*Shannon K. McWeeney,1*Lucy C. Pemberton,1,2Marc M. Loriaux,1Kristina Vartanian,1,3Stephanie G. Willis,1
Gregory Yochum,1Beth Wilmot,1Yaron Turpaz,4,5Raji Pillai,4Brian J. Druker,1,6Jennifer L. Snead,1Mary MacPartlin,1
Stephen G. O’Brien,2Junia V. Melo,7Thoralf Lange,8ChristinaA. Harrington,1,3and Michael W. N. Deininger1,8
1Oregon Health & Science University Knight Cancer Institute, Portland;2Academic Haematology, University of Newcastle upon Tyne, Newcastle upon Tyne,
United Kingdom;3Gene Microarray Shared Resource, Oregon Health & Science University, Portland;4Clinical Programs, Pathwork Diagnostics, Redwood City,
CA;5Integrative Computational Sciences, Lilly Singapore Centre for Drug Discovery, Singapore;6Howard Hughes Medical Institute, Chevy Chase, MD;
7Department of Haematology, Institute of Medical & Veterinary Science,Adelaide,Australia; and8Department of Hematology, University of Leipzig, Leipzig,
In chronic-phase chronic myeloid leukemia
(CML) patients, the lack of a major cytoge-
netic response (< 36% Ph?metaphases) to
imatinib within 12 months indicates failure
and mandates a change of therapy. To iden-
we performed gene expression array profil-
ing of CD34?cells from 2 independent co-
horts of imatinib-naive chronic-phase CML
metaphases within 12 months of imatinib
than .1 and fold difference 1.5 or more, we
identified 885 probe sets with differential
expression between responders and nonre-
sponders, from which we extracted a 75-
probe set minimal signature (classifier) that
separated the 2 groups. On application to a
prospectively accrued validation set, the
classifier correctly predicted 88% of re-
lished studies revealed overlap of classifier
gesting that chronic-phase CML patients
destined to fail imatinib have more ad-
vanced disease than evident by morpho-
ing more aggressive therapy upfront to the
patients most likely to benefit while sparing
good-risk patients from unnecessary toxic-
Imatinib is an effective therapy for the majority of patients with
mately 20% to 30% of patients fail imatinib and require alternative
treatments.1,2The cytogenetic response at 12 months is a powerful
prognosticator of outcome. In a large trial of patients treated with
standard-dose imatinib (400 mg daily), the projected rates of
event-free survival were 97% and 93%, respectively, for patients
who had attained a complete cytogenetic response (CCyR, 0%
Philadelphia chromosome-positive [Ph?] metaphases) or major
cytogenetic response (MCyR, ? 36% Ph?metaphases), but only
81% in patients with less than MCyR at 12 months.1In view of the
high risk of progression, an expert panel convened by the European
Leukemia Net has concluded that lack of MCyR at 12 months
(herein referred to as primary cytogenetic resistance) defines
imatinib failure and warrants a change in the therapeutic strategy.3
More intensive therapy upfront has been proposed to improve
the rates of MCyR.4Because most patients will do well on standard
therapy, it would be desirable to direct early treatment intensifica-
tion to high-risk patients. The best clinical predictor of primary
cytogenetic resistance is the Sokal risk score.5In the International
Randomized Interferon versus STI571 (IRIS) study, the projected
rate of CCyR at 48 months was only 69% of patients with a high
Sokal risk compared with 91% with low risk and 84% with
intermediate risk.6However, for clinical decisions, a more reliable
prognosticator is needed. Based on the promising results of gene
expression profiling for response prediction in various hematologic
malignancies,7-11we had previously attempted to predict MCyR by
microarray analysis of unselected blood or bone marrow white
cells collected before therapy but found no significant differences
between responders and nonresponders.12This led us to hypoth-
esize that detecting a signature associated with primary cytogenetic
resistance might require analyzing a more primitive cell compart-
ment. We therefore performed gene expression profiling on CD34?
cells collected before imatinib therapy from 2 independent groups
of chronic-phase CML patients, an initial training set of late
chronic-phase patients, and a prospectively accrued validation set
of newly diagnosed chronic-phase patients. Here we report the
identification of a gene classifier of CD34?CMLcells that predicts
MCyR with high accuracy.
The training set was retrospectively selected from CML patients treated at
Oregon Health & Science University between 1998 and 2004. Most of the
patients had failed prior interferon-?–based therapy and were treated on
phase 2 studies of imatinib before its regulatory approval. Eligibility
*S.K.M. and L.C.P. contributed equally to this manuscript.
The online version of this article contains a data supplement.
The publication costs of this article were defrayed in part by page charge
payment. Therefore, and solely to indicate this fact, this article is hereby
marked ‘‘advertisement’’ in accordance with 18 USC section 1734.
© 2010 by TheAmerican Society of Hematology
315BLOOD, 14 JANUARY 2010?VOLUME 115, NUMBER 2
For personal use only.on November 4, 2015. by guest
criteria were a diagnosis of CML in chronic phase (based on the criteria of
the IRIS trial), availability of bone marrow mononuclear cells (MNCs)
stored immediately before initiating imatinib therapy, and availability of at
least 1 year of follow-up, including karyotyping. Responders were defined
as those patients with at least a partial cytogenetic response within
12 months of therapy and nonresponders as all other patients. Because this
response definition is inherently imprecise given the routine sampling of
only 20 metaphases and may therefore misclassify responses, the training
set focused on patients with CCyR during their first year of imatinib therapy
as opposed to patients who had not achieved even a minor cytogenetic
response (ie, remained at least 66% Ph?) during that time, thereby reducing
noise by enriching the training set for the extremes of the response
spectrum. Of 51 samples initially processed, 36 were included in the final
analysis, whereas the remainder was excluded because they failed to meet
the minimum quality requirements for microarray analysis (see “RNA
extraction and gene expression profiling”). The second group of patients
(validation set) consisted of 42 consecutive newly diagnosed chronic-phase
patients treated with imatinib at the University of Newcastle (United
Kingdom) or University of Leipzig (Germany). In 23 of these patients, the
microarray analysis was successful. The majority of these patients were
followed with metaphase karyotyping; however, response was assessed by
fluorescence in situ hybridization in 7 of 17 responders and 2 of 6 nonre-
sponders. In these patients, CD34?cells were selected from peripheral
blood collected at diagnosis. The study was approved by the institutional
review board of all participating institutions, and all subjects provided
written informed consent in accordance with the Declaration of Helsinki.
Isolation of CD34?cells
CD34?cell selection, the cells were thawed at 37°C and washed in Dulbecco
phosphate-buffered saline containing 0.1% human albumin (Baxter Healthcare
Corporation), 1% recombinant DNase (Pulmozyme; Genentech), and 2.5mM
Kit (Miltenyi Biotec). Next, the cells were resuspended in Hank’s balanced salt
solution with 0.5% fetal bovine serum, 2% N-2-hydroxyethylpiperazine-N?-2-
ethanesulfonic acid, and 1% recombinant human DNase (Genentech), stained
with CD34-fluorescein isothiocyanate (FITC) and CD45-peridinin chlorophyll
protein (PerCP)–Cy5.5 monoclonal antibodies (BD Biosciences), and placed in
Hanks balanced salt solution containing 0.5% fetal bovine serum, 2% N-2-
hydroxyethylpiperazine-N?-2-ethanesulfonic acid, and 1% recombinant human
DNase. For the identification of dead cells, propidium iodide (Roche Diagnos-
ABD FACSAria (BD Biosciences) was used to sort CD34?cells. Gates
on forward scatter (FSC) and side scatter, followed by FSC-width (FSC-W)
and FSC-height (FSC-H), were used to exclude dead cells and debris. Next,
gates were set on propidium iodide-negative cells to ensure that only viable
cells were selected. Finally, on the CD34-FITC and CD45-PerCP-Cy5.5
histogram, CD45-PerCP-Cy5.5 dim cells that brightly coexpressed CD34-
FITC were selected.The procedure was regarded as a success if greater than
1000 CD34?cells were isolated, with a purity of greater than 80% CD34?
cells by flow cytometry. An example of the sorting strategy is shown in
supplemental Figure 1 (available on the Blood website; see the Supplemen-
tal Materials link at the top of the online article).After sorting, CD34?cells
were placed in PicoPure extraction buffer (Arcturus) and stored at ?80°C
until processed further. Small aliquots of CD34?cells were also stored for
fluorescence in situ hybridization (FISH) to assess the proportion of
BCR-ABL?cells. In the case of the validation set, MNCs were isolated
from peripheral blood using density gradient centrifugation. CD34?cells
were isolated from the MNC using MiniMACS columns (Miltenyi Biotec),
following the instructions of the manufacturer.
RNA extraction and gene expression profiling
RNA extraction was performed with the PicoPure RNA Isolation Kit
(Arcturus) once all cell sorting had been completed. Samples were
Drop Technologies), and the quality of the RNA was assessed using the
Agilent 2100 Bioanalyzer (Agilent Technologies). Only samples with
electropherograms showing a size distribution pattern predictive of accept-
able microarray assay performance were processed further. Details of the
quality assessment procedure will be reported elsewhere (K.A.V., H. Paik,
C. Runyon, B. Tompkins, L. Crossman, M.W.N.D., C.A.H. Factors
influencing the optimization and standardization of the Affymetrix Gene-
Chip expression assay small sample amplification protocol in the microar-
ray core laboratory (poster). Annual Meeting of the Association of
Biomolecular Resource Facilities, February 2005 ). To generate sufficient
cRNA target for microarray hybridization, we used the GeneChip Eukary-
otic Small Sample Target Labeling Assay Version ll (Affymetrix), with
inputs from 5 to 20 ng of total RNA. Control experiments in the microarray
core laboratory demonstrated high-quality microarray data from inputs as
low as 5 ng (C.A.H., personal communication). After successful amplifica-
tion, 10 ?g of labeled target cRNA was hybridized to HG-U133 Plus 2.0
GeneChip arrays (Affymetrix). Arrays were scanned using a laser confocal
scanner (Agilent Technologies), and the image processing and expression
analysis were performed using Affymetrix GCOS, Version 1.2 software.
For quality assurance/quality control purposes, the parameters ?1and ?2
were set to 0.05 and 0.065 (Affymetrix defaults), respectively. These
parameters set the point at which a probe set was called present (P),
marginal (M), or absent (A). Minimal quality control parameters for
inclusion in the study included P more than 30%, average signal in keeping
with the average signal of other samples within that hybridization group (ie,
the group of samples hybridized as a batch), and a GAPDH 3?/5? ratio of
less than or equal to 3.62. Overall, the process of CD34?cell selection,
RNAextraction, and array hybridization was successful in 36 of 51 patients
(71%). The average present call rate in this group was 41.5% (range,
38.8%-47.1%). FISH for BCR-ABLwas successful in 28 of the 36 samples.
The median percentage of BCR-ABL?CD34?cells was found to be 98.5%
(range, 33%-100%). The 23 samples of the validation set were processed in
an identical fashion approximately 18 months after the training set. For
consistency, similar amounts of input RNAwere used.
Standard statistical methods
Differences in the distribution of patient demographics/treatment history
were examined by categorical data analysis in the training set using the
SPSS software package.
Microarray data analysis
Low-level analysis of the Affymetrix data was conducted using the Robust
Multiarray Average (RMA) algorithm.13Only Perfect Match intensities
were used. Parameters for RMA included model-based background correc-
tion, quantile normalization, and median polish. Transcript-by-transcript
(ie, unique Affymetrix Probe set IDs) analysis of variance to determine
differential expression between NR and R was performed on the training set
(N ? 36). All P values were false discovery rate adjusted. With respect to
feature, selection was based on effect size (fold change [FC] ? 1.5) and
statistical significance (P ? .1) to minimize false negatives. Data were
further filtered based on threshold expression level and variability (based on
coefficient of variation). Class prediction was performed using the nearest
shrunken centroids algorithm.14Testing of the classifier was performed on
an independent, blinded validation set (N ? 23). The raw (cel) and
normalized data were deposited in National Center for Biotechnology
Information gene ontology (GO) database (GSE14671).
Structural analysis of the classifier
With regard to downstream analysis of the classifier, overrepresented GO
and pathway annotations were identified in the classifier transcripts using
categorical data analysis (with adjustment for the nested multiple compari-
sons). Known protein-protein interactions were examined for classifier
members as well as with other genes using the Metacore database. In
addition to examining functional enrichment, potential sub-networks (or
“small networks”) in the classifier were examined using known and curated
protein-protein interactions from the MetaCore database. These sub-
networks were ranked based on statistical significance and the number of
316 MCWEENEY et alBLOOD, 14 JANUARY 2010?VOLUME 115, NUMBER 2
For personal use only.on November 4, 2015. by guest
normalization, and summaries of high density oli-
gonucleotide array probe level data. Biostatistics.
14. Tibshirani R, Hastie T, Narasimhan B, Chu G. Di-
agnosis of multiple cancer types by shrunken
centroids of gene expression. Proc Natl Acad Sci
U S A. 2002;99(10):6567-6572.
15. Ekins S,Andreyev S, RyabovA, et al.Acombined
approach to drug metabolism and toxicity assess-
ment. Drug Metab Dispos. 2006;34(3):495-503.
16. Zheng C, Li L, Haak M, et al. Gene expression
profiling of CD34? cells identifies a molecular
signature of chronic myeloid leukemia blast crisis.
17. O’Brien SG, Guilhot F, Larson RA, et al. Imatinib
compared with interferon and low-dose cytara-
bine for newly diagnosed chronic-phase chronic
myeloid leukemia. N Engl J Med. 2003;348(11):
18. Deininger M, Buchdunger E, Druker BJ. The de-
velopment of imatinib as a therapeutic agent for
chronic myeloid leukemia. Blood. 2005;105(7):
19. Jamieson CH,Ailles LE, Dylla SJ, et al. Granulo-
cyte-macrophage progenitors as candidate leuke-
mic stem cells in blast-crisis CML. N Engl J Med.
ProcNatlAcadSciU S A.2007;104(9):3324-3329.
21. YongAS, Szydlo RM, Goldman JM,Apperley JF,
Melo JV. Molecular profiling of CD34? cells iden-
tifies low expression of CD7, along with high ex-
pression of proteinase 3 or elastase, as predic-
tors of longer survival in patients with CML.
22. Soverini S, Colarossi S, GnaniA, et al. Contribu-
tion ofABL kinase domain mutations to imatinib
resistance in different subsets of Philadelphia-
positive patients: by the GIMEMAWorking Party
on Chronic Myeloid Leukemia. Clin Cancer Res.
23. Kaneta Y, Kagami Y, Katagiri T, et al. Prediction of
sensitivity to STI571 among chronic myeloid leu-
kemia patients by genome-wide cDNAmicroarray
analysis. Jpn J Cancer Res. 2002;93(8):849-856.
24. McLean LA, Gathmann I, Capdeville R,
Polymeropoulos MH, Dressman M. Pharmaco-
genomic analysis of cytogenetic response in
chronic myeloid leukemia patients treated with
imatinib. Clin Cancer Res. 2004;10:155-165.
25. Villuendas R, Steegmann JL, Pollan M, et al.
Identification of genes involved in imatinib resis-
tance in CML: a gene-expression profiling ap-
proach. Leukemia. 2006;20(6):1047-1054.
26. Frank O, Brors B, FabariusA, et al. Gene expres-
sion signature of primary imatinib-resistant
chronic myeloid leukemia patients. Leukemia.
27. Weisberg E, Wright RD, McMillin DW, et al.
Stromal-mediated protection of tyrosine kinase
inhibitor-treated BCR-ABL-expressing leukemia
cells. Mol Cancer Ther. 2008;7(5):1121-1129.
28. Korkolopoulou P, Viniou N, Kavantzas N, et al.
Clinicopathologic correlations of bone marrow
angiogenesis in chronic myeloid leukemia: a mor-
phometric study. Leukemia. 2003;17(1):89-97.
29. Thomas J, Wang L, Clark RE, Pirmohamed M.
Active transport of imatinib into and out of cells:
implications for drug resistance. Blood. 2004;
30. Crossman LC, Druker BJ, Deininger MW, et al.
hOCT 1 and resistance to imatinib. Blood. 2005;
31. Wang L, GiannoudisA, Lane S, et al. Expression
of the uptake drug transporter hOCT1 is an im-
portant clinical determinant of the response to
imatinib in chronic myeloid leukemia. Clin Phar-
macol Ther. 2008;83(2):258-264.
33. Dulucq S, Bouchet S, Turcq B, et al. Multidrug
resistance gene (MDR1) polymorphisms are as-
sociated with major molecular responses to
standard-dose imatinib in chronic myeloid leuke-
mia. Blood. 2008;112(5):2024-2027.
34. Hatziieremia S, Jordanides NE, Holyoake TL,
Mountford JC, Jorgensen HG. Inhibition of MDR1
does not sensitize primitive chronic myeloid leu-
kemia CD34? cells to imatinib. Exp Hematol.
35. Guerzoni C, Bardini M, Mariani SA, et al. Induc-
ible activation of CEBPB, a gene negatively regu-
lated by BCR/ABL, inhibits proliferation and pro-
motes differentiation of BCR/ABL-expressing
cells. Blood. 2006;107(10):4080-4089.
37. Radich JP, Dai H, Mao M, et al. Gene expression
changes associated with progression and re-
sponse in chronic myeloid leukemia. Proc Natl
Acad Sci U S A. 2006;103(8):2794-2799.
38. Talpaz M, Silver RT, Druker BJ, et al. Imatinib in-
duces durable hematologic and cytogenetic re-
sponses in patients with accelerated phase
chronic myeloid leukemia: results of a phase 2
study. Blood. 2002;99(6):1928-1937.
39. Sawyers CL, HochhausA, Feldman E, et al. Ima-
tinib induces hematologic and cytogenetic re-
sponses in patients with chronic myelogenous
leukemia in myeloid blast crisis: results of a
phase II study. Blood. 2002;99(10):3530-3539.
40. Kantarjian H, Sawyers C, HochhausA, et al. He-
matologic and cytogenetic responses to imatinib
mesylate in chronic myelogenous leukemia.
N Engl J Med. 2002;346(9):645-652.
41. O’Hare T, Eide CA, Deininger MW. Bcr-Abl kinase
domain mutations, drug resistance, and the road
to a cure for chronic myeloid leukemia. Blood.
GENE EXPRESSION SIGNATURE OF CD34?CELLS325
BLOOD, 14 JANUARY 2010?VOLUME 115, NUMBER 2
For personal use only. on November 4, 2015. by guest
online October 16, 2009
2010 115: 315-325
Michael W. N. Deininger
Mary MacPartlin, Stephen G. O'Brien, Junia V. Melo, Thoralf Lange, Christina A. Harrington and
Willis, Gregory Yochum, Beth Wilmot, Yaron Turpaz, Raji Pillai, Brian J. Druker, Jennifer L. Snead,
Shannon K. McWeeney, Lucy C. Pemberton, Marc M. Loriaux, Kristina Vartanian, Stephanie G.
response in chronic-phase chronic myeloid leukemia patients treated
cells to predict major cytogenetic
A gene expression signature of CD34
Articles on similar topics can be found in the following Blood collections
Updated information and services can be found at:
(1413 articles)Myeloid Neoplasia
(4194 articles)Clinical Trials and Observations
Information about reproducing this article in parts or in its entirety may be found online at:
Information about ordering reprints may be found online at:
Information about subscriptions and ASH membership may be found online at:
Copyright 2011 by The American Society of Hematology; all rights reserved.
of Hematology, 2021 L St, NW, Suite 900, Washington DC 20036.
Blood (print ISSN 0006-4971, online ISSN 1528-0020), is published weekly by the American Society
For personal use only.on November 4, 2015. by guest