Section I: Expanded Methods

RNA extraction, labeling, hybridization, and data analysis
Mononuclear cell suspensions from diagnostic BM aspirates or peripheral blood (PB) samples were prepared from each patient and an aliquot cryopreserved. RNA was extracted using the Trizol reagent (Gibco BRL Life Technologies, Gaithersburg, MD) following the manufacture´s recommended protocol. RNA integrity was assessed by electrophoresis on the Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA).

First and second strand cDNA were synthesized from 5-15 µg of total RNA using the SuperScript Double-Stranded cDNA Synthesis Kit (Gibco Life Technologies) and the oligo-dT₂₄-T7 (5´-GGC CAG TGA ATT GTA ATA CGA CTC ACT ATA GGG AGG CGG-3´) primer according to the manufacturer´s instructions. cRNA was synthesized and labeled with biotinylated UTP and CTP by in vitro transcription using the T7 promoter coupled double stranded cDNA as template and the T7 RNA Transcript Labeling Kit (ENZO Diagnostics Inc., Farmingdale NY). Briefly, double stranded cDNA synthesized from the previous steps was washed twice with 70% ethanol and resuspended in 22 µl RNase-free H₂O. The cDNA was incubated with 4 µl of 10X each reaction buffer, 1µl of biotin labeled ribonucleotides, 2 µl of DTT, 1µl of RNase inhibitor mix and 2 µl 20X T7 RNA Polymerase for 5 hr at 37°C. The labeled cRNA was separated from unincorporated ribonucleotides by passing through a CHROMA SPIN-100 column (Clontech, Palo Alto, CA) and precipitated at -20°C for 1 hr to overnight.

The cRNA pellet was resuspended in 10 ml Rnase-free H₂O. and 10.0 mg was fragmented by heat and ion-mediated hydrolysis at 95°C for 35 minutes in 200 mM Tris-acetate, pH 8.1, 500 mM KOAc, 150 mM MgOAc. The fragmented cRNA was hybridized for 16 hr at 45°C to HG_U95Av2 oligonucleotide arrays (Affymetrix, Santa Clara, CA) containing 12,600 probe sets from full length annotated genes together with additional probe sets designed to represent EST sequences. Arrays were washed at 25°C with 6 X SSPE (0.9M NaCl, 60 mM NaH₂PO₄, 6 mM EDTA + 0.01% Tween 20) followed by a stringent wash at 50°C with 100 mM MES, 0.1M NaCl₂, 0.01% Tween 20. The arrays were then stained with phycoerythrin conjugated streptavidin (Molecular Probes, Eugene, OR). Arrays were scanned using a laser confoCal scanner (Agilent, Palo Alto, CA) and the expression value for each gene was calculated using Affymetrix Microarray software (MAS 4.0). The signal intensity for each gene was calculated as the average intensity difference (AID), represented by [Σ(PM - MM)/(number of probe pairs)], where PM and MM denote perfect-match and mismatch probes, respectively. Expression values were normalized across the sample set by scaling the average of the fluorescent intensities of all genes on an array to a constant target intensity of 2500, then any AID over 45,000 was capped to a value of 45,000. All AID´s less than 100, including negative values and absent calls were converted to a value of 1. In addition, a variation filter was used that eliminated any probe set in which less than 1% of the samples had a present call, or if the Max AID - Min AID across the sample set was less than 100. The average intensity differences for each of the remaining genes were analyzed. For some metrics the data was log transformed prior to analysis. The minimum quality control values required for inclusion of a sample´s hybridization data in the study were 10% or greater present calls, a GAPDH/Actin 3´ /5´ ´ ratio <5, and use of a scaling factor that was within 3 standard deviations from the mean of the scaling values of all chips analyzed. The average percent present calls for our overall dataset was 29.7%, and for each of the genetic subgroups was BCR-ABL (31.1%), E2A-PBX1 (28.9%), Hyper >50 (31%), MLL (29.8%), T-ALL (29.1%), TEL-AML1 (28.5%), Novel (30.2%), others (31.1%). In addition, we required each sample to have >75% blasts. The average percentage blasts for the overall dataset used to define the genetic subtypes was 93%, and for each genetic subtype was BCR-ABL (92%), E2A-PBX1 (96%), Hyper >50 (93%), MLL (93%), T-ALL (91%), TEL-AML1 (92%), Novel (95%), and others (94%).

The data was analyzed using the MAS 4.0 software. Recently, a new version of the Affymetrix software became available (Affymetrix MAS 5.0). Affymetrix reports an overall 94% correlation between data analyzed using these two software programs. We have performed a limited analysis to directly compare the results of our data analyzed with these two different software programs. We took 74 randomly selected cases and reanalyzed the primary data using the MAS 5.0 software. Using the default parameters set by Affymetrix, the average percent present call was lower for the MAS 5.0 analyzed data as compare to the values obtained using MAS 4.0; however, with exceedingly rare exception, all of the 271 Chi-squared genes selected as discriminators of the seven genetic subtypes (see below) remained within the new dataset. Moreover, when we used these 271 in a 2-D hierarchical clustering analysis (as described below), the data looked identical to the results obtained using the MAS 4.0 dataset (data not shown).

Reproducibility of microarray data
The reproducibility of the Affymetrix microarray system was assessed by comparing the gene expression profiles of the following samples: (i) RNA extracted from duplicate cryopreserved diagnostic leukemic samples from 23 patients; (ii) Single RNA samples from 13 patients analyzed on two separate arrays. Results from these analyses are presented in Table 2. As shown, the mean number of probe sets that displayed a >2-fold difference in expression between separately extracted but paired RNA samples was 144, and for single RNA samples analyzed on two separate occasions was 133. Moreover, very few probe sets were found to have a >3-fold difference in expression levels between replicate samples. The observed number of probe sets showing a difference in expression values represents <2% of the total number of probe sets on the microarray, and thus these data suggest that the Affymetrix microarray system has a very high degree of reproducibility.

Table 2. Comparison data from replicate bone marrow sample

Comparison of microarray data between PB and BM leukemia samples
Matched BM and PB samples that contained >80% leukemic blasts were obtained from 10 patients and the RNA extracted and assessed by microarray analysis. As shown in Table 3, a very high level of correlation was observed for the expression profiles of BM and PB, with only 189 probe sets having a >2-fold difference in expression. More importantly, no genes were found to be consistently over- or under-expressed in one sample type. These data suggest that there are minimal differences in the gene expression profiles of leukemic blasts obtained from BM or PB, and raise the possibility that diagnostic gene expression profiling might be possible on samples obtained from the PB.

Table 3. Comparison of peripheral blood and bone marrow samples

Real-time RT-PCR results
Real-time RT-PCR assays (Taqman; Perkin-Elmer/Applied Biosystems, Foster City, CA) were developed to independently determine the level of mRNA for five genes that were found by microarray analysis to be predictive of either T-lineage ALL (CD3δ, CD3D antigen delta polypeptide TiT3 complex; MAL, mal T-Cell differentiation protein; and PRKCQ, protein kinase C theta) or E2A-PBX1 expressing ALL (MERTK, c-MER proto-oncogene tyrosine kinase and KIAA802). RNA samples analyzed included four samples each of E2A-PBX1 and T-ALL, and two samples each from the remaining subtypes (BCR-ABL, MLL, TEL-AML1, Hyperdiploid >50, Hyperdiploid 47-50, Hypodiploid, Pseudodiploid, and Normal). Whenever possible, the forward and reverse primers were designed in different exons so that DNA contamination would not be a concern. In the case of MAL where this was not clear, the RNA was treated for 15 minutes at room temperature with 1.0 unit of DNase I (Invitrogen, Carlsbad, California) using the Invitrogen protocol to remove any contaminating DNA.

Thirty-three ng of RNA from each sample was reverse transcribed using random hexamers and Multiscribe Reverse Transcriptase (Perkin-Elmer/Applied Biosystems) in a total volume of 10 µl. Real Time PCR was performed on a PE Applied Biosystems 7700 prism using oligonucleotide primers and probes sequences designed using Primer Express. The following primers and probes were used: (1) MERTK, 5´ -GGC GTG CTA ACT GTT CCA GG-3´ (forward primer), 5´ -CCT TTG TCA TTG TGG GCC TC-3´(reverse primer), and 5´-CAA CTG AAG ACC GCC ATC TCC GTC AG-3´(probe); (2) KIAA802, 5´- CCA AGC TGA AGG AGT CGG AC-3´(forward), 5´-AGA TTT CCT TTG TGA TTT TCT TCT TCC-3´(reverse), and 5´-TGC TCG GCC AGT GAG AAT CTC TAC CTG-3´(probe); (3) PRKCQ, 5´-AAG CTG CCA CAA GTT CGA CC-3´(forward), 5´-ATG AAG GAA CTG CAG ACC AAG AA-3´ (reverse), and 5´-CCA GAG CGA CGT TTT ATG CTG CTG AAA TC-3´(probe); (4) MAL, 5´-CCA GTG GCT TCT CGG TCT TC-3´ (forward), 5´-CAG GAG CCA CTC ACA AAC TCA A-3´(reverse), and 5´-CCA CCT TGC CCG ACT TGC TCT TCA-3´(probe); and (5) CD3δ , 5´-TCA ACA GAG CTT GTG TGT CGG-3´(forward), 5´-CAT TGC CAC TCT GCT CCT TG-3(reverse), and 5´-TGT CCA GCA AAG CAG AAG ACT CCC AAA-3´(probe). All probes were labeled at the 5´ end with FAM (6-carboxy-fluroescein) and at the 3´ end with TAMRA (6-carboxy-tetramethyl-rhodamine).

The PCR reactions were performed in a total volume of 50 µl containing 10 µl of the RT product, 300 nm each of the forward and reverse primers, 100 nM of probe, 1X master mix (PE Applied Biosystems) and 1 µl of Taq Gold. Following a 10 minute incubation at 95°C to activate the Taq Gold, samples were denatured at 95°C for 15 seconds, then annealed and extended at 60°C for 1 minute, for a total of 40 cycles. The RNA from each sample was also amplified using primers and probes to RNase P (PE Applied Biosystems) for use in normalization. Appropriate negative controls were included in each run. Standard curves were generated for T-cell markers and RNase P using MOLT4 RNA, a T-cell leukemia cell line, and for the E2A-PBX1 markers and RNase P using a leukemia cell line, 697, that contains an E2A-PBX1 fusion. The expression level of the predictive genes and RNase P were determined in each of the 24 ALL samples (Figures 7-11). A ratio was then calculated by taking the expression value for the specific gene and dividing it by the expression level of RNase P in the sample. These ratios were then compared to the values obtained from the Affymetrix data from the same RNA sample. The raw Affymetrix data were scaled as described and then normalized using the 3´GAPDH value for each sample, yielding a normalized ratio. The Taqman and Affymetrix ratios were then log transformed and compared. Since the markers selected for Taqman analysis were predictors for either E2A-PBX1 or T-ALLs, each gene was expected to have four RNA samples with high and 20 samples with low expression. For each gene evaluated, an average expression value for both the Taqman and Affymetrix data was calculated for all samples in the up-regulated group, and similarly, for the samples in the down-regulated group. These values were then compared against each other as shown in Figure 12.

Analysis of CD3δ

Figure 7. The standard curves of critical threshold values (Ct) versus log of the concentration of RNA in picograms (pg).

Table 4. Results of Taqman assay for CD3δ and RNase P on twenty-four selected ALL diagnostic bone marrow samples representing the various leukemia subgroups.

^@Critical threshold value for CD3δ                         
⁺ Critical threshold value for RNase P                  
⁺⁺ Log of the pg CD3δ   in patient sample calculated from standard curve
^% Exponentiated value of log CD3δ
^* Log of the pg RNase P in patient sample calculated from standard curve
^** Exponentiated value of log RNase P
^& Ratio of CD3δ  to RNase P

Analysis of MAL

Figure 8. The standard curves of critical threshold values (Ct) versus log of the concentration of RNA in picograms (pg).

Table 5. Results of Taqman assay for MAL and RNase P on twenty-four selected ALL diagnostic bone marrow samples representing the various leukemia subgroups.