Document wqX87OaroEprMLjrZjbEonwo3

ORIGINAL ARTICLE Leukemia (2006) 20, 21472154 & 2006 Nature Publishing Group All rights reserved 0887-6924/06 $30.00 www.nature.com/leu Gene expression profiles of AML derived stem cells; similarity to hematopoietic stem cells H Gal1,2,3, N Amariglio2, L Trakhtenbrot2, J Jacob-Hirsh2, O Margalit2, A Avigdor4, A Nagler4, S Tavor5, L Ein-Dor3, T Lapidot6, E Domany3, G Rechavi2 and D Givol1 1Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot, Israel; 2Department of Pediatric HematoOncology, Chaim Sheba Medical Center, Tel-Hashomer and Sackler School of Medicine, Tel-Aviv University, Tel-Aviv, Israel; 3Department of Physics of Complex Systems, Weizmann Institute of Science, Rehovot, Israel; 4The Division of Hematology and Bone Marrow Transplantation, Chaim Sheba Medical Center, Tel-Hashomer, Israel; 5Department of Hematology, Tel-Aviv Sourasky Medical Center, Tel-Aviv, Israel and 6Department of Immunology, Weizmann Institute of Science, Rehovot, Israel Tumors contain a fraction of cancer stem cells that maintain the propagation of the disease. The CD34 CD38 cells, isolated from acute myeloid leukemia (AML), were shown to be enriched leukemic stem cells (LSC). We isolated the CD34 CD38 cell fraction from AML and compared their gene expression profiles to the CD34 CD38 cell fraction, using microarrays. We found 409 genes that were at least twofold over- or underexpressed between the two cell populations. These include underexpression of DNA repair, signal transduction and cell cycle genes, consistent with the relative quiescence of stem cells, and chromosomal aberrations and mutations of leukemic cells. Comparison of the LSC expression data to that of normal hematopoietic stem cells (HSC) revealed that 34% of the modulated genes are shared by both LSC and HSC, supporting the suggestion that the LSC originated within the HSC progenitors. We focused on the Notch pathway since Jagged-2, a Notch ligand was found to be overexpressed in the LSC samples. We show that DAPT, an inhibitor of gamma-secretase, a protease that is involved in Jagged and Notch signaling, inhibits LSC growth in colony formation assays. Identification of additional genes that regulate LSC self-renewal may provide new targets for therapy. Leukemia (2006) 20, 21472154. doi:10.1038/sj.leu.2404401; published online 12 October 2006 Keywords: stem cells; microarray; colony formation; Notch Introduction Acute myeloid leukemia (AML) is a heterogeneous disease with variations in the cell markers, chromosomal aberrations, mutations, response to therapy and prognosis. Increasing evidence suggest that AML and other malignancies are sustained by a minor tumor subpopulation with self-renewal potential, referred to as `cancer stem cells' or `leukemic stem cells' (LSC). These cells drive the growth and dissemination of the leukemia.14 Recent studies have extended this model to other types of cancers such as breast cancer5 and Glioblastoma multiforme.6 The advances in the analysis of the origin of the leukemic cells came from studies on engraftment of patient's derived leukemia into the immunodeficient NOD-SCID mice,7,8 and following the propagation of the disease.9 These studies led to the identification of CD34 CD38 cells as the LSC fraction of AML. In addition, a high percentage of CD34 CD38 stem Correspondence: Professor D Givol, Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot 76100, Israel. E-mail: david.givol@weizmann.ac.il Received 5 May 2006; revised 6 July 2006; accepted 10 August 2006; published online 12 October 2006 cells at diagnosis was directly correlated with poor survival and significantly correlated with a high minimal residual disease frequency, especially after chemotherapy.10 An important property that is shared by both LSC and normal HSC is their ability of self-renewal, which enables the maintenance of the leukemic clone on the one hand or the normal hematopoiesis on the other hand.3,4,11,12 Some molecular mechanisms responsible for self renewal, like Bmi-1, Notch and Wnt signaling pathways, were found to be shared by both HSC and LSC.4,13,14 This suggests that the origin of LSC may be at the hematopoietc stem cell level.2 The existence of cancer stem cells is of major clinical relevance since their unique properties, such as slow mitosis, increased multidrug resistance and lower expression of Fas/Fas-L and Fas-induced apoptosis, may enable them to escape therapy15 that is based on markers or phenotype of the entire AML population. We therefore sought to examine the gene expression profiles of the LSC within the AML cell population to further understand their biology and identify new target genes that maintain and characterize LSC. For this purpose we isolated CD34 CD38 and CD34 CD38 cell populations from five AML patients and analyzed their gene expression profiles using Affymetrix Hu133A microarrays. We found 409 genes that were over or under expressed between the two populations, sorting the modulated genes showed a clear distinction between the cell populations. The GO (Gene Ontology) analysis of the modulated genes in LSC revealed a decreased expression of DNA repair, signal transduction and cell cycle regulating genes and an increased expression of genes related to protein degradation, cell adhesion, transcription and the Notch pathway. We therefore tried to inhibit the growth of LSC colonies in methyl cellulose by inhibiting the Notch pathway, an approach that may be of clinical value. Understanding the mechanism of action of genes involved in LSC has significant implications for future research and will potentially lead to new targets for therapy and diagnostics. Materials and methods Sample collection and isolation of LSC Leukemic peripheral blood (PB) samples from five AML patients were collected after obtaining informed consent. Samples were used in accordance with the procedures approved by the human experimentation and ethics committee of the Chaim Sheba Medical Center. Disease diagnosis and classification were Leukemic stem cell gene expression profiling H Gal et al 2148 according to French-American-British (FAB) criteria. Mono- elements remained unchanged. This randomization was re- nuclear cells (MNC) were purified from the samples by Ficoll peated 105 times, yielding 105 lists of modulated genes, above Hypaque density gradient centrifugation, resulting in approxi- or below twofold, in three (or more) out of five samples. The mately 13 108 cells per sample. To determine the CD34 and maximal list-size of those 105 lists of over- (and under-) CD38 content, cells were labeled with anti-human CD34- expressed genes generated by the random model was smaller fluorescein isothiocyanate (FITC) anti-CD38-phycoerythrin (PE) than the list-sizes of 148 over- and 261 underexpressed genes and IgG isotype control monoclonal antibodies (Miltenyi Biotec, obtained from the data, and thus provide an upper bound (105) Auburn, CA, USA) and the CD34 and CD38 expression levels for the P-values for our findings. An estimate of the P-values was were determined for each sample using FACS-Calibur (Becton obtained by an approximate analytical calculation (see Supple- Dickinson, Franklin Lakes, NJ, USA). Background levels were mentary Method S1 and Figure S1 for more details). The genes determined with PE-FITC labeled IgG control. Data acquisition were classified into functional categories according to the and analysis were performed with CellQuest software (Becton David17 and GeneCards databases.18 Before data analysis, we Dickinson). Each phenotype was generated by analysis of performed normalization to eliminate noise that is due to 10 000 AML cells. Cell populations CD34 CD38 and similarities within the pairs of samples of each patient. We CD34 CD38 were sorted using a FACSVantage flow performed this normalization by scanning all the genes and cytometer (Becton Dickinson) after which the cells were spun subtracting from each pair of samples (i.e., CD34 CD38 and down, resuspended by vigorous pipeting in Trizol (200 ml/ CD34 CD38 ) the mean expression (over the pair of 106cells) and stored at 701C until RNA extraction. samples) of the particular gene. We then applied the SPIN (Sorting Points Into Neighborhoods) algorithm, an unsupervised analysis tool for organization and visualization of the data.19 RNA extraction and microarray hybridization Total RNA (410 mg) of the CD34 CD38 and the CD34 CD38 cell populations of each patient was extracted Quantitative real time-polymerase chian reaction using TRIZOL (Invitrogen, Carlsbad, CA, USA), according to the QRT-PCR assays were used to determine the expression of eight manufacturer's instructions. The quality of the total RNA were modulated genes: CD38, BUB1B, IGBP7, CCL4, HES1, BCL11A, analyzed using an agarose gel. All experiments were performed RBPMS and LIMK2. Reactions were performed using the SYBR using Affymetrix Human Hu133A oligonucleotide arrays con- Green PCR Master mix with the 7900HT ABI platform (Perkin- taining 22 215 probe sets (PS) as described in the Affymetrix Elmer/Applied Biosystems, Foster City, CA, USA), as described human_datasheet.pdf (http://www.affymetrix.com/support/tech previously.22 Primers (Danyel Biotech, Rehovot, Israel) were nical/datasheets/human_datasheet.pdf) (Affymetrix, Santa Clara, designed according to Primer-Express software. The primers CA, USA). Total RNA from each sample was used to prepare used appear in supplementary Table S5. Samples were normal- biotinylated target cRNA, with minor modifications from the ized to the housekeeping gene beta-2-microglobulin (b2M), manufacturer's recommendations (Affymetrix Expression Man- whose levels of expression were not changed significantly ual). The target cRNA generated from each sample was according to the microarray data (data not shown). processed according to the manufacturer's recommendation using an Affymetrix GeneChip Instrument System (Affymetrix Expression Manual). Arrays were then washed and stained with Colony formation in methyl cellulose streptavidin-phycoerythrin before being scanned on an Affyme- The LSC (CD34 CD38) cells from sample 4 were isolated by trix GeneChip scanner. When scaling the expression values to a magnetic beads. This AML sample contained very low levels of target intensity of 150, scaling factors for all arrays were within CD38 cells (4.6%) and after separation with anti-CD34 acceptable limits (0.8551.533). Scanned output files were beads, the isolated fraction was 94.9% CD34 CD38 (similar analyzed by the probe level analysis package MAS 5.0. results were obtained when separation was done with FACS). Cells (1 105) were plated in 35 mm plates, with varying con- centrations of DAPT, in a medium of 0.9% methyl cellulose, Data analysis RPMI medium supplemented with penicillin-streptomycin, 10% The expression data for each AML patient is represented by pairs FCS, 2 mmol/l L-glutamine, and a mixture of 1% BSA with of samples, CD34 CD38 and CD34 CD38 , respectively. 5 U/ml human erythropoietin, 10 ng/ml GM-CSF, 10 ng/ml IL-3 The complete gene expression data is available at: http://www. and 100 ng/ml SCF. Assays were performed in duplicates, and weizmann.ac.il/physics/complex/compphys/downloaddata.htm. colonies or inhibition of colony growth were scored, and Gene expression values o10 were adjusted to 10 to eliminate photographed microscopically on day 14. noise from the data and subsequently all values were log2- transformed. The expression ratio for each gene in the pair of samples, CD34 CD38 and CD34 CD38 , was deter- Comparisons with other HSC profiles mined for each AML patient. Only genes that were `Present' in We compared our LSC gene expression data with normal HSC the `Present/Absent' call provided by the Affymetrix program, populations, obtained from three previous microarray studies; in at least one sample were selected, remaining with 14 125 Georgantas et al.,20 Ivanova et al.21 and Toren et al.22 valid genes. Valid genes with expression ratios above or Georgantas et al.20 described modulated genes in human below twofold in at least three out of five patients were CD34 CD38 cells, from bone marrow (BM), cord blood selected. This resulted in 148 and 261 genes, showing over- or (CB), and PB. These lists include a total of 2205 and 1999 over- underexpression in LSC, respectively. and underexpressed genes, respectively, which met the twofold The statistical significance of the lists of modulated genes was threshold criteria when compared with the CD34 CD38 tested by a random model, described in detail in the cells. Ivanova et al.21 described 822 human homologs for Supplementary Information (Method S1 and Figure S1). Briefly, murine HSC-related genes that are expressed in fetal liver. in this model, a binary matrix of modulated gene expression Analysis by Toren et al.22 of the Ivanova et al.21 data generated ratios was randomized such that the total number of matrix a list of 1905 human HSC-related genes, as described Leukemia previously.22 Toren et al.22 determined lists of HSC-related genes in CD133 cells. They obtained 244 over- and 224 underexpressed genes, by at least twofold in gene expression profiles of HSC CD133 , compared with the differentiated CD133 that were derived from CB and mobilized PB. The HSC lists from these three studies were compared with the 148 overand 261 underexpressed genes, obtained from our study (Supplementary Tables S1 and S2). The hypergeometric distribution23 was used to obtain the chance probability of observing the number of overlapping genes between our lists and the HSC data sets. Results Heterogeneity of the AML samples The karyotypic and surface markers of the five samples, as shown in Table 1, illustrate the heterogeneity of the AML samples. Nevertheless, all samples contain a variable-size Leukemic stem cell gene expression profiling H Gal et al fraction (336%) of CD34 CD38 cells that represent the LSC fraction. The AML samples were sorted into highly purified CD34 CD38 and CD34 CD38 cell fractions that were subsequently subjected to microarray analysis. Sorting results of three AML samples and the purity of the isolated fractions are shown in Figure 1. The content and purity of each fraction, for all AML patients, is shown in Table 1. Gene expression profiling of LSC reveals similarity with HSC Human Affymetrix Hu133A oligonucleotide arrays were used to compare the gene expression profiles of the highly enriched LSC, CD34 CD38, to their differentiated counterparts, CD34 CD38 cells, obtained from the AML patients. Using filtering criteria, which selected genes that were modulated by at least twofold in the CD34 CD38 versus CD34 CD38 cell fractions, in at least three out of five AML patients, we remained with a total of 409 genes. Of these 409 genes 148 2149 Table 1 Characterization of AML samples and expression of CD34 and CD38 Sample/FAB CD34+ CD34+CD38 Purity of sorted No. subtype Sex in MNC (%) in MNC (%) CD34+CD38 (%) 1 AML/M4 2 AML/M4 3 AML/M4 4 AML/M2 5 AML/M2 F F M F F 71 81 23 48 62 3 12 2.8 36 29 490 490 490 86 490 Abbreviations: FAB, French-American-British criteria for subtypes; MNC, mononuclear cells; ND, not determined. Karyotype inv(16)(p13q22) t(9;11)(p22;q23) trisomy 8 ND t(3;8) Figure 1 Representative analysis of FACS isolated AML subpopulations. AML cells were fractionated and analyzed for CD34 and CD38 content. (a) CD34 and CD38 content of three AML samples, prior to sorting. A1 and A2 quadrants containing the cell fractions CD34 CD38 and CD34 CD38, respectively. (b) Analysis of the sorted CD34 CD38 populations. (c) Analysis of the sorted CD34 CD38 populations. % Indicates the purity of the sorted fractions. Leukemia Leukemic stem cell gene expression profiling H Gal et al 2150 were overexpressed and 261 were underexpressed in the LSC data sets of gene expression, obtained from normal HSC.2022 fraction, respectively (Supplementary Tables S1 and S2). We found that 34 (23%) out of the 148 genes overexpressed in The statistical significance of obtaining this number of over- the LSC, were reported in different data sets to be typical to and underexpressed genes, as calculated by an analytic approximation to the distribution generated by the random model (see Materials and methods) is extremely low; PB3.2 1065 and PB9.8 10223, with FDR16 of 27 and normal HSC (Table 2). The probability of observing this overlap by chance using hypergeometric distribution23 was found to be low (P 5.5 103). Similarly, 104 (39%) out of the 261 genes that were underexpressed in LSC (i.e. over-expressed in 19%, for the over- and underexpressed genes, respectively. This CD34 CD38 ) showed similar behavior in normal HSC means that the number of genes that show modulated expression between the LSC and non-stem cell populations of (Supplementary Table S2), with a chance probability of P 8.6 1012. Thirty selected genes out of the 104 genes are AML is highly significant and cannot be obtained by chance. presented in Table 3. These findings suggest that some of the Overexpressed genes (73%) and 81% of the underexpressed `signature' genes of LSC are consistent with the cell stage of HSC genes are not due to chance (for more details, see Supplemen- or progenitor cells. tary Method S1 and Figure S1). The results of the microarray hybridization were further validated by QRT-PCR analysis of a group of eight genes, LSC show underexpression of cell cycle and DNA selected from the 409 modulated genes. A good correlation repair genes between the two methods was observed, as shown in Supplementary Figure S2. Analysis of the 409 modulated genes in the AML samples by GO annotation using DAVID17 revealed underexpression in genes For the visualization of the data, we applied the SPIN (Sorting Points Into Neighborhoods) analysis tool19 on the filtered genes controlling signal transduction pathways, DNA repair and cell cycle genes (Figure 3). These genes include: Cyclin A2, Cyclin and the results are summarized in the expression matrix of B2, Cyclin B1, CDC20, CDC2, CDCA8, CDC6, CDCD45A, Figure 2. The expression matrix shows a clear distinction CDC25C, BUB1 and BUB1B (Supplementary Table S2). In spite between the LSC (CD34 CD38), and their differentiated of the overexpression of cyclin D1 which is known to be counterpart CD34 CD38 cells. We searched for genes that activated in a variety of tumors, the underexpression of many are modulated in LSC in our data and compared them with three cell cycle genes is in line with the known slow cell division rate (quiescence) characteristic of stem cells.24 This is also consistent with the underexpression of several transcription factors like E2F1, E2F2, E2F8, EGR2 and the differentiation antigen MNDA (Supplementary Table S2). Interestingly, several genes known to be oncogenic are overexpressed, such as ETS1, MAF, MLL, GATA-1 and BCL11A (Supplementary Table S1). Another characteristic of LSC gene expression is the under-expression of important DNA repair genes such as: LIG1, FEN1, RAD51, POLM and TOP2A (Supplementary Table S2). These findings are consistent with the increasing chromosomal aberrations and mutations that are typical of AML. The heterogeneity of AML does not yet permit a list of signature genes typical of AML. However, it is possible to identify some genes that may be responsible for the stemness of the malignant stem cells. Interestingly, such genes may also be expressed in both LSC and normal HSC. For example, IGF1R, MLL, VEGFB and JAG2 are part of such a group (Table 2). Jagged-2, a ligand of the Notch receptor was found to be highly expressed in our LSC population. Notch pathway has been linked with the self-renewal and maintenance of HSC stem cells.2527 Notch proteins are mutated in a variety of cancers, including leukemia progenitors.28 It has been shown that other Notch ligands such as Jagged-1 and Delta-1 affect the growth and differentiation of primary AML cells29 and are important factors in the malignancy of a variety of cancers including intestinal and hematopoietic cancers.30 This result suggests that the inhibition of the Notch pathway may affect the properties and self-renewal of the LSC. To further study the role of this pathway, we analyzed the effect of inhibitors of Notch in AML samples. Figure 2 Expression matrix of the 409 modulated genes. Each row represents a single gene, and each column represents the CD34 CD38 ( /) or the CD34 CD38 ( / ) cell population of a particular AML patient (AML1AML5). The colors indicate the relative expression levels of the genes in the AML samples, according to the color code shown on the right. The expression matrix shows the partitioning of the LSC and the non-stem populations (CD34 CD38 ) in all AML patients. Effects of inhibitor of Notch pathway on LSC colony growth The Notch transmembrane receptor interacts with the transmembrane ligands Delta and Jagged, resulting in proteolytic cleavage by the secretase family that facilitate the signaling by the Notch pathway. Several enzymes are involved in this cleavage, one of which is the gamma secretase complex, which Leukemia Leukemic stem cell gene expression profiling H Gal et al Table 2 Genes overexpressed in leukemic CD34+CD38 versus CD34+CD38+, and in HSC versus hematopoietic differentiated cells Gene Name Symbol HSC data sourcea AML1 Fold changeb AML2 AML3 AML4 Insulin-like growth factor 1 receptor Neurotrophic tyrosine kinase, receptor 3 G protein-coupled receptor 1 Neurofibromin 2 Mitogen-activated protein kinase kinase 5 MAD, homolog 6 (Drosophila) Interferon regulatory factor 6 Myeloid/lymphoid or mixed-lineage leukemia B-cell CLL/lymphoma 11A Jagged 2 Vascular endothelial growth factor B Slit homolog 2 (Drosophila) Amelogenin (amelogenesis imperfecta 1, X linked) Actin binding LIM protein 1 LIM domain kinase 2 Discs, large homolog 3 (Drosophila) Bullous pemphigoid antigen 1 Integrin, beta 4 Bridging integrator 1 Myosin X Dihydrolipoamide branched chain transacylase Aldo-keto reductase family 1, member C1 Topoisomerase (DNA) III beta Cyclin D1 RNA binding protein with multiple splicing PFTAIRE protein kinase 1 Protein kinase, DNA-activated, catalytic polypeptide B4Gal-T4 Hypothetical protein MGC14817 Hypothetical protein FLJ11000 Cathepsin Z Peroxisome biogenesis factor 1 ATP synthase lipid-binding protein, mitochondrial precursor Breast epithelial mucin-associated antigen IGF1R NTRK3 GPR1 NF2 MAP2K5 MADH6 IRF6 MLL BCL11A JAG2 VEGFB SLIT2 AMELX ABLIM1 LIMK2 DLG3 BPAG1 ITGB4 BIN1 MYO10 DBT AKR1C1 TOP3B CCND1 RBPMS PFTK1 PRKDC B4GALT4 MGC14817 FLJ11000 CTSZ PEX1 ATP5G2 BPHL I I G I I I I G G, T I I I G G G I G, T I I I I G I I I, G, T I I I, G G G I I, T I I 2.47 5.67 2.04 1.06 2.85 2.78 1.00 3.07 0.35 2.91 2.98 1.12 0.84 0.98 2.09 0.46 3.91 1.00 3.17 18.55 2.65 2.02 0.81 2.40 2.34 0.90 2.01 2.04 3.34 4.21 0.85 3.85 3.12 0.39 2.51 5.38 0.73 0.51 4.54 2.46 15.53 1.68 7.86 1.13 4.80 4.65 4.06 4.32 2.51 0.68 0.41 1.08 1.59 3.97 1.37 2.18 3.96 1.01 3.74 1.00 0.77 2.83 3.03 1.73 0.18 2.77 2.22 2.56 0.92 0.30 6.04 2.15 2.21 2.55 1.48 0.93 2.69 2.61 3.43 0.25 0.51 0.37 0.79 7.14 1.94 0.36 2.42 3.72 0.41 0.59 1.18 5.21 2.30 2.95 1.82 1.61 2.17 2.70 2.29 2.42 1.31 2.41 Abbreviations: HSC, hematopoietic stem cells; LSC, leukemic stem cells. aData sets were obtained from: G, Georgantas;20 I, Ivanova;21 T, Toren.22 bThe indicated fold change is the ratio of CD34+CD38 versus CD34+CD38+ cell populations of each AML patient. 1.53 5.25 4.36 2.61 2.11 2.99 1.00 0.44 1.97 1.21 1.24 2.25 2.29 2.13 0.89 1.14 4.30 2.43 2.88 2.41 1.45 2.20 2.59 1.00 1.31 2.22 3.23 2.11 3.57 3.29 0.64 2.50 1.58 0.41 AML5 1.39 2.67 2.04 2.89 2.20 3.78 1.77 2.06 2.30 0.31 2.74 3.13 3.83 1.92 2.24 4.82 2.46 3.11 1.00 2.35 3.04 2.69 1.02 2.96 3.53 0.95 2.15 4.44 2.44 5.67 2.67 2.11 0.38 2.17 2151 cleaves the Notch receptor within the membrane. Inhibition of this enzyme intervenes with the Notch signaling pathway and many inhibitors of gamma-secretase are being used mainly in Alzheimer's disease. We attempted to affect the growth of LSC colonies in methyl cellulose by the gamma-secretase inhibitor DAPT (N-[N-(3,5-difluorophenacetyl)-L-alanyl]-S-phenyl glycine t-butyl ester).31 The LSC from AML sample 4 were isolated by magnetic beads and 1 105 cells were mixed with methyl cellulose, medium and cytokines and incubated for 14 days, as described in Materials and methods. Several photographs were taken at similar position from each plate and the comparison of colonies with and without DAPT is shown in Figure 4. It is shown that the treatment with DAPT affected the number of colonies and reduced the size of the large colonies. This effect was dose dependent and reached approximately 50% inhibition at 1.6 105 M (Figure 4a). Discussion The `cancer-stem-cell hypothesis' implies that tumors are composed of a heterogeneous population of cells, within which resides a small population of cancer stem cells that are responsible for the maintenance and propagation of the tumors. These cells possess the stem cell properties of self-renewal and of inhibited differentiation, but lost the controls operating in normal stem cells, or gained mutations that endowed them with tumorigenicity. In this model, in the case of AML, the LSC exhibit similar phenotype to HSC (e.g. CD34 CD38), but also show differences from HSC such as the expression of CD123 (IL3 receptor) that is present only on LSC.32 Hence, LSC may represent transformed HSC or progenitor cells that acquire self-renewal by this transformation.12 It was indeed demonstrated recently that some leukemia oncogenes, generated by chromosomal translocations are able, upon transfection, to confer self-renewal to hematopoietic progenitors and convert them into leukemia cells like AML.33 An important question is what is the similarity between LSC and HSC? What may be the cellular program that confers selfrenewal property on somatic cells that are normally destined for differentiation but change course towards cancer? It is expected that therapies would have to be targeted to the cancer stem cells for successful treatment of cancer. Currently failure of cancer treatment may be due to the fact that therapies are aimed at the bulk of the cancer cells and not specifically at cancer stem cells and that LSC are more resistant to such chemotherapy. Indeed, poor survival of cancer patients was correlated with high stem cell frequency in AML.10 In this study, we aimed to identify the gene expression profile of leukemic stem cells by comparing the gene expression of CD34 CD38 cells to that of CD34 CD38 from AML Leukemia Leukemic stem cell gene expression profiling H Gal et al 2152 Table 3 Genes under expressed in leukemic CD34+CD38 versus CD34+CD38+, and in HSC versus hematopoietic differentiated cells Gene Name Symbol HSC data Sourcea AML1 Fold Changeb AML2 AML3 AML4 AML5 CD14 antigen Myeloid cell nuclear differentiation antigen Elastase 2, neutrophil CCAAT/enhancer binding protein (C/EBP), delta Lamin B1 Ribonucleotide reductase M2 polypeptide Insulin-like growth factor binding protein 7 Chemokine (CC motif) ligand 4 Cyclin B2 Topoisomerase (DNA) II alpha 170 kDa Baculoviral IAP repeat-containing 5 (survivin) CD86 antigen (CD28 antigen ligand 2, B7-2 antigen) BUB1 budding uninhibited by benzimidazoles 1, beta (yeast) CDC45 cell division cycle 45-like (S. cerevisiae) Leukocyte immunoglobulin-like receptor, B4 Chemokine (CC motif) ligand 5 G protein-coupled receptor 65 G protein-coupled receptor 109B CDC20 cell division cycle 20 homolog (S. cerevisiae) CD38 antigen (p45) Growth differentiation factor 3 Ligase I, DNA, ATP-dependent BUB1 budding uninhibited by benzimidazoles 1 (yeast) Gardner-Rasheed feline sarcoma viral (v-fgr) oncogene homolog Polymerase (DNA-directed), alpha RAD51 homolog (RecA homolog, E. coli) (S. cerevisiae) Cyclin B1 Cyclin-dependent kinase inhibitor 3 Endothelial cell growth factor 1 (platelet-derived) Mitogen-activated protein kinase 13 CD14 MNDA ELA2 CEBPD LMNB1 RRM2 IGFBP7 CCL4 CCNB2 TOP2A BIRC5 CD86 BUB1B CDC45L LILRB4 CCL5 GPR65 GPR109B CDC20 CD38 GDF3 LIG1 BUB1 FGR POLA2 RAD51 CCNB1 CDKN3 ECGF1 MAPK13 T G, T G, T G G G G, I T G G G, I T G I T T G, T G G, I G I G, I G T G, I G, I G G T I 2.4 2.1 9.5 1.1 1.0 2.6 1.2 3.3 3.4 4.6 6.8 6.6 1.7 1.4 2.5 3.7 3.0 2.6 3.0 2.8 1.5 1.3 2.0 2.5 1.3 1.3 2.3 2.0 2.6 2.2 2.8 5.1 24.0 4.7 13.8 3.2 2.0 3.6 0.7 3.8 0.2 5.0 6.5 6.4 5.9 1.2 3.1 5.1 1.0 4.4 4.4 2.8 0.4 2.8 2.6 2.4 3.0 1.5 3.2 2.0 1.0 0.7 4.7 3.2 3.4 13.4 11.9 2.5 7.8 4.8 5.3 1.0 2.4 3.8 2.5 2.8 0.4 1.3 2.8 1.6 3.4 2.5 4.2 0.9 2.3 3.0 3.4 3.3 2.2 0.7 106.4 10.1 0.5 14.3 0.8 3.2 10.1 8.0 2.5 7.3 5.0 1.8 6.0 2.7 3.0 2.6 5.1 1.5 3.2 2.3 0.4 4.4 2.8 4.1 4.0 1.0 2.5 2.5 1.0 2.9 3.2 33.9 2.9 0.7 2.0 2.4 1.6 1.5 3.1 1.2 1.5 3.4 1.0 2.7 2.0 5.1 0.4 2.8 4.7 3.3 2.8 3.0 0.8 1.5 1.1 3.7 0.8 0.7 1.0 3.0 Abbreviations: LSC, leukemic stem cells; HSC, hematopoietic stem cells. aData sets were obtained from: G, Georgantas;20 I, Ivanova;21 T, Toren.22 bThe indicated fold change is the ratio of CD34+CD38 versus CD34+CD38+ cell populations of each AML patient. The `' sign indicates under expression. The full list of underexpressed genes in LSC and HSC (104 genes) is shown in Supplementary Table S2. Figure 3 Comparison between the over- and underexpressed genes in LSC, derived from five AML patients. Functional classification was performed according to GO.17,18 The percentage of each functional group was derived with respect to the total number of over- or underexpressed genes. Figure 4 The effect of DAPT on colony formation of CD34 CD38 cells. (a). The effects of DAPT concentrations on colony formation. (b, c) Representative photomicrograph of colonies, formed by 1 105 CD34 CD38 cells, plated in methyl cellulose, in the absence (b) or in the presence (c) of DAPT (1.6 105 M). (d) A single colony of CD34 CD38 cells at high magnification ( 200). Leukemia patients. We found 409 genes that were modulated between the two populations and of these 148 and 261 were over-, or underexpressed, respectively. By comparing these lists with three datasets that defined the genes expressed in normal HSC,2022 we found that 23% of the overexpressed and 39% of the underexpressed genes in LSC were present in normal HSC (Tables 2 and 3). Classification of the modulated genes in LSC revealed decreased expression of DNA repair genes, which is consistent with the increasing chromosomal aberrations that are typical to AML. Furthermore, a large number of cell cycle genes were found to be underexpressed in LSC (Supplementary Table S2). This is in line with the slow division rate, characteristic of adult stem cells.24 A relevant point is the sensitivity of stem cells to chemotherapy, which is directed against rapidly dividing cells. For example, 5-fluorouracil treatment of human leukemia which destroy most of the cells, spare a small fraction of resistant cells. Upon transplantation into SCID mice, these spared cells can reconstitute the leukemia, indicating the resistance of LSC to chemotherapy.34 Additionally in other systems, normal stem cells were found to be mitotic quiescent and differentiationinhibited.24 Analysis of LSC gene expression may identify new targets for therapy that may be directed at the stem cell fraction. We focused on the Jagged-2 gene, a ligand of the Notch signaling pathway that was found to be overexpressed in LSC in our data. The oncogenic effect of Notch signaling is well known and was reviewed recently.35 Jagged-1, a Notch ligand that is similar to Jagged-2 was shown to transform fibroblasts36 and is considered to be involved in self-renewal of HSC through the activity of the Notch pathway. We use DAPT, the gamma-secretase inhibitor of the Notch pathway, to study the effect of such inhibition on leukemic stem cells colony formation. We show that the treatment with DAPT reduced the number of colonies in LSC and reduced the size of large colonies, probably by inhibiting their proliferation. Previous experiments demonstrated that various inhibitors of the Notch pathway and particularly DAPT analogs, suppress the growth of T-ALL cell lines.28,37 Furthermore, activating Notch1 mutations occur in more than 50% of some leukemias,38 underscoring the involvement of Notch1 in the self-renewal mechanism of LSC. Clinical studies provide many examples of a positive response of tumors to drugs without statistically significant improvements in survival.39 A well known example is the case of CML (chronic myelogenous leukemia) and imatinib (STI571 Gleevec, Novartis, 2001). Although imatinib is highly active against differentiated CML progenitors, this drug may have a limited activity against the CML stem cells and therefore will not prevent relapse.40 Similar examples are known for a variety of other cancers39 and indicate the need for expanding the study of cancer stem cell population as targets for new therapy. Our work and that of others shows that the AML stem cells are different from the majority of the leukemic cells and may respond differently to therapy. Understanding the mechanism of action of genes involved in LSC has significant implications for future research on leukemia and may lead to identifying novel targets for therapy and diagnosis. Acknowledgements We thank the Kahn Family Foundation for their generous support. This research was partially supported by the Wolfson Family Charitable Trust on Tumor Cell Diversity, by the Israel Academy of Leukemic stem cell gene expression profiling H Gal et al Science, and by grants from Ruth & Allen Zeigler for Stem Cell Research. References 1 Passegue E, Jamieson CH, Ailles LE, Weissman IL. Normal and leukemic hematopoiesis: are leukemias a stem cell disorder or a reacquisition of stem cell characteristics? Proc Natl Acad Sci USA 2003; 100 (Suppl 1): 1184211849. 2 Hope KJ, Jin L, Dick JE. Human acute myeloid leukemia stem cells. Arch Med Res 2003; 34: 507514. 3 Pardal R, Clarke MF, Morrison SJ. Applying the principles of stemcell biology to cancer. Nat Rev Cancer 2003; 3: 895902. 4 Reya T, Morrison SJ, Clarke MF, Weissman IL. Stem cells, cancer, and cancer stem cells. Nature 2001; 414: 105111. 5 Al-Hajj M, Wicha MS, Benito-Hernandez A, Morrison SJ, Clarke MF. Prospective identification of tumorigenic breast cancer cells. Proc Natl Acad Sci USA 2003; 100: 39833988. 6 Singh SK, Clarke ID, Hide T, Dirks PB. Cancer stem cells in nervous system tumors. Oncogene 2004; 23: 72677273. 7 Lapidot T, Sirard C, Vormoor J, Murdoch B, Hoang T, CaceresCortes J et al. A cell initiating human acute myeloid leukaemia after transplantation into SCID mice. Nature 1994; 367: 645648. 8 Bonnet D, Dick JE. Human acute myeloid leukemia is organized as a hierarchy that originates from a primitive hematopoietic cell. Nat Med 1997; 3: 730737. 9 Rombouts WJ, Martens AC, Ploemacher RE. Identification of variables determining the engraftment potential of human acute myeloid leukemia in the immunodeficient NOD/SCID human chimera model. Leukemia 2000; 14: 889897. 10 van Rhenen A, Feller N, Kelder A, Westra AH, Rombouts E, Zweegman S et al. High stem cell frequency in acute myeloid leukemia at diagnosis predicts high minimal residual disease and poor survival. Clin Cancer Res 2005; 11: 65206527. 11 Till JE, McCulloch EA. A direct measurement of the radiation sensitivity of normal mouse bone marrow cells. Radiat Res 1961; 14: 213222. 12 Cozzio A, Passegue E, Ayton PM, Karsunky H, Cleary ML, Weissman IL. Similar MLL-associated leukemias arising from selfrenewing stem cells and short-lived myeloid progenitors. Genes Dev 2003; 17: 30293035. 13 Taipale J, Beachy PA. The Hedgehog and Wnt signalling pathways in cancer. Nature 2001; 411: 349354. 14 Park IK, Qian D, Kiel M, Becker MW, Pihalja M, Weissman IL et al. Bmi-1 is required for maintenance of adult self-renewing haematopoietic stem cells. Nature 2003; 423: 302305. 15 Costello RT, Mallet F, Gaugler B, Sainty D, Arnoulet C, Gastaut JA et al. Human acute myeloid leukemia CD34+/CD38 progenitor cells have decreased sensitivity to chemotherapy and Fas-induced apoptosis, reduced immunogenicity, and impaired dendritic cell transformation capacities. Cancer Res 2000; 60: 44034411. 16 Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J Roy Stat Soc 1995; 57: 289300. 17 Dennis Jr G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC et al. DAVID: database for annotation, visualization, and integrated discovery. Genome Biol 2003; 4: P3. 18 Rebhan M, Chalifa-Caspi V, Prilusky J, Lancet D. GeneCards: a novel functional genomics compendium with automated data mining and query reformulation support. Bioinformatics 1998; 14: 656664. 19 Tsafrir D, Tsafrir I, Ein-Dor L, Zuk O, Notterman DA, Domany E. Sorting points into neighborhoods (SPIN): data analysis and visualization by ordering distance matrices. Bioinformatics 2005; 21: 23012308. 20 Georgantas RW, Tanadve V, Malehorn M, Heimfeld S, Chen C, Carr L et al. Microarray and serial analysis of gene expression analyses identify known and novel transcripts overexpressed in hematopoietic stem cells. Cancer Res 2004; 64: 44344441. 21 Ivanova NB, Dimos JT, Schaniel C, Hackney JA, Moore KA, Lemischka IR. A stem cell molecular signature. Science 2002; 298: 601604. 22 Toren A, Bielorai B, Jacob-Hirsch J, Fisher T, Kreiser D, Moran O et al. CD133-positive hematopoietic stem cell `stemness' genes 2153 Leukemia Leukemic stem cell gene expression profiling H Gal et al 2154 contain many genes mutated or abnormally expressed in leukemia. 32 Jordan CT, Upchurch D, Szilvassy SJ, Guzman ML, Howard DS, Stem Cells 2005; 23: 11421153. Pettigrew AL et al. The interleukin-3 receptor alpha chain is a 23 Tavazoie S, Hughes JD, Campbell MJ, Cho RJ, Church GM. unique marker for human acute myelogenous leukemia stem cells. Systematic determination of genetic network architecture. Nat Leukemia 2000; 14: 17771784. Genet 1999; 22: 281285. 33 Huntly BJ, Gilliland DG. Blasts from the past: new lessons in stem 24 Guasch G, Fuchs E. Mice in the world of stem cell biology. Nat cell biology from chronic myelogenous leukemia. Cancer Cell Genet 2005; 37: 12011206. 2004; 6: 199201. 25 Karanu FN, Murdoch B, Gallacher L, Wu DM, Koremoto M, 34 Terpstra W, Ploemacher RE, Prins A, van Lom K, Pouwels K, Sakano S et al. The notch ligand jagged-1 represents a novel Wognum AW et al. Fluorouracil selectively spares acute myeloid growth factor of human hematopoietic stem cells. J Exp Med 2000; leukemia cells with long-term growth abilities in immunodeficient 192: 13651372. mice and in culture. Blood 1996; 88: 19441950. 26 Varnum-Finney B, Xu L, Brashem-Stein C, Nourigat C, Flowers D, 35 Sjolund J, Manetopoulos C, Stockhausen MT, Axelson H. The Bakkour S et al. Pluripotent, cytokine-dependent, hematopoietic Notch pathway in cancer: differentiation gone awry. Eur J Cancer stem cells are immortalized by constitutive Notch1 signaling. Nat 2005; 41: 26202629. Med 2000; 6: 12781281. 36 Ascano JM, Beverly LJ, Capobianco AJ. The C-terminal PDZ-ligand 27 Duncan AW, Rattis FM, DiMascio LN, Congdon KL, Pazianos G, of JAGGED1 is essential for cellular transformation. J Biol Chem Zhao C et al. Integration of Notch and Wnt signaling in hemato- 2003; 278: 87718779. poietic stem cell maintenance. Nat Immunol 2005; 6: 314322. 37 Weng AP, Nam Y, Wolfe MS, Pear WS, Griffin JD, Blacklow SC 28 O'Neil J, Calvo J, McKenna K, Krishnamoorthy V, Aster JC, Bassing et al. Growth suppression of pre-T acute lymphoblastic leukemia CH et al. Activating Notch1 mutations in mouse models of T-ALL. cells by inhibition of notch signaling. Mol Cell Biol 2003; 23: Blood 2006; 107: 781785. 655664. 29 Tohda S, Kogoshi H, Murakami N, Sakano S, Nara N. Diverse 38 Weng AP, Ferrando AA, Lee W, Morris JPT, Silverman LB, effects of the Notch ligands Jagged1 and Delta1 on the growth and Sanchez-Irizarry C et al. Activating mutations of NOTCH1 in differentiation of primary acute myeloblastic leukemia cells. Exp human T cell acute lymphoblastic leukemia. Science 2004; 306: Hematol 2005; 33: 558563. 269271. 30 van Es JH, Clevers H. Notch and Wnt inhibitors as potential new 39 Huff CA, Matsui W, Smith BD, Jones RJ. The paradox of response drugs for intestinal neoplastic disease. Trends Mol Med 2005; 11: and survival in cancer therapeutics. Blood 2006; 107: 431434. 496502. 40 Angstreich GR, Matsui W, Huff CA, Vala MS, Barber J, Hawkins AL 31 Dovey HF, John V, Anderson JP, Chen LZ, de Saint Andrieu P, Fang et al. Effects of imatinib and interferon on primitive LY et al. Functional gamma-secretase inhibitors reduce beta- chronic myeloid leukaemia progenitors. Br J Haematol 2005; 130: amyloid peptide levels in brain. J Neurochem 2001; 76: 173181. 373381. Supplementary Information accompanies the paper on the Leukemia website (http://www.nature.com/leu) Leukemia