The NUP98-HOXA10 and NUP98-HOXD13 fusion gene as well as the N51S-ND13 dead homeodomain mutant gene constructs have been described elsewhere [19]. In short, the cDNAs were subcloned into the murine stem-cell virus (MSCV) 2.1 vector upstream of the internal ribosomal entry site (IRES) sequence linked to the gene encoding the enhanced green fluorescence protein (EGFP; Clontech). As a control, the MSCV vector carrying only the IRES-GFP cassette (GFP virus) was used. Production of high-titer, helper-free retrovirus was carried out by standard procedures [28]. Constructs have been validated by sequencing and correct expression and transmission were confirmed by western blot and Southern blot analysis.

Mice were bred and maintained at the British Columbia Cancer Research Centre animal facility. Donors of primary BM cells were older than 12-weeks (C57Bl/6Ly-Pep3b×C3H/HeJ) F1 (PepC3) mice. Primary mouse BM cells were transduced as previously described [29]. Briefly, BM cells were harvested from mice treated 4 days previously with 150 mg/kg 5-fluorouracil (Faulding, Underdaler, Australia) and prestimulated for 48 hours in Dulbecco modified Eagle medium (DMEM) supplemented with 15% fetal bovine serum (FBS), 10 ng/mL human interleukin-6 (hIL-6), 6 ng/mL murine interleukin-3 (IL-3), and 100 ng/mL murine stem cell factor (mSF) (StemCell Technologies, Vancouver, BC, Canada). Cells were infected by co-cultivation with irradiated (4000 cGy x-ray) GP+E86 viral producer cells with the addition of 5 µg/mL protamine sulfate (Sigma, Oakville, ON, Canada). Loosely adherent and nonadherent cells were harvested from the co-cultures after 2 days and were cultured for 24 hours in the same medium without protamine sulfate.

The single cell suspensions collected were blocked for 10 min on ice with 5 µg/ml anti mouse CD16/CD32 (Fc Block, BD Pharmingen) in Phosphate Buffered Saline (STI)+2% Fetal Bovine Serum (PF). Cells were washed once with PF and then incubated on ice for 20 min with the primary mAb. Cells were then washed once, incubated with the secondary antibody if needed, washed again, and then analysed by flow cytometry using a FACSCalibur™ flow cytometer and CELLQuest™ software (BD Pharmingen). GFP+ Sca-1+Lin- cells were sorted using a FACSVantage™ (BD Pharmingen). Purity>90% were confirmed by re-analysis of sorted cells. The forward versus side scatter profile was used to gate on viable cells and an unstained sample was used to determine appropriate gating for expression. Monoclonal antibodies (mAbs) were all purchased from PharMingen (San Diego, CA) (phycoerythrin [PE]–labeled Gr-1, B220, Ter-119, CD4, CD5 and CD8).

Sorted cells were lyzed in Trizol™ (Invitrogen) and total RNA was extracted according to the manufacturer instructions. One hundred ng of total RNA from each sample were then double linear amplified with the ENZO BioArray High Yield RNA Transcript Labeling kit and the GeneChip Eukaryotic Small Sample Target Labeling Assay, Version II protocol (Affymetrix, Santa Clara, CA) to produce target for hybridization to Affymetrix MOE430 according to the manufacturer's instructions and performed at the Genome Science Centre, BC Cancer Agency, Vancouver, Canada. All experiments were performed in biological triplicate.

Gene array data (CEL-files) were imported into GeneSpring® software version 7.3 (Silicon Genetics, Redwood City, CA). The GC-RMA method [30] was used for normalization and data was processed as follows: all values below 0.01 were set to 0.01, all of the genes in each sample were divided by the median of the specified list of 100 positive control genes present on the MOE430 chip and all samples were then normalized against the median of the MIG control samples. Each measurement for each gene in those specific samples was divided by the median of that gene's measurements in the corresponding control samples. We judged genes to be differentially expressed when the difference in expression in the ND13 and NA10 condition vs. the GFP control or the non-leukemic ND13(N51S) condition was at least 50%; and the extent of difference in expression was significantly different in the Student's t-test (p<0.05). Classification of genes into functional categories and to analyze signaling pathways was done by collecting annotations and keywords with the Onto-Express software [31] (http://vortex.cs.wayne.edu/ontoexpress), Affymtrix NetAffx (http://www.affymetrix.com/analysis/index.affx) and the Gene Ontology Tool and KEGG maps included in the GeneSpring 7.3 Software. Unigene and RefSeq IDs were used in the analysis to exclude redundant genes included in the array probe sets. The array data is deposited at Gene Expression Omnibus (GEO), (http://www.ncbi.nlm.nih.gov/geo) and the MIAME (minimal information about a microarray experiment) guidelines was followed for data presentation.

Non-amplified RNA from the transduced murine primary bone marrow samples was used for validation with quantitative RT-PCR (qRT-PCR). RNA was isolated using Trizol™ and the samples were then treated with DNase I (amplification grade, Invitrogen). Complementary DNA (cDNA) was generated by reverse transcription (RT) with the iScript cDNA Synthesis Kit (BioRad Inc., Hercules, CA). Gene transcripts were quantified by real-time PCR using the iCycler apparatus (Bio-Rad Inc., Hercules, CA) and were detected with SYBR Green as flurochrome (IQ™ SYBR® Green Supermix, BioRad Inc.). Gene sequences for primer design were obtained from the NCBI Reference Sequences database (http://www.ncbi.nlm.nih.gov/RefSeq/). Primers were chosen using the Primer3 software (http://www.broad.mit.edu/cgi-bin/primer/primer3_www.cgi) and the specificity of all primer pairs was tested with electronic PCR using the mouse genome and the mouse transcript database (http://www.ncbi.nlm.nih.gov/sutils/e-pcr/reverse.cgi). The relative expression changes were determined with the 2^−ΔΔCT method [32] and the housekeeping glyceraldehyde-3-phosphate dehydrogenase (GAPDH) gene transcript was used to normalize the results. Primer sequences (5′ to 3′); Pbx1(NM_008783) forward primer ATCGGGGACATTTTACAGCA and reverse primer AGGCTTCATTCTGTGGCAGT; Pbx2 (NM_017463) TGAAGCAAAGCACCTGTGAG and AGTGGCCTGTTTGCTGAAGT; Pbx3 (NM_016768) AGAGCCAAATTGACCCAGAT and ATGGGACGCGTTCTACTCTG; HoxA5 (NM_010453) CGCAAGCTGCACATTAGTCA and AGGTAGCGGTTGAAGTGGAA; HoxA7 (NM_010455) GAAGCCAGTTTCCGCATCTA and CGTCAGGTAGCGGTTGAAAT; HoxA9 (NM_010456) ACAATGCCGAGAATGAGAGC and GTTCCAGCGTCTGGTGTTTT; HoxB4 (NM_010459) CTGGATGCGCAAAGTTCAC and TCCTTCTCCAACTCCAGGAC; Meis1 (NM_010789) GCACAGGTGACGATGATGAC and AGGGTGTGTTAGATGCTGGAA, Meis2 (NM_010825) AACGACGCCTTGAAAAGAGA and GCTCGCACTTCTCAAAAACC; Meis3 (NM_008627) CAGCGACAGCTTGAAGAGAG and GCCAGCTCACACTTCTCAAA; Flt3 (NM_010229) ATCCTTCCCCAACCTGACTT and TTGCCACCCATGTTCTGATA; Evi1 (NM_007963) GGAGGAGGACTTGCAACAAA and GACAGCATGTGCTTCTCCAA; Anxa1 (NM_010730) CACAGAGCCACCAGGATTTT and CGTTCGGAAATTGACATGAA; Tgfbi (NM_009369) GGAAGCTTCACCATCTTTGC and ATGTTGACGTTGCTCACCAG; c-Jun (NM_010591) TCCCCTATCGACATGGAGTC and TTTTGCGCTTTCAAGGTTTT; Csf2rb (NM_007781) CCTGGAACAAGGGAAGTTCA and CAATGCAGGCTGTAGCTGTC; Ptprf (NM_011213) TGGCCATCTCTTCATTAGGC and ACAGGCTCGGTATTTCCAGA; Gapdh (NM_008084) AACTTTGGCATTGTGGAAGG and ATGCAGGGATGATGTTCTGG.

The part of the study involving patient samples and healthy volunteers was performed in accordance to the Declaration of Helsinki and with approval by the local ethics committee at Göteborg University and informed written consent was obtained from all participants. All samples were collected at diagnosis between year 2000 and 2006 and stored at the department of Clinical Chemistry and Transfusion Medicine at Sahlgrenska University Hospital. The analysis included 34 adult patients, 20 females and 14 males, with de novo AML, representing FAB subclasses M0-M5. Mean age at diagnosis was 56 year (range 26 to 83). Four healthy volunteers were donors of normal bone marrow that was pooled for assay normalization.

RNA from patient and healthy volunteer samples was isolated using Trizol™ (Invitrogen, Cat.No. 15596-026). Complementary DNA (cDNA) was generated from 500 ng RNA by reverse transcription (RT) with random primers and the Superscript II enzyme and RNase inhibitor (Invitrogen) in a reaction volume of 20 µL. The RT reaction was incubated at 42°C for 50 minutes followed by 15 minutes at 70°C. Before the enzymes were added the mix was preheated at 65°C for 10 minutes. All assays were performed on an ABI Prism 7900 HT real-time PCR-system with ABI SDS Software 2.2.3 (Applied Biosystems, Foster City, CA, USA). For the TLDA 20 µL of cDNA, corresponding to 25 ng starting RNA, was mixed with 30 µL water and 50 µL TaqMan Universal PCR Master Mix (Applied Biosystems, Stockholm, Sweden); 100 µL was loaded per port. Thermal cycling conditions were 50°C for 2 minutes, 94.5°C for 10 minutes, 97°C for 30 seconds and 59.7°C for 1 minute. The relative expression changes were determined with the 2^−ΔΔCT method [32] and the housekeeping beta-glucuronidase (GUSB, Hs99999908_m1) gene transcript was used to normalize the results. The following genes were analyzed (TaqMan assay ID number is indicated, Applied Biosystems); HOXA5 (Hs00430330_m1), HOXA7 (Hs00600844_m1), HOXA9 (Hs00365956_m1), HOXA10 (Hs00538183_m1), PBX1 (Hs00231228_m1), PBX2 (Hs00855025_s1), PBX3 (Hs00608415_m1), MEIS1 (Hs00180020_m1), MEF2C (Hs00231149_m1), JAG2 (Hs00171432_m1), PRNP (Hs00175591_m1), DDX4 (Hs00251859_m1) and HLF (Hs00171406_m1).

Microsoft^® Excel in combination with the Excel plug-in software Analyse-It^® v1.73 was used for the statistical calculations. Pearson regression was used for comparison between gene array and qRT-PCR results. Spearman rank correlation was used to test possible associations. Non-parametric Kruskal-Wallis 1-way ANOVA were used to evaluate differences between groups.

Adult murine bone marrow cells transduced with vectors carrying ND13, NA10, ND13(N51S) or an empty GFP control vector were isolated on the basis of GFP expression by FACS 24 hours post-transduction. Viable transduced cells were further enriched for primitive hematopoietic cells by exclusion of cells expressing linage markers (Gr-1, B220, Ter-119, CD4, CD5 and CD8) and selection for cells expressing the stem cell antigen-1 (Sca-1). This resulted in an overall recovery of 0.25–6% of all cells with a purity of 90–95% for GFP expression. The cells were kept on ice during sorting and immediately lyzed in Trizol for RNA extraction. Three independent experiments were performed for each of the four different conditions included in the study.

After extraction, RNA was amplified and analyzed using the Affymetrix GeneChip MOE430A array containing 23,000 probe sets. The Gene Chip robust multi-array analysis (GC-RMA) was used for initial normalization and the GFP control was used for per gene normalization. Pearson correlation coefficient between the experimental replicates ranged between r = 0.92–0.99 suggesting low inter-experimental variation and the 3′-to-5′ ratios for Gapdh and Actin in all samples were less than 3.0 (ranging from 1.7–2.7), indicating that no serious bias was introduced by the RNA amplification procedure. Our main interest was to define the subset of genes that could explain the transforming and cell expansion potential of NUP98-HOX fusions in the Sca-1+, GFP+, Lin- cell population and that also could be direct NUP98-Hox binding target genes. First, all genes that were significantly differentially expressed between ND13, NA10 or the presumably non DNA-binding ND13(N51S) and the GFP control were identified. Genes were considered differentially expressed if they had a change in expression level compared to control of at least 50% and the extent of difference in expression was statistically significant (p<0.05) in a parametric Welsh-ANOVA t-test. More genes were activated than repressed by all NUP98-HOX fusions, 560 activated and 43 repressed genes with ND13, 414 activated and 20 repressed genes with NA10 and 204 activated and 82 repressed genes with ND13(N51S), (Figure 1). ND13 and NA10 had relatively more activation compared to the ND13(N51S) mutant and there was only a significant overlap on activated genes between ND13 and NA10 (Figure 1). Furthermore, of the 170 differentially expressed genes that were induced by both ND13 and NA10, 74 of these genes also differed significantly between ND13 and NA10 and the functionally inert ND13(N51S) mutant genes with at least a 50% difference in expression level (Table 1). These results suggest that ND13 and NA10 mainly act as transcriptional activators in undifferentiated BM cells and that this effect is dependent on an intact homeodomain. Moreover, the gene array results also indicate that a relatively small number of activated genes can be linked to NUP98-Hox induced effects on primitive hematopoietic cell function.

To confirm the fidelity of the microarray data a subset of 15 genes was selected for validation using quantitative RT-PCR (Table 2). Quantitative real-time RT-PCR was used to measure relative abundances. The housekeeping gene GAPDH was used as an endogenous control to normalize the data. The fold change was calculated between the GFP control and the NA10, ND13 and ND13(N51S) samples respectively. The majority of the individual fold changes determined from the gene array was verified by qRT-PCR (Table 2). Overall the analysis revealed a good correlation between the gene array data and the qRT-PCR results (Pearson correlation, r = 0.83), with a tendency that the PCR results showed greater changes than what the array suggested. Taken together the validation analysis provides confidence in our approach to identify differentially expressed genes with a high likelihood of exhibiting true expression level changes.

The 74 genes that were both induced by ND13 and NA10 and whose expression was dependent on an intact homedomain were classified into gene ontology categories according to involvement in different biological processes. The genes were separated into 12 main categories (Figure 2). A relatively high number of differentially expressed genes were classified as being involved in development and differentiation (16 genes), cell cycle, cell growth and/or cell proliferation (13 genes) or signal transduction (19 genes). At least 16 genes were classified as being involved in transcriptional regulation (Table 1). Thus, processes potentially important for cell self-renewal, expansion and transformation.

Several of the genes induced by ND13 and NA10 were Hox or Hox cofactors (Hoxa5, Hoxa7, Hoxa9 and Pbx3, Table 1). No significant change of gene expression could be seen for Pbx1, Pbx2 or Meis1, also verified with Q-RT-PCR (Table 2), but all three genes were expressed in the Sca1+, Lin- BM cell population in concordance with published results [33]. Meis2 and Meis3 could not be detected either by gene array or Q-RT-PCR (data not shown). These findings are basically in line what have been reported in other studies [34]–[36]. Besides the Hox or Hox co-factors several other putative target genes of ND13 and NA10 are involved in cell development and proliferation. This included induction of the DEAD-box protein gene Ddx4, the hepatic leukemia factor (Hlf), the MADS box transcription enhancer factor 2C (Mef2c), the prion protein gene (Prnp) and Jagged-2 (Jag2). The Drosophila ortholog of Ddx4, VASA, has a central role in germ cell development and is conserved in invertebrates and vertebrates. Ddx4 is a member of the DEAD box family of ATP-dependent RNA helicases, the same gene family as Ddx10 belongs to, which has been found in translocations with NUP98 [37]. The prion protein has been shown to be present on human CD34+ bone marrow (BM) stem cells [38] and was recently shown to be expressed on long-term hematopoietic stem cells and to be important in hematopoietic stem cell self-renewal [39]. The HLF gene has been found in translocations together with E2A in human acute lymphoblastic leukemia [40] and enforced HLF expression has been reported to enhance both HSC engraftment and to inhibit apoptosis [41]. The MEF2 family of regulatory proteins are involved in myogenesis and the Mef2c gene is important for normal morphogenesis [42]. Jagged-2 (Jag2) is a ligand that activates NOTCH1 and related receptors that are critical for various cell fate decisions. Furthermore, the tyrosine kinase receptor Flt3 was found to be induced by ND13 and NA10 in Sca1+, Lin- primary BM cells. We and others have previously shown that Flt3 expression is induced by Meis1 in a context with either high expression of NUP98-Hox fusions [43] or HOXA9 [44] and the present finding further strengthen the conclusion that the Flt3 is a direct target gene of Hox and Hox co-factors in primary HSCs or progenitor cells.

To discern what target genes identified in the microarray analysis might be involved in leukemic transformation and to investigate if they are associated with Hox and Hox co-factor expression, 34 de nova AML samples collected at diagnosis from adult patients were analyzed. Complete karyotype and FLT3-ITD status was known for all subjects and represented FAB subclass M0-M5 morphologically. Six patients had favorable, 19 had intermediate and 9 had unfavorable cytogenetics (Table 3). Quantitative real-time RT-PCR with TaqMan® Low density array (TLDA) was used to measure gene expression, allowing analysis of all the selected genes in several samples at the same time lowering assay variation and increasing reproducibility. We tested a subset of Hox and Hox cofactor genes, HOXA5, HOXA7, HOXA9, HOXA10, PBX1, PBX2, PBX3, and MEIS1, and possible Hox target genes FLT3, HLF, JAG2, MEF2C, DDX4, and PRNP. The gene expressions in leukemia samples were normalized with the housekeeping gene GUS as an endogenous control and the expression levels were calculated relative to pooled normal bone marrow (Table 3).

Two of the selected genes could not be detected in the majority of the samples in either normal or leukemia bone marrow (HOXA10 and DDX4) and were therefore excluded from further analysis. Analysis of the selected Hox A cluster and Hox cofactor genes revealed a high degree of co-expression between these in leukemia samples, where correlation coefficient between HOXA5, HOXA7, HOXA9 and MEIS1 ranged between 0.83 and 0.92 (Spearman rank correlation, p<0.0001 in all cases, e.g. in Figure 3A). PBX3 gene expression also correlated with these (r = 0.67–076, p<0.0001) but PBX2 and PBX1 did not. Furthermore, patients with intermediate or unfavorable cytogenetics had significantly higher expression of MEIS1, PBX3, HOXA5, HOXA7 and HOXA9 compared to patients with favorable cytogenetics (non-parametric Kruskal-Wallis ANOVA, e.g. HOXA7 p = 0.0013 and p = 0.018 respectively) but there was no significant difference between patients with intermediate and unfavorable cytogenetics. These findings suggest a tight co-regulation of these factors and define a subset of homeodomain transcription factors linked to leukemic development and are in concordance with other reports [45], [46].

Among the selected possible direct target genes, FLT3 gene expression showed the closest association with HOXA5, HOXA7, HOXA9 and MEIS1 (Spearman, r-value between 0.45–0.57, p–value 0.016–0.0020) but no significant correlation was seen with PBX3 expression (r = 0.22, p = 0.20). Furthermore, FLT3 expression showed a weak but significant correlation with bone marrow blast counts (r = 0.36, p = 0.049). Moreover, FLT3 expression was significantly higher in patients with intermediate or unfavorable cytogenetics (Kruskal-Wallis, p = 0.047 and p = 0.046 respectively) but there was no significant association between FLT3 expression and FLT3-ITD status (p = 0.74). Thus, FLT3 expression was associated with HOXA5, HOXA7, HOXA9 and MEIS1 in human leukemia and this gene expression profile was confined to leukemia with either intermediate or unfavorable cytogenetics.

Expression of the HLF gene correlated significantly with HOXA7 and HOXA9 expression (Spearman, r = 0.46, p = 0.0060 and r = 0.43, p = 0.011 respectively), but not with HOXA5 (r = 0.24, p = 0.17), MEIS1 (r = 0.25, p = 0.15) or PBX3 (r = 0.25, p = 0.16). The same association was seen for JAG2, with a positive correlation with HOXA7 and HOXA9 gene expression (Spearman, r = 0.46, p = 0.0062 and p = 0.0068, respectively) but not with HOXA5, MEIS1 or PBX3 (r = 0.31, p = 0.074, r = 0.32, p = 0.061 and r = −0.02, p = 0.98 respectively). Furthermore, JAG2 and HLF gene expression correlated (r = 0.47, p = 0.0052, Figure 3B) and there was also a significant correlation between JAG2 and FLT3 expression (r = 0.39, p = 0.024). Altogether, these findings suggest that HOXA7 and HOXA9 could be directly involved in JAG2 and HLF gene regulation and possibly, unlike FLT3, independent of MEIS1.

The expression of the MEF2C and PRNP genes showed a tight co-association (r = 0.70, p<0.0001) and both showed significantly higher expression in patients with unfavorable vs. favorable cytogenetics (Kruskal-Wallis ANOVA, p = 0.017 and p = 0.026 respectively). However, their expression did not correlate with HOXA5, HOXA7, HOXA9 or MEIS1 and only weakly with PBX3 expression in human leukemia (Spearman, r = 0.37, p = 0.033 and r = 0.44, p = 0.078 respectively). This suggests that their induction may be restricted to NUP98-Hox fusions and that they are not regulated by the native HOXA5, HOXA7, HOXA9 or MEIS1 in human bone marrow cells.