Expression of HOXB genes is significantly different in acute myeloid leukemia with a partial tandem duplication of MLL vs. a MLL translocation: a cross-laboratory study

Cancer Genetics 204 (2011) 252e259

Expression of HOXB genes is significantly different in acute myeloid leukemia with a partial tandem duplication of MLL vs. a MLL translocation: a cross-laboratory study Hsi-Che Liu a,b, Lee-Yung Shih c,d, Mei-Ju May Chen e, Chien-Chih Wang f, Ting-Chi Yeh a, Tung-Huei Lin d, Chien-Yu Chen e,g, Chih-Jen Lin f, Der-Cherng Liang a,* a

Division of Pediatric Hematology-Oncology, Mackay Memorial Hospital, Taipei, Taiwan; b Mackay Medical College, Taipei, Taiwan; c Division of Hematology-Oncology, Department of Internal Medicine, Chang Gung Memorial Hospital, Taipei, Taiwan; d School of Medicine, Chang Gung University, Taoyuan, Taiwan; e Genome and Systems Biology Degree Program, National Taiwan University, Taipei, Taiwan; f Department of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan; g Department of Bio-industrial Mechatronics Engineering, National Taiwan University, Taipei, Taiwan In acute myeloid leukemia (AML), the mixed lineage leukemia (MLL) gene may be rearranged to generate a partial tandem duplication (PTD), or fused to partner genes through a chromosomal translocation (tMLL). In this study, we first explored the differentially expressed genes between MLL-PTD and tMLL using gene expression profiling of our cohort (15 MLL-PTD and 10 tMLL) and one published data set. The top 250 probes were chosen from each set, resulting in 29 common probes (21 unique genes) to both sets. The selected genes include four HOXB genes, HOXB2, B3, B5, and B6. The expression values of these HOXB genes significantly differ between MLL-PTD and tMLL cases. Clustering and classification analyses were thoroughly conducted to support our gene selection results. Second, as MLL-PTD, FLT3-ITD, and NPM1 mutations are identified in AML with normal karyotypes, we briefly studied their impact on the HOXB genes. Another contribution of this study is to demonstrate that using public data from other studies enriches samples for analysis and yields more conclusive results. Keywords Acute myeloid leukemia, MLL-PTD, HOXB genes, gene expression profiling ª 2011 Elsevier Inc. All rights reserved.

Rearrangements of the mixed lineage leukemia (MLL) gene located in chromosome band 11q23 are commonly involved in acute leukemia. In acute myeloid leukemia (AML), the MLL gene may be either fused to a variety of partner genes through chromosomal translocation (tMLL) (1) or rearranged to generate a partial tandem duplication (PTD) (2). In our previous study, MLL-partial tandem duplication (MLL-PTD) was detected in 7.1% of de novo AML (3). The majority of MLL-PTD was observed in AML cases lacking major translocations.

Received March 25, 2010; received in revised form January 18, 2011; accepted February 7, 2011. * Corresponding author. E-mail address: [email protected] 2210-7762/$ - see front matter ª 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.cancergen.2011.02.003

In particular, 36/47 cytogenetically characterized MLL-PTD cases were AML with normal karyotypes (NK) (3). In contrast to the tMLL fusion protein, the MLL-PTD protein retains the SET domain, which functions as a lysine-directed histone methyltransferase and impacts the transcriptional regulation of HOX genes (4,5). In a murine model, the MLL-PTD serves as a gain-of-function allele and leads to proliferation and self-renewal advantages to hematopoietic stem/progenitor cells without causing leukemia (6). It seems apparent that AML with tMLL and AML with MLL-PTD are two subtypes with different molecular aberrations. MLL-PTD was reported to be an adverse prognostic molecular marker identified in AML with NK (7). However, Whitman et al. (8) suggested that intensive consolidation therapy may contribute to better outcomes for AML with

HOXB gene expression in AML MLL-PTD. With regard to the impact of concurrent biomarkers in treatment outcomes, Whitman et al. highlighted a higher percentage of mutant FLT3 internal tandem duplication (FLT3-ITD) and wild-type nucleophosmin (NPM1) in patients who relapsed. Using real-time quantitative PCR, the expression level of MLL-PTD was shown to be a stable and useful marker of minimal residual disease (9). We conducted the first study to compare clinical outcome between patients with tMLL and MLL-PTD, and demonstrated that there were no differences in complete remission rate, remission duration, event-free survival, and overall survival rates (3). Based on gene expression profiling, tMLL was shown to be a distinct subtype in AML (10). Some studies demonstrated that AML with MLL-PTD could be discriminated from AML with tMLL (11,12). However, Ross et al. (10) showed there was no distinct expression signature for AML with MLL-PTD, and they failed to define a clear relationship between MLL-PTD and tMLL. In this paper, rather than defining the specific expression signature of MLL-PTD, we investigated whether there are differentially expressed genes between MLL-PTD and tMLL. We used our cohort as well as published data to explore and validate these findings. Clustering and classification analysis were thoroughly performed. Then we briefly studied the impact of the co-existing mutations FLT3-ITD and NPM1 for differentiation between MLL-PTD and tMLL. It is worth noting that the public data used here were originally used for other investigations. Our cross-laboratory approach helps enrich data for analysis and validation.

Materials and methods Patients and samples In this study, the basis for tMLL patient selection was according to the four most common subtypes from our previous study d MLL-AF6, MLL-AF9, MLL-AF10 and MLL-ELL (3). A total of 15 MLL-PTD and 10 tMLL patients were included. Their clinical features are summarized in Table 1. All patients, except one, were adults. The diagnosis and classification of AML were established according to FrencheAmericaneBritish (FAB) classification (13e15). The karyotypes were interpreted according to the International System for Human Cytogenetic Nomenclature (16). All analyzed samples were obtained from bone marrow (BM) aspirates with informed consent. BM mononuclear cells were enriched with Ficoll-Hypaque density gradient centrifugation and cryopreserved until they were tested. The samples contained more than 80% blast cells after thawing.

Molecular characterization Genomic DNA was extracted from frozen BM mononuclear cells with a DNA extraction kit (Puregene Gentra System, Minneapolis, MN) according to the manufacturer’s instructions. RNA was extracted and reverse transcribed to cDNA using methods described in our previous work (17). The identification of MLL rearrangement and characterization of MLL fusion partners were described previously (3). Briefly, Southern blot analysis was performed to screen for a

253 Table 1

Clinical features of our 25 patients

Total number Gender Male Female Median age (yr) (range) FAB subtypes M0 M1 M2 M4 M5a M6 Molecular genetics MLL-AF10 MLL-ELL MLL-AF6 MLL-AF9 a

MLL-PTD

tMLL

15

10

4 11 55 (22e84)

4 6 42a (22e78)a

1 4 7 2 0 1

0 2 1 3 4 0 1 2 3 4

Exclusion of one 2-year-old child.

MLL rearrangement (18). Reverse-transcriptase polymerase chain reaction (RT-PCR) was used to detect the common fusion transcripts from MLL-AF6, MLL-AF9, MLL-AF10, and MLL-ELL (19e21). The genomic DNA PCR assay was performed to identify the FLT3-ITD. GeneScan analysis (Applied Biosystems, Perkin-Elmer, Foster City, CA) to determine the allelic distribution was then performed for those PCR products with aberrant bands (22). Direct sequencing of PCR products was also carried out in each sample with a FLT3-ITD. The amplification of exon 12 in NPM1, purification and sequencing of PCR products, and further confirmation of NPM1 mutations were performed according to the methods described by Falini et al. (23).

Gene expression profiling Total RNA was extracted with Trizol reagent (Gibco, Grand Island, NY) according to the manufacturer’s instructions. The concentration and purity/integrity of the extracted RNA were determined by the Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA) using 1 mg of total RNA as the starting material. cDNA synthesis, biotin-labeled cRNA (target) synthesis, HG-U133 Plus 2.0 array hybridization, staining, and scanning were all performed according to the standard protocols supplied by Affymetrix (Santa Clara, CA). During the experiments, the cRNA yields and fragmentations were checked to ensure that quality control criteria were met. Affymetrix GeneChip Operating Software was used to compute signal values from the image files. All 25 samples passed the minimal quality control parameters, and the minimal percentage of probe sets with “present calls” was 36.3%. The complete microarray data set is available at Gene Expression Omnibus (accession no. GSE15013).

Three public data sets Studies in recent years have accumulated a huge amount of AML microarray samples in public databases. The emergence

254

H.-C. Liu et al.

of common guidelines MIAME (Minimum Information About a Microarray Experiment; (24)), adapted by scientific journals, has led to the possible universal use of public data. In this study, we used three public data sets in addition to our own data (Table 2). The first one, from Ross et al. (10), was a childhood study using the Affymetrix HG-U133A microarray. This data set included AML cases with MLL-PTD, tMLL, and NK. The second data set, from Valk et al. (25, 26), included adult AML samples in subtypes of tMLL, NK, and mutations of FLT3-ITD/NPM1. The Affymetrix HG-U133A microarray was used in the second data set as well. The last data set, generated by Bullinger et al. (27), included cDNA arrays of adult AML subtypes of MLL-PTD, tMLL, NK, and FLT3-ITD mutation. Table 2 summarizes the key genetic characteristics of all reference data including the sample numbers of analyzed subtypes in each study. Out of all four data sets, three of them provide MLL-PTD labels (this study, Ross et al., and Bullinger et al.) and three of them have information about the FLT3-ITD mutation (this study, Valk et al., and Bullinger et al.). Moreover, two of them have NPM1 mutation status (this study and Valk et al.). Note that these three public data sets were not used for analyzing the relationship between MLL-PTD and tMLL, nor were they used for studying the impact of concurrent mutations on MLL-PTD/tMLL. After expression values were downloaded from the referred public websites, values from each HG-U133A array were rescaled by setting the 2% trimmed mean of all the genes in an array to be 500, as suggested in the Affymetrix Microarray Suite 5.0 software (Affymetrix statistical algorithms description document, 2002 Rev 3). For cDNA microarrays, we used the raw data directly without rescaling.

Gene selection Two data sets, our cohort (HG-U133 Plus 2.0) and the Ross et al. cohort (HG-U133A), were first used to investigate whether AML with MLL-PTD can be distinguished from AML with tMLL by gene expression profiling. Since two generations of Affymetrix microarrays (HG-U133A and HG-U133 Plus 2.0) were used, it was important to identify a subset of common genes (28). HG-U133 Plus 2.0 contains all the probe sets used in HG-U133A, so the identification of common genes is straightforward. We selected 22,277 probe sets shared by both generations. The Student’s t-test, invoked with the setting of assuming an equal variance, was performed to find differentially expressed genes. The significant probe sets were obtained by sorting their P values Table 2

in ascending order. Significance analysis of microarray (SAM; (29)) was also used to select genes differentially expressed between MLL-PTD and tMLL. In addition to using individual sets for finding genes, the Student’s t-test and SAM were applied to the combined set derived from the two original data sets. To functionally classify the differentiated genes, Gene Ontology (GO) classification was performed with the BiNGO 2.3 (30) plug-in for the Cytoscape (31). Selected enriched GO terms were those that comprised at least three genes with a P < 0.05 obtained by hypergeometric statistics (cluster versus the whole annotation bank). All experimental analyses in this paper, unless specified otherwise, were conducted by software packages included in the R-project (32).

Clustering For validating results, hierarchical clustering analysis was conducted on different data sets based on several gene subsets of interest. In the clustering algorithm, Pearson’s correlation coefficient was used for measuring object similarity and average linkage was employed for cluster similarity. For each gene, the expression values of the different probe sets were averaged before clustering analysis. The cDNA set was used to validate the gene selection process described above. The expression values of genes were pre-processed according to the method in Bullinger et al. (27). Some expression values were missing, so several samples or genes were removed before the analysis.

Classification To further estimate the differential power of the selected genes, classification analysis was performed by the k-Nearest Neighbors (KNN) application (33). Our cohort and the study of Ross et al. were alternatively taken as training and testing sets. For the cross-laboratory data prediction, some studies have shown that considering a gene’s rank within an array is better than the expression value (28). Thus, we used gene rank information to prepare training and testing sets. The expression value of a gene was substituted with its rank within a single array. Some values were replaced by the mean of ranks if ties happened among a group of probe sets. Since the performance of KNN depends on the parameter k, a leave-one-out cross-validation (LOO CV) procedure was applied to the training set for acquiring a suitable k. For any given k, LOO CV sequentially selected one training instance for validation; i.e.,

Microarray datasets and genetic characteristics of selected patient cohorts Genetic subtype

Study reference This study Ross et al. (10) Valk et al. (25, 26) Bullinger et al. (27)b

Microarray platform U133 Plus 2.0 U133A U133A cDNA

Total samples 25 130 285 119

MLL-PTD 15 13 e 10

tMLL 10 23 19 8

NPM1þ 0 e 95 e

FLT3-ITDþ 9 (8) e 78 30

a

NK 15 (15) 16 (5) 119 45 (3)

Abbreviation: e, not available. a Numbers in parentheses in the FLT3-ITDþ and NK columns indicate the number of MLL-PTD included. b For the cDNA data set, some expression values are missing. In clustering analysis, according to the subtypes of samples and the subset of genes to be investigated, data with no missing information are retained.

HOXB gene expression in AML KNN predicted the class of the held instance by the dominant class of its k nearest neighbors in the remaining set. The distance between any two data instances was measured by the Euclidean metric. Only odd integers were checked to search for the best k, because an even number may cause both classes to be equally dominant. Finally, the value k with the best LOO CV accuracy was applied to predict the independent testing data.

Results To identify the differentially expressed genes between MLL-PTD and tMLL, two data sets were used. The first one was our own data set, containing 25 AML samples of which 15 were MLL-PTD and 10 were tMLL. Among the 15 MLL-PTD samples, all had NK without NPM1 mutation and 8 were FLT3-ITD positive. In the 10 tMLL samples, there was one FLT3-ITD mutation and one NPM1 mutation. Clinical characteristics and molecular genetics of each sample are listed in Supplementary Table 1. The second was the data set from Ross et al. (10), which included 13 MLL-PTD and 23 tMLL. Details of these two sets are shown in Table 2. By Student’s t-test, the top 250 significant probe sets to differentiate MLL-PTD and tMLL were selected from each set. Then, the two lists of top 250 probe sets were compared and 29 probes sets, corresponding to 21 unique genes, were found to be in common. These probe sets are referred to as the “29-probe list” throughout the rest of this paper. GO analysis revealed enrichment in the 29-probe list in biological processes including transcription regulation (HOXB2, B3, B5, B6, ATXN1, CTNND1, and MLL), cell proliferation (CLEC11A, Table 3

255 DAB2, EPS15, FSCN1, and IGFBP7), and cell death (ATXN1, MLL, and TETRAN ). More information on the differentially expressed genes is shown in Table 3. SAM identified 25 probes overlapping in each set at a false discovery rate (FDR) level below 5%. Furthermore, 17/25 probes were common to the 29-probe list. To further verify the 29-probe list, additional analyses were conducted on the two combined data sets. Among the top 50 probe sets selected by the t-test, 22 probes overlapped with the 29-probe list. All of these genes were significant according to the method proposed by Benjamini and Hochberg (34), which is implemented in the R multitest package (FDR < 5%). Using the SAM method on the combined set with FDR < 5%, 21 probes were repeated in the 29-probe list. The lists of probes by the above analyses are shown in Supplementary Table 2.

Validation of the selected genes: clustering Figures 1A and 1B, respectively, show the clustering results of using the 21 gene expression values of our own data set and the data set from Ross et al. (10). MLL-PTD and tMLL samples are clearly well separated. To further validate our analysis, we applied these 21 genes to cluster the cDNA data set of Bullinger et al. By mapping the 21 selected genes to the cDNA data, 10 genes were matched. After removing cDNA arrays containing missing values, we obtained a subset of 12 samples (4 MLL-PTD and 8 tMLL). A clustering analysis shows that MLL-PTD and tMLL cases can be separated into two groups (Supplementary Figure 1).

List of 21 significantly and differentially expressed genes between MLL-PTD and tMLL

Symbol

Description

Reported function

AK2 ATXN1 C11orf32

Adenylate kinase 2 Ataxin 1 Sortilin-related receptor, L (DLR class) A repeats containing (SORL1) Chromosome 5 open reading frame 23 C-type lectin domain family 11, member A Chloride intracellular channel 4 Clathrin, heavy chain-like 1 Catenin (cadherin-associated protein), delta 1 Disabled homolog 2, mitogen-responsive phosphoprotein Epidermal growth factor receptor pathway substrate 15 Fascin homolog 1, actin-bundling protein Homeobox B2 Homeobox B3 Homeobox B5 Homeobox B6 Insulin-like growth factor binding protein 7 Kinectin 1 (kinesin receptor) Muscleblind-like Myeloid/lymphoid or mixed-lineage leukemia (trithorax homolog) Natriuretic peptide receptor C/guanylate cyclase C (atrionatriuretic peptide receptor C) Major facilitator superfamily domain containing 10 (MFSD10)

Nucleic acid metabolic process Cell death, RNA processing, transcription regulation Lipid transport

C5orf23 CLEC11A CLIC4 CLTCL1 CTNND1 DAB2 EPS15 FSCN1 HOXB2 HOXB3 HOXB5 HOXB6 IGFBP7 KTN1 MBNL1 MLL NPR3 TETRAN

Unknown Cell proliferation Cell differentiation, chloride transport Protein transport Wnt receptor signaling pathway, transcription regulation Cell proliferation Cell proliferation, protein transport Cell proliferation Transcription regulation Transcription regulation Transcription regulation Transcription regulation Cell growth, cell proliferation Microtubule-based movement RNA processing Histone methyltransferase activity, DNA methylation, cell death, transcription regulation Signal transduction Cell death, transmembrane transport

256

H.-C. Liu et al.

Figure 1 (A) Clustering analysis on our cohort using the 21 significant genes. Hierarchical clustering was performed on the 25 patients (columns) of our cohort using the 21 selected genes (rows). (B) Clustering analysis on the data set of Ross et al. using the 21 significant genes. A total of 36 samples (columns) are hierarchically clustered using the 21 selected genes (rows). The normalized expression value for each gene is represented by a color. Red and green cells indicate high and low expression, respectively. The genetic information of each sample is represented by a black (positive) or white (negative) box below the dendrogram.

HOXB gene expression in AML

257

Validation of the selected genes: classification Two settings of classification analysis were performed. One used all of the probe sets as features, and the other used the selected 21 significant genes. Each experiment was evaluated according to accuracy, determined by the percentage of cases from the test set whose AML subtypes (MLL-PTD or tMLL) were correctly predicted. Table 4 shows the accuracy for classifying MLL-PTD and tMLL samples. For each pair of training and testing sets, data was prepared using either the 21 selected genes or all of the genes shared by the HG-U133A and HG-U133 Plus 2.0 platforms. The high accuracy of using 21 significant genes validates our gene selection procedure.

HOXB genes, rather than HOXA genes, are selected in the significant gene set In MLL-PTD cases, 14/21 significant genes were overexpressed and 7 were underexpressed. Two MLL probe sets were chosen as up-regulated. Interestingly, none of the HOXA genes was selected, but four of nine HOXB genes, including HOXB2, B3, B5 and B6, had high expression values in MLL-PTD cases. These HOXB genes were consistently selected as differentially expressed genes in other analyses mentioned above (Supplementary Table 2). In addition, HOXB7, though not in the top 250, was also ranked high (data not shown). The impact of co-existing genetic subtypes was studied briefly. In clustering our own data, Figure 1A shows that the nine FLT3-ITDepositive cases (including one in tMLL) cannot be distinguished by HOXB expression. By analyzing the data from Valk et al. and Bullinger et al., clustering results (Supplementary Figure 2, Supplementary Table 3, and Supplementary Figure 3) show that some FLT3-ITDepositive cases (negative cases) had underexpressed (overexpressed) values of the selected HOXB genes. In clustering the NK cases of the Valk data, HOXB genes did not show uniformly high expression (Supplementary Figure 2 and Supplementary Table 3). Moreover, by applying the HOXB genes to tMLL and NK cases in the data of Ross et al., these two subtypes cannot be distinctly separated (Supplementary Figure 4).

Discussion Past microarray studies regarding the possible distinction between MLL-PTD and tMLL Gene expression patterns are a useful tool for looking for biological differences in AML cases with a MLL-PTD and tMLL. Table 4 Classification analysis of our cohort and the data set from Ross et al. (10) for classifying MLL-PTD and tMLL samples Accuracy Training / Testing

21 selected genes

All genesa

This study / Ross et al. Ross et al. / This study

94.4% 92.0%

80.6% 88.0%

a

Genes shared by HG-U133A and HG-U133 Plus 2.0 platforms.

Studies in recent years have accumulated a huge amount of microarray data from AML samples in public databases. Major subtypes of AML were well classified and predicted by gene expression profiling (10,25,27). Some genetic aberrations (e.g. MLL-PTD), however, failed to have extensive studies, due in part to relatively limited case numbers. In the abstract report by Schnittger et al. (11), specific expression profiling was found to differentiate 10 tMLL cases from 15 MLL-PTD cases, as well as 30 NK-AML cases without a MLL-PTD, respectively. Rozovskaia et al. (12) also suggested that molecular differences exist between the two subtypes by clustering 3 MLL-PTD and 12 tMLL cases. Rather than identifying differentially expressed genes between MLL-PTD (13 cases) and tMLL (23 cases), Ross et al. (10) performed two different comparisons: first, tMLL alone versus other AML samples, and second, the combination of tMLL and MLL-PTD versus other AML samples. Fewer discriminating genes were found when using the combined group of MLL-PTD and tMLL. They thus suggested that MLL-PTD might be distinct from tMLL at a molecular heterogeneous level.

Bioinformatics exploration of the 21 significant genes In this study, using our own cohort and the data from Ross et al., we identified 21 significant genes that can possibly differentiate MLL-PTD from tMLL. This result was supported by extensive clustering and classification analyses. Furthermore, some of these genes have been reported to be related to tumorigenesis. For example, the down-regulation of DAB2 may play an important role in ovarian carcinogenesis (35). CLIC4, another differentially expressed gene found to be significant, is a member of a family of intracellular chloride channels. The loss of CLIC4 in tumor cells and the gain in tumor stroma have been found in renal, ovarian, and breast cancers, and indicated malignant progression (36). Two probe sets of MLL genes were chosen as up-regulated in MLL-PTD. Interestingly, none of the HOXA genes, which were thought to be the possible mechanism of leukemogenesis in AML patients with tMLL (37,38), were selected. Instead, four HOXB genes (HOXB2, B3, B5, and B6) were differentially expressed in our analysis. The overexpression of these genes was reported to be associated with AML cases with NPM1 mutations (39). The four HOXB genes were also shown to be significant in differentiating AML with NPM1 from AML with tMLL (40). In addition, HOXB2 and B5 were included in a 20-gene classifier to predict FLT3-ITD mutation in NK-AML samples (41). In a recent study, HOXB2, B3, B5, and B6 were shown to be down-regulated in NK-AML patients with CEBPA mutations (42). HOXB genes were suggested to have a role in erythroid or granulopoietic (eosinophilic) lineage maturation and hematopoietic progenitor lineage commitment (43). The study of Giampaolo et al. (44) provided novel evidence that HOXB genes are coordinately expressed and play a key functional role in the early stages of hematopoiesis. The function of HOXB genes is bilineage (B3, B4, and B5) or restricted to granulopoiesis (B6). The same study reported that the HOXB3 gene is particularly important. Giampaolo et al. (45) further demonstrated by RT-PCR that the HOXB6

258 gene was expressed at a very low incidence in AML with tMLL (1/9 cases), but frequently expressed (18/49 cases) in NK-AML. They concluded that abnormalities of HOXB6 expression may promote the development of the leukemic phenotype.

Impact of other genetic subtypes on HOXB genes The cytogenetics of AML at diagnosis provides the most important prognostic information, but 40e50% of patients present with normal karyotypes (46). Similar to MLL-PTD, mutations in FLT3-ITD and NPM1 have been identified in NK-AML and were often not mutually exclusive. The presence of these mutations was reported to carry prognostic information (47). It is thus interesting to preliminarily study the behavior of HOXB genes in these subtypes of NK-AML. Our analysis of three large-scale datasets suggests that FLT3-ITD/normal karyotypes may not cause the overexpression of HOXB genes in MLL-PTD. Since the coexistence of MLL-PTD and NPM1 mutation is rare (47), the impact of NPM1 mutation on the HOXB genes in differentiating MLL-PTD from tMLL is difficult to evaluate. We conclude that AML with MLL-PTD and AML with tMLL can be differentiated on the basis of the gene expression signature of 21 selected genes, including four HOXB genes (B2, B3, B5, and B6). Clustering and classification analyses were thoroughly conducted to support our gene selection results. In addition, our preliminary analysis shows that the overexpression of HOXB genes in MLL-PTD may not be related to the FLT3-ITD/normal karyotype status. We also demonstrated that public data originally considered for other purposes can be extended effectively for additional analyses.

Acknowledgments We thank the National Microarray and Gene Expression Analysis Core Facility of the National Research Program for Genomic Medicine, Taiwan, for performing target labeling and hybridization of microarray experiments. The work was supported by grants MMH-E-98009 from Mackay Memorial Hospital, NHRI-EX96-9434SI from the National Health Research Institute, and NSC97-2314-B-182-011-MY3 from the National Science Council, Taiwan. We thank anonymous reviewers for valuable comments. We also thank Hao-Han Liaw and Yun-Chieh Sung for their great help in collecting the required information and data for revision.

Supplementary data Supplementary data associated with this article can be found online at 10.1016/j.cancergen.2011.02.003.

References 1. Ziemin-van der Poel S, McCabe NR, Gill HJ, et al. Identification of a gene, MLL, that spans the breakpoint in 11q23 translocations associated with human leukemias. Proc Natl Acad Sci USA 1991; 88:10735e10739. 2. Schichman SA, Caligiuri MA, Gu Y, et al. ALL-1 partial duplication in acute leukemia. Proc Natl Acad Sci USA 1994;91:6236e6239.

H.-C. Liu et al. 3. Shih LY, Liang DC, Fu JF, et al. Characterization of fusion partner genes in 114 patients with de novo acute myeloid leukemia and MLL rearrangement. Leukemia 2006;20: 218e223. 4. Basecke J, Whelan JT, Griesinger F, et al. The MLL partial tandem duplication in acute myeloid leukaemia. Br J Haematol 2006;135:438e449. 5. Li ZY, Liu DP, Liang CC. New insight into the molecular mechanisms of MLL-associated leukemia. Leukemia 2005;19: 183e190. 6. Dorrance AM, Liu S, Yuan W, et al. Mll partial tandem duplication induces aberrant Hox expression in vivo via specific epigenetic alterations. J Clin Invest 2006;116:2707e2716. 7. Schnittger S, Kinkelin U, Schoch C, et al. Screening for MLL tandem duplication in 387 unselected patients with AML identify a prognostically unfavorable subset of AML. Leukemia 2000;14: 796e804. 8. Whitman SP, Ruppert AS, Marcucci G, et al. Long-term diseasefree survivors with cytogenetically normal acute myeloid leukemia and MLL partial tandem duplication: a Cancer and Leukemia Group B study. Blood 2007;109:5164e5167. 9. Weisser M, Kern W, Schoch C, et al. Risk assessment by monitoring expression levels of partial tandem duplications in the MLL gene in acute myeloid leukemia during therapy. Haematologica 2005;90:881e889. 10. Ross ME, Mahfouz R, Onciu M, et al. Gene expression profiling of pediatric acute myelogenous leukemia. Blood 2004;104: 3679e3687. 11. Schnittger S, Kohlmann A, Haferlach T, et al. Acute myeloid leukemia (AML) with partial tandem duplication of the MLL-gene (MLL-PTD) can be discriminated from MLL-translocation based on specific gene expression profiles (abstract). Blood 2002;100: abstract 1202. 12. Rozovskaia T, Ravid-Amir O, Tillib S, et al. Expression profiles of acute lymphoblastic and myeloblastic leukemias with ALL-1 rearrangements. Proc Natl Acad Sci USA 2003;100:7853e7858. 13. Bennett JM, Catovsky D, Daniel MT, et al. Proposed revised criteria for the classification of acute myeloid leukemia. Ann Intern Med 1985;103:620e625. 14. Bennett JM, Catovsky D, Daniel MT, et al. Criteria for the diagnosis of acute leukemia of megakaryocyte lineage (M7). A report of the French-American-British Cooperative Group. Ann Intern Med 1985;103:460e462. 15. Bennett JM, Catovsky D, Daniel MT, et al. Proposal for the recognition of minimally differentiated acute myeloid leukemia (AML-MO). Br J Haematol 1991;78:325e329. 16. Mitelman F. ISCN. An International System for Human Cytogenetic Nomenclature. Basel: S Karger; 1995. 17. Liang DC, Shih LY, Yang CP, et al. Molecular analysis of fusion transcripts in childhood acute myeloid leukemia in Taiwan. Med Pediatr Oncol 2001;37:555e556. 18. Thirman MJ, Gill HJ, Burnett RC, et al. Rearrangement of the MLL gene in acute lymphoblastic and acute myeloid leukemias with 11q23 chromosomal translocations. N Engl J Med 1993; 329:909e914. 19. Repp R, Borkhardt A, Haupt E, et al. Detection of four different 11q23 chromosomal abnormalities by multiplex-PCR and fluorescence based automatic DNA-fragment analysis. Leukemia 1995;9:210e215. 20. Chaplin T, Bernard O, Beverloo HB, et al. The t(10;11) translocation in acute myeloid leukemia (M5) consistently fuses the leucine zipper motif of AF10 onto the HRX gene. Blood 1995; 86:2073e2076. 21. Rubnitz JE, Behm FG, Curcio-Brint AM, et al. Molecular analysis of t(11;19) breakpoints in childhood acute leukemias. Blood 1996;87:4804e4808. 22. Shih LY, Huang CF, Wu JH, et al. Internal tandem duplication of FLT3 in relapsed acute myeloid leukemia: a comparative

HOXB gene expression in AML

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33. 34.

35.

analysis of bone marrow samples from 108 adult patients at diagnosis and relapse. Blood 2002;100:2387e2392. Falini B, Mecucci C, Tiacci E, et al. Cytoplasmic nucleophosmin in acute myelogenous leukemia with a normal karyotype. N Engl J Med 2005;352:254e266. Brazma A, Hingamp P, Quackenbush J, et al. Minimum information about a microarray experiment (MIAME)-toward standards for microarray data. Nat Genet 2001;29:365e371. Valk PJM, Verhaak RGW, Beijen MA, et al. Prognostically useful gene-expression profiles in acute myeloid leukemia. N Engl J Med 2004;350:1617e1628. Verhaak RGW, Goudswaard CS, van Putten W, et al. Mutations in nucleophosmin (NPM1) in acute myeloid leukemia (AML): association with other gene abnormalities and previously established gene expression signatures and their favorable prognostic significance. Blood 2005;106:3747e3754. €hner K, Bair E, et al. Use of gene-expression Bullinger L, Do profiling to identify prognostic subclasses in adult acute myeloid leukemia. N Engl J Med 2004;350:1605e1616. Liu HC, Chen CY, Liu YT, et al. Cross-generation and crosslaboratory predictions of Affymetrix microarrays by rank-based methods. J Biomed Inform 2008;41:510e519. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA 2001;98:5116e5121. Maere S, Heymans K, Kuiper M. BiNGO: a Cytoscape plugin to assess overrepresentation of Gene Ontology categories in Biological Networks. Bioinformatics 2005;21:3448e3449. Shannon P, Markiel A, Ozier O, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res 2003;13:2498e2504. Development Core Team of R. R: A language and environment for statistical computing. R Foundation for Statistical Computing; Vienna, Austria; 2005. Available at: http://www.R-project.org. Cover T, Hart P. Nearest neighbor pattern classification. IEEE Trans Information Theory 1967;13:21e27. Klipper-Aurbach Y, Wasserman M, Braunspiegel-Weintrob N, et al. Mathematical formulae for the prediction of the residual beta cell function during the first two years of disease in children and adolescents with insulin-dependent diabetes mellitus. Med Hypotheses 1995;45:486e490. Santin AD, Zhan F, Bellone S, et al. Gene expression profiles in primary ovarian serous papillary tumors and normal ovarian epithelium: identification of candidate molecular markers for

259

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

46. 47.

ovarian cancer diagnosis and therapy. Int J Cancer 2004;112: 14e25. Suh KS, Crutchley JM, Koochek A, et al. Reciprocal modifications of CLIC4 in tumor epithelium and stroma mark malignant progression of multiple human cancers. Clin Cancer Res 2007;13:121e131. Armstrong SA, Staunton JE, Silverman LB, et al. MLL translocations specify a distinct gene expression profile that distinguishes a unique leukemia. Nat Genet 2002;30:41e47. Kohlmann A, Schoch C, Dugas M, et al. New insights into MLL gene rearranged acute leukemias using gene expression profiling: shared pathways, lineage commitment, and partner genes. Leukemia 2005;19:953e964. Andreeff M, Ruvolo V, Gadgil S, et al. HOX expression patterns identify a common signature for favorable AML. Leukemia 2008; 22:2041e2047. Mullighan CG, Kennedy A, Zhou X, et al. Pediatric acute myeloid leukemia with NPM1 mutations is characterized by a gene expression profile with dysregulated HOX gene expression distinct from MLL-rearranged leukemias. Leukemia 2007;21: 2000e2009. €hner K, Kranz R, et al. An FLT3 gene-expression Bullinger L, Do signature predicts clinical outcome in normal karyotype AML. Blood 2008;111:4490e4495. Marcucci G, Maharry K, Radmacher MD, et al. Prognostic significance of, and gene and microRNA expression signatures associated with, CEBPA mutations in cytogenetically normal acute myeloid leukemia with high-risk molecular features: a Cancer and Leukemia Group B Study. J Clin Oncol 2008;26:5078e5087. Giampaolo A, Pelosi E, Valtieri M, et al. HOXB gene expression and function in differentiating purified hematopoietic progenitors. Stem Cells 1995;13(Suppl 1):90e105. Giampaolo A, Sterpetti P, Bulgarini D, et al. Key functional role and lineage-specific expression of selected HOXB genes in purified hematopoietic progenitor differentiation. Blood 1994;84: 3637e3647. Giampaolo A, Felli N, Diverio D, et al. Expression pattern of HOXB6 homeobox gene in myelomonocytic differentiation and acute myeloid leukemia. Leukemia 2002;16:1293e1301. zek K, Heerema NA, Bloomfield CD. Cytogenetics in acute Mro leukemia. Blood Rev 2004;18:115e136. € hner K, Krauter J, et al. Mutations and treatment Schlenk RF, Do outcome in cytogenetically normal acute myeloid leukemia. N Engl J Med 2008;358:1909e1918.