Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Whole-genome DNA methylation characteristics in pediatric precursor B cell acute lymphoblastic leukemia (BCP ALL)

  • Radosław Chaber,

    Roles Conceptualization, Methodology, Writing – original draft

    Affiliation Institute of Nursing and Health Sciences, Faculty of Medicine, University of Rzeszow, Rzeszow, Poland

  • Artur Gurgul,

    Roles Formal analysis, Methodology, Visualization, Writing – original draft

    Affiliation National Research Institute of Animal Production, Laboratory of Genomics, Balice, Poland

  • Grażyna Wróbel,

    Roles Methodology

    Affiliation Department of Paediatric Bone Marrow Transplantation, Oncology and Hematology, Medical University of Wroclaw, Wroclaw, Poland

  • Olga Haus,

    Roles Methodology, Validation, Writing – review & editing

    Affiliation Department of Clinical Genetics, Faculty of Medicine, Collegium Medicum in Bydgoszcz, Bydgoszcz, Nicolaus Copernicus University in Torun, Torun, Poland

  • Anna Tomoń,

    Roles Methodology

    Affiliation Institute of Nursing and Health Sciences, Faculty of Medicine, University of Rzeszow, Rzeszow, Poland

  • Jerzy Kowalczyk,

    Roles Methodology

    Affiliation Department of Pediatric, Hematology, Oncology and Bone Marrow Transplantation, Medical University of Lublin, Lublin, Poland

  • Tomasz Szmatoła,

    Roles Investigation, Methodology

    Affiliation National Research Institute of Animal Production, Laboratory of Genomics, Balice, Poland

  • Igor Jasielczuk,

    Roles Formal analysis, Investigation, Methodology

    Affiliation National Research Institute of Animal Production, Laboratory of Genomics, Balice, Poland

  • Blanka Rybka,

    Roles Methodology

    Affiliation Department of Paediatric Bone Marrow Transplantation, Oncology and Hematology, Medical University of Wroclaw, Wroclaw, Poland

  • Renata Ryczan-Krawczyk,

    Roles Methodology

    Affiliation Department of Paediatric Bone Marrow Transplantation, Oncology and Hematology, Medical University of Wroclaw, Wroclaw, Poland

  • Ewa Duszeńko,

    Roles Investigation

    Affiliation Department of Hematology, Medical University, Wroclaw, Poland

  • Sylwia Stąpor,

    Roles Methodology

    Affiliation Department of Hematology, Medical University, Wroclaw, Poland

  • Krzysztof Ciebiera,

    Roles Formal analysis, Investigation, Methodology, Writing – original draft

    Affiliation Institute of Informatics, University of Warsaw, Warsaw, Poland

  • Sylwia Paszek,

    Roles Investigation, Methodology, Writing – original draft

    Affiliation Centre for Innovative Research in Medical and Natural Sciences, Laboratory of Molecular Biology, Faculty of Medicine, University of Rzeszow, Rzeszow, Poland

  • Natalia Potocka,

    Roles Investigation, Methodology, Writing – original draft

    Affiliation Centre for Innovative Research in Medical and Natural Sciences, Laboratory of Molecular Biology, Faculty of Medicine, University of Rzeszow, Rzeszow, Poland

  • Christopher J. Arthur,

    Roles Conceptualization, Formal analysis, Methodology, Visualization

    Affiliation School of Chemistry, University of Bristol, Bristol, United Kingdom

  •  [ ... ],
  • Izabela Zawlik

    Roles Conceptualization, Methodology, Project administration, Supervision, Writing – original draft, Writing – review & editing

    Affiliations Centre for Innovative Research in Medical and Natural Sciences, Laboratory of Molecular Biology, Faculty of Medicine, University of Rzeszow, Rzeszow, Poland, Department of Genetics, Institution of Experimental and Clinical Medicine, University of Rzeszow, Rzeszow, Poland

  • [ view all ]
  • [ view less ]

Whole-genome DNA methylation characteristics in pediatric precursor B cell acute lymphoblastic leukemia (BCP ALL)

  • Radosław Chaber, 
  • Artur Gurgul, 
  • Grażyna Wróbel, 
  • Olga Haus, 
  • Anna Tomoń, 
  • Jerzy Kowalczyk, 
  • Tomasz Szmatoła, 
  • Igor Jasielczuk, 
  • Blanka Rybka, 
  • Renata Ryczan-Krawczyk


In addition to genetic alterations, epigenetic abnormalities have been shown to underlie the pathogenesis of acute lymphoblastic leukemia (ALL)—the most common pediatric cancer. The purpose of this study was to characterize the whole genome DNA methylation profile in children with precursor B-cell ALL (BCP ALL) and to compare this profile with methylation observed in normal bone marrow samples. Additional efforts were made to correlate the observed methylation patterns with selected clinical features. We assessed DNA methylation from bone marrow samples obtained from 38 children with BCP ALL at the time of diagnosis along with 4 samples of normal bone marrow cells as controls using Infinium MethylationEPIC BeadChip Array. Patients were diagnosed and stratified into prognosis groups according to the BFM ALL IC 2009 protocol. The analysis of differentially methylated sites across the genome as well as promoter methylation profiles allowed clear separation of the leukemic and control samples into two clusters. 86.6% of the promoter-associated differentially methylated sites were hypermethylated in BCP ALL. Seven sites were found to correlate with the BFM ALL IC 2009 high risk group. Amongst these, one was located within the gene body of the MBP gene and another was within the promoter region- PSMF1 gene. Differentially methylated sites that were significantly related with subsets of patients with ETV6-RUNX1 fusion and hyperdiploidy. The analyzed translocations and change of genes’ sequence context does not affect methylation and methylation seems not to be a mechanism for the regulation of expression of the resulting fusion genes.


Acute lymphoblastic leukemia (ALL) is the most common pediatric cancer with 3–5 cases per 100,000 with a peak of incidence 2–5 years [1]. Most of these cases (80–85%) originate from precursor B cells, BCP ALL [2]. The initiation of BCP ALL (as with other ALL’s) is driven by genetic alterations including point mutations, chromosome amplifications or translocations which ultimately lead to abnormal expression of key genes responsible for cell proliferation and differentiation [3]. Indeed, leukemic cells in ~75% of patients contain some chromosome abnormalities. Some of these abnormalities have prognostic significance. For example, genetic changes with good prognosis include hyperdiploidy with greater than 50 chromosomes or translocation t(12;21) ETV6-RUNX1 (TEL-AML1) which are detected in about a quarter of cases of childhood BCP ALL [4]. Hypodiploidy with fewer than 44 chromosomes and chromosomal rearrangements including translocation t(9;22) BCR-ABL1, rearrangement of MLL at 11q23 to a diverse range of fusion partners and internal amplification of chromosome 21 (iAMP 21) are less common (1–6% of cases) but are prognostic of a significantly higher risk of relapse and poor outcome. There are also chromosomal abnormalities as t(1;19)(q23;p13) or TCF3-PBX1 (formerly known as E2A-PBX1) (4% cases) with uncertain prognostic significance [4,5].

In addition to identifiable chromosomal abnormalities, many other recurring genetic alterations have been reported. The most prevalent submicroscopic alterations in pediatric ALL occur in CDKN2A/B (30–40%), IKZF1 (15%), PAX5 (20%), and ETV6 (10–15%). Most of these lesions have not been shown to impact prognosis. Mutations or deletions of IKZF1, located at 7p12, have, however, been shown to be independent predictors of outcome [58].

In addition to these genetic alterations, epigenetic abnormalities such as DNA methylation, post-translational histone protein modifications and interaction with non-coding RNA (miRNA or siRNA) have been shown to underlie the pathogenesis of ALL [9]. DNA methylation is a reversible process of attaching methyl residues to cytosine by the action of DNA methyltransferases—DNMT1, DNMT3a, DNMT3b. The methylation of cytosine’s located among CpG dinucleotides is significant for the control of gene expression. Such CpG sites may be dispersed amongst the whole genome but they frequently appear together in, so-called, CpG islands that are found within 70% of gene promoters [10] and are particularly common within the promoters of genes responsible for cell cycle control and cellular metabolism [11]. CpG islands are generally not methylated and methylation can inhibit gene expression both directly, through the prevention of transcription factors or RNA I polymerase binding, and indirectly by impacting on histone modification or chromatin remodeling proteins [12]. Aberrant methylation of CpG islands for the genes associated with DNA repair (hMLH1, MGMT), cell cycle control (p16INK4A, p15INK4B, p14ARF), apoptosis (DAPK) or detoxification (GSTP1) [11] can initiate carcinogenesis [11,13,14] including leukemogenesis [1417].

The purpose of this study was to describe and characterize the whole genome DNA methylation profile in children with BCP ALL and to compare this with methylation profile obtained in normal bone marrow samples. The secondary aim was to correlate the methylation landscape with some biological and clinical features as characteristic of patients, cytogenetic aberrations and prognosis based on treatment protocol criteria.

Material and methods


Samples of bone marrow were obtained from 38 patients with pediatric BCP ALL at the time of diagnosis. The characteristics of patients are shown in Table 1. Ethics Committee approval was obtained from the Institutional Review Board of the Medical University of Lodz (number, RNN/226/11/KE). Informed consent has been obtained from parents/legal guardians of all the participating children. Various genetic aberrations were detected among most of the patients. The patients were stratified into prognostic groups according to the BFM ALL IC 2009 protocol [18]. This stratification is based on the initial clinical features including patient age, white-blood cells count at diagnosis, presence of specific genetic aberrations, the response to steroids at day 8, the cytomorphological response in bone marrow at day 15 and 33 and the minimal residual disease level at day 15.

Samples of normal bone marrow cells were examined as controls. These bone marrow samples were obtained during routine diagnostic procedures of other diseases (sideropenic anemia, localized Wilms tumor and localized lipofibromatosis) from 4 patients (2 male, 2 female, age 5–17 years). In each case, the bone marrow as well as microscopic examination of peripheral blood smears revealed no pathology.

Samples and DNA methylation profiling

DNA was purified using QIAamp DNA Blood Mini Kit (QIAGEN), examined for integrity by agarose gel electrophoresis and quantified using Qubit 2.0 fluorimeter using a double stranded DNA (BR) assay (Thermo Fisher Scientific). About 500 ng of the sampled DNA was analyzed on a Infinium® MethylationEPIC BeadChip (Illumina, San Diego, CA). The analysis comprised bisulfite conversion of DNA with EZ DNA Methylation Kit (Zymo Research) using modified thermal conditions (as recommended by the supplier). Purified converted DNA was then used as an input to Infinium HD Assay Methylation Protocol (Illumina). The hybridized and stained arrays were ultimately scanned using HiScanSQ system (Illumina). The Infinium MethylationEPIC BeadChip used enabled the analysis of more than 850,000 methylation sites per sample which cover the broad content categories, including sites: within known CpG islands, outside of CpG islands, Non-CpG methylated sites identified in human stem cells (CHH sites), differentially methylated (DM) sites identified in tumor versus normal (multiple forms of cancer) and across several tissue types, FANTOM5 enhancers, ENCODE open chromatin and enhancers, DNase hypersensitivity sites and miRNA promoter regions. Moreover, the array content covers >90% of content contained on the previous Illumina HumanMethylation450 BeadChip.

Data quality control and analysis

The raw intensity data were assessed for quality using BeadArray Controls Reporter (Illumina). Next, the obtained IDAT files were analyzed using the Chip Analysis Methylation Pipeline (ChAMP) [19] for EPIC array data. During initial data handling, probes with p-value <0.01 and with fewer than 3 beads in at least 5% of samples were excluded. Additionally, non-CpG probes, SNP-related probes, multi-hit probes and probes located in chromosome X and Y were also removed. The beta values for 800,619 probes per sample were then calculated and checked for quality by evaluation of a beta multidimensional salting (MDS) plot and a density plot across the study groups. The beta value is defined as a proportion of DNA methylation at a particular CpG site (also called the methylation beta-value (β)) which is ascertained by taking the ratio of the methylated (C) to unmethylated (T) signals, using the formula: β = intensity of the methylated signal/(intensity of the unmethylated signal + intensity of the methylated signal + 100). A β-value of 0 represents an unmethylated CpG site and a β-value approaching 1 represents a fully methylated CpG site.

The obtained beta values were then normalized using the BMIQ method [20] and the singular value decomposition method (SVD) implemented by Teschendorff et al. (2009) [21] was used to identify the most significant causes of variation, including technical variation (batch effects like: slide, array or sample well). Batch effects (if detected for separate comparisons) were removed using the ComBat algorithm which uses an empirical Bayes method designed to correct data for technical variation [22].

Differential probes methylation between groups was calculated using the champ.DMP() function which uses the Limma package [23] to calculate the p-value for differential methylation using a linear model. The DMP detecting t-test p-values were corrected for multiple testing using the Benjamin-Hochberg procedure [24].

Differential methylation analysis was performed in three setups: to identify the general differences between all BCP ALL patients and the controls, second to identify sites/regions differentially methylated between SRG/IRG and HRG patients and third detecting methylation sites associated with recurrent genetic abnormalities. Additional pairwise comparisons have been made within BCP ALL patients group to detect sites which may be associated with confounding factors such as age (≤6 vs. >6 years of age) or gender (males vs. females). Differences in the distribution of CpG sites across different genomic classes (gene bodies, IGR, promoters, CpG islands etc.) were evaluated using Chi-square test.

Functional genes annotation and analysis

The genes associated with specific differentially methylated (DM) sites were analyzed in terms of molecular functions, biological presses, cellular components, pathways and phenotypes using WebGestalt (WEB-based GEne SeT AnaLysis) toolkit, exploiting an information obtained from GO, KEGG, WikiPathways, Human Phenotype Ontology and PharmGKB databases. Enrichment analysis was performed with respect to all known human genes (genome), identifying enriched categories with corrected p-value (according to Benjamin-Hochberg procedure) lower than 0.01 and requiring at least 4 genes per enriched category.


General assay performance and CpG statistics

The general assay performance was validated using control probes and no issues were found with the samples. After initial filtration, the beta values of 800,619 probes were normalized and batch effect was identified (slide, array and sample well) within the data by evaluation of components of variation. After normalization for technical variation, SVD indicated that the only source of variation amongst the sample groups was inter-group differentiation. Among the valid probes, representing individual CpG sites, 295,458 were located within genes (genes body), 228,710 in intergenic regions (IGR) and 67,841 in genes’ 5’UTR. Promoter associated sites, defined as: TSS200 (0–200 nucleotides (nt) upstream of the transcription start site) and TSS1500 (200–1500 nt upstream of TSS) were represented by 98,881 and 60,483 CpGs, respectively. 150,080 analyzed probes were located within known CpG islands, 142,693 in “shores” (up to 2000 bp from island) and 453,065 probes were located outside CpG islands (open sea).

Genome wide differential methylation analysis between BCP ALL and control

The DNA methylation status of 800,619 CpG loci distributed across the genome was interrogated in 38 pediatric BCP ALL patients and four non-leukemic control samples. A common set of 118,871 probes, mapped to 17,885 unique genes (promoters and gene bodies) or intergenic regions, were found to be differentially methylated (adjP<0.05; DM) in the patient’s relative to controls, suggesting the presence of distinct BCP ALL-associated DNA methylation signature (S1 Appendix). The absolute Δβ value for separate DM probes (calculated as an absolute difference between averaged β-values for separate groups) ranged from 0.009 to 0.730 with a mean of 0.216. Applying principal component analysis (PCA) to the differentially methylated site data shows clear separation of leukemic and control methylation profiles, with a higher level of profile variation observed amongst the BCP ALL samples (Fig 1). DM sites are uniformly distributed across the genome (Fig 2 as a Manhattan plot of -log10p-values for differentially methylated probes).

Fig 1. PCA based on 118,871 sites differentially methylated between BCP ALL and control samples.

Fig 2. Genomic distribution of CpGs with significant differences in methylation level between BCP ALL and control in gene context.

Of the 118,871 sites differentially methylated between leukemic and control samples, 67,115 (56%) were hypermethylated in BCP ALL samples and 51,756 (44%) were hypomethylated. The annotation analysis and comparison of hypo- and hypermethylated CpG features showed that genome-wide hypermethylation in leukemic DNA is more commonly associated with CpG islands and regions in close vicinity of transcription start site (TSS200 and 1st exon), whereas hypomethylation is more common in gene bodies and regions outside CpG islands (open-sea) (Fig 3, Table 2). The differences in distribution of hyper- and hypomethylated sites among separate genomic regions were statistically significant for all CpG context classes (as shown by Chi-square test; p<0.001).

Fig 3. Distribution of hyper- and hypomethylated sites differentially methylated between leukemic and control samples.

Table 2. Genomic distribution of probes differentially methylated between leukemic and control samples with respect to CpG island and gene context.

Differential methylation analysis of genes’ promoters in BCP ALL and control samples

To characterize changes in methylation profile of genes’ promoters accompanying leukemia progression, we focused on a differentially methylated CpGs (BCP ALL vs. CTR) located in regions upstream from transcription start sites (TSS200, TSS1500), within CpG islands or in their direct vicinity (island shore; ±2000 bp). DM sites were additionally filtered to remove those with absolute Δβ value between groups lower than 0.3. This resulted in selection of 5,465 DM CpGs of which 2,611 (47.8%) were in a range of 200 bp from TSS and the remaining 2,854 were in regions from 200 to 1500 bp from TSS. The majority of the sites (3,451; 63.1%) were positioned within CpG island cores and the remaining were in regions upstream or downstream of the islands (island shores; 2,014 CpGs). The average absolute Δβ value for the promoters-associated sites after filtration was 0.399 (±0.076).

The vast majority of the promoter-associated DM sites (4733; 86.6%) were hypermethylated in leukemia when compared to healthy control. Hypermethylated CpGs were located both in TSS200 (53.4%) and TSS1500 (46.6%) promoter regions and were mainly found within CpG islands (72.5%), whereas hypomethylated sites occurred predominantly in regions positioned further away from TSS (TSS1500; 88.7%) and outside of CpG islands (island shores; 97.5%).

Unsupervised hierarchical clustering (based on Euclidean distance) of the promoter methylation profiles showed clear separation of the leukemic and non-leukemic samples into two visible clusters (Fig 4) proving the existence of distinct methylation patterns amongst genomic promoters. Some variation was also observed within BCP ALL group with visible separation of three leukemic samples.

Fig 4. Unsupervised hierarchical clustering of promoter regions-associated methylation profiles in leukemic and control samples.

Identification of a leukemia-characteristic methylation profile

To find the most uniform and distinct from physiological methylation signature for leukemia, additional filtering criteria of all probes differentially methylated between BCP ALL and control were applied. The procedure aimed to identify CpGs with the lowest methylation level variation amongst BCP ALL samples (beta values standard deviation <0.05 within BCP ALL group) and with the largest difference in methylation level with respect to the control (delta beta absolute value >0.3). This filtration step allowed a set of 265 DM CpGs to be identified, with the most uniform and distinct from the methylation profile of the non-leukemic samples. As expected, unsupervised hierarchical clustering based on these probes enabled clear separation of the leukemic and non-leukemic methylation profiles and demonstrated high profile uniformity across BCP ALL samples (Fig 5) with minor discrepancies found for two leukemic samples diagnosed as BCP ALL with prodromal, preleukemic phase. This observation was also confirmed by PCA which showed that with the use of filtered probes set, low and lower than in control group level of methylation profile variation was observed in BCP ALL (Fig 6).

Fig 5. Unsupervised hierarchical clustering of methylation profiles with probes selected to minimize variation among leukemic samples.

Fig 6. PCA based on filtered probes set minimizing the variation of methylation profiles in leukemic samples.

The potentially leukemia-characteristic DM sites were distributed on all 22 autosomes and were predominately located in gene bodies (48.7%) or intergenic regions (26.8%) and most (83.0%) were distributed in regions outside of CpG islands (open sea). The hypermethylated sites (n = 177; 66.8%) showed similar regional distribution as hypomethylated with slightly increased share of probes located within gene bodies. The detailed statistics on the selected CpG are presented in S2 Appendix.

CpG associated with leukemia risk (SRG/IRG vs. HRG)

To find CpG sites which methylation level is associated with leukemia risk, the BCP ALL cases were stratified into two groups, encompassing HRG and combined SRG/IRG patients according to the BFM ALL IC 2009 protocol [18]. Statistical analysis showed, that of the 800,619 CpG analyzed sites, only 14 differed in methylation level between the groups. Most of these sites (n = 10; 71.4%) were hypomethylated in HRG. The average absolute Δβ for these sites was low with a mean of 0.232 (±0.058). Eight of the differentially methylated CpGs were located in intergenic regions and only two were found within promoter regions (TSS1500). Another three DM CpG were positioned within gene bodies. The list of DM sites between HRG and SRG/IRG is presented in S3 Appendix.

The most important CpG sites for prognosis were identified using a lasso penalized logistic regression [25]. This identified seven sites which are correlated with prognosis. Six of these relate to high-risk group stratification with poorer prognosis when they were hypomethylated and one if hypermethylated (Fig 7a). Among these selected CpG two were located within genes body or were promoter region-related (corresponding to MBP and PSMF1 gene) and five are found in intergenic regions.

Fig 7. Results of Ridge Regression of a high risk of a patient in the analyzed data set.

Each curve corresponds to a variable. It shows how much this variable contributes to the prediction of high risk of patient (a), hyperdiploidy (b) and t(12;21) ETV6-RUNX1 (c) aberrations in analyzed data set. Numbers on the left are coefficients and numbers on top are total count of variables selected for an L1 Norm (statistical parameter) shown on the bottom.

Age and gender-related methylation changes in leukemia patients

Age and gender-related changes in CpG methylation in leukemia patients were identified in two separate comparisons by the stratification of BCP ALL group according to sex and age (≤6 and >6 years of age). A MDS analysis based on 1000 the most variable sites did not show clear stratification of methylation profiles according to these two conditions (Fig 8).

Fig 8. Multidimensional scaling analysis based on 1000 the most variable sites with respect to: A—gender, B—age.

The identification of DM probes between separate subgroups allowed 113 sites (connected with 50 different genes) that differ in methylation depending on sex and 66 (36 genes) connected with age. Of the gender-related sites, 77 (68.1%) were hypermethylated in females with low level of methylation differences between sexes, expressed by a mean absolute Δβ of 0.143. The age-related changes were mostly linked with hypermethylation of CpGs, with almost 85% of probes showing higher methylation levels in older patients. The extent of age related methylation was higher than this observed for sex, with mean absolute delta beta of 0.329 (±0.073). Only three of the sex-related DM sites and seven of the age-related sites have been found among previously detected 118,871 CpGs differing in methylation levels between BCP ALL cases and controls, suggesting that age and sex are not important confounding factors for the methylation profile differentiation of pediatric leukemia.

Changes in methylation patterns associated with recurrent genetic abnormalities in childhood ALL

To find methylation profiles distinctive for genetic BCP ALL subtypes, patients were stratified into groups with ETV6-RUNX1 (n = 7), TCF3-PBX1 (n = 3), IGH (n = 3), hyperdiploidy (n = 13) and triploidy (n = 3) cytogenetic variants. Differentially methylated sites were identified in pairwise comparisons of specific genetic subtype with all other patients with known cytogenetic status, including eight patients without major chromosomal aberrations. Differential methylation analysis identified from 117 to 6,977 sites for the separate genetic subtypes. Of these, only a few (up to 67) overlapped between the separate groups. All analyzed subtypes were associated with hypomethylation with respect to the remaining patients with the highest proportion of hypomethylated sites found in IGH, hyperdiploidy and triploidy aberration carriers (from 72 to 77%) (Table 3). The distribution of DM sites across the genome was generally similar for separate subtypes, however, some evidence for an increased number of DM sites in the TSS1500 promoter regions in IGH carries have been found. The highest differences in methylation level compared to remaining BCP ALL cases, measured as average Δβ (0.384) were observed for TCF3-PBX1 genetic subtype (Table 3). The DM sites were related to 74 to 2963 unique genes depending on genetic abnormality and total number of detected DM sites. These genes are associated with a broad range of KEGG pathways, of which the top ten are presented in Table 4. The list of DM probes along with their annotation and associated genes is presented in S4 Appendix.

Table 3. Statistic of sites differentially methylated between specific genetic subtypes of leukemia and remaining BCP ALL patients with known cytogenetic status.

Table 4. Top ten KEGG pathways connected with genes containing sites differentially methylated between specific genetic subtype and remaining BCP ALL patients.

To determine the abilities of the sites that differed between separate genetic subtypes, from each comparison the top 100 DM sites (with the lowest adjusted p-value) were selected and a common panel of 500 sites was created with potential differentiation and diagnostic abilities. Unsupervised hierarchical clustering of the methylation level of selected probes allowed visible separation of ETV6-RUNX1 and TCF3-PBX1 profiles and classification of the remaining subtypes into mixed clusters with clear homogeneity (Fig 9). The three dimensional PCA plot allowed to clearly distinguish genetic leukemia subtypes associated with ETV6-RUNX1 and TCF3-PBX1 variants with visible separation of clusters including hyperdiploidy cases and the separation of a single case with IGH/triploidy aberrations (Fig 10). The list of probes used to distinguish the genetic BCP ALL subtypes are presented in S5 Appendix.

Fig 9. The unsupervised hierarchical clustering of the methylation level of a panel of 500 probes with the largest differences in methylation level between different leukemia genetic subtypes.

Fig 10. Principal component analysis 3D plot based on 500 probes with the highest differences in methylation level among genetic subtypes and remaining pre-B ALL patients.

The plot is presented in three different layouts to enable visualization of separate clusters.

An additional analysis was performed to explore the methylation profile of genes that are substrates of the most common chromosomal translocations found in the studied BCP ALL cases. The methylation profile of ETV6 and RUNX1 (TEL-AML1) genes was found to be variable across gene bodies, but showed visible demethylation within CpG islands located in promoter regions (both in BCP ALL and control patient groups) (Fig 11). Several sites across the ETV6 and RUNX1 genes differ significantly at a pointwise level between BCP ALL and the control group. Few of these however, differ between BCP ALL patients and BCP ALL cases with ETV6-RUNX1 aberration. Within the RUNX1 gene, demethylation was observed within two alternative gene promoters with clear hypermethylation of the gene body. The most pronounced differences in methylation level between BCP ALL and the control group were observed within the CpG island located in the central part of the RUNX1 gene, which is an alternative gene promoter (P2). This region, however, did not show noticeable differences in methylation level between BCP ALL and BCP ALL-ETV6-RUNX1 carriers.

Fig 11. Methylation profile of ETV6 and RUNX1 genes in ETV6-RUNX1 subtype patients, remaining pre-B ALL cases and control individuals.

Red square marks the CpG fragments with large differences in methylation level between pre-B ALL and control samples.

A similar methylation pattern (demethylation of promoters and visible hypermethylation of gene bodies) was found for the TCF3 and PBX1 (E2A-PBX1) genes involved in t(1;19)(q23;p13) chromosomal rearrangement. The methylation profile of the genes was similar between BCP ALL, control and TCF3-PBX1 rearrangement carriers, except the north-shore part of the CpG island located near the transcription start site of the TCF3 gene shows hypermethylation in control group with respect to the BCP ALL patients (genome-wide adjusted p<0.05) (Fig 12).

Fig 12. Methylation profile of TCF3 and PBX1 genes in TCF3-PBX1 subtype patients, remaining pre-B ALL cases and control individuals.

Red square marks the CpG fragments with large differences in methylation level between pre-B ALL and control samples.

It was possible using the lasso penalized logistic regression analysis to extract some DM sites related significantly with subsets of patients with ETV6-RUNX1 and hyperdiploidy (Fig 7b and 7c). Among these sites, some were localized with promoter regions or gene bodies. The most important functional DM sites in particular subgroups are summarized in Table 5. These DM sites can potentially distinguish the groups with analyzed chromosomal aberrations after being confirmed in future studies with a larger patient cohort. The other cytogenetic aberrations were not analyzed using this method due to an insufficient number of affected patients.

Table 5. The most important DM sites in patients from particular BCP ALL subgroups extracted using the lasso penalized logistic regression analysis.


A distinct DNA methylation pattern of leukemic samples

The normal hematopoietic tissue in bone marrow consists of a variety of cell types including the blood cells and their precursors at different developmental stages derived from pluripotential, self-renewing stem cells and the microenvironment’s cells like adventitial reticular cells, endothelial cells, macrophages, adipocytes, bone lining cells (e.g., osteoblasts). During blood-cell development, pluripotent stem cells undergo either self-renewal or differentiation into multilineage committed progenitor cells: myeloid stem cells which can then differentiate into erythroid and myeloid (including granulocytes, megakaryocytes and macrophages) or lymphoid cell line. The neoplastic transformation of early lymphoblastic progenitor cells in BCP ALL dramatically changes their phenotype resulting in rapid proliferation and expansion. These abnormal lymphoblasts quickly dominate in bone marrow cavities; comprising up to 100% of the bone marrow cells. Due to this, the population of bone marrow cells obtained from patients with BCP ALL is practically homogenous compared to normal samples, where many cells of different origin and of different developmental stage occur. In this second, instance the poorly differentiated lymphocytic and myelocytic progenitors consist of up to 5% bone marrow cellularity. Consequently, the differences of methylation profiles between BCP ALL and normal bone marrow controls observed here-in can be explained by the heterogeneity of cells types in normal bone marrow, epigenetic control of the natural progenitor’s development and with methylation changes dependent with neoplastic transformation in leukemic samples.

The clear separation of DNA methylation profile between leukemic and normal bone marrow cells agrees with previous results [2631]. Abnormal DNA methylation has been confirmed in many cancers including hematological neoplasms [32,33]. The role of DNA methylation in hematopoietic regulation has also been established [34]. For example, DNA methylation within bone marrow stem cells is important for their self-renewal and methylation deficiency leads to their differentiation into myeloerythroid progenitors [35]. Epigenetic control also plays a key role in lymphoid B-cells development as shown by Almamun et al. [36].

The differences in DNA methylation pattern between leukemic and normal bone marrow samples in our groups arise from many sites– 118,871 probes mapped to 17,885 unique genes or intergenic regions. Among them, a slight domination of hypermethylated over hypomethylated loci (56:44) in BCP ALL samples was revealed. Moreover, these DM sites are more commonly associated with CpG islands and in close vicinity of transcription start sites. Interestingly, the promoter-associated sites with different methylation were mainly (in 86,6% cases) hypermethylated in leukemia when compared to healthy controls. This may result in altered expression of many genes. The prevalence of differential hypermethylated promoter-associated CpG islands compare to non-diseased samples was detected also in other studies [26,28,30].

Functional analysis of differential methylated genes (DMGs)

CpGs with distinct leukemic methylation were associated with 184 different genes (mainly gene bodies). Those genes are enriched in several GO biological processes, of which the most interesting appears to be: leukocyte activation (14 genes; adjP = 0.0097) or immune system process (32 genes; adjP = 0.0068). The genes were also over-represented in six different KEGG pathways, including the “pathways in cancer” term (6 genes; adjP = 0.0043). When the WikiPathways terms were interrogated, the most enriched was associated with the integrated pancreatic cancer pathway (5 genes; adjP = 0.0011).

The most meaningful results were however obtained, when terms from the PharmGKB database of disease phenotypes were analyzed. Several consistent disease phenotypes have been enriched by the analyzed genes, including: lymphoma, B-cell lymphoma, lymphoproliferative disorders, lymphatic diseases, lymphoid leukemia and leukemia (Table 6; S6 Appendix).

Table 6. Selected disease phenotypes enriched by genes associated with uniform leukemia methylation pattern.

When 124 genes related with hypermethylated CpGs were screened separately for their functions, only a single KEGG pathway was found to be significant and is related to protein processing in endoplasmic reticulum (4 genes; adjP = 0.0030). The hypermethylated genes were also enriched in several disease phenotypes with one which could be directly connected with cancer, namely: hepatocellular carcinoma phenotype (4 genes; adjP = 0.0075).

When the remaining 60 genes associated with hypomethylated sites were analyzed separately, no enriched biological processes, molecular functions or pathways were found. However, all previously mentioned lymphoma/leukemia associated disease phenotypes were enriched by this gene category.

The analyzed promoter regions were associated with 2045 different genes. The annotation analysis showed, that hypermethylated probes were found within putative promoter regions of 1540 genes and hypomethylated sites were associated with 518 genes. When the genes were analyzed jointly in terms of their functional relevance, several general biological processes or processes connected with nervous system functioning and development have been detected, including e.g.: nervous system development, single-multicellular organism process, neuron differentiation, neurogenesis, cell-cell signaling and many others. Among the top ten significantly enriched KEGG pathways there were those associated, for example, with neuroactive ligand-receptor interaction (68 genes; adjP = 4.21e-30), calcium signaling (36 genes; adjP = 6.73e-13) or pathways in cancer (34 genes; adjP = 6.29e-05). Among the top ten enriched disease phenotypes, no leukemia-related disorders were found. When genes with hyper- and hypomethylated promoters were analyzed separately, however, we find that the hypermethylated genes are enriched in disease phenotypes associated with nervous system and mental disorders, and that the hypomethylated genes are enriched several disease phenotypes, like e.g.: lymphoproliferative disorders, lymphoid leukemia, B-Cell lymphoma, Lymphoid leukemia NOS, lymphatic diseases, general leukemia or B-Cell leukemia (Table 6 and S7 Appendix).

The results of other studies have revealed that the spectrum of genes with potentially aberrated methylation in BCP ALL is very wide and includes many biological processes. These processes include, among others, growth and proliferation’s regulation [26,2931], apoptosis [26,29,30,37] hematopoiesis and lymphocyte B development [27,28,36] and immune response [38]. It must be noted that alteration of promoter DNA methylation does not have to translate to major changes in gene expression in any case [26].

The impact of methylation signature on prognosis in BCP ALL

The discovery of new molecular prognostic factors is important for further therapy individualization and optimization. The meaningful differences in genes methylation between HR and IR/SR BCP ALL patients were depended with both: promoters as well as gene bodies regions. Promoter region-related sites were connected with PSMF1 and DIP2B genes, whereas differential methylation of CpGs within gene bodies was found in NGFR, ADAMTS20 and MBP genes. Although there was not enough data for conclusive results we also tried to extract the differences between methylation of single/multiple genes in patients stratified by risk group. We observe only minimal differences between count of DM CpG sites in HRG and IRG/SRG patients but it was possible to distinguish seven DM sites which could be the biomarkers of the HR group. The risk prediction based on these 7 DM sites might be determined early, before therapy response assessment. Two of these 7 sites, are localized in gene body/promoter regions, and thus may have potential functional roles in regulating the expression of MBP and PSMF1 genes. [Table 5]. The protein encoded by the classic MBP gene is a major constituent of the myelin sheath of oligodendrocytes and Schwann cells in the nervous system. However, MBP-related transcripts are also present in the bone marrow and the immune system [39]. The MBP has also an important role to play in apoptosis [40]. It has been shown that MBP SNP rs3794845 is significantly associated with childhood ALL risk [41]. PSMF1 gene encodes a protein that inhibits the activation of the proteasome by the 11S and 19S regulators (provided by apoptotic and cell cycle mechanisms). Proteasome inhibitors (for example bortezomib) enhance many conventional therapies, also in ALL [42]. Although there are some limitations including small sample size and a lack of genes’ expression information supporting the aberrated DM sites methylation, our study identified two genes with potential impact on childhood ALL prognosis.

Global promoter methylation profiling has been previously demonstrated to all subgrouping and prognostication of pediatric ALL in T cells phenotype ALL [43,44]. Takeuchi et al [45] revealed that patients from medium/high risk group had multiple genes methylated compared to those from low risk but they included patients with both, B- and T-cells ALL together. No specific methylation profile characteristic for HRG was observed in our study. The relationship between methylation and outcome prognosis or relapse risk in pediatric BCP ALL is unproven. Some authors [28,30,46] have identified some genes with aberrant methylation relevant to relapse prediction but others [38] reached opposing conclusions and didn’t find any statistically significant associations between methylation level of any CpGs and subsequent relapse of BCP ALL.

Epigenetic landscape and cytogenetic subtypes in childhood BCP ALL

Our study confirmed the distinguishable differences in the general methylation pattern between the analyzed cytogenetic subsets of patients. It was possible to clearly separate genetic leukemia subtypes associated with ETV6-RUNX1 and TCF3-PBX1 variants with visible separation of clusters including hyperdiploidy cases and separation of the single case with IGH/triploidy aberrations. These findings are consistent with observations from other studies [30,38,47] where the cytogenetic subtypes showed a clear correlation with methylation profile but this clustering was not absolute.

The lack of clear differences in methylation level of genes involved in chromosomal translocations between BCP ALL patients with- or without specific rearrangement, suggest that the translocation and change of genes’ sequence context does not affect methylation and that methylation seems not to be a mechanism for the regulation of expression of the resulting fusion genes. The methylation profile of some of these genes, however, show signs of alteration between healthy and leukemia samples.

Methylation sites with potential significant different methylation status specifically depended on analyzed cytogenetic status. Only a few of these sites are in gene bodies or promoter regions. Their functional role stays unknown but they may be potentially used as biomarker of ALL patients with hyperdiploidy or t(12;21) and should be confirmed in future studies with a larger cohort of patients. Genes specifically correlated with analyzed cytogenetic subsets (listed in Table 5) may be suitable targets for the searching new individualized BCP ALL therapy.


The clear separation of DNA methylation profile between leukemic and normal bone marrow cells was confirmed. The analyzed translocations and change of genes’ sequence context does not affect methylation thus methylation seems not to be a mechanism for the regulation of expression of the resulting fusion genes. Moreover, it was possible to distinguish some DM sites which could be the biomarkers of the BFM ALL IC 2009 high risk group as well as hyperdiploidy and t(12;21) ETV6-RUNX1 carriers.

Supporting information

S1 Appendix. A set of 118,871 probes differentially methylated (adjP<0.05) in all ALL_B patients relative to controls.


S2 Appendix. CpG sites with low variation among ALL_B group and differentially methylated with respect to control.


S3 Appendix. A list of CpG sites differentially methylated between HRG and SRG/IRG leukemia patients.


S4 Appendix. The list of probes differentially methylated between genetic subtypes and remaining BCP ALL patients, along with their annotation and associated genes.


S5 Appendix. The list of 500 probes with the highest differences in methylation level among genetic subtypes and remaining BCP ALL patients.


S6 Appendix. Disease phenotypes enriched by genes associated with uniform CpG methylation profile in leukemia patients.


S7 Appendix. Disease phenotypes enriched by genes with hypomethylated CpG sites in leukemia samples across genome (promoters and gene bodies).



  1. 1. Ward E, DeSantis C, Robbins A, Kohler B, Jemal A. Childhood and adolescent cancer statistics. CA. Cancer J. Clin. 2014;64: 83–103. pmid:24488779
  2. 2. Silverman Lewis B. Acute lymphoblastic leukemia. In: Orkin SH, Fisher DE, Look AT, Lux S, Ginsburg D, Nathan D, editors. Oncology of Infancy and Childhood. 1st ed. Philadelphia: Saunders;2009. p. 297–330.
  3. 3. Iacobucci I, Mullighan CG. Genetic Basis of Acute Lymphoblastic Leukemia. J Clin Oncol. 2017 Mar 20;35(9):975–983. pmid:28297628
  4. 4. Kanwar VS, Satake N, Yoon JM, Reikes Willert J. Pediatric Acute Lymphoblastic Leukemia. 2017.
  5. 5. Teachey DT, Hunger SP. Predicting relapse risk in childhood acute lymphoblastic leukaemia. Br J Haematol. 2013; 162(5):606–20. pmid:23808872
  6. 6. Hunger SP, Raetz EA, Loh ML, Mullighan CG. Improving outcomes for high-risk ALL: translating new discoveries into clinical care. Pediatric blood & cancer. 2011;56: 984–993.
  7. 7. Moorman AV. The clinical relevance of chromosomal and genomic abnormalities in B-cell precursor acute lymphoblastic leukaemia. Blood reviews. 2012;26: 123–135. pmid:22436535
  8. 8. Mullighan CG. Molecular genetics of B-precursor acute lymphoblastic leukemia. The Journal of Clinical Investigation. 2012;122: 3407–3415. pmid:23023711
  9. 9. Dawson MA, Kouzarides T. Cancer epigenetics: from mechanism to therapy. Cell. 2012;6; 150(1):12–27. pmid:22770212
  10. 10. Saxonov S, Berg P, Brutlag DL. A genome-wide analysis of CpG dinucleotides in the human genome distinguishes two distinct classes of promoters. Proc Natl Acad Sci. 2006;103: 1412–1417. pmid:16432200
  11. 11. Esteller M. CpG island hypermethylation and tumor suppressor genes: a booming present, a brighter future. Oncogene. 2002; 21: 5427–40. pmid:12154405
  12. 12. Bogdanovic O, Veenstra GJ. DNA methylation and methyl-CpG binding proteins: developmental requirements and function. Chromosoma. 2009;118: 549–65. pmid:19506892
  13. 13. Qu Y, Dang S, Hou P. Gene methylation in gastric cancer. Clin Chim Acta. 2013; 424: 53–65. pmid:23669186
  14. 14. Schulze I, Rohde C, Scheller-Wendorff M, Bäumer N, Krause A, Herbst F, et al. Increased DNA methylation of Dnmt3b targets impairs leukemogenesis. Blood. 2016 Mar 24;127(12): 1575–86. pmid:26729896
  15. 15. Bergmann AK, Castellano G, Alten J, Ammerpohl O, Kolarova J, Nordlund J, et al. DNA methylation profiling of pediatric B-cell lymphoblastic leukemia with KMT2A rearrangement identifies hypomethylation at enhancer sites. Pediatr Blood Cancer. 2017 Mar;64(3). pmid:27786413
  16. 16. Florean C, Schnekenburger M, Grandjenette C, Dicato M, Diederich M. Epigenomics of leukemia: from mechanisms to therapeutic applications. Epigenomics. 2011 Oct;3(5): 581–609. pmid:22126248
  17. 17. Nordlund J, Bäcklin CL, Wahlberg P, Busche S, Berglund EC, Eloranta ML, et al. Genome-wide signatures of differential DNA methylation in pediatric acute lymphoblastic leukemia. Genome Biol. 2013 Sep 24;14(9): r105. pmid:24063430
  18. 18. Campbell M, Castillo L, Riccheri C, Janic D, Jazbec J, Kaiserova E, et al. A Randomized Trial of the I-BFM-SG for the Management of Childhood non-B Acute Lymphoblastic Leukemia Final Version of Therapy Protocol from August-14-2009. ALL IC-BFM 20/04/2017
  19. 19. Morris TJ, Butcher LM, Teschendorff AE, Chakravarthy AR, Wojdacz TK, Beck S. “ChAMP: 450k Chip Analysis Methylation Pipeline.” Bioinformatics. 2014;30(3): 428–430. pmid:24336642
  20. 20. Teschendorff AE, Marabita F, Lechner M, Bartlett T, Tegner J, Gomez-Cabrero D, et al. A beta-mixture quantile normalization method for correcting probe design bias in illumina infinium 450 k dna methylation data. Bioinformatics. 2013;29(2): 189–196. pmid:23175756
  21. 21. Teschendorff AE, Menon U, Gentry-Maharaj A, Ramus SJ, Gayther SA, Apostolidou S, et al. An epigenetic signature in peripheral blood predicts active ovarian cancer. PLoS One. 2009;4(12): e8274. pmid:20019873
  22. 22. Johnson WE, Li C, Rabinovic A. Adjusting batch effects in microarray expression data using empirical bayes methods. Biostatistics. 2007;8(1): 118–127. pmid:16632515
  23. 23. Smyth GK. Limma: linear models for microarray data. Bioinformatics and Computational Biology Solutions Using R and Bioconductor,. Springer, New York, NY. 2005; pp. 397–420.
  24. 24. Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society. Series B (Methodological). 1995;Vol. 57, No. 1: 289–300.
  25. 25. Tibshirani R. Regression shrinkage and selection via the lasso. J. Royal. Statist. Soc B. 1996; Vol. 58, No. 1: 267–288.
  26. 26. Chatterton Z, Morenos L, Mechinaud F, Ashley DM, Craig JM, Sexton-Oates A, et al. Epigenetic deregulation in pediatric acute lymphoblastic leukemia. Epigenetics. 2014 Mar;9(3):459–67. pmid:24394348
  27. 27. Wong NC, Ashley D, Chatterton Z, Parkinson-Bates M, Ng HK, Halemba MS, et. al. A distinct DNA methylation signature defines pediatric pre-B cell acute lymphoblastic leukemia. Epigenetics. 2012 Jun 1;7(6):535–41. pmid:22531296
  28. 28. Nordlund J, Bäcklin CL, Wahlberg P, Busche S, Berglund EC, Eloranta ML, et. al. Genome-wide signatures of differentia DNA methylation in pediatric acute lymphoblastic leukemia. Genome Biol. 2013 Sep 24;14(9): r105. pmid:24063430
  29. 29. Nordlund J, Milani L, Lundmark A, Lönnerholm G, Syvänen AC. DNA methylation analysis of bone marrow cells at diagnosis of acute lymphoblastic leukemia and at remission. PLoS One. 2012;7(4): e34513. pmid:22493696
  30. 30. Milani L, Lundmark A, Kiialainen A, Nordlund J, Flaegstad T, Forestier E, et al. DNA methylation for subtype classification and prediction of treatment outcome in patients with childhood acute lymphoblastic leukemia. Blood. 2010 Feb 11;115(6): 1214–25. pmid:19965625
  31. 31. Wang MX, Wang HY, Zhao X, Srilatha N, Zheng D, Shi H, et al. Molecular detection of B-cell neoplasms by specific DNA methylation biomarkers. Int J Clin Exp Pathol. 2010 Jan 28;3(3): 265–79. pmid:20224725
  32. 32. Boultwood J, Wainscoat JS. Gene silencing by DNA methylation in haematological malignancies. Br J Haematol. 2007 Jul;138(1): 3–11. pmid:17489980
  33. 33. Vadakedath S, Kandi V. DNA Methylation and Its Effect on Various Cancers: An Overview. J Mol Biomark Diagn. 2016;S2: 017.
  34. 34. Ji H, Ehrlich LI, Seita J, Murakami P, Doi A, Lindau P, et al. Comprehensive methylome map of lineage commitment from haematopoietic progenitors. Nature. 2010 Sep 16;467(7313): 338–42. pmid:20720541
  35. 35. Bröske AM, Vockentanz L, Kharazi S, Huska MR, Mancini E, Scheller M, et al. DNA methylation protects hematopoietic stem cell multipotency from myeloerythroid restriction. Nat Genet. 2009 Nov;41(11): 1207–15. pmid:19801979
  36. 36. Almamun M, Levinson BT, Gater ST, Schnabel RD, Arthur GL, Davis JW, et al. Genome-wide DNA methylation analysis in precursor B-cells. Epigenetics. 2014 Dec;9(12): 1588–95. 4161/15592294.2014.983379. pmid:25484143
  37. 37. Vilas-Zornoza A, Agirre X, Martín-Palanco V, Martín-Subero JI, San José-Eneriz E, Garate L, et al. Frequent and simultaneous epigenetic inactivation of TP53 pathway genes in acute lymphoblastic leukemia. PLoS One. 2011 Feb 28;6(2): e17012. pmid:21386967
  38. 38. Gabriel AS, Lafta FM, Schwalbe EC, Nakjang S, Cockell SJ, Iliasova A, et al. Epigenetic landscape correlates with genetic subtype but does not predict outcome in childhood acute lymphoblastic leukemia. Epigenetics. 2015;10(8): 717–26. pmid:26237075
  39. 39. [Internet].
  40. 40. Lu ML, Sato M, Cao B, Richie JP. UV irradiation-induced apoptosis leads to activation of a 36-kDa myelin basic protein kinase in HL-60 cells. Proc Natl Acad Sci USA. 1996;93: 8977–8982. pmid:8799139
  41. 41. Han S, Lan Q, Park AK, Lee KM, Park SK, Ahn HS, et al. Polymorphisms In innate immunity genes and risk of childhood leukemia. Hum Immunol. 2010 Jul;71(7): 727–30. pmid:20438785
  42. 42. McCloskey SM, McMullin MF, Walker B, Irvine AE. The therapeutic potential of the proteasome in leukemia. Hematol Oncol 2008; 26: 73–81. pmid:18324639
  43. 43. Borssén M, Haider Z, Landfors M, Norén-Nyström U, Schmiegelow K, Åsberg AE, et al. DNA Methylation Adds Prognostic Value to Minimal Residual Disease Status in Pediatric T-Cell Acute Lymphoblastic Leukemia. Pediatr Blood Cancer. 2016 Jul;63(7): 1185–92. pmid:26928953
  44. 44. Borssén M, Palmqvist L, Karrman K, Abrahamsson J, Behrendtz M, Heldrup J, et al. Promoter DNA methylation pattern identifies prognostic subgroups in childhood T-cell acute lymphoblastic leukemia. PLoS One. 2013 Jun 6;8(6): e65373. pmid:23762353
  45. 45. Takeuchi S, Matsushita M, Zimmermann M, Ikezoe T, Komatsu N, Seriu T, et al. Clinical significance of aberrant DNA methylation in childhood acute lymphoblastic leukemia. Leuk Res. 2011 Oct;35(10): 1345–9. pmid:21592569
  46. 46. Roman-Gomez J, Jimenez-Velasco A, Agirre X, Castillejo JA, Navarro G, Garate L, et al. Promoter hypermethylation and global hypomethylation are independent epigenetic events in lymphoid leukemogenesis with opposing effects on clinical outcome. Leukemia. 2006; 20: 1445–1448. pmid:16688225
  47. 47. Figueroa ME, Chen SC, Andersson AK, Phillips LA, Li Y, Sotzen J, et al. Integrated genetic and epigenetic analysis of childhood acute lymphoblastic leukemia. J Clin Invest. 2013 Jul;123(7): 3099–111. pmid:23921123