Childhood myelodysplastic syndromes (MDS) are a heterogeneous group of stem cell disorders characterized by peripheral blood cytopenia, ineffective hematopoiesis, hyper- or hypocellular bone marrow (BM) and propensity to evolve into acute myeloid leukemia (AML) in approximately 30–40% of cases.1, 2, 3
Childhood MDS are rare diseases and several differences between adult and pediatric MDS have been recognized.1 MDS in pediatric patients is classified, using morphological criteria, into refractory cytopenia of childhood (RCC), refractory anemia with excess of blasts (RAEB) and RAEB in transformation (RAEB-t). Different from adult MDS classification according to World Health Organization (WHO) indications, pediatric MDS patients with a blast count of 20–30% are classified as MDS and not as AML. As relative blasts count alone is not sufficient to differentiate AML from MDS, morphological evaluation remains crucial for diagnosis of MDS.1
Gene expression profile (GEP) analysis has proved to be a powerful tool for the identification of gene signatures associated with distinct leukemia subtypes and has helped to classify these diseases, to stratify patients into different risk classes and to identify deregulated genes involved in leukemia development.4, 5, 6 Only few studies on GEP of pediatric MDS have been published so far,7 mainly due to the rarity of these disorders. Recently, we reported that GEP analysis of whole BM specimens of juvenile myelomonocytic leukemia patients distinguishes clinical subgroups with prognostic relevance.8
Here, we report results of GEP analysis on 32 BM pediatric specimens with a diagnosis of MDS according to the MDS classification proposed for pediatric patients.1 Furthermore, specimens of 16 pediatric patients with a diagnosis of de novo AML with normal karyotype and 8 healthy BM donors, were also included in the analysis. Through GEP analysis, we aimed to characterize subgroups of MDS and to identify the genes and molecular pathways related to the progression of pediatric MDS into AML.
For microarray experiments, GeneChips Human Genome U133 Plus 2.0 (Affymetrix, Santa Clara, CA, USA) was used. Total RNA was extracted from total BM using TRIZOL (Invitrogen, Karlsruhe, Germany). RNA quality and purity were assessed on the Agilent Bioanalyzer 2100 (Agilent Technologies, Waldbronn, Germany); good quality RNA was purified (RNeasy Mini Kit, Qiagen, Hilden, Germany) and RNA concentration was determined using the NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies Inc., Wilmington, DE, USA). In vitro transcription, hybridization and biotin labeling were performed according to Affymetrix One Cycle Target Labelling protocol (Affymetrix).
CEL files can be found at GEO repository datasets (GEO, http://www.ncbi.nlm.nih.gov/geo; Series accession number GSE29326). Gene expression data were analyzed using R package (http://www.R-project.org). Supervised analyses were performed using shrinkage test9 and multiplicity corrections were used to control FDR (false discovery rate); probes with local FDR lower than 0.05 are considered significant.
All MDS and AML specimens were collected at time of diagnosis and processed in the laboratory of Oncohematology (Padova, Italy). The patient's parents or their legal guardian provided written informed consent following the tenets of the Declaration of Helsinki.
MDS patients were distinguished in RCC (16 patients, 50%), RAEB (8 patients, 25%) and RAEB-t (8 patients, 25%). On the whole, 16 patients were classified as low risk (RCC) and 16 as high risk (RAEB+RAEB-t). In one specimen, morphological classification was RCC based on a low percentage of blasts (2%); however this patient was considered high risk due to flow cytometry immunophenotype that pointed to features typical of RAEB specimens.10
Cytogenetic data were available for 30 patients; 15 showed a normal karyotype, whereas either monosomy of chromosome 7 or del(7q) was detected in 10 patients (5 RCC, 2 RAEB and 3 RAEB-t); one patient (RCC) harbored a trisomy of chromosome 8; other patient-specific aberrations are reported in Supplementary Table 1. Even if some studies indicate monosomy 7 as a factor associated with evolution from low risk to high risk MDS or AML,11 its prognostic value remained debated.2 In view of this consideration, we decided to maintain RCC patients with monosomy 7 in the low-risk group. Indeed, outcome of patients (Supplementary Table 1) confirmed the good prognosis for RCC patients with monosomy 7, the only patient that presented evolution into AML having had a subsequent favorable outcome.
In our recent study on GEP-based classification of juvenile myelomonocytic leukemia patients, using a diagnostic classifier (DC) model algorithm developed during the MILE study,12 we recognized two clinically relevant patient groups with an AML-like and a non AML-like signature.8 When applying the same DC model to pediatric MDS specimens, we identified also for MDS patients two different signatures: 59.3% of samples received an AML-like call and 37.5% a non AML-like call. One patient received a tie between classes.
MDS patients’ blood smears with an AML-like signature were reexamined to exclude misclassification of AML and diagnosis of MDS was confirmed in all cases. Moreover, unsupervised hierarchical clustering analysis of the 32 MDS specimens together with 16 AML and 8 healthy BM samples distinguished the MDS and healthy BM specimens from AML (Figure 1a). Retrospective analysis revealed that the only MDS patient who clustered with the AML samples harbored a t(11;19) translocation typical of MLL-rearranged AML. At time of diagnosis, the identification of this translocation was not available for differential classification between MDS and AML.
Comparing clinical variables and classification based on gene expression, we observed a correlation between DC call (AML-like or non AML-like) and both evolution into AML (P=0.0001) and stratification by group of risk (P=0.0007) of MDS patients using χ2 (the P-value of chi-square was corrected with multiplicity correction using Holm's method). Other clinical variables such as specific karyotype aberrations, type of treatment received and overall survival were not correlated with DC signatures (Supplementary Table 2).
All MDS patients who evolved into AML showed an AML-like signature, indeed the AML-evolution curve exhibits a significant difference in probability of evolution to AML between the two groups of MDS patients. None of the MDS patients with a non AML-like signature showed evolution to AML; but the probability of time to evolution in AML for MDS patients with an AML-like signature was 73% (P=0.0004; Figure 1b). Using a log rank test to compare time to evolution in AML with GEP DC call, morphological classification and risk stratification DC call showed the most significant P-value (Supplementary Table 3). Four specimens that had been stratified as low risk were classified by the DC algorithm as AML-like; apparently the GEP-based DC classification recognized AML features even in these samples with a low blast count. Three of these patients evolved to AML, underscoring the power of the DC model to identify patient groups according to the probability of time to evolution to AML. This result also clearly indicates that GEP analysis using the DC model on whole BM samples can stratify pediatric MDS patients in clinically relevant subgroups, as previously also reported for adult MDS specimens.6
To identify gene pathways that characterize AML-like and non-AML-like signatures, a supervised gene expression analysis was performed (Supplementary Table 4). MDS patients with an AML-like signature showed overexpression of PROM1, MEIS1, WT1, GATA2, HOXB3, HOXA cluster genes, FLT3, ANGPT1, KIT, RUNX1 genes. Gene set enrichment analysis showed an enrichment of genes involved in AML pathways in MDS samples with the AML-like signature, suggesting a molecular predisposition to evolve in AML already at diagnosis.13, 14 Four genes (Supplementary Table 4) of the AML-like signature significantly predicted evolution to AML when we employed Prediction Analysis for Microarrays using the minimal gene number threshold. The sensitivity and specificity in predicting evolution into AML was 85.7% and 95%, respectively, with an overall accuracy of 91%. Remarkably, in non AML-like patients we identified, among the most upregulated genes, OLFM4, a pro-apoptotic gene involved in myeloid differentiation (Supplementary Table 4).15
In conclusion, in this study we demonstrated how GEP analysis applied to whole BM specimens of pediatric MDS could identify, already at diagnosis, patients with high risk to progress into AML. Application of GEP can contribute to further improving risk stratification of pediatric MDS patients and may be added to standard diagnostic protocols.
Gene Expression Omnibus
This work was supported by grants from Fondazione Città della Speranza, Padova, Italy (GB, MG), Associazione Italiano Ricerca sul Cancro (GteK, FI, GB), PRIN/programmi di ricerca rilevante interesse nazionale, Roma (GteK, GB). We thank Marta Campo Dell’ Orto for initial study design, Ken Mills for discussing DC model data of pediatric MDC specimens and Anna Leszl for providing cytogenetic data.
SB, GteK wrote the manuscript; SB, LT and MG performed the experiments and analyzed the data; SB designed the study and the experiments; MZ, LS collected patient data and coordinated morphological evaluation of MDS subtypes; FL, GB and GteK supervised the study; FL, GB revised the manuscript.
About this article
Supplementary Information accompanies the paper on the Leukemia website (http://www.nature.com/leu)