Abstract

Background: Patients with chronic phase CML who achieve a complete cytogenetic response (CCR) have a low risk of disease progression. Since patients unlikely to achieve a CCR may benefit from more aggressive therapy, it would be clinically advantageous to identify those patients prior to therapy. Based on the hypothesis that cytogenetic refractoriness may be a property of leukemic progenitor cells, we explored the potential of gene expression profiling of CD34+ cells as a tool for predicting failure to achieve CCR.

Methods: Fifty-one patients with CML who either achieved a CCR within 1 year of imatinib therapy (R, n = 35), or remained at least 65% Ph+ (NR, n = 16) were included in the study. Pre-imatinib CD34+ cells were FACS isolated from cryopreserved bone marrow mononuclear cells. RNA was extracted from the CD34+ cells and samples with >=5ng of high quality total RNA were amplified and labelled with the Affymetrix two cycle cDNA synthesis and IVT labeling protocol using <20ng input RNA; 10μg of labelled target cRNA were hybridized to Affymetrix HG-U133 Plus 2.0 GeneChip® arrays. Gene-by-gene ANOVA determined differential expression between NR and R. Low-level analysis was done using PLIER (Affymetrix) and RMA. Ranked p-values of n-way ANOVA results were used to select candidate classifiers and 90 classification models were tested. Best model selection and conserved accuracy estimate were done by 1 and 2 level nested cross validation respectively. Following hierarchical clustering and Principal Component Analysis (PCA), we identified the gene onotologies and associated pathways of the classifiers that had been selected in high frequency in the tested classification models.

Results: The selection procedure yielded CD34+ cells with a median purity of 95.9% and median cell number of 1.0x104. Despite relatively low numbers of input cells, successful hybridization was achieved for 36 patients (24 R and 12 NR). Partial separation of NR and R was seen following hierarchical clustering and PCA, with R vs. NR being identified as a significant source of variation. Regardless of the specific low-level analysis approach, the classifiers performed similarly (conserved estimate of accuracy: 64% PLIER, 65% RMA). Certain biological pathways were associated with those classifiers who were selected in high frequency in the tested classification models (table). The majority of transcripts associated with these pathways were common to both low-level analysis methods, while additional unique transcripts for each pathway were identified with each method.

Conclusions: (i) Gene expression profiling of CD34+ cells selected from cryopreserved bone marrow is feasible. (ii) This pilot study suggests that transcript profiling may have clinically useful predictive value in identifying patients unlikely to respond to imatinib. A larger study is needed to determine the predictive accuracy and obtain a clear understanding of the clinical relevance of those values. (iii) The classifiers are dominated by transcripts associated with PI3K and MAPK signalling, cell cycle, cell adhesion and nucleic acid metabolism.

PathwayRMA Unique TranscriptsPLIER Unique TranscriptsShared Transcripts
Cell Cycle 14 
Complement & Coagulation Cascades 
Oxidative Phosphorylation 
Phosphatidylinositol Signaling System 15 
Purine Metabolism 13 
Pyrimidine Metabolism 
PathwayRMA Unique TranscriptsPLIER Unique TranscriptsShared Transcripts
Cell Cycle 14 
Complement & Coagulation Cascades 
Oxidative Phosphorylation 
Phosphatidylinositol Signaling System 15 
Purine Metabolism 13 
Pyrimidine Metabolism 

Author notes

Corresponding author