Comparative Gene Expression Analysis Reveals Similarities and Differences of Chronic Myeloid Leukemia Phases

Abstract

Chronic myeloid leukemia (CML) is a slowly progressing blood cancer that primarily affects elderly people. Without successful treatment, CML progressively develops from the chronic phase through the accelerated phase to the blast crisis, and ultimately to death. Nowadays, the availability of targeted tyrosine kinase inhibitor (TKI) therapies has led to long-term disease control for the vast majority of patients. Nevertheless, there are still patients that do not respond well enough to TKI therapies and available targeted therapies are also less efficient for patients in accelerated phase or blast crises. Thus, a more detailed characterization of molecular alterations that distinguish the different CML phases is still very important. We performed an in-depth bioinformatics analysis of publicly available gene expression profiles of the three CML phases. Pairwise comparisons revealed many differentially expressed genes that formed a characteristic gene expression signature, which clearly distinguished the three CML phases. Signaling pathway expression patterns were very similar between the three phases but differed strongly in the number of affected genes, which increased with the phase. Still, significant alterations of MAPK, VEGF, PI3K-Akt, adherens junction and cytokine receptor interaction signaling distinguished specific phases. Our study also suggests that one can consider the phase-wise CML development as a three rather than a two-step process. This is in accordance with the phase-specific expression behavior of 24 potential major regulators that we predicted by a network-based approach. Several of these genes are known to be involved in the accumulation of additional mutations, alterations of immune responses, deregulation of signaling pathways or may have an impact on treatment response and survival. Importantly, some of these genes have already been reported in relation to CML (e.g., AURKB, AZU1, HLA-B, HLA-DMB, PF4) and others have been found to play important roles in different leukemias (e.g., CDCA3, RPL18A, PRG3, TLX3). In addition, increased expression of BCL2 in the accelerated and blast phase indicates that venetoclax could be a potential treatment option. Moreover, a characteristic signaling pathway signature with increased expression of cytokine and ECM receptor interaction pathway genes distinguished imatinib-resistant patients from each individual CML phase. Overall, our comparative analysis contributes to an in-depth molecular characterization of similarities and differences of the CML phases and provides hints for the identification of patients that may not profit from an imatinib therapy, which could support the development of additional treatment strategies.

Keywords: accelerated phase; blast crisis; chronic myeloid leukemia (CML); chronic phase; comparative transcriptome analysis; computational cancer genomics.

Figure 1

Principal component analysis of genome-wide CML expression data. Individual patient samples are represented by colored dots. The color of a dot defines the CML phase of the sample (yellow: chronic phase, red/pink: accelerated phase, light blue: blast crisis).

Figure 2

Differential gene expression analysis and functional annotation analysis of altered genes based on pairwise comparisons of the three CML phases. Volcano plots quantifying the average expression changes of genes (x-axis) in relation to the obtained significance of gene expression difference (y-axis) are shown in the first row. The gray dotted line represents the significance cutoff (q-value: 0.05). Underexpressed genes are represented by blue dots and overexpressed genes by red dots. The corresponding numbers of under- and overexpressed genes in different functional annotation categories (TF: transcription factor or co-factor, KI: kinase, PP: phosphatase, SP: signaling pathway gene, MP: metabolic pathway gene, OG: oncogene, TS: tumor suppressor gene, CC: cancer census gene, EG: essential gene) are shown in the bar plots in the second row. Enriched genes in a specific functional category are highlighted by asterisks (Fisher’s exact test: ‘**’: FDR-adjusted p ≤ 0.01, ‘*’: FDR-adjusted p ≤ 0.05). The comparisons of accelerated to chronic phase, blast to accelerated phase, and blast to chronic phase are shown in the columns (a–c).

Figure 3

Gene expression alterations affecting cancer-relevant signaling pathways. Percentages of under- and overexpressed genes observed for specific signaling pathways are shown for the pairwise comparisons of the three CML phases. Differentially expressed genes at the q-value cutoff of 0.05 were considered and overrepresented pathways were marked by asterisks separately for an enrichment of under- or overexpressed genes (Fisher’s exact test: ‘**’: FDR-adjusted p ≤ 0.01, ‘*’: FDR-adjusted p ≤ 0.05). Signaling pathway alterations for the comparisons of accelerated to chronic phase, blast to accelerated phase, and blast to chronic phase are shown in the subpanels (a–c), respectively.

Figure 4

Differentially expressed genes distinguishing the three CML phases. Heatmap of individual CML gene expression profiles. CML samples were clustered based on the similarity of their gene expression profiles (columns) and their corresponding gene-specific expression values are visualized (rows). All differentially expressed genes of the pairwise comparisons of the three CML phases at the level of the q-value cutoff of 0.05 are included. The heatmap represents log${}_2$-ratio expression values of each specific CML sample in comparison to a pool of reference samples visualizing underexpressed genes with values clearly less than zero in blue shades, unchanged genes with values about zero in gray, and overexpressed genes with values clearly greater than zero in red shades of the specific sample. The colored line below the column dendrogram at top of the heatmap represents the phase-specific label of each sample (yellow: chronic phase, red: accelerated phase, light blue: blast crisis). Lilac, orange and green lines right to the row dendrogram of the heatmap mark known signaling pathway genes, cancer census genes and genes that are part of both categories, respectively.

Figure 5

Network-based visualization of genes with increased network connectivity and corresponding expression behavior for pairwise comparisons of CML phases. Nodes represent selected genes with at least two outgoing links to other genes that were present in at least 90 of the 100 learned networks. Node sizes are proportional to their out-degrees and nodes are colored according to expression change observed for the pairwise comparison of CML phases (blue: underexpressed, gray: unchanged, red: overexpressed). Predicted activating links between genes are shown in orange. Links can represent direct or indirect regulatory dependencies or may only represent correlations between expression levels of genes. The network (a) shows expression alterations of the selected genes with increased connectivity to other signature genes for the comparison of the accelerated to the chronic phase. The network (b) shows the expression behavior of these genes for the comparison of the blast to the accelerated phase.

Figure 6

Comparison of transcriptomes of imatinib-resistant patients to patients of the three CML phases. (a) Heatmap of pairwise correlations between genome-wide gene expression profiles of imatinib-resistant patients and individual patients of the three CML phases. The imatinib-resistant patients are shown in the rows of the heatmap clustered according to the similarity of their correlation profiles (green: imatinib-resistant). The individual patients from the three CML phases are shown in the columns of the heatmap in phase-wise order (yellow: chronic phase, red: accelerated phase, light blue: blast crisis). The heatmap represents negative correlations in blue shades, correlations about zero in gray shades, and positive correlations in red shades. (b) Bar plot of percentages of under- and overexpressed genes of specific signaling pathways comparing imatinib-resistant patients to patients in blast crisis. Differentially expressed genes at the q-value cutoff of 0.05 with an average absolute expression change of |log${}_2$-ratio| ≥ 1 were considered. Overrepresented pathways are marked by asterisks for an enrichment of under- or overexpressed genes (Fisher’s exact test: ‘**’: FDR-adjusted p < 0.01). Identical enrichment patterns were also observed for the comparison to the chronic and accelerated phase (Figure S3).