Genome‐wide <scp>DNA</scp> methylation profiling predicts relapse in childhood B‐cell acute lymphoblastic leukaemia

Although the five-year survival of childhood acute lymphoblastic leukaemia (ALL) exceeds 80%, a group of patients presents poor prognosis due to early relapse (van den Berg et al, 2011). To date, treatment strategies have been defined by cytogenetically-based subtype categorization. However, ALL patients without chromosomal translocations associated with poor prognosis lack diagnostic markers that would allow specific therapies to be developed. DNA methylation alteration is a frequent event in cancer and is potentially very useful in the diagnosis, prognosis and prediction of drug response (Rodríguez-Paredes & Esteller, 2011). Hence, we attempted to characterize childhood B-ALLs without Philadelphia (BCR-ABL1) and MLL translocations on the basis of the DNA methylation profile of more than 450 000 CpG sites with the aim of providing a means to improve the accuracy of prognosis and treatment strategies. All the obtained DNA methylation data have been deposited in the Gene Expression Omnibus (GEO) database in the following link: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=bfsbfcigsakcuty&acc=GSE39141 We derived genome-wide DNA methylation profiles of 29 childhood B-ALL patients and four normal B-cell samples (NBC) using the Infinium 450 K DNA methylation Bead assay (450 K) (Deneberg et al, 2011). Twenty-five patient samples were obtained at the time of diagnosis (including 10 cytogenetically normal, eight hyperdiploid, five pseudodiploid and two unclassified samples) and four samples at disease relapse (Table 1). Profiling varying 11 112 CpG sites (SD>0·25) within all samples analysed clearly distinguished healthy B-cell specimens from B-ALL patient samples (Fig 1A). Overall, B-ALL samples had a significantly greater variance in DNA methylation level than NBC samples (Wilcoxon test, P < 0·01) with 11% and 56% showing a standard deviation greater than 0·05% in healthy and ALL samples, respectively (Fig 1B). We further noted that most of the variation in ALL samples was a loss of DNA methylation located outside of the CpG rich (CpG island; CGI) context (62%, SD>0·05). However, the highest variance (SD≥0·25) was DNA hypermethylation found within the CGI context (Fig 1C, Fig S1), consistent with previous results obtained with lower coverage platforms (Milani et al, 2010). To obtain closer insight into the nature of variant sites, we determined differentially methylated CpG sites (dmCpGs) between healthy and cancer specimens (Table S1). We identified 3,414 dmCpGs, 88·2% (3,014) and 11·8% (400) of which respectively lost and gained DNA methylation in cancer samples (Fig 1D; Fig S2). Despite the predominantly hypomethylated dmCpGs, CGIs in gene promoters significantly gained DNA methylation at dmCpGs (Wilcoxon test; P < 0·01). Interestingly, the CGIs flanking CpG-poor regions (CpG island shores) were hypomethylated in ALL samples (Wilcoxon test; P < 0·01), consistent with previous studies identifying both regions as being highly variable in different cancer types, including leukaemia (Milani et al, 2010; Irizarry et al, 2009). Analysing variant CpG sites in an unsupervised manner, we identified two clearly distinct DNA methylome profiles in B-ALL patients (Fig 1A). While 14 samples (ALL-1) displayed highly aberrant methylation levels compared with the control, 15 samples (ALL-2) showed close similarities to healthy B-cells. Most strikingly, the five with disease-relapse-associated samples were all present in ALL-2 (5/15; X2 test, P < 0·01), presenting a signature with a significant association between DNA methylation and disease-free survival (log-rank Mantel-Cox test, P < 0·01; Fig 1E) and suggesting a possible application in future therapy strategies by taking into account epigenetically defined groups as previously suggested (Milani et al, 2010). Considering the presence of distinct DNA methylation subtypes in childhood B-ALLs and their potential application in clinical practice, we extracted a DNA methylation profile represented by 20 661 dmCpGs that distinguished the two groups (Table S2). In total, we detected 17 182 hypo- and 3 479 hypermethylated CpG sites in ALL-2 compared with ALL-1, respectively; differing in variance and associated with unique genomic features (Fig 1F-H). Confirming the signature in a 10-fold cross-validation model (area under the curve: 89·5), we concluded that the signature reliably detected both B-ALL subtypes and is thus an important tool for future disease diagnosis. In order to determine the affected biological and disease-associated pathways, we analysed the gene ontology (GO) of hyper- and hypomethylated CpG sites located in gene promoters. We noticed an enrichment (GO, level 5) of genes related to lymphocyte (including B-cell) differentiation in 672 gene promoters presenting higher methylation level in ALL-2 (false discovery rate [FDR]<0·05, Table S3). These genes include FOXP1, TCF3, BLNK, CD79A, RAG1 and RAG2, also associated with chromosomal translocations in ALLs and mutation in B-cell maturation-defective syndromes. We suggest that defects in the B-cell differentiation process display a unique property of samples present in ALL-2. The 2 608 genes associated with promoters showing less methylation in ALL-2 were highly enriched in developmental genes (FDR<0·05), including 97 HOX genes (Table S4). From the epigenetic point of view, developmental genes are associated with the Polycomb (Pc) complex and are marked by the histone modification H3K27me3. Interestingly, 43% (281 out of 654) of Polycomb target genes (PcTG) displayed lower methylation in ALL-2 gene promoters and were significantly enriched compared with promoters gaining methylation (X2 test, P < 0·01) (Lee et al, 2006). Accordingly, we found 24% (2181 out of 9052) of hypomethylated CpG sites associated with and significantly enriched in the Polycomb histone mark H3K27me3 (X2 test, P < 0·01, Fig S3) (Ernst et al, 2011). Taken together, the GO and PcTG analyses suggest that hypomethylation of genes involved in developmental processes are a unique feature of patient samples present in ALL-2. Hypermethylation of PcTG (Deneberg et al, 2011) has been previously identified as good prognostic markers in acute myeloid leukaemia patients, supporting the predictive potential of DNA methylation signatures in leukaemia. In addition, overexpression of HOX genes has been related to oncogenic transformation (Bach et al, 2010) and an increase in stem cell self-renewal (Ross et al, 2012) in leukaemia cells. Overall, genome-wide screening of DNA methylation in normal B-cells and primary B-ALL samples revealed distinct profiles, but most importantly defined two previously unknown B-ALL subtypes. Furthermore, we hypothesize that epigenetic changes mediate an undifferentiated stem cell-like phenotype of a newly identified B-ALL subtype that is possibly associated with drug resistance, resulting in disease relapse and presenting a signature of potential clinical use in the future. The work was supported by a Childhood Research Grant from the Asociación Española Contra el Cáncer (AECC) and the European Community's Seventh Framework Programme (FP7/2007-2013) under the HEALTH-F5-2011-282510 – BLUEPRINT. The funding sources had no role in the study design, collection, analysis and interpretation of data, or in the writing of the report. All authors performed the research, designed the research study, contributed essential reagents or tools and analysed the data. JS, HH and ME wrote the paper. The authors declare no conflicts of interest. Fig S1. DNA methylation level in different genomic features. Fig S2. Differentially methylated CpG site (dmCpG) between B-cell (NBC) and B-ALL (ALL) samples. Fig S3. Relative number of hyper- and hypomethylated differentially methylated sites between ALL-1 and ALL-2 considering their overlap with loci occupied by different histone marks, H2A.Z and CTCF. Table S1. 3,414 differentially methylated CpG sites comparing healthy B-cells and B-ALL samples. Table S2. 20,661differentially methylated CpG sites comparing B-cell group ALL-1 and ALL-2. Table S3. Gene ontology analysis of hypomethylated gene promoters in group ALL-2. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.