Expression and prognosis analysis of DNMT family in acute myeloid leukemia

DNA methyltransferases (DNMTs) by regulating DNA methylation play crucial roles in the progression of hematologic malignancies, especially for acute myeloid leukemia (AML). Accumulating investigations have identified the high incidence of DNMT3A mutation in AML, and it is correlated with poor prognosis. Although a few studies have shown the expression of DNMTs and their clinical significance in AML, the results remain to be discussed. Herein, we systemically analyzed the DNMTs expression and their relationship with clinic-pathological features and prognosis in AML patients. DNMTs expression especially for DNMT3A/3B was closely associated with AML among various human cancers. DNMT3A expression was increased in AML patients, whereas DNMT3B expression was decreased. Significant associations between DNMT3A/B expression and clinic-pathological features/gene mutations were observed. Kaplan-Meier analysis showed that DNMT3A expression was associated with better overall survival (OS) and leukemia-free survival (LFS) among whole-cohort AML, and independently affected OS determined by Cox repression multivariate analysis. Notably, patients that received hematopoietic stem cell transplantation (HSCT) showed significantly better OS and LFS in DNMT3A lower-expressed groups, whereas patients in DNMT3A higher-expressed groups did not. By bioinformatics analysis, DNMT3A expression was found to be positively correlated with several leukemia-associated genes/microRNAs, and DNMT3A was identified as direct targets of miR-429 and miR-29b in AML. Collectively, our study demonstrated that DNMT3A/3B showed significant expression differences in AML. DNMT3A expression acted as a potential prognostic biomarker and may guide treatment choice between chemotherapy and HSCT in AML.


INTRODUCTION
DNA methylation, as the most common epigenetic modification, plays a crucial role in tissue-and stagespecific gene regulation, genomic imprinting, and X-chromosome inactivation, and has shown to be essential for normal mammalian development [1]. Also, accumulating studies have proved that both global DNA hypomethylation and hypermethylation occur frequently in tumorigenesis [2]. The hypermethylation of CpG AGING islands at the promoter regions is often associated with the inactivation of tumor suppressor genes (TSGs) [3]. In hematopoietic disorders, aberrant DNA hypermethylation is proved to be involved in leukemogenesis [4]. For example, Spencer et al demonstrated that CpG island hypermethylation mediated by DNMT3A was a consequence of acute myeloid leukemia (AML) progression [5]. In addition, aberrant DNA methylation was also regarded as a dominant mechanism in progression from myelodysplastic syndromes (MDS) to AML [6]. Furthermore, deregulated DNA methylation in MDS and AML has led to the approval for the clinical use of hypomethylating agents (HMAs) in both MDS and AML patients [7].
DNA methyltransferases (DNMTs) are the main key effectors of DNA methylation by catalyzing the transfer of methyl groups from S-adenosyl-lmethionine to the 5'-position of cytosine residing in the dinucleotide sequence cytosine-guanine [8]. The DNMTs include three major members (DNMT1, DNMT3A and DNMT3B), among which DNMT3A and DNMT3B catalyze cytosine methylation of mammalian genomic DNA to establish de novo DNA methylation patterns, whereas DNMT1 maintains a methylation state through DNA replication [9]. Recent studies have demonstrated that DNMTs play vital roles in the progression of hematologic malignancies, especially AML [10]. Trowbridge et al showed that haploinsufficiency of DNMT1 impaired leukemia stem cell (LSC) function through derepression of bivalent chromatin domains [11]. More importantly, high incidence of DNMT3A mutation was identified in AML and DNMT3A mutation correlated with poor prognosis in AML [12]. Functional studies showed that DNMT3A was essential for hematopoietic stem cell (HSC) differentiation and mutated DNMT3A initiated AML [13][14], suggesting DNMT3A acted as a tumor suppressor gene. Although DNMT3B was rarely mutated in AML [15], studies have proved that loss of DNMT3B accelerated MLL-AF9 leukemia progression and increased expression of DNMT3B in LSC delayed leukemogenesis [16][17]. A few studies have shown the expression of DNMTs and their clinical significance in AML, but the results remain to be discussed [18][19][20][21]. Herein, we systemically analyzed DNMTs expression and their relationship with clinic-pathological features and prognosis in patients with AML.

DNMTs expression associated with AML among human cancer cell lines
Using the Cancer Cell Line Encyclopedia (CCLE) databases, we found that expression of DNMTs was highly expressed in AML cell lines among 40 types of human cancer cell lines ( Figure 1A). Moreover, DNMTs expression was also closely associated with myeloid cell lines revealed by The Human Protein Atlas (HPA) databases ( Figure 1B). The detailed comparison of DNMTs expression in 14 types of AML cell lines was assessed by using the European Bioinformatics Institute (EMBL-EBI) website, which was shown in Figure 1C.

DNMTs expression associated with AML patients among human cancers
We further determined expression of DNMTs in AML patients by using Gene Expression Profiling Interactive Analysis (GEPIA) dataset. Aberrant expression of DNMT3A and DNMT3B was observed in AML among 33 types of human cancers, whereas DNMT1 did not show significant difference in AML ( Figure 2A). DNMT3A expression was significantly increased in AML patients, whereas DNMT3B expression was markedly decreased in AML patients ( Figure 2B). Moreover, DNMT1 expression was slightly associated with both DNMT3A and DNMT3B expression, whereas DNMT3A expression was positively correlated with DNMT3B expression in AML patients ( Figure 2C).

Association between DNMT3A/3B expression and clinical characteristics in AML
Since aberrant expression of DNMT3A and DNMT3B was identified in AML, we further explore their clinical significance in patients with AML. Clinical implication of DNMT3A and DNMT3B was obtained by the comparison of clinical/laboratory characteristics of the AML patients between two groups divided based on median level of DNMT3A and DNMT3B transcript (Table 1). Interestingly, DNMT3A overexpression was associated younger age and lower white blood cells (WBCs) (P=0.008 and 0.063, respectively), higher peripheral blood (PB) blasts (P=0.006). Among the distribution of French-American-British (FAB) and cytogenetic subtypes, DNMT3A overexpression was associated with higher frequency of FAB-M0/M2, t(8;21), and -7/del(7) (P=0.009, 0.028, 0.007, and 0.007, respectively), whereas lower frequency of FAB-M4/M5 and normal karyotype (P=0.004, 0.001, and 0.000, respectively). In addition, DNMT3B underexpression was correlated with higher WBCs and lower PB blasts (P=0.041 and 0.006, respectively). Among the distribution of FAB and cytogenetic subtypes, DNMT3B underexpression was associated with higher frequency of FAB-M4/M5 and inv(16) (P=0.000, 0.000, and 0.001, respectively), but lower frequency of FAB-M1 and complex karyotype (P=0.000 and 0.000, respectively).

Correlation between DNMT3A/3B expression and gene mutations in AML
Among the common gene mutations in AML, patients with DNMT3A overexpression showed lower NPM1 and DNMT3A mutation rates (P=0.000 and 0.002, respectively), whereas cases with DNMT3B underexpression presented lower frequency of IDH1 and TP53 mutation (P=0.038 and 0.005, respectively). In addition, we further compared the expression of DNMT3A and DNMT3B in mutation and wild-type groups of these genes ( Figure 3).

Prognostic value of DNMTs expression in AML
We next evaluate the prognostic effect of DNMTs expression on survival in AML. By Kaplan-Meier analysis, only DNMT3A overexpression was associated with longer overall survival (OS) and leukemia-free survival (LFS) in whole-cohort AML (P=0.001 and 0.003, respectively, Figure 4). In order to confirm the independent prognostic value of DNMT3A expression on both OS and LFS, we performed Cox regression analysis adjusting for prognosis-related factors. By Cox regression multivariate analysis, DNMT3A expression could act as an independent prognostic biomarker for OS in whole-cohort AML (Table 2). However, no significant differences were observed in either DNMT1 or DNMT3B groups ( Figure 4).

DNMT3A expression may guide treatment choice between chemotherapy and HSCT
Because lower DNMT3A expression predicted poor clinical outcome in AML, we intended to investigate whether patients with lower DNMT3A expression could AGING benefit from HSCT. We compared OS and LFS between patients with and without HSCT in both lower and higher DNMT3A expression groups. In lower DNMT3A expression groups, patients who received HSCT showed significantly longer OS and LFS than patients who were not treated with HSCT among both total AML ( Figure 5). However, in higher DNMT3A expression groups, there were no significant differences in OS and LFS between two groups ( Figure 5). Taken together, AML patients with lower DNMT3A expression could benefit from HSCT, whereas those with higher DNMT3A expression did not. Therefore, we deduced that DNMT3A expression pattern may guide treatment choice between chemotherapy and HSCT.

Molecular signature correlated with DNMT3A expression in AML
To gain insights into the biological function of DNMT3A in AML, we first compared the transcriptomes of lower and higher DNMT3A expression groups. A total of 972 differentially expressed genes (DEGs) were identified including 428 positively correlated genes and 544 negatively correlated genes (FDR<0.05, P<0.05, |log2 FC|>1.5; Figure 6A; Supplementary Table 1). Positively correlated genes such as MN1, BAALC, CD34, and H19 have been reported as proto-oncogenes in AML [22][23][24]. Furthermore, the Gene Ontology (GO) analysis was also showed in Figure 6B.
Obviously, further studies are required to confirm the direct connections between DNMT3A and miR-429 by luciferase assay.

DISCUSSION
In this study, we systemically analyzed DNMTs expression and their relationship with clinic-pathological features and prognosis in patients with AML. We found that DNMT3A expression was increased in AML, whereas DNMT3B expression was decreased in AML. Although previous study showed DNMTs overexpression and its negative prognostic effects in AML [18][19][20][21], recent researches revealed that DNMT3B expression was decreased in AML blasts, whereas DNMT1 and DNMT3A expression showed no significant differences [16,33]. Additionally, the potential roles of DNMT3A and DNMT3B in AML remained poorly defined. Peters et al previously showed the tumor suppressor functions of DNMT3A and DNMT3B in the prevention of malignant mouse lymphopoiesis, but not in the development of myeloid malignancies [34]. However, ectopic DNMT3B expression was reported to delay leukemogenesis [17], and the loss of DNMT3B accelerated MLL-AF9 leukemia progression [16]. These studies demonstrated that DNMT3B played a crucial role in AML development, but may not act as a cancer-related driver gene during leukemogenesis. As for DNMT3A, it was indicated that DNMT3A loss progressively impaired HSC differentiation [13]. Notably, loss of DNMT3A and endogenous KRAS G12D/+ cooperated to regulate hematopoietic stem and progenitor cell functions in leukemogenesis [35]. In our study, we also observed the significant associations of DNMT3A expression with other molecular events such as NPM1 and DNMT3A mutations. These studies suggested that DNMT3A generally not worked independently in the development of AML, and it may cooperate with other molecular events.

AGING
The prognostic value of DNMT3A mutation in AML has been systemically revealed. Increasing studies showed that DNMT3A mutations were independently associated with poor outcome in AML patients with an intermediate-risk cytogenetic profile or CN-AML [12,36]. Moreover, loss-of-function of DNMT3A caused by mutations or underexpression predicted response to the HMAs decitabine treatment in AML [37]. In this study, low DNMT3A expression was observed to act as an independent prognostic biomarker in AML and also helpful for the selecting treatment choice between chemotherapy and HSCT. Interestingly, although two recent reports showed that high DNMT3B expression was a poor prognostic biomarker in AML [19,20], we did not observe the association of aberrant DNMT3B expression with AML survival. The differences may be caused by the specific cell population selection and different ethnics. Obviously, prospective studies are needed to confirm and expand our results before DNMT3A expression pattern can be used routinely as a potential prognostic biomarker guiding treatment choice for newly diagnosed AML.
Despite that the role of DNMT3A in regulation of DNA methylation is well-known, the potential mechanism regulating DNMT3A was poorly investigated. Jost et al reported that aberrant DNA hypermethylation within the DNMT3A gene was frequently observed in AML, and was associated with downregulation of DNMT3A mRNA transcript 2 [38]. Moreover, DNMT3A was also identified as a direct target of a number of microRNAs such as miR-30a-3p, miR-133a-3p, miR-450, miR-29a/b/c, and miR-129-5p [39][40][41][42][43]. From our study, we observed the direct association of two microRNAs miR-29b and miR-429 with DNMT3A in AML.
Although several studies showed the direct link between DNMT3A and miR-29b in other human cancers, little studies showed the direct correlation between miR-429 and   AGING DNMT3A in any type of human cancers. Interestingly, our pervious study disclosed that miR-429 expression was decreased in AML [32], which presented the opposite expression pattern of DNMT3A in AML. Accordingly, further studies are required to confirm the direct associations of DNMT3A with miR-429 by luciferase assay.
In summary, although we analyzed the expression and prognosis analysis of DNMTs expression only by public databases, our study demonstrated that DNMT3A and DNMT3B showed significant expression differences in AML. Moreover, DNMT3A expression acted as a potential prognostic biomarker and may guide treatment choice between chemotherapy and HSCT in AML.

GEPIA dataset
DNMTs expression in AML patients and normal controls was analyzed by the Gene Expression Profiling Interactive Analysis (GEPIA) web (http://gepia.cancer-pku.cn/), whose data obtained from The Cancer Genome Atlas (TCGA) and the Genotype-Tissue Expression (GTEx) projects [47].

Patients from TCGA datasets
A total of 173 AML patients with available DNMTs expression data from TCGA were identified and included in this study [15]. Clinical and molecular characteristics were collected including age, sex, white blood cell (WBC) counts, peripheral blood (PB) blasts, bone marrow (BM) blasts, French-American-British (FAB) subtypes and the frequencies of genetic mutations. After induction chemotherapy, consolidation treatment included chemotherapy (100 patients received) and hematopoietic stem cell transplantation (HSCT) (73 patients received).

Bioinformatics analyses
The details for the identification of microRNAs targeting DNMT3A were reported as our previous study [48,49].

Statistical analyses
SPSS 22.0 were used for statistical analyses and figures creation. Mann-Whitney's U test was used for the comparison of continuous variables, whereas Pearson Chisquare analysis or Fisher exact test was applied for the comparison of categorical variables. The prognostic effect of DNMTs expression was evaluated using leukemia-free survival (LFS) and overall survival (OS) analyzed though Kaplan-Meier analysis and Cox regression analysis. The two-tailed P value < 0.05 in all statistical analyses was defined as statistically significant.

Ethical approval and consent to participate
The present study approved by the Ethics Committee and Institutional Review Board of the Affiliated People's Hospital of Jiangsu University. Written informed consents were obtained from all enrolled individuals prior to their participation.

AUTHOR CONTRIBUTIONS
Jing-dong Zhou conceived and designed the experiments, Ting-juan Zhang, Liu-chao Zhang, Zi-jun Xu and Jing-dong Zhou analyzed the data, all authors read and approved the final manuscript.

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.

FUNDING
The work was supported by National Natural Science foundation of China (81900166, 81970118), Zhenjiang Clinical Research Center of Hematology (SS2018009). Very thanks for Prof. Jun Qian providing technical and financial support.