We assessed total RNA integrity by electrophoresis using the Agilent Bioanalyzer (Agilent, Palo Alto, California). Additional inclusion criteria for accepting expression array results that were at least 10% of genes on an array were expressed (“present call” by MAS 5.0), GAPDH and Actin 3’/5’ ratio less than 5, and a scaling factor within four standard deviations of the mean of all 120 chips analyzed in this study. We isolated mononuclear (leukemia) cells from each bone marrow aspirate by centrifugation over Ficoll. The median percent blasts (post-Ficoll) was 97% in the diagnostic bone marrow samples and 95% in post-treatment samples. All post-treatment samples, except four, had greater than 65% blasts. The four with less than 65% blasts (i.e. 41%, 49%, 60%, 62%) had 78%, 94%, 94% and 98% blasts in their diagnostic samples. The percent blast was not determined in three post-treatment samples, but all had greater than 98% blasts at diagnosis. While differences in post-treatment expression could be influenced by selection of cells that favored survival after treatment, the delayed cytotoxicity of these antimetabolites, coupled with the modest drop in circulating blast counts at the time the post-treatment samples were obtained (median 9.7% decrease across all samples), indicate that selection could not explain the magnitude of changes in gene expression that we observed. Furthermore, the percentage of ALL blasts in the pre- and post-treatment samples did not differ (median of 97% at diagnosis versus 95% post-treatment).
We extensively assessed reproducibility of gene expression within our facility, as previously described (Cancer Cell 1, 133-143 (2002)). Additionally, we tested duplicate cryopreserved bone marrow samples from two patients obtained after treatment; RNA was extracted approximately six months apart. Replicate analysis of the same sample produced similar results, as depicted in Fig. 7 (i.e., R2=0.87 and 0.92).
Expression level of two separate aliquots of post-treatment bone marrow, obtained from two different patients, were extracted and analyzed independently, revealing a high level of concordance between replicate analyses.