High DNA Methyltransferase DNMT3B Levels: A Poor Prognostic Marker in Acute Myeloid Leukemia

It has been recently shown that DNA methyl transferase overexpression is correlated with unfavourable prognosis in human malignancies while methylation deregulation remains a hallmark that defines acute myeloid leukemia (AML). The oncogenic transcription factor EVI1 is involved in methylation deregulation and its overexpression plays a major role for predicting an adverse outcome. Moreover, the identification of DNMT3A mutations in AML patients has recently been described as a poor prognostic indicator. In order to clarify relationship between these key actors in methylation mechanisms and their potential impact on patient outcomes, we analysed 195 de novo AML patients for the expression of DNMT3A, 3B (and its non-catalytic variant 3BNC) and their correlations with the outcome and the expression of other common prognostic genetic biomarkers (EVI1, NPM1, FLT3ITD/TKD and MLL) in adult AML. The overexpression of DNMT3B/3BNC is (i) significantly correlated with a shorter overall survival, and (ii) inversely significantly correlated with event-free survival and DNMT3A expression level. Moreover, multivariate analysis showed that a high expression level of DNMT3B/3BNC is statistically a significant independent poor prognostic indicator. This study represents the first report showing that the overexpression of DNMT3B/3BNC is an independent predictor of poor survival in AML. Its quantification should be implemented to the genetic profile used to stratify patients for therapeutical strategies and should be useful to identify patients who may benefit from therapy based on demethylating agents.


Introduction
Methylation-specific gene alteration is the major mechanism involved in inappropriate gene activation or silencing in leukemic cells and has been shown to be a universal feature occurring in all acute myeloid leukemia (AML) patients [1,2]. DNA methyltransferases (DNMTs) are the main key effectors of DNA methylation by catalysing the transfer of a methyl group from the ubiquitous methyl donor S-adenosyl methionine to the 59-position of cytosine residing in the dinucleotide sequence cytosine-guanine [3,4]. At least three prototype-related structure DNMTs are known. DNA methyltransferase 3 (DNMT3A, DNMT3B) are thought to act as de novo DNMTs mostly implicated in somatic alterations [5]. DNMT3A is particularly required for the methylation of imprinted and single copy genes while DNMT3B is specialized in the methylation of pericentric satellite repeats [6,7,8]. Isoforms of DNMT3B can be divided into those who do not alter catalytic activity of DNMT (3B1, 3B2, 3B6) and others (3B3, 3B7, 3B8) may be inactive in catalysis [9] but could act as a rheostat in modulating DNMT3B and/or 3A [10,11,12]. The mechanism(s) by which cancer cells acquire alteration in DNA methylation is unknown but aberrant transcription of the DNMT3B gene is frequent [13]. DNMT3B7, a truncated non catalytic DNMT3B isoform particularly expressed in human tumors, has been shown to accelerate lymphomagenesis, increase chromosomal rearrangements and shows more locus specific perturbations in DNA methylation patterns in mice when Dnmt3b7 transgenic mice are bred with Em-Myc transgenic mice [14]. Transcriptional activation of the human EVI1 (Ecotropic Virus Integration site 1) gene located on 3q26.2, has been reported in up to 10% of AML patients and is an independent indicator of adverse prognosis [15,16,17]. Although most patients with 3q26 rearrangements (inv(3)(q21q26.2)/ t(3,3)(q21;q26.2) overexpress EVI1 (EVI1+), its level, through unknown mechanisms, is also elevated in about 10% of AML patients with no 3q aberrations [18]. EVI1 functions as a transcription repressor complex recruiting diverse proteins involved in chromatin remodelling.
High levels of EVI1 are associated with aberrant epigenetic signatures containing differentially hypermethylated genes, with an overrepresentation of EVI1 binding sites in their promoters [19]. DNMT3A mutations that abrogate its enzymatic activity are relatively common in de novo AML (22%) and in myelodysplasic syndromes (8%) [20,21]. Their identification in AML patients has been recently described as a poor prognostic marker associated with disease progression and poor survival [22,23,24,25]. Moreover, DNMT3A was found to be highly expressed in primary EVI1+ AMLs as compared to other AML and a direct recruitment of DNMT3A and 3B by EVI1 has recently been demonstrated [1,19]. The deregulation of DNMT3B expression clearly contributes to tumorogenesis and tumor suppressor gene hypermethylation [26]. In addition, a high expression level of DNMTs and their variants has been reported and described as a poor prognostic marker in various malignancies [27,28].
Overall, these data provide a rationale that has prompted us to analyse a series of 195 de novo AML patients in order to establish at first time, the relationships between these strong mediators of gene expression (DNMT3A/B and EVI1) and other well-known biomarkers such as Nucleophosmin (NPM1), fms-like tyrosine kinase-3 (FLT3) mutations, partial tandem duplications of mixedlineage leukemia gene (MLL/PTD) and HOXA9 expression level. We studied in a second time their effect on survival outcomes.

Patients
Ethic statement. Written informed consent was obtained from all patients and the procedures followed were in accordance to the Helsinki declaration as revised in 2008. Samples were stored in the Biological Ressource Center Bank according to ''the Comité de Protection des Personnes''. The review board protocol of the Hospices Civils de Lyon approved this study.
Samples from 195 consecutive newly diagnosed AML patients (excluding acute promyelocytic leukemia) admitted at the Hematology department of Lyon University Hospital and treated onto clinical trials were referred to our laboratory between September 2002 and April 2011 (Table 1). AML was diagnosed according to the French-American-British (FAB) and World Health Organization classification of tumours criteria [29,30]. There were 88 females and 107 males. Median age was 53 (range: 18-73 years). The main patient characteristics at diagnosis of AML are shown in Table 1 and in Figure 1. Induction consisted in one cytarabineand anthracycline-based chemotherapy course (6salvage therapy). Patients achieving complete remission (CR) received then consolidation courses according to the trial in which they were included. Fifty-one patients with an HLA-compatible donor received allogeneic hematopoietic stem cell transplantation in first remission.

Cytogenetic Analyses
Cytogenetic R and G-banding analyses were performed according to standard methods. The definition of a cytogenetic clone and descriptions of karyotypes followed the International System for Human Cytogenetic Nomenclature. To be considered cytogenetically normal, at least 20 metaphases from bone marrow sample at diagnosis had to be evaluated. Abnormalities were categorized and classified into 3 categories (favourable, intermediate, and unfavourable) according to the British Medical Research Council's classification [31]. In the overall cohort of 195 patients, 9 cytogenetic analyses failed and 93 normal cytogenetic (NC) karyotypes were identified in at least 20 metaphase cells evaluated from diagnostic bone marrow of each patient (Table 1). Ecotropic Viral Integration Site 1 gene (EVI1/1D) expression were detected as previously described [32,33,34,35].

Molecular Assessments
RT/RQ-PCR of DNMT (3A/3B and 3BNC) and HOXA9 mRNA. Reverse transcription (RT) was performed as previously described [36] for 195 AML patients and 11 normal bone marrow donors. The targeted sequence for DNA corresponded to the mRNA of the two DNMT genes DNMT3A: ENST00000321117; DNMT3B: ENST00000328111 and 3B NC : ENST00000456297. For ease and because all catalytic (3B1, 3B2, 3B6) and noncatalytic (NC: 3B3, 3B7, 3B8) DNMT3B spliceoforms showed respectively the same 39 end nucleotide sequences, we used the term 3B and 3B NC for each subgroup. DNMT (3A, 3B and 3B NC ) and HOXA9 transcripts were amplified on the same cDNA using primers depicted in Table 2  reference for RNA quality (namely, the ABL gene) as it was previously shown to be the most adequate for quantitative analysis in AML [37].    Table 2. Sequences of the different primers and probes. instrument (Roche Applied Sciences). HRM analysis is a suitable method in routine laboratories for the detection of the DNMT3A mutations located between amino acids V867 and R891 and especially for the most common mutation R882H/C with a 2.5%sensitivity ( Figure S1). All mutations were confirmed on another RT-PCR product and by direct sequencing. HRM analysis of DNMT3B catalytic region, highly similar to DNMT3A (Figure 2), were performed as described for the DNMT3A except for the use of primers from DNMT3B (Table 2).
Mutational status of DNMT3A (exon 8 to exon 23) in EVI1+ patients. In order to verify total absence of DNMT3A mutation in functional domain between exon 8 and exon 23 from 32 AML patients showing overexpression of EVI1 (Table 1), a first round of PCR was performed on cDNA (5 ml), using primers DNMT3AF 7-8 and DNMT3AR 23 ( Table 2). Thermocycling conditions used were 3 minutes at 94uC followed by 35 cycles of denaturation at 94uC for 1 minute, annealing at 60uC for 1 minute, extension at 68uC for 4 minutes, and a final extension step at 68uC for 10 minutes. Two nested PCR were performed using one fiftieth of the first PCR product described above, with the same thermocycling conditions and with forward (DNMT3AF 8 and DNMT3AF [11][12] ) and reverse (DNMT3AR 16-17 and DNMT3AR 23INT ) primers, respectively ( Table 2). Amplicons (1078 and 1297 bp) were directly sequenced in both strands using PCR primers and DNMT3AR [14][15] and DNMT3AF 16-17 depicted in Table 2. All mutations were confirmed on another RT-PCR product. Moreover, absence of mutations (identified at diagnosis) was verified on complete remission samples whenever available.

Statistical Analyses
Complete remission (CR) was defined according to Cheson's criteria [38]. Overall survival (OS) was calculated from the date of diagnosis until the date of death. Event-free survival (EFS) was measured from the date of diagnosis until the date of the first event (morphological relapse or death). Comparisons of patients' characteristics (covariates) were performed using the Fisher's exact test for categorical variables, the Mann-Whitney U-test for continuous variables and by Spearman's rank correlation for quantitative variables. Quantitative variables as expression levels of DNMT3A, 3B, 3B NC , HOXA9 and EVI1 were also analysed as binary variables using median expression levels (DNMTs and HOXA9) or delta CT method (DCTEVI1-ABL,7) as cut off point in AML patients. OS and EFS rates were estimated by the Kaplan-Meier method and compared using the log-rank test. Covariates tested in multivariate Cox models were sex, age, DNMTs, HOXA9, EVI1 expressions, cytogenetic risk group defined as previously described [39]. A p value,.05 was considered as statistically significant. In all univariate and multivariate analyses provided, the patients were censored at the time of transplantation but the same analyses were performed in the overall population without censoring allografted patients. All statistical analyses were performed in the normal karyotype subgroup (93 patients). Clinical data for first remission, overall survival and event-free survival analyses were available for all patients.

Expression Analysis of DNMT3A, 3B, 3B NC and Correlation with other Biomarkers
Among the whole population, Spearman's rank order correlation showed that DNMT3A expression was higher in young patients (p = 0.019; for,60 years) and in patients with low levels of DNMT3B (p = 0.002) and HOXA9 (p = 0.005) whereas DNMT3B was highly related to 3B NC (p,0.0000) and HOXA9 expression (p = 0.0003). These results were confirmed using Mann-Whitney U-test and Fisher's exact test (two tailed) when categorical variables were used for statistical analysis (Figure 3). EVI1 overexpression was significantly related to AML patients with NPM1 wild type status (p = 0.001) and to MLL abnormalities duplications (p = 0.039) or rearrangements (p = 0.001). Conversely to EVI1, HOXA9 overexpression was more frequently found in NPM1 (p,0.0000) and FLT3-ITD/TKD mutated patients (p = 0.0004; Figure 3). It should be noted that no correlation was observed at the transcriptional level between EVI1 and (i) DNMT3A (p = 0.38), (ii) DNMT3B (p = 0.5) or (iii) DNMT3B NC (p = 0.27).

DNMT3 Mutation Analysis
Twenty-six mutated patients (13%) were identified in DNMT3A exon 23 and were confirmed by direct sequencing. Furthermore, HRM method allowed the identification of two new mutations onto amino acids R882 (R882S) and Q886 (Q886E). All new mutations have been verified on another RT-PCR product. Moreover, the Q886E mutation was not found in an available remission sample before transplantation which can invalidate the hypothesis of a polymorphism (data not shown). Morphologic data are summarized in Table 1. A slight association was found with monocytic differentiation (FAB M5 p = 0.04) but not confirmed towards myelomonocytic differentiation (FAB M4/M5; p = 0.27). As previously described [40] DNMT3A exon 23 mutations were mostly distributed in the normal karyotype category (p = 0.01) and exclusively in the intermediate cytogenetic risk group, with a strong association with NPM1 mutations (p,0.000), HOXA9 expression (p = 0.001) and FLT3 mutations (p = 0.049) and inversely related to EVI1 expression (p = 0.008). No DNMT3A exon 23 mutations were detected in high risk patients, particularly in EVI1+ AML patients. Entire exon 8 to exon 23 sequencing of DNMT3A detected 3 further mutations in DNMT3A catalytic domain (V636E; D702G; R736C).
In order to screen this domain in DNMT3B, HRM analysis was performed from 137 patients' samples. No change was observed in the catalytic active variant of DNMT3B (amino acids 808-836) and especially at amino acid R823 of DNMT3B, one of the most common mutated amino acids involved in hereditary syndrome characterized by ICF and which corresponds to the amino acid R882 of DNMT3A (Figure 2).

Impact of HOXA9, EVI1,and DNMTs Expression Levels on Overall Survival (OS) and Event-free Survival (EFS)
HOXA9 overexpression had no prognostic impact whereas patients with low HOXA9 expression level had a better OS than Patients who showed an overexpression of EVI1 had significant lower EFS and OS ( Figure 5) and when we stratified on age, this observation was more important in young patients (EFS: p = 0.002 and OS: p = 0.037). It has been recently described that patients who showed a total lack of EVI1 expression might have a good prognosis [41], In our series, patients with absolutely no EVI1 expression did not show better survival than those with basal EVI1 expression, even in patients below 60 years (data not shown).
No significant differences were observed for the DNMT3A expression in the whole cohort population but its overexpression was related to a significant better EFS (p = 0.01) and OS (p = 0.012) in the normal karyotype AML subgroup. There was no association between DNMT3A exon 23 mutations and their prognostic impact among the whole population, patients with normal cytogenetic profiles or those with FLT3ITD mutations independently of age. Patients with an overexpression of DNMT3B and 3B NC had worse EFS (p = 0.006) and OS (p = 0.045), respectively ( Figure 6A and 7B) and a trend to a worse OS (p = 0.056) and to a worse EFS (p = 0.07), respectively ( Figure 6B and 7A).
It should be noted that the prognostic impact of these markers was erased by the transplantation when considering the small cohort of patients allografted in first remission.Multivariate analysis showed a significant negative impact of age, unfavourable complex karyotype (more than 3 abnormalities, cytogenetic group 3) and DNMT3B/B NC levels on EFS and OS ( Figure 8A and 8B). In addition we observed a positive impact of NPM1 mutations on EFS. Regarding the normal karyotype AML subgroup only a negative trend for the DNMT3B expression on EFS and OS (p = 0.07) was detected.

Discussion
The deregulation of DNMT expression clearly contributes to tumorogenesis and its overexpression has been described as a poor prognostic marker in various malignancies [26,27,28]. Aberrant DNA hypermethylation feature has been directed by EVI1 in AML [1] which can suggest a strong relationship between these proteins. EVI1 has been recognized as one of the most aggressive oncogene associated with AML [15,16,17]. Even if it has been recently shown in a large cohort of AML patients that its poor prognosis is independent from the EVI1 spliceoform expressed [17], the poor impact of the EVI1+/1D has always been confirmed and should be relevant for stratifying patients in therapeutic protocols. Our results confirm that EVI1 overexpression has an adverse prognostic impact, either in young or elderly AML patients. It is of note that EVI1 is rarely overexpressed in mutated NPM1 patients. An association between EVI1 overexpression and MLL abnormalities has been previously noted by other [17]. EVI1  is frequently up-regulated in bone marrow cells transformed by the MLL oncoproteins and could be one of its targets which could explain their high association in leukemogenesis [42]. It has been previously described [41], that the total lack of EVI1 expression might have a good prognosis, in our series this hypothesis was not confirmed.
A direct recruitment of DNMT3A and 3B by EVI1could reflect strong relationship between these proteins [1,19]. In our series, EVI1+ is not associated with high expression level of DNMTs at mRNA level which suggest functional interaction between these proteins rather than a common regulatory mechanism at their transcriptional levels. However, EVI1+ is highly associated with absence of DNMT3A exon 23 mutations which could suggest a functional DNMT3A among EVI1+ patients. This hypothesis may be modulated by the identification of 3 mutated patients in the catalytic domain of DNMT3A. As previously described, we confirm here the high frequency of DNMT3A exon 23 mutations. These mutations were strongly associated to AML with a normal karyotype and with NPM1 mutations, and to a lower extend to those with FLT3 abnormalities and monocytic involvement. Moreover, we showed that DNMT3A exon 23 mutations were highly associated with HOXA9 expression. The poor prognostic impact of DNMT3A mutations was not confirmed, probably due to the small number of patients [40]. Constitutional mutations in the catalytic domain of DNMT3B have been described and are responsible for hereditary syndrome characterized by ICF (immunodeficiency, instability of the centromeric region of chromosomes and facial abnormalities) in humans [43]. More recently, all tested ICF mutations are responsible of altered catalytic properties of DNMT3B [44]. No change was observed in the catalytic active variant of DNMT3B which could strongly suggest that mutation of DNMT3B may be not a common event In AML.
DNMT3A and 3B transcripts seem to be inversely expressed in AML patients and related to HOXA9 expression. DNMT3B and 3B NC are highly related since they have the same promoter and are splicing variants. We have shown worse prognostic outcome associated with high DNMT3B/3B NC expression levels. The functions of DNMT3B NC are not clearly defined. These truncated proteins, lacking the catalytic domain, could act as a rheostat in modulating DNMT3B and/or 3A. However they have been recently involved in various human tumours. Thus, forced expression of DNMT3B7, which could be the main isoform quantified in our assays, led to altered DNA methylation levels, particularly in hematologic malignancies. It has recently been shown in MYC-transgenic mice that overexpression of the noncatalytic Dnmt3b7 isoform or inactivation of the catalytically active Dnmt3b accelerated lymphomagenesis [14,45]. Overex-  pression of catalytically inactive isoforms seems to have similar consequences as inactivation of active Dnmt3b isoforms and suggest possible oncogenic functions of catalytically inactive Dnmt3b isoforms, perhaps acting as dominant negative isoforms. In this context, we have tried to evaluate EFS and OS independently in cohorts of patients that had high active DNMT3B but low DNMT3B NC and vice versa. We did not find significant prognostic impact, probably related to the fact that DNMT3B/ 3B NC are highly correlated (p,0.0000) and therefore the number of patients in each cohort is too small (24 versus 18 patients, data not shown). In human AML, overexpression of DNMT3B and 3B NC both seem to be correlated with poor prognosis and could participate together to the oncogenic methylation alterations in leukemic cells.
In spare cases, we have detected a significant prognostic impact of one molecular marker on EFS and not on OS (as DNMT3B which has shown only a trend to a worse OS) or vice versa (as DNMT3BNC or HOXA9). The clinical significance of these observations could be related to the efficiency of relapse treatments which improved OS, or conversely, related to very high mortality rates in first line setting that would principally affected OS and this, according to the expression of the considered marker.
We could not explain the relationship between HOXA9 and DNMT3 expressions or with NPM1 mutations but homeotic proteins are largely involved in leukemogenesis, in MLL transduction pathway and more recently, one of this protein (HOXB3) has been shown to regulate DNMT3B expression in human cancer cell lines [46]. Nevertheless, we have shown that DNMT3A overexpression was associated with a favourable outcome in normal karyotype AML subgroup and DNMT3B (and 3B NC ) overexpression with a worse outcome. More recently DNMT3B expression level has been shown to be an adverse prognosis marker in diffuse large B-cell lymphomas [27]. In this setting, patients with DNMT overexpression were characterized by aggressive disease and poor prognosis, probably in relation to the hypermethylation of important genes in homeostasis although no target gene have been clearly identified in AML. In the future, these observed results could suggest that treatment specifically targeting methylation as cytosine analog drugs which interfere with methylation (5azacytidine and decitabine (5-aza-20-deoxycytidine) should be clinically evaluated in those specific patients.
In conclusion this study represents the first report showing the prognostic impact of DNMT3A, DNMT3B and 3B NC overexpression in AML. DNMT3B (3B NC ) overexpression represents a new independent poor prognostic marker in AML. This should be useful for identifying patients who may benefit from demethylating agents. Figure S1 HRM analysis of DNMT3A exon 23 mutations. A. HRM profiles of 8 patients (in duplicate) harbouring DNMT3A exon 23 mutations (red) compared to 20 negative patients (blue). B. The sensitivity of the test (, 2.5%) was obtained from the dilution of R882H mutated samples in non mutated cDNA (undiluted: heterozygous rate at 50%); Analysis was performed on the LC480 Roche device. Positive detected dilutions in red. (TIF)