36 (10) GASDD45G,FANCG, XRN2

2 � 10�7

Bold functions are up-regulated in progression; italic functions are down-regulated.*The number of genes present on the array with the specific keyword; inparentheses are the percentage of significant genes found differentiallyexpressed among the keyword class.

Table 2. Promoter sequences differentially representedin progression

PromotersP value in progression

genesP value in

phase genes

MZF-1 �10�15 �10�15

EF1� 9 � 10�11 1 � 10�9

SPI-B 1 � 10�10 2 � 10�11

Yin Yang 1 � 10�9 7 � 10�11

AHR-ARNT 2 � 10�9 5 � 10�10

2796 � www.pnas.org�cgi�doi�10.1073�pnas.0510423103 Radich et al.

However, there were gene expression features that were uniqueto the imatinib relapsers. Examples of genes unique to imatinibresistance failure are those of serine threonine kinases (CTRL,MAP21K14, CLK3), mitogen-activated protein kinase (MKNK2),the tyrosine kinase oncogene FYN, and the drug exporter ABCC3.Fig. 10, which is published as supporting information on the PNASweb site, compares relapsed imatinib cases against blast crisis cases,and several areas (boxed) indicate genes that are expressed inreverse direction in these two states (Tables 12 and 13, which arepublished as supporting information on the PNAS web site, listsgenes in the first box. See Table 12 for the rest of the imatinibresistance genes).

DiscussionThe biology and treatment of CML is dictated by the phase ofdisease, as the efficacy of all therapies (transplantation, IFN, orimatinib) are most successful when used during the chronicphase, in comparison to accelerated phase and�or blast crisis.Understanding the biology of progression may provide clinicaldiagnostic markers of progression and offer insights into newstrategies for treatment. The data presented here suggest thatthe progression of chronic phase CML to advanced phase CMLis a two-step process, with progression associated a block ofdifferentiation and apoptosis, a shift toward turning on expres-sion of genes involved in the nucleosome, with alterations in celladhesion, and activation of alternative signaling pathways. Inaddition, it appears that relapse after initial successful treatmentwith imatinib may be associated with gene expression patternssimilar to advanced phase CML, suggesting that the process of

progression persists in a subpopulation of CML cells duringtherapy.

In our analysis, we first used unsorted CML samples to developgene signatures associated with the phase of disease. This data setgives a more global picture of the transformation of disease, and isimportant because any diagnostic assay for progression wouldoptimally be performed on whole blood or marrow, rather thanselected cell subpopulations. To control for any gene signatures thatmight be common to an immature blast phenotype, we subtracteda normal CD34� signature from the CML samples. This left us witha progression gene set that should be highly enriched in genesassociated with CML progression, but not reflecting normal CD34�

cell gene expression. This gene set may be more indicative of thebiology of CML progression, and give more insight to pathways thatmight be exploited by new therapeutic agents. Moreover, theprogression gene set may contain genes able to distinguish earlyevidence of progression in immunophenotypically similar immaturecells.

The demonstration that the gene expression pattern betweenaccelerated and blast phases are very similar suggests that thecrucial steps in progression are at the transition of chronic toaccelerated phase, before obvious morphologic, cytogenetic, orclinical evidence of progression. This has clinical implications,because these patients might benefit from more aggressive therapy.In addition, the observation that gene signatures of blast crisis canbe seen in accelerated phase patients by cytogenetic criteria onlyand blast crisis cases in remission (both of which have similarly lowblast counts as patients with chronic phase disease), demonstratestwo important points. First, it points out the difficulty of correlating

Table 3. Characteristics of cases treated with imatinib

CodePhase of disease at

initiation of imatinibPhase of disease

at time of sample Cytogenetics Mutation(s)

2 Chronic Relapsed chronic 100% Ph�; NEW:t(1;12) in all cells M244V4 Chronic Relapsed chronic 100% Ph� T315I5 Chronic Relapsed chronic 80% Ph� F359V6 Chronic Relapsed chronic 100% Ph� M351T, F317L8 Chronic Relapsed chronic 100% Ph� M351T

10 Chronic Relapsed chronic 100% Ph� E255K11 Chronic post-allo BMT Relapsed chronic 100% Ph� T315I12 Chronic Relapsed chronic 100% Ph� L248V14 Chronic Relapsed chronic 100% Ph� M351T, H396R

BMT, bone marrow transplantation.

Fig. 3. Gene expression in patients with resistance toImatinib failure cases. Cases were ranked and sorted bythe correlation of summed gene expression, with casesrepresenting the most ‘‘chronic phase-like’’ and most‘‘blast crisis-like’’ forming the boundaries of gene ex-pression patterns. Cases that had a poor response toimatinib are designated by a blue dot in the ‘‘IM-cases’’box to the right of the heat map. The red arrows pointto the two cases that had the T315I Abl mutation.

Radich et al. PNAS � February 21, 2006 � vol. 103 � no. 8 � 2797

MED

ICA

LSC

IEN

CES

morphology with the biology of the disease. Secondly, the pene-tration of a progression gene expression signature into a ‘‘chronicphase’’ appearing bone marrow suggests that progression is notmerely an absolute block of differentiation, but that abnormal geneexpression signals persist in normal appearing differentiated cells.This is critically important because it provides the basis for testingfor progression genes in unsorted bone marrow samples from‘‘chronic phase’’ patients.

Bcr-Abl has a wholesale range of biological activities. Critical inthe transformation process is the activation of the Ras�mitogen-activated protein kinase (MAPK) pathways, which has broadeffects on changes in cell adhesion (through Rho), proliferation(MAPK pathway), and apoptosis (through Akt) (16, 17). Imatinib’sefficacy results from a blockade of these effects. Imatinib is lesssuccessful for advanced-phase disease; this may be because theseleukemia have become less dependent on the pathways that ima-tinib blocks. Thus, we found that the MAPK pathway was relativelyunderexpressed in advanced disease compared to chronic phase,but other signaling pathways, including those involving cytokines(IL3RA, SOCS2), alternative ras pathways (Rras2), and thoseinvolved in cell adhesion (WNT��-catenin) are activated. Theactivation of these pathways may allow progression even in the faceof therapeutic blockade of Bcr-Abl activated pathways. In addition,abl2 (Abl related gene, or ARG) is also up-regulated in progression.As opposed to abl1, ARG is a cytoplasmic protein (18). Thesignaling targets of ARG are unknown, and although broadlyexpressed in tissue, its only known functional role apparent fromknockout mouse models appears to be in the nervous system (19,20). ARG has been associated with myeloid leukemia in the contextof TEL�ARG translocations (21, 22). ARG shares �90% homol-ogy of its tyrosine kinase domain with abl1, and ARG tyrosinekinase activity is inhibited by imatinib at similar drug concentra-tions (18, 23). However, given that Bcr-Abl amplification is con-sidered to play a role in imatinib resistance (24, 25), ARG over-expression in blast crisis could theoretically contribute to therelative resistance to imatinib found with progressive disease, eitheras independently acting on aberrant signaling, or acting as animatinib ‘‘sink.’’

Two recent observations on the molecular biology of progressionin CML are relevant to this study. First, activation of the WNT��-catenin pathway was observed in primary cell samples frompatients with CML (26). Secondly, it was recently observed thatmice deficient in Jun B develop a disease much like CML (27). Ourdata complement these findings, as we found broad dysregulationof WNT��-catenin pathway as well as decreased Jun B expression.A link between these pathways may be the gene MDFI (I-mfa), aninhibitor of myogenic basic helix–loop–helix transcription factors(28). MDFI interacts with axin, which is involved in binding andmodulating free �-catenin, and thus changes in MDFI may influ-ence �-catenin mediated gene activation (29). In addition, MDFIand axin interaction also influences WNT and Jun signaling (30).

Our analysis suggests that genes controlled by MZF1 and EF1�may be particularly important in progression. MZF1 is a memberof the Kruppel family of zinc finger proteins, and was originallycloned from a cDNA library from a blast crisis CML patient (31).MZF1 appears to play a critical role in hematopoietic stem celldifferentiation, including modulation of CD34 and c-myb expres-sion (32, 33). MZF1�/� knockout mice display an increase inhematopoietic progenitor proliferation, which continues in long-term culture conditions (32). These data suggest MZF1 deregula-tion may disrupt normal differentiation, promoting the progressionto advanced disease. EF1� is related to the Smad zinc fingerproteins that play an important role in TGF-� gene regulation.EF1� has been shown to compete with basic helix–loop–helixactivators, and is implicated in modulation of MyoD regulatedpathways (34, 35). It is not known whether EF1� directly influencesMDFI expression. Of note is that both MZF1 and EF1� have beenshown to influence cadherin expression (36–38). Thus, the further

study of the control of MZF1 and EF1� may be particularly fruitfulin understanding the molecular mechanisms of CML progression.

Given that efficacy of treatment in CML (be it with IFN,imatinib, or transplantation) is so intimately associated with phaseof disease, one wonders whether those chronic phase patients inwhom imatinib therapy is ineffective have genetic features ofadvanced phase invisible to routine pathological and cytogeneticexamination. Imatinib failures are a reasonable setting to explorethis possibility. Although imatinib can cause cytogenetic remissionsin the majority of chronic phase cases, treatment failure, especiallysecondary to point mutations, is an increasingly important problem.It has previously been demonstrated that the probability of devel-oping a point mutation depends largely on the time from diagnosisto initiation of therapy (8). This finding implies that the geneticmechanisms that lead to point mutations are relentless, and there-fore the treatment of ‘‘late’’ chronic phase patients (i.e., � 1 yearfrom diagnosis) may be undermined by genetic changes that havealready occurred. Branford et al. (8) demonstrated that patientswho developed P-loop point mutations had a very poor outcome,with approximately half dying within a year of relapse. Theseobservations are in keeping with our demonstration that manyimatinib failures have gene expression changes similar to advanceddisease, despite their benign pathological appearance. The predic-tive power of genetic markers of response and progression will beimportant to test prospective in future clinical studies of imatiniband similar compounds.

Several of the genes found in the progression set might serve asearly markers of progression in diagnostic assays, and may serve astherapeutic targets, as well. For example, PRAME was originallyidentified as a tumor antigen recognized by cytotoxic T cells againsta melanoma surface antigen (13, 39). Like similar antigens MAGE,BAGE, or GAGE, PRAME is expressed in some solid tumors;unlike these other antigens, however, it has been found to beoverexpressed in �25% of leukemia, and has been found to beinduced by Bcr-Abl in CML cell lines (14, 40). Indeed, PRAMEoverexpression has been described as one of the few features thatcharacterize the transient myeloproliferative syndrome of Down’ssyndrome from the progressive acute megakaryoblastic leukemiafound in that disorder (41). Very recently, insights into the functionof PRAME make it an attractive target for potential therapy.PRAME is a nuclear protein, which seems to act as a dominantrepressor of retinoic acid receptor (RAR) signaling (42). In cell linemodels, the forced overexpression of PRAME blocked RAR-mediated differentiation, growth arrest, and apoptosis. Thus, ther-apeutic maneuvers to block the PRAME�RAR effect may allowfor differentiation and cell death, and be particularly effective inattacking advanced phase CML.

In comparison to other types of leukemia, there have been fewpapers exploring the use of microarrays on the biology of CML(43–45). It is difficult to make a direct comparison of these studiesand ours, given the different types of samples obtained (someunsorted, some AC133� or CD34� selected), the difference in thearray platforms, etc. Nonetheless, in general, the functional changesof progression (changes in differentiation, apoptosis, cell adhesion,and inflammatory response) remained as common themes acrossthese studies. Moreover, in keeping with a recent publication thatfound that the high expression of elastase (ELA2) was associatedwith a long period of indolent disease (45), in our data, ELA2-decreased expression is associated with progression.

These findings have the potential to influence therapy ofCML. Patients who present with gene expression patterns sug-gestive of advanced phase disease might benefit by moving tomore aggressive (i.e., transplantation) or other investigationaltherapies. Moreover, PCR assays of individual genes (or smallsets of genes) may be used to monitor patients early in the courseof imatinib therapy for signs of progression to advanced disease.Lastly, new targets to interfere with progression, or reverseimatinib resistance, may be exploited.

2798 � www.pnas.org�cgi�doi�10.1073�pnas.0510423103 Radich et al.

Materials and MethodsPatient Samples. All samples were obtained under the auspices ofinstitutional review board approve protocols. Samples came fromthe Fred Hutchinson Cancer Research Center, the SouthwestOncology Group (SWOG) Myeloid Repository, the University ofOregon Health Sciences Center, the University of California, LosAngeles, or the University of Chicago. RNA extraction was eitherperformed immediately, or in the case of samples stored in a liquidnitrogen repository, after thawing. All RNA samples were qualitytested by analysis of ribosomal RNA peaks using an ABI Bioalyzer.The definition of chronic, accelerated and blast crisis was based onthe criteria of Sokal et al. and the International Bone MarrowTransplant Registry (46, 47). Thus, chronic phase was defined as�10% blasts, the accelerated phase was defined as 10–30% blastsor �10% blasts with clonal evolution, and blast crisis was definedas �30% blasts.

Amplification, Labeling, and Hybridization. The procedures of RNAamplification, labeling, and the hybridization to arrays, as well asthe specifics of the array platforms, has been published (48).

Analytic Methods and Results. Each individual sample was hybrid-ized to a pool of chronic phase samples. The log10 (Ratio) ofintensity of individual samples to the pool were used for thesubsequent analysis. Before selecting features by ANOVA, 25,000genes on the array were first screened for evidence of differentialregulation by requiring P value of regulation �1% in more thanthree experiments, where P value of regulation is based on theplatform error model. Features differentiating progression stageswere selected by ANOVA test.

Functional Annotation of Gene Lists. Genes represented on themicroarray were annotated by assignment to GO Biological Process

or Molecular Function categories (www.ebi.ac.uk�GOA), or toKEGG pathways (www.genome.jp�kegg�pathway.html). Gene lists(input sets) were queried for enrichment of members of specificfunctional classes or pathways relative to the background frequency.The significance (P value) of enrichment was computed by using thehypergeometric probability distribution. Reported in each case arethe numbers of genes in the input set (input gene count), numberof genes in a particular category or pathway in the input set (overlapgene count), and number of genes in a particular category amongall genes present on the array (set gene count). The total numberof unique genes on the array is 24,132.

Methods to Common Promote Site Analysis. The common promotersites were based on the predictions derived from the database(oPOSSUM) by Wasserman et al. (www.cisreg.ca). The hypergeo-metric P value for enrichment of a particular binding site wascomputed by comparing the number of genes with the binding sitefrom a signature gene set to that from a background set (i.e., allgenes represented on the microarray).

Controls. We compared gene expression signatures from the sitescontributing samples to confirm that there were no site-signaturesconfounding the analysis.

Because some samples of blast crisis came from peripheral bloodrather than bone marrow, we compared three samples in whichsimultaneous samples were available from bone marrow and pe-ripheral blood. Gene expression was extremely well correlated (r �0.97–0.99; Fig. 11, which is published as supporting information onthe PNAS web site).

This work was supported in part by National Institutes of Health GrantsCA-18029 and CA-85053 (to J.P.R.) and the Howard Hughes MedicalInstitute (B.D. and C.S.).

1. Faderl, S., Talpaz, M., Estrov, Z. & Kantarjian, H. M. (1999) Ann. Intern. Med. 131, 207–219.2. Radich, J. P., Gooley, T., Bensinger, W., Chauncey, T., Clift, R., Flowers, M., Martin, P.,

Slattery, J., Sultan, D. & Appelbaum, F. R. (2003) Blood 102, 31–35.3. Kantarjian, H. M., O’Brien, S., Smith, T. L., Rios, M. B., Cortes, J., Beran, M., Koller, C.,

Giles, F. J., Andreeff, M., Kornblau, S., et al. (1999) J. Clin. Oncol. 17, 284–292.4. Druker, B. J., Talpaz, M., Resta, D. J., Peng, B., Buchdunger, E., Ford, J. M., Lydon, N. B.,

Kantarjian, H., Capdeville, R., Ohno-Jones, S. & Sawyers, C. L. (2001) N. Engl. J. Med. 344,1031–1037.

5. Hughes, T. P., Kaeda, J., Branford, S., Rudzki, Z., Hochhaus, A., Hensley, M. L., Gathmann,I., Bolton, A. E., van Hoomissen, I. C., Goldman, J. M. & Radich, J. P. (2003) N. Engl. J. Med.349, 1423–1432.

6. Cortes, J., O’Brien, S. & Kantarjian, H. (2004) Blood 104, 2204–2205.7. Shah, N. P., Nicoll, J. M., Nagar, B., Gorre, M. E., Paquette, R. L., Kuriyan, J. & Sawyers,

C. L. (2002) Cancer Cell 2, 117–125.8. Branford, S., Rudzki, Z., Walsh, S., Parkinson, I., Grigg, A., Szer, J., Taylor, K., Herrmann,

R., Seymour, J. F., Arthur, C., et al. (2003) Blood 102, 276–283.9. Calabretta, B. & Perrotti, D. (2004) Blood 103, 4010–4022.

10. Shet, A. S., Jahagirdar, B. N. & Verfaillie, C. M. (2002) Leukemia 16, 1402–1411.11. Wada, C., Shionoya, S., Fujino, Y., Tokuhiro, H., Akahoshi, T., Uchida, T. & Ohtani, H.

(1994) Blood 83, 3449–3456.12. Mori, N., Takeuchi, S., Tasaka, T., Lee, S., Spira, S., Ben-Yehuda, D., Mizoguchi, H.,

Schiller, G. & Koeffler, H. P. (1997) Leukemia 11, 151–152.13. van Baren, N., Chambost, H., Ferrant, A., Michaux, L., Ikeda, H., Millard, I., Olive, D.,

Boon, T. & Coulie, P. G. (1998) Br. J. Haematol. 102, 1376–1379.14. Watari, K., Tojo, A., Nagamura-Inoue, T., Nagamura, F., Takeshita, A., Fukushima, T.,

Motoji, T., Tani, K. & Asano, S. (2000) FEBS Lett. 466, 367–371.15. Soverini, S., Martinelli, G., Rosti, G., Bassi, S., Amabile, M., Poerio, A., Giannini, B.,

Trabacchi, E., Castagnetti, F., Testoni, N., et al. (2005) J. Clin. Oncol. 23, 4100–4109.16. Faderl, S., Talpaz, M., Estrov, Z., O’Brien, S., Kurzrock, R. & Kantarjian, H. M. (1999)

N. Engl. J. Med. 341, 164–172.17. Ren, R. (2002) Oncogene 21, 8629–8642.18. Kruh, G. D., Perego, R., Miki, T. & Aaronson, S. A. (1990) Proc. Natl. Acad. Sci. USA 87,

5802–5806.19. Koleske, A. J., Gifford, A. M., Scott, M. L., Nee, M., Bronson, R. T., Miczek, K. A. &

Baltimore, D. (1998) Neuron 21, 1259–1272.20. Perego, R., Ron, D. & Kruh, G. D. (1991) Oncogene 6, 1899–1902.21. Cazzaniga, G., Tosi, S., Aloisi, A., Giudici, G., Daniotti, M., Pioltelli, P., Kearney, L. &

Biondi, A. (1999) Blood 94, 4370–4373.22. Iijima, Y., Ito, T., Oikawa, T., Eguchi, M., Eguchi-Ishimae, M., Kamada, N., Kishi, K.,

Asano, S., Sakaki, Y. & Sato, Y. (2000) Blood 95, 2126–2131.23. Okuda, K., Weisberg, E., Gilliland, D. G. & Griffin, J. D. (2001) Blood 97, 2440–2448.24. le Coutre, P., Tassi, E., Varella-Garcia, M., Barni, R., Mologni, L., Cabrita, G., Marchesi,

E., Supino, R. & Gambacorti-Passerini, C. (2000) Blood 95, 1758–1766.25. Weisberg, E. & Griffin, J. D. (2000) Blood 95, 3498–3505.

26. Jamieson, C. H., Ailles, L. E., Dylla, S. J., Muijtjens, M., Jones, C., Zehnder, J. L., Gotlib,J., Li, K., Manz, M. G., Keating, A., Sawyers, C. L. & Weissman, I. L. (2004) N. Engl. J. Med.351, 657–667.

27. Passegue, E., Wagner, E. F. & Weissman, I. L. (2004) Cell 119, 431–443.28. Chen, C. M., Kraut, N., Groudine, M. & Weintraub, H. (1996) Cell 86, 731–741.29. Nelson, W. J. & Nusse, R. (2004) Science 303, 1483–1487.30. Kusano, S. & Raab-Traub, N. (2002) Mol. Cell. Biol. 22, 6393–6405.31. Hromas, R., Collins, S. J., Hickstein, D., Raskind, W., Deaven, L. L., O’Hara, P., Hagen,

F. S. & Kaushansky, K. (1991) J. Biol. Chem. 266, 14183–14187.32. Gaboli, M., Kotsi, P. A., Gurrieri, C., Cattoretti, G., Ronchetti, S., Cordon-Cardo, C.,

Broxmeyer, H. E., Hromas, R. & Pandolfi, P. P. (2001) Genes Dev. 15, 1625–1630.33. Perrotti, D., Melotti, P., Skorski, T., Casella, I., Peschle, C. & Calabretta, B. (1995) Mol. Cell.

Biol. 15, 6075–6087.34. Funahashi, J., Sekido, R., Murai, K., Kamachi, Y. & Kondoh, H. (1993) Development

(Cambridge, U.K.) 119, 433–446.35. Sekido, R., Murai, K., Funahashi, J., Kamachi, Y., Fujisawa-Sehara, A., Nabeshima, Y. &

Kondoh, H. (1994) Mol. Cell. Biol. 14, 5692–5700.36. Guaita, S., Puig, I., Franci, C., Garrido, M., Dominguez, D., Batlle, E., Sancho, E., Dedhar,

S., De Herreros, A. G. & Baulida, J. (2002) J. Biol. Chem. 277, 39209–39216.37. Le Mee, S., Fromigue, O. & Marie, P. J. (2005) Exp. Cell. Res. 302, 129–142.38. Miyoshi, A., Kitajima, Y., Sumi, K., Sato, K., Hagiwara, A., Koga, Y. & Miyazaki, K. (2004)

Br. J. Cancer 90, 1265–1273.39. Matsushita, M., Ikeda, H., Kizaki, M., Okamoto, S., Ogasawara, M., Ikeda, Y. & Kawakami,

Y. (2001) Br. J. Haematol. 112, 916–926.40. Paydas, S., Tanriverdi, K., Yavuz, S., Disel, U., Baslamisli, F. & Burgut, R. (2005) Am. J.

Haematol. 125, 257–261.41. McElwaine, S., Mulligan, C., Groet, J., Spinelli, M., Rinaldi, A., Denyer, G., Mensah, A.,

Cavani, S., Baldo, C., Dagna-Bricarelli, F., et al. (2004) Br. J. Haematol. 125, 729–742.42. Epping, M. T., Wang, L., Edel, M. J., Carlee, L., Hernandez, M. & Bernards, R. (2005) Cell

122, 835–847.43. Nowicki, M. O., Pawlowski, P., Fischer, T., Hess, G., Pawlowski, T. & Skorski, T. (2003)

Oncogene 22, 3952–3963.44. Ohmine, K., Ota, J., Ueda, M., Ueno, S., Yoshida, K., Yamashita, Y., Kirito, K., Imagawa,

S., Nakamura, Y., Saito, K., et al. (2001) Oncogene 20, 8249–8257.45. Yong, A. S., Szydlo, R. M., Goldman, J. M., Apperley, J. F. & Melo, J. V. (2005) Blood 107,

205–212.46. Savage, D. G., Szydlo, R. M., Chase, A., Apperley, J. F. & Goldman, J. M. (1997) Br. J.

Haematol. 99, 30–35.47. Sokal, J. E., Cox, E. B., Baccarani, M., Tura, S., Gomez, G. A., Robertson, J. E., Tso, C. Y.,

Braun, T. J., Clarkson, B. D., Cervantes, F., et al. (1984) Blood 63, 789–799.48. Hughes, T. R., Mao, M., Jones, A. R., Burchard, J., Marton, M. J., Shannon, K. W.,

Lefkowitz, S. M., Ziman, M., Schelter, J. M., Meyer, M. R., et al. (2001) Nat. Biotechnol.19, 342–347.

Radich et al. PNAS � February 21, 2006 � vol. 103 � no. 8 � 2799

MED

ICA

LSC

IEN

CES