Hypermethylation and down-regulation of DLEU2 in paediatric acute myeloid leukaemia independent of embedded tumour suppressor miR-15a/16-1

The FAB subtype M5 (monocytic/blastic leukaemia) is a distinct subtype with characteristic chromosomal abnormalities including t(8; 16), +8 and various translocations involving 11q23 and the MLL locus such as t(9;11), t(10;11), t(11;19) and others[27]. AML subtype M5 also has a high proportion of cytogenetically normal (CN-AML) patients[27], and those with complex karyotypes[28]. A combination of these factors add to the overall unfavourable outcome of paediatric M5 diagnosis[29]. Genome-scale methylation profiling of AML M5 bone marrow samples identified 3,352 significantly differentially methylated probes (DMPs) between paediatric AML FAB M5 (n = 20) and matching non-leukaemic (n = 17) samples. Applying more stringent feature selection criteria of an adjusted p-value <0.01 and ∆β of >0.4 reduced the number of DMPs to 137 (Additional file1).

The list of DMPs included several localising to the long non-coding RNA DLEU2 and the embedded miRNA cluster miR-15a/16-1[19, 20], previously implicated in adult leukaemic[19, 20, 30]. To date, disruption of this region has not been observed in paediatric cancers and as such we chose to focus on DLEU2 in subsequent analysis. A total of three DLEU2 DMPs had an adjusted p-value <0.01 and ∆β of >0.4 (cg05394800, cg20529344, cg12883980). These probes were located within three CpG islands (Chr 13: 50, 690,000-50,708,000: UCSC human hg19 assembly) at the DLEU2/Alt1 transcriptional start site (TSS) and ‘north shore’ (Figure 1), a region up to 2 kb upstream from the DLEU2/Alt1 TSS CpG island under investigation[31]. Henceforth this will be referred to as the DLEU2 promoter.

A divergent DNA methylation profile was observed between all cases of paediatric AML with elevated methylation (mean 64% (48-80% CI) across significant probes and non-leukaemic samples (23% methylation; 7-39% CI of mean) (p < 0.001 for all significant probes) (Figure 1; Additional file2). DNA methylation levels at CpG sites interrogated by cg05394800, cg20529344, cg12883980 were confirmed using SEQUENOM EpiTYPER MassArray (Additional file3 and Additional file4) in the discovery sample set of 20 leukaemic patients and 17 non-leukaemic controls, as well as a validation set of paediatric AML M5 patients with heterogeneous cytogenetic diagnoses (n = 19) (CN-AML (n = 5), MLL (n = 7), RUNX1 (n = 2), WT1 (n = 1)) (Additional file5). This analysis confirmed specific DLEU2 promoter hypermethylation in paediatric AML patients (p < 0.001) (Additional file4), with additional genes in this region, including DLEU1, hsa-miR-15a/16-1 microRNA cluster, TRIM13, KCNRG and hsa-miR-3613 showing no differences in methylation between cases and controls (Additional file6). This highlights the potential regional specificity of the observed DLEU2 methylation change in association with AML.

We observed a significant down-regulation of DLEU2 gene expression in paediatric AML (0.07 Fold Change (FC); p = 0.014), and a significant inverse correlation between promoter DNA methylation and gene expression levels (p = 0.0001, Additional file7). Recent studies in adult CLL have also identified a negative correlation between DLEU2 promoter methylation (DLEU2/Alt1) and gene expression[32]. Interestingly there was no change in gene expression for any other genes in this region (TRIM13 = 1.75 FC; DLEU1 = 1.05 FC. Figure 2A).

The miR-15a/16-1 cluster has been described to have potent tumour suppressor activity, targeting numerous oncogenic and cell cycle regulatory genes[19, 33]. The cluster is embedded within intron 4 of DLEU2, and has been speculated that expression is driven by the DLEU2 promoter[20, 30]. However, we found no correlation between DLEU2 expression and miR-15a/16-1 expression in paediatric AML, nor down-regulation of the miR-15a/16-1 miRNA cluster in relation to increasing DLEU2 promoter DNA methylation (Figure 2B). In contrast to previous reports for adult leukaemia, no significant change in mature miR-16 expression was observed between paediatric AML and control samples, a result we have reported elsewhere[34]. These observations are independent of the homologous miR-16 cluster embedded in SMC4 on chromosome 3q26, which shows no significant expression or DNA methylation changes in association with AML (Additional file8). Similar results have been reported for adult CLL[32]. Taken together this data suggests an alternate mode of regulation for the miR-15a/16-1 cluster, outside of the DLEU2 promoter region.

Interestingly, recent research has indicated the processing mechanisms of miRNA may be affected by cancer, such that mature miRNA expression becomes disassociated from precursor miRNA levels, and also from the levels of the host gene[35]. We found the miR-15a/16-1 primary precursor transcript is in fact expressed up to three-fold higher in AML patients compared to non-leukaemic counterparts (Figure 2B). The increase in primary transcript appears to correspond to an increase in mature miR-16-1* (2.52 FC) and miR-15a* (2.2 FC) expression, and a moderate increase in miR-15a (1.5 FC). However, individual patient samples show differential degrees of over-expression of one, or all, of the mature species despite the common precursor (Additional file9).

Canonical miR-15a/16-1 microRNA species target numerous oncogenic and integral cell cycle regulatory genes such as BCL-2, MCL-1, CCND1, CDK6, BMI-1, RASSF5, IGSF4, c-MYB and WNT3A[33, 36‐42]. To identify the potential significance of increasing alternate miR-15a/16-1 transcripts over their canonical counterparts in paediatric AML, we investigated miRNA target genes using prediction tools, and undertook gene ontology analysis. We found miR-16-1* is three-fold enriched for targeting components of the RNA processing and splicing machinery such as hnRNP regulators, SRRM1 and NUDT2. Down-regulation of these genes by increased expression of a targeting miRNA may be contributing to miRNA/host gene disassociation in DLEU2/miRNA-15a/16-1 expression (Additional file10). One target of miR-15a* was found to be ADORA2A, an important G-protein receptor critical in tissue-specific and systemic inflammatory responses[43, 44]. We have identified the down-regulation of ADORA2A expression within paediatric AML, not mediated through DNA methylation, but potentially through the up-regulation of the miR-15a/16-1 cluster (data not shown). Moreover, miR-15a* and miR-16-1* are 60–90 fold enriched for targeting genes involved in the intrinsic apoptotic pathway such as BCL-2 L11, PPIF and DNM1L (Additional file10). The inhibition of these genes to act upon the cell, mediated through targeting miRNA, can potentially encourage continued cell growth and perpetuation of AML, through negating apoptotic signalling. The over expression of alternate mature miRNA in paediatric AML may also have consequences for integral cell cycle pathways, but through different mechanisms to their canonical counterparts.

In addition to assessing the association of DLEU2 promoter DNA methylation and miR-15a/16-1 miRNA expression with paediatric AML, we also assessed correlations with distinct clinical and diagnostic variables. We found no significant association between DLEU2 methylation/expression with sex, age of disease onset or relapse status nor common gene abnormalities such as FLT3 or MLL, as has been found in a number of other association studies[6, 45]. DLEU2 methylation clustering based on diagnostic FAB subtype revealed a wide range of values (M5a: 16-91%; M5b: 48-90%; ‘Other’ subtypes: 51-93%. Additional file11), indicating mean methylation in isolation does not stratify paediatric AML subtype. Additionally, we found there to be no correlation between percentage of leukaemic blasts at patient diagnosis and the DNA methylation status at the DLEU2 promoter region (data not shown).

Previous adult leukaemia studies have identified novel methylation subgroups, independent of cytogenetic profile[4, 46]. We identified a subset of patients displaying low DLEU2 promoter methylation comprising patients with trisomy 8 and/or chromosome 11q abnormalities, independent of MLL rearrangements (separated into a t(11)/+8 subgroup - Figure 3A, Additional file11). This delineation identified t(11)/+8 subgroup as having comparable DLEU2 methylation levels to non-leukaemic specimens (Figure 3B), and significantly lower methylation compared to the remaining paediatric AML samples (M5a/b p < 0.05; M1/M2/M4 p < 0.0001 Figure 3B). Classifying t(11)/+8 patients as a separate subgroup further delineated M5a, M5b and M1/M2/M4 AML patients from non-leukaemic counterparts according to average methylation status (M5a MM = 0.81, M5b MM = 0.78, M1/M2/M4 MM = 0.87. p < 0.0001. Figure 3B).

Cases carrying 11q rearrangements are the most heterogeneous of paediatric AML[27, 29, 47], linked to 50–104 translocation fusion partners to date[29, 46, 48]. Trisomy 8 is a frequently reported aberration in adult and paediatric AML[29]. However little is known about the gain of chromosome 8 in isolation and its relationship to disease onset. As such this has been speculated to be a disease modulating secondary event[49]. Lower DNA methylation in the t(11)/+8 subgroup may potentially confer a better prognosis, as it is well documented that hypermethylation of the DLEU2 region is involved in leukaemic transformation in adults[19, 20, 30]. Analysis of the t(11)/+8 subgroup in isolation also revealed no significant miRNA expression differences relative to non-leukaemic samples (Figure 4A), and it is well documented that miR-15a/16-1 abnormalities are also associated with leukaemic transformation in adults[19, 20]. We additionally identified a trend towards decreased risk of relapse in t(11)/+8 subgroup compared to other subtypes, and a trend towards better survival outcomes (Additional file12). These analyses combined may elucidate a connection between DLEU2 promoter DNA methylation (Figure 3B), miRNA expression (Figure 4A) and prognostic outcomes for this subgroup of paediatric AML (Additional file12).

The classification of t(11)/+8 cases as an independent subgroup revealed a significant increase in pri-miR-15a expression in traditional FAB subtype M5b from non-leukaemic (20.29-fold, p < 0.001; Figure 4C) and from M5a/M1/M2/M4 (Figure 4B). The increase translates to a >2-fold increase in miR-15a, 4.87-fold increase in miR-15a*, and 9.86-fold increase in miR-16-1* expression. Based on DLEU2 DNA methylation status, in conjunction with miRNA expression analysis, we speculate that pri-miR-15a expression alone may be a useful biomarker to distinguish M5b FAB subtype from all other AML subtypes.

This study was approved by the Royal Children’s Hospital (RCH), Melbourne, Ethics Committee (HREC reference #27138E). Samples used consisted of snap frozen bone marrow specimens and archived bone marrow biopsies taken at diagnosis, patient remission/follow-up or at relapse from paediatric acute myeloid leukaemia (AML) cases. The diagnosis of AML was established according to the criteria of the French–American–British (FAB) classification by standard morphological and cytological methods. The evaluation for mutations in MLL, FLT3, RUNX1 and WT1 genes were assessed in a small number of patients as a part of the initial assessment. No patient recorded any chromosome 13 deletions or abnormalities listed in their clinical diagnoses. The median percentage of leukaemic blasts at patient diagnosis is 88% (67–93 95% CI).

Patient samples used in our study consisted of archived, air-dried bone marrow smear slides. The utility of these samples for DNA methylation and miRNA expression analysis has been outlined previously[34, 50]. Cryogenically frozen patient bone marrows were also used where available. All patients were <18 years of age. We chose to focus specifically on the FAB subtype M5 (a and b) as it is a common subtype. Our cohort included comparable ratios between males and females, and also similar numbers of children that relapsed or not. (Additional file5). Control samples from bone marrow of unrelated and unaffected children were analysed in parallel, as well as multiple cell lines, including adult leukaemia (REH[51], CCRF-CEM[52]), paediatric AML (Kasumi-1[53], THP-1[54], MV-4-11 and AML-193[55]) obtained from American Type Culture Collection (ATCC) and subjected to characterisation using the ATCC Proficiency Standard program. Fluorescent Activated Cell Sorted (FACS) isolated haematopoietic progenitor cell populations from unrelated and unaffected paediatric donors (CD19+, CD33+, CD34+ and CD45+ populations, from herein known as CD sorted cells), were also used in this study. Our ‘non-leukaemic’ group consisted of all individual CD sorted cell populations as well as non-leukaemic remission and follow-up slides and bone marrow from paediatric patients. We compared the DLEU2/Alt1 DNA methylation for all ‘non-leukaemic’ specimens, and found no significant differences across these non-leukaemic samples (Additional file13).

Genomic DNA was extracted using the phenol/chloroform method and bisulphite converted in accordance with manufacturer’s protocols, as reported previously[23]. DNA extracts were checked for quality and quantity using a NanoDrop® ND-1000 spectrophotometer (Thermo Fisher Scientific Inc., Scoresby, Victoria, Australia). All DNA was stored at -80°C.

Bisulphite conversion of genomic DNA was performed using the MethylEasy Xceed Bisulfite Modification Kit (Human Genetic Signatures, Sydney, AUST). The converted samples were processed by the Australian Genome Research Facility (AGRF, Melbourne, Australia) and analysed using the Illumina HumanMethylation450 (HM450) Bead Chip arrays according to manufacturer’s protocol. Illumina Genome Studio software was used to extract the raw M-values and probe intensities for downstream processing. Samples with a probe p-detection value of p <0.05 were retained. Data were quantile normalized using lumi[56] and analysed using LIMMA[57]. All analysis was performed using the R statistical software package on M-values that were converted to β-values (0 = unmethylated, 1 = fully methylated) for reporting and biological interpretation. To eliminate sex bias, probes hybridizing to sex chromosomes were filtered out, leaving 366,553 probes common to all samples in the final dataset. Further filtering was based on the degree of difference in β between sample groups, indicated as ∆β calculated using the formula:

SEQUENOM MassARRAY® EpiTYPER® was used to measure locus-specific methylation. Sixty-one samples were analysed, including the 42 analysed on the HM450 platform (Additional file5), an additional 3 cell lines, 8 leukaemic and 8 non-leukaemic whole bone marrow aspirate samples. Primers for analysis were designed using SEQUENOM EpiDesigner software (http://​www.​epidesigner.​com) and the sequences for these are listed in Additional file14. Gene ontology and pathway analysis of genes associated with significantly altered DNA methylation probes were analysed through the use of IPA (Ingenuity® Systems, http://​www.​ingenuity.​com) and GOrilla[58].

Small RNA extraction was performed using TRIzol® (Ambion®) for patient bone marrow samples, mononuclear cells and cell lines in accordance with the manufacturer’s instructions, or the High Pure miRNA Isolation Kit (Roche) for archived slide samples as previously described[34]. Before RNA extraction of fresh bone marrow aspirates, samples were processed using Ficoll-Paque™ (GE Healthcare, Piscataway USA) to isolate the mononuclear cell population. This was immediately cryo-frozen or stored in RNAlater® (Ambion® by Life Technologies, Mulgrave, Victoria Australia) for later extraction. The concentration and purity of all RNA samples was assessed using the NanoDrop® ND-1000 spectrophotometer (Thermo Fisher Scientific Inc., Scorsby, Victoria, Australia). All RNA was stored at -80°C.

Seventy samples were used in the interrogation of miRNA expression, 64 of which were also used for HM450 and SEQUENOM methylation analysis (Additional file5). TaqMan® MicroRNA Reverse Transcription kit and singleplex TaqMan ® microRNA Assays (Applied Biosystems, Life Technologies) (assays listed in Additional file15) were utilised according to the manufacturer’s instructions before routine quantitative real-time PCR (qRT-PCR) was performed using Applied Biosystems 7300 Sequence Detection System. A subset of 29 high quality samples which included all cell lines as previously described, CD19+, CD33+, CD34+, CD45+ as well as 8 leukaemic and 9 non-leukaemic patient whole bone marrows, were used for gene and pri-miR expression analysis. This was due to longer RNA species often being degraded in archived specimens[59, 60]. The SuperScript® VILO™ cDNA Synthesis Kit (Life Technologies) and TaqMan® gene expression assays were used for gene/pri-miR expression as per manufacturer’s instructions (Additional file15 and Additional file16) before performing qRT-PCR analysis according to the manufacturer’s instructions. All qRT-PCR samples were analysed in duplicate.

Expression data was normalized using the synthetic small nucleolar RNA, C/D Box 44 (RNU44) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) for gene/pri-miR expression; and RNU44 and hsa-miR-26b for miRNA analysis, previously identified as having the greatest stability in our samples[34]. Data were analysed using DataAssist™ software (v.3.0, Applied Biosystems, Life Technologies: http://​www.​lifetechnologies​.​com/​au/​en/​home/​technical-resources/​software-downloads/​dataassist-software), and all p-values were adjusted using Benjamini-Hochberg False Discovery Rate.

All analyses used the combined average of non-leukaemic primary bone marrow tissue and CD sorted non-leukaemic cell expression values as the Reference Group (calibrator). Fold Change (FC) was calculated using the Livak method of 2^-∆∆C_t[61] plotting fold change and associated p-values. The fold change reported here is the difference of the means of each group, and a fold change of >2 between disease and non-leukaemic groups were considered noteworthy. microRNA gene target prediction was assessed using microRNA.org[62, 63] and miRWalk[64]. The top 500 target genes for each miRNA (Additional file17) were used for gene ontology/pathway analysis (Additional file10) by IPA (Ingenuity® Systems, http://​www.​ingenuity.​com) and GOrilla[58].

Where groups were compared with more than one variable measured (non-repeated measures), a Two-Way ANOVA with Bonferroni Post-testing was performed. Where groups were compared with only one variable measured, a Kruskal-Wallis test was performed with Dunns Post-testing. These analyses were undertaken on normalized data. Correlation assessments utilised Spearman Rank tests. Two-tailed p-values were utilised for assessment of statistical significance, and results with p-values p < 0.05 were considered to be statistically significant. Survival and relapse curves were estimated by the Kaplan-Meier method. Statistical analyses were performed with GraphPad PRISM (Version 5) for Windows, (GraphPad Software, USA, http://​www.​graphpad.​com). CI refers to 95% Confidence Intervals.