An Approach to Virtual Karyotyping for Unbalanced Rearrangements Based on Gene Expression Data.

Effects of gene dosage on gene expression have been demonstrated in several leukemia entities. We asked the question whether unbalanced chromosome aberrations can be reliably identified based on gene expression data. Therefore, we performed genomic arrays (Affymetrix 10K arrays) and gene expression arrays (Affymetrix U133A+B or 2.0 plus) in 33 AML with a complex aberrant karyotype. For comparison gene expression data of 100 AML with normal karyotype were used. Based on this data set an algorithm was developed for the detection of higher and lower expression of groups of genes located in chromosomal bands (CB) in individual patients. Therefore, a “normal” expression status was defined for each CB based on gene expression data of AML with normal karyotype. Next for each individual case the gene expression status of each CB in relation to the earlier defined “normal” expression status was calculated and if a significant deviation was observed called “+” or “-”. First the new algorithm was tested for prediction of 5q deletions and their localization as well as for −7 and 17p deletion comprising 64 CB. Thus, overall 2112 pairs of genomic (GD) and gene expression data (GED) were compared in these 33 cases with AML. Based on GD 1011 CB were identified as deleted, 648 of these showed a significantly reduced gene expression. In addition gene expression was not reduced in 898 not deleted CB resulting in 1546/2112 (73.2%) concordant results between GD and GED. Thus, the sensitivity for prediction of deletion based on gene expression was 64% and the specificity 82%. In the next step gene expression data was correlated to karyotype data (KD) obtained by chromosome analysis. Therefore, 48 AML cases with available GED and a karyotype showing unbalanced chromosome aberrations such as trisomies of chromosomes 4 (n=2), 8 (n=14), 9 (n=1), 10 (n=1), 11 (n=3), 13 (n=8), 14 (n=3), 19 (n=1), 21 (n=3), and 22 (n=2) and monosomies of chromosomes 7 (n=7) and 13 (n=1) as well as gain of 1q (n=2), loss of 3q (n=1), 5q (n=11), 7q (n=4), 9q (n=3), 12p (n=1), and 10q (n=1) were evaluated. First we performed in this second cohort an analysis focusing again on loss of 5q, 7 and 17p. Based on KD 607 CB were identified as deleted, 390 of these showed a significantly reduced gene expression, in addition gene expression was not reduced in 2232 not deleted CB resulting in 2622/3408 (76.9%) concordant results between KD and GED. Thus, the sensitivity for prediction of deletion based on gene expression was 64% and the specificity 80%. Next an overall analysis on the whole genome broken down into 653 CB (excluding X and Y chromosome) was performed in the second cohort. Based on KD 1002 CB were gained, 820 lost and 29522 not affected by unbalanced rearrangements. A significantly higher gene expression was found in 570/1002 gained CB, a lower expression in 501/820 lost CB and an unchanged expression in 17366/29522 CB not affected, resulting in 18437/31344 (58.8%) concordant results between KD and GED. Thus, the sensitivity for prediction of deletion based on gene expression was 61% and the specificity 82% and for gain 57% and 81%. In conclusion, we present an approach for predicting unbalanced karyotype changes based on gene expression. It could be demonstrated genome wide that an association between gene dosage and gene expression exists.