The derivation of diagnostic markers of chronic myeloid leukemia progression from microarray data

Abstract

Currently, limited molecular markers exist that can determine where in the spectrum of chronic myeloid leukemia (CML) progression an individual patient falls at diagnosis. Gene expression profiles can predict disease and prognosis, but most widely used microarray analytical methods yield lengthy gene candidate lists that are difficult to apply clinically. Consequently, we applied a probabilistic method called Bayesian model averaging (BMA) to a large CML microarray dataset. BMA, a supervised method, considers multiple genes simultaneously and identifies small gene sets. BMA identified 6 genes (NOB1, DDX47, IGSF2, LTB4R, SCARB1, and SLC25A3) that discriminated chronic phase (CP) from blast crisis (BC) CML. In CML, phase labels divide disease progression into discrete states. BMA, however, produces posterior probabilities between 0 and 1 and predicts patients in "intermediate" stages. In validation studies of 88 patients, the 6-gene signature discriminated early CP from late CP, accelerated phase, and BC. This distinction between early and late CP is not possible with current classifications, which are based on known duration of disease. BMA is a powerful tool for developing diagnostic tests from microarray data. Because therapeutic outcomes are so closely tied to disease phase, these probabilities can be used to determine a risk-based treatment strategy at diagnosis.

Figure 1

Experimental overview. A summary of our overall approach is shown. The merits of BMA include that the model is multivariate and considers multiple biomarkers simultaneously; that the uncertainty of gene selection is accounted for by averaging over multiple good models; that posterior probabilities are generated for all selected genes and models; and that it produces a high predictive accuracy with relatively few genes.

Figure 2

Model selection by BMA. This figure illustrates the membership of the 6 signature genes in the 21 models selected by the iterative BMA algorithm on the CML progression microarray data. The 6 signature genes are shown on the vertical axis, whereas the 21 models are shown in the horizontal axis. The widths of the columns are proportional to the posterior probabilities of the selected models. A red entry indicates that the corresponding gene is included in a given model.

Figure 3

Differential gene expression of the 6-gene signature in patient cases. Heatmaps of the 6-gene signature selected by the iterative BMA algorithm. (A) Heatmap of the gene expression of the 6-gene signature in the training set of CP and BC CML cases using microarray-based gene expression analyses. (B) Heatmap of the gene expression in the test set, also by microarray-based gene expression analyses. This group is made up of AP by cytogenetic criteria only (AP cyto); AP by cytogenetic and blast count criteria (AP); and BC after a return to second CP (BC-rem). Green indicates differentially decreased expression of each gene in each case compared with its expression in a pool of CP patient samples and red indicates differentially increased expression. The more saturated the color, the greater is the degree of differential expression. (C) A graphic depiction of the predicted posterior probability of all patient samples is shown. The samples are represented on the horizontal axis, whereas the predicted probabilities are represented on the vertical axis. For the training data consisting of 72 CP (group 0) and BC (group 1) cases, the average predicted probabilities over 100 cross validation runs are shown. For the test data, composed of 21 patients before allogeneic transplantation, the predicted probabilities of the 6 signature genes averaged over 21 models are shown. This “intermediate phase” test set is composed of group 2 AP by cytogenetic criteria only; group 3 AP by cytogenetic and blast count criteria; and group 4 BC after a return to second CP. For 17 of 21 patients, posttransplantation outcomes were available. Using a posterior probability threshold of 0.3, we found that 7 of 11 patients with predicted posterior probabilities ≥ 0.3 died of relapse or treatment-related mortality after transplantation versus only 1 of 6 with a predicted posterior probability < 0.3 (OR = 9).

Figure 4

The 6-gene signature discriminates early from late CP in independent patient samples by QPCR. Results of leave-one-out cross validation comparing diagnostic early CP (class 0, n = 45) and late CP (class 1, n = 22) patient samples (A) including graphic representation (B). Both early CP patients misclassified at diagnosis as late CP (indicated in the red circles) subsequently did poorly on imatinib mesylate therapy.