Clinical Cancer Research Bridging the Lab and the Clinic in Cancer Medicine Infection and Cancer: Biology, Therapeutics, and Prevention
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
Cancer Research Clinical Cancer Research
Cancer Epidemiology Biomarkers & Prevention Molecular Cancer Therapeutics
Molecular Cancer Research Cancer Prevention Research
Cancer Prevention Journals Portal Cancer Reviews Online
Annual Meeting Education Book Meeting Abstracts Online
This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Moos, P. J.
Right arrow Articles by Carroll, W. L.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Moos, P. J.
Right arrow Articles by Carroll, W. L.
Clinical Cancer Research Vol. 8, 3118-3130, October 2002
© 2002 American Association for Cancer Research

Molecular Oncology, Markers, Clinical Correlates

Identification of Gene Expression Profiles That Segregate Patients with Childhood Leukemia1

Philip J. Moos2, Elizabeth A. Raetz2, Marlee A. Carlson, Aniko Szabo, Fiona E. Smith, Cheryl Willman, Qi Wei, Stephen P. Hunger and William L. Carroll3

Center for Children at the Huntsman Cancer Institute [P. J. M., E. A. R., M. A. C., F. E. S., W. L. C.], Division of Pediatric Hematology/Oncology, Departments of Pediatrics [E. A. R., W. L. C.] and Oncological Sciences [P. J. M., A. S., W. L. C.], University of Utah School of Medicine, Salt Lake City, Utah 84112; University of New Mexico, Albuquerque, New Mexico 87131 [C. W.]; and University of Colorado Health Sciences Center [Q. W., S. P. H.] and the Children’s Hospital [S. P. H.], Denver, Colorado 80218


   ABSTRACT
Top
 ABSTRACT
INTRODUCTION
 PATIENTS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
To identify genes whose expression correlated with biological features of childhood leukemia, we prospectively analyzed the expression profiles of 4608 genes using cDNA microarrays in 51 freshly processed bone marrow samples from children with acute leukemia, over a 24-month period, at a single institution. Two supervised methods of analysis were used to identify the 20 best discriminating genes between the following cohorts: acute myelogenous leukemia (AML) versus acute lymphoblastic leukemia (ALL); B-lineage versus T-lineage ALL; newly diagnosed B-lineage standard-risk versus high-risk ALL; and B-lineage leukemia harboring the TEL-AML1 fusion versus patients without a molecularly characterized translocation. These methods identified overlapping sets of genes that segregated patients within described subgroups. Cross-validation demonstrated that the majority of patients could be correctly classified based on these genes alone, and hierarchical clustering grouped patients with similar clinical and biological disease features. The potential for select genes to discriminate patients was validated using real-time PCR in samples that were analyzed by microarray profiling and in other uniformly processed leukemic marrow samples. As expected, microarray technology can successfully segregate patients defined by traditional measures such as immunophenotype and cytogenetic alterations. However, among specific subgroups, this preliminary analysis also suggests that microarrays can identify unanticipated similarities and diversity in individual patients and thus may be useful in augmenting risk-group stratification in the future.


   INTRODUCTION
Top
 ABSTRACT
 INTRODUCTION
PATIENTS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Leukemia is the most common pediatric malignancy in the United States, and ALL4 accounts for the vast majority (75%) of new cases diagnosed annually (1) . The outcome for children with ALL has improved dramatically over the past three decades (2 , 3) , but despite significant progress in treatment, 25% of children with ALL and 50–60% of children with AML develop recurrent disease and their prognosis remains poor (4 , 5) .

Although cure rates have improved considerably, curative therapy is often associated with serious side effects. Risk-based therapy attempts to tailor treatment to maximize the chances for cure and minimize side effects (6 , 7) . Using this approach, patients are segregated into different risk groups based upon a combination of clinical and biological features of their disease, and subsequent therapy is tailored according to their predicted risk for treatment failure. Age, presenting WBC count, and abnormalities in either chromosome number or structure are the most important features presently used to stratify children with ALL into good-, standard-, high-, or very high-risk groups (8, 9, 10, 11, 12) . Similar approaches have been used in the treatment of childhood AML.

Although risk-group stratification has been very useful in classifying patients and improving their outcomes, variability in clinical course exists among individuals within a single risk group and among those with the same prognostic features. For example, the majority of children who suffer a relapse while being treated on contemporary therapy were originally defined as standard risk. Also, some patients whose blasts express the TEL-AML1 fusion transcript, which has been associated with a very favorable outcome, fail treatment for reasons that are poorly understood (13, 14, 15, 16, 17, 18, 19) . In fact, Donadieu et al. (20) have found that the most significant prognostic factors in childhood ALL explain no more than 4% of the variability in prognosis, suggesting that undiscovered molecular mechanisms dictate clinical behavior. Discovery of additional molecular prognosticators may therefore enhance current risk-based classification schemes.

Emerging technologies, such as DNA microarrays, facilitate gene expression analysis on an unprecedented scale, allowing the surveillance of thousands of transcripts simultaneously (21, 22, 23) . Important clinical applications of this technology include refinement of diagnoses, prediction of prognosis, study of disease progression, and the identification of relevant pathways for the design of novel therapeutics (24) . We have undertaken a project investigating patterns of gene expression using microarray analysis in pediatric leukemia in an effort to better understand clinical behavior within established subgroups of patients. This was a single-institution, prospective pilot study with a primary goal being the identification and validation of genes that could reliably differentiate known subgroups of patients, as well as highlight subsets of patients with distinctive clinical or biological features of their disease. As the effect of cryopreservation on gene expression analyses has not been defined as of yet, we chose to analyze only samples that were processed within 1 h of the time that they were obtained.

cDNA microarrays were used to analyze the expression patterns of 4608 genes in leukemic blasts from children, and gene expression profiles were identified that collectively characterized patients within defined disease subgroups. Using two complementary methods of analysis, we identified the 20 genes that best discriminated the following cohorts: AML versus ALL; B versus T-lineage ALL; standard versus high-risk ALL; and B-lineage leukemia harboring the TEL-AML1 fusion versus that without molecularly defined translocations. We propose that the set of genes in common between the two methods could be used to segregate pediatric leukemia patients. Although this analysis has identified genes that accurately differentiate patient cohorts, our results also show differences in the gene expression profiles of patients with similar features of their disease. We hypothesize that this level of molecular diversity may be responsible for differences in biological behavior among established subgroups of patients. Prospective microarray analyses may augment current risk-based classification schemes in pediatric leukemia.


   PATIENTS AND METHODS
Top
 ABSTRACT
 INTRODUCTION
 PATIENTS AND METHODS
RESULTS
 DISCUSSION
 REFERENCES
 
Patient mRNA Preparation.
cDNA microarray technology was used to prospectively determine patterns of gene expression in leukemia samples from 51 children with leukemia collected at the time of initial diagnosis or subsequent relapse through an Institutional Review Board approved protocol from 11/1998 to 11/2000 (Table 1)Citation . This cohort represented ~50% of all patients with newly diagnosed leukemia seen at Primary Children’s Medical Center during this time period. Heparinized bone marrow aspirates or, in some cases, peripheral blood samples were obtained from patients undergoing an initial evaluation for leukemia or suspected relapse. Patients were enrolled into the study if additional bone marrow could be aspirated at the time of diagnosis or if they had >50% circulating peripheral blasts. Analyses were performed on 44 marrow samples and 8 peripheral blood samples. All samples were processed into RNA within 1 h of acquisition to avoid artifacts associated with processing and/or storage. Enrichment for leukemic blasts was accomplished using Ficoll-Paque centrifugation. Mononuclear cells were solubilized with Trizol for total RNA purification, and mRNA was purified from DNase-treated total RNA samples using Oligotex oligo-dT beads (Qiagen, Chatsworth, CA). The arbitrary reference control sample used in all comparisons was the HL-60 cell line.


View this table:
[in this window]
[in a new window]
  		Table 1 Patient characteristics

 
Molecular Characterization of Leukemia Specimens.
Single-round or nested RT-PCR assays were performed to detect fusion transcripts produced by specific nonrandom chromosomal rearrangements in 24 of 25 patients with newly diagnosed B-lineage ALL. RT-PCR assays were performed using methods described previously (25, 26, 27, 28, 29) . Additionally, the BCR-ABL minor (m) and major (M) transcripts were detected using the following primers: M-BCR-ABL first-round sense, 5'-GGC AAG AGT TAC ACG TTC CTG; M-BCR-ABL antisense, 5'-GGT TGG GGT TCA TTT TCA CTG; M-BCR-ABL second-round sense, 5'-GGT TGG GGT TCA TTT TCA CTG; M-BCR-ABL antisense, 5'-AGA TGC TGA CCA ACT CGT GTG; m-BCR-ABL first-round sense, 5'-GAA CTC GCA ACA GTC CTT CG; m-BCR-ABL antisense, 5'-GGT TGG GGT TCA TTT TCA CTG; m-BCR-ABL second-round sense, 5'-CAC GCC GCAGTG CCA; and m-BCR-ABL antisense, 5'-GGT TGG GGT TCA TTT TCA CTG. ALL samples were tested for the presence of BCR-ABL, E2A-PBX1, MLL-AF4, and TEL-AML1 fusion transcripts, whereas AML specimens were tested for AML1-ETO, CBFß-SMMHC, and PML-RARA fusion transcripts. The single case of acute undifferentiated leukemia was tested with both the ALL and AML panels. Additional molecular studies were performed in 3 infants. MLL gene rearrangements were detected by Southern blot analysis as part of the Children’s Cancer Group protocol, CCG-1953, for infant leukemia. RT-PCR for the MLL/AF4, MLL/AF9, and MLL/ENL gene products was also performed in these patient samples.

Gene Expression Profiling.
cDNA hybridization probes were prepared by reverse transcription of poly(A) RNA with end-labeling with dCTP conjugated to fluorescent dyes. The hybridization probes consisted of reference cell (HL-60) cDNA labeled with Cy5-dCTP and patient cell cDNA labeled with Cy3-dCTP. The Cy3- and Cy5-cDNA hybridization probes were diluted in buffer containing 10% (w/v) dextran sulfate, 50% (v/v) formamide, 0.3 M NaCl, 10 mM Tris (pH 8.0), 1 mM EDTA, 1x Denhardt’s, 10 mM DTT, and 0.5 mg/ml nonhomologous DNA. Probes generated from patient marrow samples and from the reference cell line were hybridized simultaneously to a glass slide containing 4608 separate cDNAs deposited in duplicate using a Molecular Dynamics Gen III robotic spotter. Eight hundred of the cDNA clones were selected by investigators and included primarily genes involved in transcription, apoptosis, and development. The remaining cDNAs were randomly selected from IMAGE consortium clones. All differentially expressed cDNAs identified in our comparative analyses were independently sequence verified. After hybridization at 42°C for 36 h in a humidified chamber, nonbound labeled cDNA hybridization probes were removed by washing with 2x SSC, 0.1% SDS at room temperature, followed by washing with 0.5x SSC, 0.1% SDS at 45°C. Fluorescence from dried microarray slides was detected with a Molecular Dynamics Array Scanner at two respective emission wavelengths.

Microarray Analysis.
Gene expression for each channel was quantified by ArrayVision software. Therefore, a single channel (patient) analysis was performed as follows. Within each array, the minimum value was subtracted from each element to adjust for hybridization, Cy-dye incorporation, and background, and all arrays were grid normalized to account for spotting differences for each pin (30) . The average of the duplicate spotting for each array element was then log2 transformed before analysis to stabilize variance and improve normality. Cohorts were compared using a t test (two tailed and homoscedastic) and the entropy-dependent statistic Infoscore (31) . The 20 genes that best discriminated the cohorts by each method were selected for further analysis. We evaluated the gene selection procedure through a cross-validation strategy. LOOCV was used to assess the quality of the genes for discriminating leukemia cohorts by measuring the Euclidean distance metric with a k-nearest neighbor (k-NN) approach (k = 9; Ref. 32 ). In other words, the Euclidean distance between each individual patient sample was calculated to all of the other samples using these genes. In most comparisons, the number of samples between the two cohorts was not equal, and there could be a disproportionate number of votes for the larger cohort. Still, we used the k-NN method because it was the least parametric and required the fewest number of assumptions with the following modification. With unequal sample sizes, m < n, to be assigned to the larger cohort (n), the proportion of k neighbors should be at least: Po = n/(m + n), i.e., more than expected by random chance. In other words, the proportion of votes required to be assigned to the larger cohort has to be greater than the fraction of samples of the larger cohort relative to all samples being compared. We estimated the random prediction rate by randomly permuting the samples labels (five times), used the same algorithm, and averaged the results to determine a cross-validation prediction rate for cohorts without differences. Hierarchical clusters were generated using Spotfire to replace the relative expression values with trimmed mean, Z-scores to accentuate the patterns of expression, and the Stanford University Microarray package that groups samples and target genes based upon similarity of gene expression (33) .

Verification of Gene Expression Using Real-Time PCR.
The results from array analysis were verified using quantitative PCR for a limited number of target genes. Real-time PCR was performed using the following sense and antisense primer pairs: 5'-ATT TCG ACC GCG ATG ATG TGG and 5'-GAA CCC AGG GCA TGA AGA TCC for ferritin light chain (Hs.111334, M11147); 5'-CTT CCA GGT GAA TCC AAC AAG and 5'-AAT CCA TGT TGA CCC AAG TCC for proteoglycan (Hs.1908, W25334); 5'-CAA CAT GCA GAT CTG GAC CAC and 5'-CTT TAG TAC CAG GGT CAT GGC for endoglin (Hs.76753, W24164); 5'-AAT CTT CCT TCC TGA GGG GTC and 5'-GGG ACA AAT GAG TGC CCA TTG for MHC (MHC) class I (Hs.1845, AA149083); 5'-AGG CCC GCA ACA AGT TCG AG and 5'-TGG CAC TGA ACT TGT CGT GG for protein kinase C substrate 80-K-H (Hs.1432, NM_002743); and 5'-CCT CAG AGA ATC ACA CCT TCC C and 5'-ACC CTC TCC TGA CCC TGA AG for glyoxylate reductase/hydroxypyruvate reductase (Hs.155742, W39164).

First-strand cDNA was generated in a 40-µl reverse transcription reaction and was purified using a QIAquick PCR purification kit (Qiagen, Chatsworth, CA). Real-time PCR was performed in a fluorescence temperature cycler (LightCycler LC24; Idaho Technology, Idaho Falls, ID; Refs. 34 , 35 ). Individual 10-µl reactions contained 3 mM MgCl2, 0.25 mg/ml BSA, 50 mM Tris (pH 8.3), 200 µM deoxynucleotide triphosphates, 5 µM each primer, 1:30,000 dilution in TE (10 mM Tris 1 mM EDTA, pH 8) of SYBR Green I (Molecular Probes), and 1 µl of a 1:1:8 mix of Taq DNA Polymerase, TaqStart Antibody (Clontech Laboratories, Palo Alto, CA), and Taq Enzyme Diluent (Idaho Technology) and template. Each reaction contained 50 ng of cDNA purified from individual patient RT reactions as template. After an initial 95°C denaturation, the amplification program consisted of 40 four-step cycles of heating at 20°C/s to 95°C with a 0-s hold, cooling at 20°C/s to 55°C with a 0-s hold, heating at 10°C/s to 72°C with a 15-s hold, and heating at 20°C/s to 85°C with a 0-s hold. Fluorescent product detection was undertaken at the last step of each cycle. After amplification, a melting profile was acquired by first cooling at 20°C/s to 72°C and slowly heating at 0.2°C/s to 95°C with fluorescence data collection at 0.2°C intervals.

Standards for determining PCR product copy number were prepared from diluted PCR products. PCR products were purified using a QIAquick PCR purification kit (Qiagen). The copy number (copies/µl) for each standard was calculated based on the size of the PCR product and absorbance at 260 nm. Transcript copy numbers for patient samples were calculated by comparing the fluorescence of PCR products of unknown concentration with the fluorescence of serial dilutions of the external standard. Only fluorescence values measured in the log-linear phase of amplification were used for sample quantification. A "best-fit" log-linear fluorescence curve was calculated for each sample, and these curves in turn were used to estimate the number of transcripts in individual patient samples. All samples were run in triplicate and analyzed as follows. The real-time PCR data were log-transformed to stabilize variance and improve normality. The statistical assessment to evaluate the real-time PCR was a repeated measures analysis that was performed separately for the samples on and off the slides (SAS). The effect size estimates and confidence intervals were transferred back to the original scale through exponentiation; therefore, they represent multiplicative effects.


   RESULTS
Top
 ABSTRACT
 INTRODUCTION
 PATIENTS AND METHODS
 RESULTS
DISCUSSION
 REFERENCES
 
Bone Marrow versus Peripheral Blood.
Eight peripheral blood samples and 43 patient marrow samples were used for this analysis. To determine the similarity between blood and marrow samples, gene expression analyses were compared in paired samples in several patients (nos. 43, 44, and 49). In all cases, peripheral blood samples clustered next to the appropriate marrow sample when arranging the samples by gene expression correlation coefficients, illustrating the homogeneity of blasts in the peripheral blood and the marrow. Fig. 1Citation shows the Euclidean distance matrix for these samples.



View larger version (49K):
[in this window]
[in a new window]
  		Fig. 1. The Euclidean distance measurements were performed on 3 patients’ (nos. 43, 44, and 49) bone marrow (BM) samples at diagnosis and subsequent peripheral blood (0, 1, or 6 h) samples taken during initial chemotherapy. These samples demonstrate common expression profiles from blasts collected from the bone marrow and peripheral blood. The entire 4608 element array was used to generate the upper distance matrix, and the lower distance matrix represents the genes that were in common between the two methods used for identifying genes in the subsequent analyses between patient cohorts. The value in each box is the distance measurement, and the boxes are shaded from black to white as the distance increases.

 
Microarray Analysis of Established Patient Subgroups.
The ability of two approaches, t tests and Infoscore, were tested to identify individual genes that discriminate patients with AML and ALL. This initial analysis included 32 patients with B-lineage ALL, 7 patients with T-lineage ALL, and 10 patients with AML. A LOOCV approach was used to determine the error rate of selecting the top 20 discriminating genes between ALL and AML cohorts. These methods could predict the clinical classification at overall rates of 87.7 and 75%, respectively (Table 2)Citation . ALL patients were predicted correctly nearly 95% of the time when using t tests but only 77% for Infoscore, but AML patients were misclassified more frequently with t tests than Infoscore. Still, the same 19 genes were found in >50% of the comparisons. Perhaps the reduced prediction rate for AML is partially attributable to the heterogeneous FAB classifications of AML patients (Table 1)Citation .


View this table:
[in this window]
[in a new window]
  		Table 2 AML versus ALL LOOCV: LOOCV of gene selection method (Po = 0.796)

 
To determine the usefulness of the top discriminating genes when using all AML and ALL patients in this analysis for predicting cohort status, the 20 best discriminating genes for each methodology were identified with high significance ranging from P = 0.000003 to 0.0005 (Fig. 2)Citation . These 20 discriminating genes were used as the inputs for hierarchical clustering to assess the reliability of this method as a classification tool for known subtypes of leukemia (Fig. 2)Citation . After the selection of the top 20 discriminators, 1 patient sample each with acute undifferentiated leukemia and transient myeloproliferative disorder were added to the AML versus ALL comparison. In determining gene profiles that distinguished lymphoid from myeloid disease, the patients clustered on more explicit branches when using the best 20 discriminators by t test. All AML patients segregated on one major branch. When clustering using the best 20 genes by Infoscore, most of the AML patients segregated in one branch. Eleven genes were commonly identified by both methods (Fig. 2)Citation .



View larger version (68K):
[in this window]
[in a new window]
  		Fig. 2. A, hierarchical cluster analysis of the 20 best array elements discriminating AML and ALL as determined by t test. B, hierarchical cluster analysis of the 20 best array elements discriminating AML and ALL as determined by Infoscore. Genes found in common among both methods are represented in bold. C, real-time PCR validation of select genes in common between A and B. The red bar represents the geometric mean of the number of transcripts. AUL, acute undifferentiated leukemia; PRG, proteoglycan (W25334); Ferrtin L chain, M11147.

 
The microarray data were validated independently by real-time PCR using the actual patient samples from the arrays, as well as other identically processed, leukemic marrow samples not included in this initial array analysis. The target genes selected for validation were those that were identified as discriminators by t test and Infoscore, independently, e.g., hematopoietic proteoglycan core protein (Hs. 1908) and the ferritin light chain (Hs.111335; Fig. 2CCitation ). In the case of hematopoietic proteoglycan core protein, expression profiles were validated in 15 AML patient samples, with 9 from the arrays and 6 marrow samples not on the arrays. Expression profiles were similarly validated in 29 ALL patient samples, with 17 from the arrays and 12 marrow samples not on the arrays (Fig. 2C)Citation . Real-time PCR indicated that patient samples that were analyzed previously with the microarray affirmed distinct expression, and an independent test confirmed that this gene was differentially expressed (P = 0.006) in additional samples. In the case of ferritin light chain, expression profiles were validated in 13 AML patient samples, with 9 from the arrays and 4 banked samples not on the arrays. Expression profiles were similarly validated in 28 ALL patient samples, with 16 from the arrays and 12 banked samples (Fig. 2)Citation . Once again, real-time PCR demonstrated that patient samples that were analyzed previously with the microarray were differentially expressed, and an independent test confirmed that this gene was expressed more highly in AML than ALL marrow samples (P = 0.033).

Within the ALL cohort, the genes that best discriminated T-lineage ALL versus B-lineage ALL were identified in a similar manner. This initial analysis included the 32 patients with B-lineage ALL and 7 patients with T-lineage ALL. Both methods were tested using LOOCV and predicted the correct classification at a rate of 89.7 and 79.5%, respectively (Table 3)Citation . The t test predicted T-lineage correctly 100% of the time, whereas Infoscore correctly predicted T-lineage at a rate of 71.4%. It is noteworthy that this method erred in predicting 2 T-cell ALL patients who had very immature immunophenotypes. The 20 best discriminating genes for each methodology were identified with high significance ranging from P = 0.000000002 to 0.0006. Hierarchical clustering of the patients demonstrated clear distinctions of T-lineage patients from the B-lineage patients using either methodology (Fig. 3)Citation . The majority (16 unique genes) of the top 20 discriminating genes were identified by both t test and Infoscore. The genes identified included those predicted to show differential expression, such as the T-cell specific surface marker, CD3. Additionally, as expected, MHC molecules displayed increased expression in B-lineage blasts. We again selected a subset of genes that were independently identified by both methods for validation. In the case of MHC class 1, expression profiles were validated in 7 T-cell patient samples (4 from the arrays; 3 banked) and in 23 B-lineage samples (13 from the arrays; 10 banked). Real-time PCR affirmed that patient samples among the two groups that were analyzed with the microarrays were different, and additional testing in off array samples confirmed this differential gene expression (P = 0.023) in an independent sample set. In the case of endoglin, expression profiles were validated in 6 T-cell patient samples (4 from the arrays; 2 banked) and in 18 B-lineage samples (12 from the arrays; 6 banked). Again, real-time PCR demonstrated that patient samples that were analyzed previously with the microarray were distinct; however, an independent test did not confirm that this gene was differentially expressed (P = 0.28). However, the test set only contained two T-cell samples resulting in a very wide confidence interval.


View this table:
[in this window]
[in a new window]
  		Table 3 T versus B-lineage ALL LOOCV: LOOCV of gene selection method (Po = 0.821)

 


View larger version (58K):
[in this window]
[in a new window]
  		Fig. 3. A, hierarchical cluster analysis of the 20 best array elements discriminating T-lineage and B-lineage ALL as determined by t test. B, hierarchical cluster analysis of the 20 best array elements discriminating T-lineage and B-lineage ALL as determined by Infoscore. C, real-time PCR validation of select genes in common between A and B. The red bar represents the geometric mean of the number of transcripts. MHC, AA149083; Endoglin, W24164.

 
Patterns of Gene Expression in Leukemic Patients in Different Risk Strata.
The transcript profiles of 13 standard-risk (by National Cancer Institute criteria) and 10 high-risk newly diagnosed B-lineage leukemic patients were also analyzed. Here, t test and Infoscore were again used to independently identify groups of genes that discriminated patients based on their predicted risk of relapse. The error rates, based on cross-validation, were higher for these cohorts, resulting in lower overall prediction rates of 61 and 65%, respectively (Table 4)Citation . Here the high-risk group was poorly predicted by both methods, at only 50%. Interestingly, the same high-risk patient samples were misclassified as standard risk by both methods. The same was found for those standard risk that were misclassified as high risk. This may indicate that the clinical criteria for segregating these cohorts are less well defined, as determined by this molecular analysis. A larger cohort comparison is necessary to determine whether this is the primary reason for the lower prediction rates. The 20 best discriminating genes for each methodology were identified with high significance ranging from P = 0.00002 to 0.0003. Cross-validation indicated good discrimination of patients based on clinical risk of relapse. The patients that were misclassified by the clinical criteria and the prediction strength are listed in Table 4Citation . However, only 5 genes were commonly identified using a combination of t test and Infoscore (Fig. 4)Citation . The majority of the gene sequences identified were not annotated. Several patient groups emerged after hierarchical clustering, some showing an admixing of standard-risk and high-risk patients (Fig. 4)Citation . We again selected genes for validation in patient samples that were independently identified by both methods (Fig. 4C)Citation . The expression of protein kinase C substrate (Hs.1432) was validated in 22 standard-risk patient samples (9 from the arrays; 13 banked) and in 21 high-risk samples (10 from the arrays; 11 banked). Real-time PCR affirmed that patient samples that were analyzed previously with the microarray were distinct; however, an independent test did not confirm that this gene was differentially expressed (P = 0.86).


View this table:
[in this window]
[in a new window]
  		Table 4 Standard risk versus high risk LOOCV: LOOCV of gene selection method (Po = 0.565)

 


View larger version (43K):
[in this window]
[in a new window]
  		Fig. 4. A, hierarchical cluster analysis of the 20 best array elements discriminating Standard Risk and High Risk as determined by t test. B, hierarchical cluster analysis of the 20 best array elements discriminating Standard Risk and High Risk as determined by Infoscore. C, real-time PCR analysis of a select gene in common between A and B. The red bar represents the geometric mean of the number of transcripts. PKCsubstrate, protein kinase C substrate (NM_002743).

 
Gene Expression Analyses of Patients with Defined Cytogenetic Abnormalities.
The t(12;21) translocation is a defined cytogenetic abnormality that generally has been associated with a favorable outcome. We compared 8 patient samples harboring the TEL-AML1 fusion transcript by PCR to 11 samples that were PCR negative using both methods. Cross-validation analysis indicated that both methods could predict the correct cohort at rates of 73.7 and 90%, respectively (Table 5)Citation . The 20 best discriminating genes for each methodology were identified with high significance ranging from P = 0.00001 to 0.0002. Eleven of the genes were common to both methodologies, and most of these are not annotated (Fig. 5)Citation .


View this table:
[in this window]
[in a new window]
  		Table 5 TEL-AML1 versus none LOOCV: LOOCV of gene selection method (Po = 0.579)

 


View larger version (39K):
[in this window]
[in a new window]
  		Fig. 5. A, hierarchical cluster analysis of the 20 best array elements discriminating patients with TEL-AML1 translocation from patients without a molecularly classified translocation as determined by t test. B, hierarchical cluster analysis of the 20 best array elements discriminating patients with TEL-AML1 translocation from patients without a molecularly classified translocation as determined by Infoscore. C, real-time PCR analysis of select genes in common between A and B. The red bar represents the geometric mean of the number of transcripts. Glyoxylate reductase, glyoxylate reductase/hydroxypyruvate reductase (W39164).

 
Hierarchical clustering indicated clear classifications of the patients with and without the translocation. The only exception was #102, a sample that also failed cross-validation by both methodologies (Fig. 5)Citation . Otherwise, all patients with the TEL-AML1 translocation clustered together in one branch. The expression of glyoxylate reductase/hydroxypyruvate reductase (Hs. 155742) was validated in 4 Tel-AML1-positive patient samples and in 7 patient samples where no translocation was identified by molecular means (Fig. 5C)Citation . Here, real-time PCR did not affirm that patient samples that were analyzed previously with the microarray were differentially expressed. This could indicate that there could be some other property of these samples such that they segregated when using multiple genes in the clustering analysis, or that the primer pairs used in real-time PCR did not have sufficient fidelity to faithfully produce the product in question, or that the sample size was too small to differentiate these cohorts. It is noteworthy that a study published recently, using a large number of patient samples, was capable of distinguishing all of the common chromosomal translocations observed in pediatric ALL (36) .


   DISCUSSION
Top
 ABSTRACT
 INTRODUCTION
 PATIENTS AND METHODS
 RESULTS
 DISCUSSION
REFERENCES
 
Traditionally, childhood leukemia has been classified according to a host of clinical parameters and biological features that correlate with prognosis (37 , 38) . This classification system has been used for risk-based treatment assignment. Universally accepted criteria for leukemia classification are age and initial WBC count at the time of diagnosis (8) . Other factors that are also used to determine risk of relapse are the rapidity of response to therapy (39 , 40) and cytogenetic abnormalities that include both alterations in chromosome number (ploidy; Refs. 41, 42, 43 ) and structure (translocations; Ref. 10 ). Although the prognostic importance of these features has been well established, unexplained variability in clinical course still exists among some individuals within defined risk-group strata. Differences in the molecular constitution of malignant cells within subgroups may help to explain this variability.

Because the biological behavior of a cancer cell is governed, in part, by its expressed genetic repertoire, the development of techniques to determine relative levels of RNA expression offer powerful insight into the molecular mechanisms of disease. Microarray methodology is a particularly useful tool for these analyses and been used successfully to study many hematopoietic and solid tumors. Golub et al. (44) have shown that array technology can reliably classify leukemias and can specifically distinguish acute myeloid from lymphoid leukemias and B- and T-lineage ALL. More recently, a large study was conducted that could distinguish T-lineage ALL and all of the common B-lineage chromosomal translocations with specific transcript profiles (36) . Molecular classification schemes have also been applied to human B-cell lymphomas and B-lineage ALL and have uncovered previously unrecognized genetic heterogeneity among patient subgroups, some of which correlate with prognosis (36 , 45) . Variation in gene expression has also been characterized recently in human breast cancers (46) and correlated with the risk of metastasis in melanomas (47) . Direct clinical applications include predicting prognosis, refining diagnosis, and following disease progression.

In our initial profiling of 51 pediatric bone marrow samples, we have used microarray technology to both segregate patient subgroups and to uncover genetic diversity among patients that fall within the same traditional risk groups. We used two complementary methods of supervised classification for our analyses that differ in their assumptions of the underlying data structure. t tests assume normally distributed data, and significance is defined by how distinct the means of the cohorts are, given the variability of the data. In contrast, Infoscore does not assume anything about the data distribution, but expression data with little dynamic range could result in the identification of genes whose expression level between cohorts may not be biologically significant. Significant overlap in the top 20 genes identified by these two methods was seen. Still, we hypothesize that the best discriminators are those differentially expressed using a combination of the two methods. To test this hypothesis, the expression of potential gene discriminators was validated by real-time PCR in both the samples on the arrays and in other uniformly processed leukemic marrow samples not included on the arrays.

Independent validation of the array data were seen in the AML versus ALL and T-cell versus B-lineage cohorts. Validation of the expression of protein kinase C substrate, however, was not seen in the standard-risk and high-risk ALL samples that were not on the arrays. This was not unexpected because the clinical distinctions among patients within different risk strata of a uniform disease type may be less than the differences among entirely different types of leukemia or different lineages. Also, a smaller sample number limited the standard-risk/high-risk comparison as well as the expected regression to the mean effect. A group of genes, rather than an individual gene, may more accurately predict risk-group assignment. In the future, we plan to use the same methods, in larger numbers of banked samples, to identify a set of genes that reliably differentiate standard-risk from high-risk ALL. Once a discriminating gene set for standard-risk and high-risk ALL is validated, patients could potentially undergo testing for gene expression at the time of initial diagnosis. This could be done efficiently using techniques such as quantitative PCR with the ultimate objective of using this information prospectively to augment risk classification schemes.

A unique aspect of our study was the exclusive use of freshly harvested bone marrow and uniform RNA processing within 1 h of sample acquisition. It is currently unknown whether differences in expression profiles are seen as a result of differences in sample banking or processing. A limitation to this method was small patient sample number, because we analyzed only samples from our own institution and, furthermore, only those in which enough additional bone marrow could be obtained at the time of initial diagnosis to perform an array without amplification. The overall goal of this initial study, however, was to pilot a method of analysis that could be used in larger sample numbers in the future.

As reported by Golub et al. (44) , we found that no single transcript could unilaterally distinguish subgroups, even between seemingly disparate subgroups such as ALL versus AML. However, we did identify and validate certain genes that did distinguish between ALL versus ALL and T-lineage versus B-lineage. Furthermore, several predicted patterns of differential gene expression were observed. Class II HLA molecules are located predominantly on B cells and macrophages, playing a major role in antigen presentation. Consistent with known endogenous patterns of expression, MHC molecules were consistently overexpressed in B-lineage ALL when compared with T-ALL. Of the top 20 genes discriminating T-lineage versus B-lineage ALL, 5 were class II MHC molecules using both methods of analysis. Similarly, the delta subunit of the T-cell antigen receptor, CD3, was highly expressed in T-ALL samples. Although the majority of the 50 genes reported to differentiate AML versus ALL by oligonucleotide array in an analysis by Golub et al. were not on our cDNA arrays, we observed some similar patterns of gene expression in our comparison of ALL versus AML samples, i.e., SNF2 was more highly expressed in ALL samples, and proteoglycan and cathepsin D were more highly expressed in AML patient samples. Differentially regulated genes from broad functional categories were also identified. For example, cell surface proteins (CD14 antigen), genes involved in cell cycle control (S-100 proteins, rhoG), transcription factors (SRY-box 4), oncogenes (v-fgr), cell adhesion molecules (integrin ß2), genes involved in metabolism and protein degradation (phosphodiesterase, cathepsin D), and genes involved in chromatin remodeling (SNF2b) were among those that were differentially expressed within the cohorts of leukemia patients.

Although limited by the small sample number for certain disease subtypes, interesting clinical correlates were suggested by this preliminary analysis. Some admixing of the gene expression profiles of newly diagnosed B-lineage ALL standard-risk and high-risk patients was seen. Of note, the three standard-risk patients that were misclassified as high-risk by both methods all showed a slow initial marrow response with day 7 marrow blast percentages ranging from 34 to 92%. Indeed, one patient has died from an early marrow relapse. In contrast, the four high-risk patients misclassified as standard risk all demonstrated rapid early marrow responses, with day 7 marrow blast percentages of <25%. Although the prognostic significance of slow early marrow response remains to be seen in standard-risk patients, it has been associated with inferior outcomes in patients with higher risk disease, unless therapy is intensified (48 , 49) . The admixing of standard-risk and high-risk patient expression profiles was not unexpected because approximately one quarter of standard-risk patients develop disease recurrence, and conversely, a significant subset of high-risk patients respond very favorably to less intensive therapy. Upon further validation of the discriminating gene set in a larger sample size, future goals include prospective testing of patient marrow samples at the time of diagnosis.

In summary, we have demonstrated that microarray technology can be successfully applied to the study of childhood leukemia, broadly segregating patients defined by traditional measures such as clinical features, immunophenotype, and cytogenetic alterations. This method can also uncover previously unrecognized diversity among individual patients in disease subgroups. This additional diversity may explain differences in clinical behavior and may help to further refine risk-group allocation. Future direct clinical applications of DNA array technology include the prospective identification of specific patterns of gene expression that predict clinical outcome more precisely. Finally, these studies of childhood leukemia may provide insight into relevant pathways for the development of novel therapeutics.


   ACKNOWLEDGMENTS
 
We acknowledge all of the patients and their families who participated in this study. In addition, we thank members of the Division of Pediatric Hematology-Oncology at Primary Children’s Medical Center for their help in obtaining marrow samples. We would also like to acknowledge the Microarray Core Facility and the Biostatistics Resource at the Huntsman Cancer Institute and Drs. Stephen Prescott, Joseph Yost, Frank Fitzpatrick, and David Jones for their helpful comments and critical review of the manuscript. We would also like to thank Drs. Bruce Camitta and Paul Gaynon of the Children’s Oncology Group ALL Strategy Committee for their encouragement and support of these studies.


   FOOTNOTES
 
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1 Supported by Translational Research Awards from the Leukemia and Lymphoma Society (to W. L. C. and S. P. H.), Director’s Challenge Grant U01 CA88361 (to W. L. C. and C. W.) from the National Cancer Institute, and NIH Mentored Clinical Scientist Development Award K08 HD01215-01 (to E. R.). Core Facility support was provided by Cancer Center Support Grant 2P30CA42014 and The Huntsman Cancer Foundation. Back

2 These authors contributed equally to this work. Back

3 To whom requests for reprints should be addressed, at Combined Division of Pediatric Oncology Mount Sinai and NYU Schools of Medicine, 1 Gustave Levy Place, New York, NY 10029. Phone: (212) 241-7022; Fax: (212) 360-6921. Back

4 The abbreviations used are: ALL, acute lymphoblastic leukemia; AML, acute myelogenous leukemia; RT-PCR, reverse transcription-PCR; LOOCV, leave-one-out cross-validation; FAB, French-American-British. Back

Received 12/ 4/01; revised 4/23/02; accepted 6/11/02.


   REFERENCES
Top
 ABSTRACT
 INTRODUCTION
 PATIENTS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 

  1. Pui C. H. Childhood leukemias. N. Engl. J. Med., 332: 1618-1630, 1995.[Free Full Text]
  2. Rivera G. K., Pinkel D., Simone J. V., Hancock M. L., Crist W. M. Treatment of acute lymphoblastic leukemia. 30 years’ experience at St. Jude Children’s Research Hospital[see comments]. N. Engl. J. Med., 329: 1289-1295, 1993.[Abstract/Free Full Text]
  3. Pui C. H., Evans W. E. Acute lymphoblastic leukemia. N. Engl. J. Med., 339: 605-615, 1998.[Free Full Text]
  4. Gaynon P. S., Qu R. P., Chappell R. J., Willoughby M. L., Tubergen D. G., Steinherz P. G., Trigg M. E. Survival after relapse in childhood acute lymphoblastic leukemia: impact of site and time to first relapse—the Children’s Cancer Group Experience. Cancer (Phila.), 82: 1387-1395, 1998.[CrossRef][Medline]
  5. Chessells J. Relapsed lymphoblastic leukemia in children: a continuing challenge. Br. J. Haematol., 102: 423-438, 1998.[CrossRef][Medline]
  6. Miller D. R., Leikin S., Albo V., Vitale L., Sather H., Coccia P., Nesbit M., Karon M., Hammond D. Use of prognostic factors in improving the design and efficiency of clinical trials in childhood leukemia: Children’s Cancer Study Group Report. Cancer Treat. Rep., 64: 381-392, 1980.[Medline]
  7. Pui C. H. Acute lymphoblastic leukemia in children. Curr. Opin. Oncol., 12: 3-12, 2000.[CrossRef][Medline]
  8. Smith M., Arthur D., Camitta B., Carroll A. J., Crist W., Gaynon P., Gelber R., Heerema N., Korn E. L., Link M., Murphy S., Pui C. H., Pullen J., Reamon G., Sallan S. E., Sather H., Shuster J., Simon R., Trigg M., Tubergen D., Uckun F., Ungerleider R. Uniform approach to risk classification and treatment assignment for children with acute lymphoblastic leukemia[see comments]. J. Clin. Oncol., 14: 18-24, 1996.[Abstract]
  9. Ribeiro R. C., Pui C. H. Prognostic factors in childhood acute lymphoblastic leukemia. Hematol. Pathol., 7: 121-142, 1993.[Medline]
  10. Williams D. L., Harber J., Murphy S. B., Look A. T., Kalwinsky D. K., Rivera G., Melvin S. L., Stass S., Dahl G. V. Chromosomal translocations play a unique role in influencing prognosis in childhood acute lymphoblastic leukemia. Blood, 68: 205-212, 1986.[Abstract/Free Full Text]
  11. Pui C. H., Crist W. M., Look A. T. Biology and clinical significance of cytogenetic abnormalities in childhood acute lymphoblastic leukemia. Blood, 76: 1449-1463, 1990.[Abstract/Free Full Text]
  12. Chessels J. M., Swansbury G. J., Reeves B., Bailey C. C., Richards S. M. Cytogenetics and prognosis in childhood lymphoblastic leukaemia: results of MRC UKALL X. Medical Research Council Working Party in Childhood Leukaemia. Br. J. Haematol., 99: 93-100, 1997.[CrossRef][Medline]
  13. Shurtleff S. A., Buijs A., Behm F. G., Rubnitz J. E., Raimondi S. C., Hancock M. L., Chan G. C., Pui C. H., Grosveld G., Downing J. R. TEL/AML1 fusion resulting from a cryptic t(12;21) is the most common genetic lesion in pediatric ALL and defines a subgroup of patients with an excellent prognosis. Leukemia (Baltimore), 9: 1985-1989, 1995.[Medline]
  14. McLean T. W., Ringold S., Neuberg D., Stegmaier K., Tantravahi R., Ritz J., Koeffler H. P., Takeuchi S., Janssen J. W., Seriu T., Bartram C. R., Sallan S. E., Gilliland D. G., Golub T. R. TEL/AML-1 dimerizes and is associated with a favorable outcome in childhood acute lymphoblastic leukemia. Blood, 88: 4252-4258, 1996.[Abstract/Free Full Text]
  15. Rubnitz J. E., Downing J. R., Pui C. H., Shurtleff S. A., Raimondi S. C., Evans W. E., Head D. R., Crist W. M., Rivera G. K., Hancock M. L., Boyett J. M., Buijs A., Grosveld G., Behm F. G. TEL gene rearrangement in acute lymphoblastic leukemia: a new genetic marker with prognostic significance. J. Clin. Oncol., 15: 1150-1157, 1997.[Abstract/Free Full Text]
  16. Maloney K., McGavran L., Murphy J., Odom L., Stork L., Wei Q., Hunger S. TEL-AML1 fusion identifies a subset of children with standard risk acute lymphoblastic leukemia who have an excellent prognosis when treated with therapy that includes a single delayed intensification. Leukemia (Baltimore), 13: 1708-1712, 1999.[CrossRef][Medline]
  17. Loh M. L., Silverman L. B., Young M. L., Neuberg D., Golub T. R., Sallan S. E., Gilliland D. G. Incidence of TEL/AML1 fusion in children with relapsed acute lymphoblastic leukemia. Blood, 92: 4792-4797, 1998.[Abstract/Free Full Text]
  18. Harbott J., Viehmann S., Borkhardt A., Henze G., Lampert F. Incidence of TEL/AML1 fusion gene analyzed consecutively in children with acute lymphoblastic leukemia in relapse. Blood, 90: 4933-4937, 1997.[Abstract/Free Full Text]
  19. Chow C. D., Dalla-Poazz L., Gottlieb D. J., Hertzberg M. S. Two cases of very late relapsing ALL carrying the TEL: AML1 fusion gene [letter]. Leukemia (Baltimore), 13: 1893-1894, 1999.[CrossRef][Medline]
  20. Donadieu J., Auclerc M. F., Baruchel A., Leblanc T., Landman-Parker J., Perel Y., Michel G., Cornu G., Bordigoni P., Sommelet D., Leverger G., Hill C., Schaison G. Critical study of prognostic factors in childhood acute lymphoblastic leukaemia: differences in outcome are poorly explained by the most significant prognostic variables. Fralle group. French Acute Lymphoblastic Leukaemia study group. Br. J. Haematol., 102: 729-739, 1998.[CrossRef][Medline]
  21. Schena M., Shalon D., Davis R. W., Brown P. O. Quantitative monitoring of gene expression patterns with a complementary DNA microarray[see comments]. Science (Wash. DC), 270: 467-470, 1995.[Abstract/Free Full Text]
  22. Brown P. O., Botstein D. Exploring the new world of the genome with DNA microarrays. Nat. Genet., 21: 33-37, 1999.[CrossRef][Medline]
  23. Duggan D. J., Bittner M., Chen Y., Meltzer P., Trent J. M. Expression profiling using cDNA microarrays. Nat. Genet., 21: 10-14, 1999.[CrossRef][Medline]
  24. Young R. A. Biomedical discovery with DNA arrays. Cell, 102: 9-15, 2000.[CrossRef][Medline]
  25. Hunger S. P., Galili N., Carroll A. J., Crist W. M., Link M. P., Cleary M. L. The t(1;19)(q23;p13) results in consistent fusion of E2A and PBX1 coding sequences in acute lymphoblastic leukemias. Blood (Baltimore), 77: 687-693, 1991.[Abstract/Free Full Text]
  26. Pallisgaard N., Hokland P., Riishoj D. C., Pedersen B., Jorgensen P. Multiplex reverse transcription-polymerase chain reaction for simultaneous screening of 29 translocations and chromosomal aberrations in acute leukemia. Blood, 92: 574-588, 1998.[Abstract/Free Full Text]
  27. Nakao M., Yokota S., Horiike S., Taniwaki M., Kashima K., Sonoda Y., Koizumi S., Takaue Y., Matsushita T., Fujimoto T., Misawa S. Detection and quantification of TEL/AML1 fusion transcripts by polymerase chain reaction in childhood acute lymphoblastic leukemia. Leukemia (Baltimore), 10: 1463-1470, 1996.[Medline]
  28. Nucifora G., Birn D. J., Erickson P., Gao J., LeBeau M. M., Drabkin H. A., Rowley J. D. Detection of DNA rearrangements in the AML1 and ETO loci and of an AML1/ETO fusion mRNA in patients with t(8;21) acute myeloid leukemia. Blood, 81: 883-888, 1993.[Abstract/Free Full Text]
  29. van der Reijden B. A., Lombardo M., Dauwerse H. G., Giles R. H., Muhlematter D., Bellomo M. J., Wessels H. W., Beverstock G. C., van Ommen G. J., Hagemeijer A., et al RT-PCR diagnosis of patients with acute nonlymphocytic leukemia and inv(16)(p13q22) and identification of new alternative splicing in CBFB- MYH11 transcripts. Blood, 86: 277-282, 1995.[Abstract/Free Full Text]
  30. Tsodikov A., Szabo A., Jones D. Adjustments and measures of differential expression for microarray data. Bioinformatics, 18: 251-260, 2002.[Abstract/Free Full Text]
  31. Hedenfalk I., Duggan D., Chen Y., Radmacher M., Bittner M., Simon R., Meltzer P., Gusterson B., Esteller M., Kallioniemi O. P., Wilfond B., Borg A., Trent J. Gene-expression profiles in hereditary breast cancer. N. Engl. J. Med., 344: 539-548, 2001.[Abstract/Free Full Text]
  32. McLachlan G. F. Discriminant Analysis and Statistical Pattern Recognition. Series in Probability and Applied Statistics, 9-15, Wiley New York 1992.
  33. Eisen M. B., Spellman P. T., Brown P. O., Botstein D. Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. USA, 95: 14863-14868, 1998.[Abstract/Free Full Text]
  34. Morrison T. B., Weis J. J., Wittwer C. T. Quantification of low-copy transcripts by continuous SYBR Green I monitoring during amplification. Biotechniques, 24: 954-958, 960, 962, 1998.[Medline]
  35. Heid C. A., Stevens J., Livak K. J., Williams P. M. Real time quantitative PCR. Genome Res., 6: 986-994, 1996.[Abstract/Free Full Text]
  36. Yeoh E-J., Ross M. E., Shurtleff S. A., Williams W. K., Patel D., Mahfouz R., Behm F. G., Raimondi S. C., Relling M. V., Patel A., Cheng C., Campana D., Wilkins D., Zhou X., Li J., Liu H., Pui C-H., Evans W. E., Naeve C., Wong L., Downing J. R. Classification, subtype discovery, and prediction of outcome in pediatric acute lymphoblastic leukemia by gene expression profiling. Cancer Cell, 1: 133-143, 2002.[CrossRef][Medline]
  37. Pui C. H., Crist W. M. Biology and treatment of acute lymphoblastic leukemia[see comments]. J. Pediatr., 124: 491-503, 1994.[CrossRef][Medline]
  38. Crist W., Boyett J., Pullen J., van Eys J., Vietti T. Clinical and biologic features predict poor prognosis in acute lymphoid leukemias in children and adolescents: a Pediatric Oncology Group review. Med. Pediatr. Oncol., 14: 135-139, 1986.[Medline]
  39. Lilleyman J. S. Clinical importance of speed of response to therapy in childhood lymphoblastic leukaemia. Leuk. Lymphoma, 31: 501-506, 1998.[Medline]
  40. Gaynon P. S., Desai A. A., Bostrom B. C., Hutchinson R. J., Lange B. J., Nachman J. B., Reaman G. H., Sather H. N., Steinherz P. G., Trigg M. E., Tubergen D. G., Uckun F. M. Early response to therapy and outcome in childhood acute lymphoblastic leukemia: a review. Cancer (Phila.), 80: 1717-1726, 1997.[CrossRef][Medline]
  41. Trueworthy R., Shuster J., Look T., Crist W., Borowitz M., Carroll A., Frankel L., Harris M., Wagner H., Haggard M., Mosijczuk A., Pullen J., Steuber P., Land V. Ploidy of lymphoblasts is the strongest predictor of treatment outcome in B-progenitor cell acute lymphoblastic leukemia of childhood: a Pediatric Oncology Group study. J. Clin. Oncol., 10: 606-613, 1992.[Abstract/Free Full Text]
  42. Ito C., Kumagai M., Manabe A., Coustan-Smith E., Raimondi S. C., Behm F. G., Murti G., Rubnitz J. E., Pui C-H., Campana D. Hyperdiploid acute lymphoblastic leukemia with 51 to 65 chromosomes: a distinct biological entity with a marked propensity to undergo apoptosis. Blood, 93: 315-320, 1999.[Abstract/Free Full Text]
  43. Heerema N., Nachman J., Sather H., Sensel M., Lee M., Hutchinson R., Lange B., Steinherz P., Bostrom B., Gaynon P., Arthur D., Uckun F. Hypodiploidy with <45 chromosomes confers adverse risk in childhood acute lymphoblastic leukemia: a report from the Children’s Cancer Group. Blood, 94: 4036-4045, 1999.[Abstract/Free Full Text]
  44. Golub T. R., Slonim D. K., Tamayo P., Huard C., Gaasenbeek M., Mesirov J. P., Coller H., Loh M. L., Downing J. R., Caligiuri M. A., Bloomfield C. D., Lander E. S. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science (Wash. DC), 286: 531-537, 1999.[Abstract/Free Full Text]
  45. Alizadeh A. A., Eisen M. B., Davis R. E., Ma C., Lossos I. S., Rosenwald A., Boldrick J. C., Sabet H., Tran T., Yu X., Powell J. I., Yang L., Marti G. E., Moore T., Hudson J., Jr., Lu L., Lewis D. B., Tibshirani R., Sherlock G., Chan W. C., Greiner T. C., Weisenburger D. D., Armitage J. O., Warnke R., Staudt L. M., et al Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling[see comments]. Nature (Lond.), 403: 503-511, 2000.[CrossRef][Medline]
  46. Perou C., Sorlie T., M. E. Molecular portraits of human breast tumors. Nature (Lond.), 406: 747-752, 2000.[CrossRef][Medline]
  47. Clark E. A., Golub T. R., Lander E. S., Hynes R. O. Genomic analysis of metastasis reveals an essential role for RhoC. Nature (Lond.), 406: 532-535, 2000.[CrossRef][Medline]
  48. Nachman J. B., Sather H. N., Sensel M. G., Trigg M. E., Cherlow J. M., Lukens J. N., Wolff L., Uckun F. M., Gaynon P. S. Augmented post-induction therapy for children with high-risk acute lymphoblastic leukemia and a slow response to initial therapy. N. Engl. J. Med., 338: 1663-1671, 1998.[Abstract/Free Full Text]
  49. Tubergen D. G., Gilchrist G. S., O’Brien R. T., Coccia P. F., Sather H. N., Waskerwitz M. J., Hammond G. D. Improved outcome with delayed intensification for children with acute lymphoblastic leukemia and intermediate presenting features: a Childrens Cancer Group Phase III trial. J. Clin. Oncol., 11: 527-537, 1993.[Abstract/Free Full Text]



This article has been cited by other articles:


Home page
 BloodHome page
S. S. Winter, Z. Jiang, H. M. Khawaja, T. Griffin, M. Devidas, B. L. Asselin, and R. S. Larson
Identification of genomic classifiers that distinguish induction failure in T-lineage acute lymphoblastic leukemia: a report from the Children's Oncology Group
Blood, September 1, 2007; 110(5): 1429 - 1438.
[Abstract] [Full Text] [PDF]

Home page
 Clin. Chem.Home page
C. Langebrake, K. Gunther, J. Lauber, and D. Reinhardt
Preanalytical mRNA Stabilization of Whole Bone Marrow Samples
Clin. Chem., April 1, 2007; 53(4): 587 - 593.
[Abstract] [Full Text] [PDF]

Home page
 Clin. Cancer Res.Home page
Y. Ge, C. L. Haska, K. LaFiura, M. Devidas, S. B. Linda, M. Liu, R. Thomas, J. W. Taub, and L. H. Matherly
Prognostic Role of the Reduced Folate Carrier, the Major Membrane Transporter for Methotrexate, in Childhood Acute Lymphoblastic Leukemia: A Report from the Children's Oncology Group
Clin. Cancer Res., January 15, 2007; 13(2): 451 - 457.
[Abstract] [Full Text] [PDF]

Home page
 Clin. Cancer Res.Home page
R. Kirschner-Schwabe, C. Lottaz, J. Todling, P. Rhein, L. Karawajew, C. Eckert, A. von Stackelberg, U. Ungethum, D. Kostka, A. E. Kulozik, et al.
Expression of Late Cell Cycle Genes and an Increased Proliferative Capacity Characterize Very Early Relapse of Childhood Acute Lymphoblastic Leukemia
Clin. Cancer Res., August 1, 2006; 12(15): 4553 - 4561.
[Abstract] [Full Text] [PDF]

Home page
 Clin. Cancer Res.Home page
S. Chiaretti, X. Li, R. Gentleman, A. Vitale, K. S. Wang, F. Mandelli, R. Foa, and J. Ritz
Gene Expression Profiles of B-lineage Adult Acute Lymphocytic Leukemia Reveal Genetic Patterns that Identify Lineage Derivation and Distinct Mechanisms of Transformation
Clin. Cancer Res., October 15, 2005; 11(20): 7209 - 7219.
[Abstract] [Full Text] [PDF]

Home page
 BloodHome page
G. Cario, M. Stanulla, B. M. Fine, O. Teuffel, N. v. Neuhoff, A. Schrauder, T. Flohr, B. W. Schafer, C. R. Bartram, K. Welte, et al.
Distinct gene expression profiles determine molecular treatment response in childhood acute lymphoblastic leukemia
Blood, January 15, 2005; 105(2): 821 - 826.
[Abstract] [Full Text] [PDF]

Home page
 BloodHome page
M. E. Ross, X. Zhou, G. Song, S. A. Shurtleff, K. Girtman, W. K. Williams, H.-C. Liu, R. Mahfouz, S. C. Raimondi, N. Lenny, et al.
Classification of pediatric acute lymphoblastic leukemia by gene expression profiling
Blood, October 15, 2003; 102(8): 2951 - 2959.
[Abstract] [Full Text] [PDF]

Home page
 ASH Education BookHome page
W. L. Carroll, D. Bhojwani, D.-J. Min, E. Raetz, M. Relling, S. Davies, J. R. Downing, C. L. Willman, and J. C. Reed
Pediatric Acute Lymphoblastic Leukemia
Hematology, January 1, 2003; 2003(1): 102 - 131.
[Abstract]