Gene Expression Profile (GEP) of Chronic Myeloid Leukemia (CML) Patients at Diagnosis: Two Distinguished Subgroups of CML Patients Identified, Based on a Molecular Signature, Irrespective of Their Sokal Risk Score

Abstract

CML is a clonal myeloproliferative disease which typically presents in chronic phase (CP), in which malignant progenitor cells proliferate rapidly but retain much of their ability to differentiate, with the disease later evolving to accelerated phase/blast crisis. Even after the introduction of imatinib, the calculation of the Sokal and the Euro prognostic scores has remained essential in clinical practice, since allow to stratify CML patients at different evolutive risk at diagnosis, guiding therapeutic decisions. More recently, numerous research efforts are ongoing to gain a better understanding about the intrinsic heterogeneity of CML, in order to identify a novel molecular signature which might characterize patients (pts) with different prognosis, and propensity to respond to treatmentIn the present study we adopted a GEP strategy in an attempt to identify genes and pathways, able to predict and/or elucidate the disease course of CP-CML pts at the time of diagnosis. To this aim, highly enriched CD34+ cells from peripheral blood obtained at diagnosis from pts with untreated Ph+ CML in CP were used throughout the study. Overall, 28 pts were included in the present analysis. They were diagnosed from August 2006 to June 2008; front-line treatment was either imatinib (12 pts) or nilotinib (16 pts).GEP was performed using the Affymetrix HG133 Plus microarray platform as per manufacturer's recommendation. Raw data was normalized using the RMA algorithm and filtered. Unsupervised analysis was performed by Multidimensional Scaling and hierarchical clustering. Top differential genes were selected by significance analysis of microarrays method (SAM), setting the FDR threshold 0.01. Genes associated with Sokal risk score were searched by various methods (Limma, ANOVA, SAM, ROC analysis, EB-arrays). Analyses were performed using R and Bioconductor, BRB array tools and MeV. To analyse the data, we have first stratified our cohort of pts according to their Sokal risk, into 3 groups (high, intermediate and low risk) of 10, 13 and 5 pts, respectively. We did not identify any significant gene signature uniquely associated with Sokal risk score, even by various supervised techniques. Neither was it possible to obtain a signature from the most extreme phenotypes, i.e. by excluding intermediate-risk pts. Interestingly, however, unsupervised analysis of the whole gene expression data set clearly identified two subgroups of pts (we have named them A and B), which included 15 and 13 pts, respectively. Those two subgroups of pts show a differential expression of a list of at least 461 probe sets, which represents the most significantly up and down regulated probes, without any false positive. Both a hierarchical clustering of these probe sets, as well as a multidimensional scaling plot showed a clear demarcation between the two subgroups of pts, and the Sokal risk score did not co-vary along with their gene expression profile. In the group A of pts, of the 461 probe sets, 317 resulted up-regulated and 144 down-regulated. Genes up-regulated are mainly involved in regulation of transcription and/or gene expression (40/317 probe sets code for zinc finger protein), whereas down-modulated genes are involved in cell differentiation, cell death and cell cycle regulation.Of interest, several genes and functional classes showed the same deregulated expression as previously described during progression to blast crisis of CML (Radich et al. PNAS 2006). Moreover, 6 probe sets (EREG, FOS, IL8, WT1, HSPA1A and DKFZp761P0423) that are part of the Radich “top ten” list of genes most significantly associated with progression in CML, are differentially expressed in our two subgroups.Overall, our data suggests the existence of two biologically distinct subgroups of CML, irrespective of Sokal score, which could be identifiable at diagnosis. Those findings further support a hypothesis that different clinical behaviours of the disease and response to treatment could be associated with different gene expression profiles of individual pts. Our plan is to follow up the pts which we have analyzed in this study and to investigate if those molecular signatures, identified through GEP could be correlated with treatment responses.