Identication of Aberrantly Methylated and Silenced Genes in Acute Myeloid Leukemia by Combined Methylation/expression Analysis

Background Aberrant genomic methylation plays an important role in pathogenic process of acute myeloid leukemia (AML) by silencing tumor suppressor genes (TSG). While the key aberrantly methylated genes and related pathways have not been well understood yet, which we aimed to reveal by combined analysis of methylation and expression datasets. The prognostic signicance was validated by survival analysis derived from TCGA database. Methods Micro-array data of GSE 15061 and GSE58477 were downloaded from Gene Expression Omnibus (GEO) database. The differentially methylated regions (DMR) and differentially expressed genes (DEG) were identied using R program (R 3.6.1). Over-representation analysis was performed to obtain the enriched biological processes and pathways. Cox hazards analysis was employed to select the genes signicantly associated with AML survival, using the data derived from the Cancer Genome Atlas (TCGA) database. Subgroup analysis, regarding induction type, was conducted to identify biomarkers for HMA treatment. Furthermore, SYNJ2 associated genome-wide gene/miRNA expression and methylation prole were explored. for HMA treatment in AML.


Introduction
Acute myeloid leukemia (AML) is characterized by unlimited clonal proliferation of myeloid precursors and failure of normal hematopoiesis (1). AML is the most common subtype of acute leukemia in adults, and accounts for 80% of cases (2). Leukemia remains one of the most common cancers in both China and the United States (3,4). In 2019, an estimated 21450 new cases and 10920 deaths caused by AML were reported in the United States (3). According to European LeukemiaNet risk strati cation in AML(5), the risk status of AML patients is according to recurrent cytogenetic and molecular abnormalities. The clinical features and genetic background have an impact on the prognosis of AML patients. A comprehensive evaluation of molecular mutations and cytogenetic markers is crucial for evaluation of risk strati cation, which also provides rationales guiding treatment decisions. While the biomarkers with prognostic value or therapeutic targets were still being explored in AML.
In addition to the genetic variables, alterations of methylation on genomic DNA play an important role in the pathological process of AML. As the most well-known mechanism, methylation of cytosine in palindromic CpG sites, which are always clustered in promoter regions, results in downregulation of gene expression (6,7). As reported by Cancer Genome Atlas Research Network, 44% of de novo AML patients harbor mutations of DNA-methylation-related genes, including DNMT3A/TET2 etc (8), leading to aberrant methylation of genomic DNA in AML patients compared with that in healthy donors (9,10). The hypomethylating agents (HMA, including azacitidine and decitabine) irreversibly inhibit DNA methyltransferase via covalent bonds, resulting in global genome demethylation (11,12), especially in hypermethylated regions (8). A growing body of evidence indicates that HMAs have clinical bene t and are well tolerated in myeloid neoplasms including MDS and AML, especially in older and/or un t patients (13,14). According to the above evidence, we assume that important tumor suppressor genes (TSGs) may be silenced by aberrant hypermethylation in AML, and this effect is potentially reversed by HMA treatment and the sequential inhibition of tumorigenesis.
Gene expression pro ling (GEP) and gene methylation pro ling (GMP) analyses are useful tools to reveal dysregulated genes and pathways; these analyses can be used to compare patients and healthy control, and to screen potential prognostic/diagnostic markers and therapeutic targets. Increasing microarray data have emerged for AML (9,(15)(16)(17)(18)(19). We combined GEP and GMP analysis to reveal the shared genes which were hypermethylated and underexpressed, which may contain potential TSGs. These results facilitated us to nd prognostic markers for AML and potential targets for HMA therapy. We then identi ed the prognostic gene signature, methylation/expression level of which was signi cantly associated with overall survival (OS), using the AML cohort dataset of TCGA database. Moreover, the correlation between survival of patients, who were receiving HMA treatment, and identi ed genes was investigated. Notably, the heterogeneity originating from different cell subpopulations may have resulted in deviation of transcriptome analysis, which used mix cell samples (20). Therefore, combined GEP and GMP analysis can eliminate the deviation of expression-only analysis, because methylation alterations are known to be heritable across cell subpopulations. The overall owchart of the study design was shown in Fig. 1. Based on our assumption, this present study may provide novel insights for gene methylation in AML.

Data source
After the review of all gene expression and methylation pro le datasets of AML patients from the GEO database (https://www.ncbi.nlm.nih.gov/gds/), we selected GSE15061 and GSE58477, because they included the most AML samples among the GEP and GMP array datasets respectively. GSE15061 included 202 AML patients and 69 healthy donors, and the platform of which is Affymetrix Human Genome U133 Plus 2.0 Array. GSE58477 contains 62 AML patients and 10 healthy donors, and the platform of which was Illumina HumanMethylation450 BeadChip. GSE40871 was used for exploring the relation of HMA treatment and identi ed genes, which consisted of methylation array pro ling part (GSE40870, Illumina HumanMethylation450 BeadChip platform) and expression array pro ling part (GSE40442, Affymetrix Human Exon 1.0 ST Array platform). GSE40442 contained 17 decitabine treated primary AML samples and 17 control samples. While GSE40870 consisted of 16 decitabine treated primary AML samples and 16 control samples. For micro-array data, expression level of a gene was calculated as the mean value (M value) of all probe sets annotating to it, while methylation level was measured by beta value.
Gene level 3 RNA-seq data (RSEM normalized count, IlluminaHiSeq , n = 173) and beta value of methylation (HumanMethylation450, n = 194) was downloaded were download from the TCGA database (https://portal.gdc.cancer.gov/), as well as corresponding clinical demographics. The cytogenetic and molecular risk strati cation was based on ELN2017 recommendations (5). The date of data downloading is 2019.12.03.
Identi cation of aberrantly methylated underexpressed genes in AML This process was conducted using the R program (R 3.6.0, Bioconductor 3.10).
The normalized methylation array data (beta value) were downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/gds/). Differentially methylated regions (DMR) were identi ed using the "min " package with the default parameters, by comparing AML and healthy control. The results were ltered with a q value < 0.05 and a mean change between AML and healthy donors of ≤ −0.2. The annotation of functional consequences was conducted by wANNOVAR (http://wannovar.wglab.org/) (21).
The normalized expression data (M value) were downloaded from the GEO database. The "limma" package was used to identify differentially expressed genes (DEGs). For discrimination of DEGs, the false discovery rate (FDR) adjustment based on Benjamin-Hochberg procedure was used, and cutoff p value is 0.05. Genes with a log2foldchange of DEG < 0.5 were de ned as underexpressed genes. The shared genes of hypomethylated DMR and underexpressed DEG were selected by Venn analysis, which were presumed to be aberrantly methylated and silenced.

Functional enrichment analysis of aberrantly methylated genes
To reveal the implication on cell signaling, the ClueGO plugin and Cytoscape software (version 3.7.3) were used to conduct functional enrichment analysis according to the Reactome pathway database and Kyoto Encyclopedia of Genes and Genomes database (KEGG). The "topGO" package and "REVIGO" package (R 3.6.0, Bioconductor 3.10) were employed to perform and summarize the Gene ontology (GO) analysis. The processes and pathways were addressed to the following parameters: right-sided hypergeometric distribution tests, with a cutoff p value < 0.05; Benjamini and Hochberg adjustment was performed for the terms (with Kappa-statistics score threshold set to 0.4; the leading term groups that were selected corresponded to the highest signi cance. The enriched GO/KEGG/REACTOME terms or pathways with an FDR-adjusted p value < 0.05, were de ned to be signi cant.
Identi cation of prognostic markers in AML cohort from TCGA database The key aberrantly methylated and silenced TSGs were inferred to be associated with pathogenic process of AML and would thus predict clinical outcomes. Therefore, we validated the association of survival and methylation/expression level of target genes and screened for the signi cant prognostic genes.
SPSS (version 25.0) was used to perform survival analysis. A Cox hazards model was used to conduct univariable and multivariable analyses. Then univariable analysis was conducted and included vital clinical features (age, race, mutation count, induction type, FAB subtypes, risk strati cation by cytogenetics, risk strati cation by molecular mutations, transplant or not and white blood cell count), expression and methylation level of individual target genes. Multivariable analysis included factors with p value < 0.2 in univariable analysis, the following stepwise selection (forward, likelihood ratio) was used to establish a prognostic model. The factors with a p value < 0.05 were selected for the next step.
Moreover, the AML cohort of TCGA was divided into 2 subgroups according to the type of induction therapy. The HMA subgroup, who received decitabine or azacitidine, included 14 subjects available for methylation-survival analysis and 15 subjects available for expression-survival analysis. While the chemotherapy subgroup, who received intense chemotherapy, included 139 subjects available for methylation-survival analysis and 154 subjects available for expression-survival analysis. Univariable and multivariable analyses were performed using SPSS with the abovementioned method.

Association of genome-wide methylation pro les and SYNJ2 methylation
We investigated the association between methylation level of SYNJ2 and genomic methylation pro le. In AML cohort of TCGA, the median beta value of SYNJ2 was used to divide the patients into SYNJ2 hyperand hypo-methylated groups. The sequential differentially methylated position and DMR analysis was performed between the 2 groups by "min " package and R.
Genome-wide gene/microRNA pro les and SYNJ2 methylation Genome-wide genes and microRNAs expression pro le, associated with SYNJ2 methylation, was investigated based on the data derived from TCGA. The differentially expressed genes/microRNA were identi ed by "DESeq2" package and R, comparing SYNJ2 hyper-and hypo-methylated groups. Then we accessed the impacted of SYNJ2-methylation on cell signaling pathway by gene set enrichment analysis (GESA), based on MSigDB database (http://software.broadinstitute.org/gsea/msigdb) (22)(23)(24). The pathways with p value < 0.05 were de ned to be signi cantly regulated by SYNJ2 methylation. The predicted target genes of DEmiR were obtained by microRNA target predicting algorithm and correlating analysis, via miRWalk 2.0 online tools (http://zmf.umm.uni-heidelberg.de/) (24,25).The interaction network of miRNA and mRNA was established.

Statistical analysis
DEGs and DMRs were identi ed by an FDR-guided Benjamini-Hochberg procedure. For univariable survival analysis, a p value < 0.05 was set as the cutoff value. The factors with a p value < 0.2 were included in the multivariable analysis. The stepwise selection (forward, likelihood ratio) method was used to establish the prognostic model, and factors with p value < 0.05 entered the next step. The characteristics of the HMA and chemotherapy groups were compared by a two-tailed t-test and/or Fisher's test.

Identi cation of aberrantly methylated and underexpressed genes
The clinical and genetic features of patients of GSE15061 and GSE58477 were described in the original article (9). In comparison with the control group, 9954 DEGs and 4536 underexpressed DEGs were identi ed in the AML group of GSE15061. A total of 1139 exonic hypermethylated DMRs were revealed in AML group of GSE58477. 198 shared genes between DEGs and DMRs were found by Venn analysis (the shared genes are list in Supplementary table 1).

Overrepresentation analysis of aberrantly methylated genes
The shared genes were revealed to be signi cantly enriched in the following biological processes (BP): antigen processing and presentation, positive regulation of NK cell cytokine product, etc. Cell component (CC) analysis demonstrated that shared genes were predominantly located in the MHC class I protein complex, recycling endosome membrane, etc. Molecular function (MF) analysis indicated that shared genes were enriched in peptide antigen binding, CD8 receptor binding, etc. Based on KEGG pathway analysis, the shared genes were signi cantly enriched in the following signaling pathways: endocytosis, cell adhesion molecules (CAMs), etc. In Reactome analysis, the shared genes were enriched in neutrophil degranulation, signaling by NOTCH3, etc. The detail results of overrepresentation analysis (ORA) were listed in Supplementary Table 2, meanwhile the top enriched pathways are shown in Figure 2.

Survival analysis
The assumption for the proportional hazards model was tested for each variable. The age variable did not meet the P-H model and a time-dependent Cox hazards model was used to adjust for age. In univariable analysis, the methylation level of 6 genes (out of 198 shared genes) signi cantly predicted unfavorable prognosis ( Table 2): CORO1A (cg06038367, CpG shore), EHD1 (cg01151584, open sea), MPO (cg14619064, CpG island), SLC11A1 (cg07015784, CpG shore), SYNJ2 (cg10977795, CpG island), and GAS2L1 (cg04929173, CpG island). While expression level of MPO and SYNJ2 were also signi cantly associated with OS (Table 2). Notably, we summarized the expression pro les of the 6 genes in previous studies by Oncomine online tools (https://www.oncomine.org/resource/) (Figure 3), indicating that all 6 genes were predominantly underexpressed in leukemia patients compared with their expression in normal controls.
In multivariable analysis, age, race, mutation count, induction type, FAB subtype (M3 or non-M3), risk strati cation by cytogenetics, risk strati cation by molecular mutations, transplant or not, WBC count, methylation level (beta value) of 6 genes, expression level (M value) of CORO1A/MPO/SLC11A1/SYNJ2 were included, according to the results of univariable analysis. The stepwise forward linear ranking (LR) selection was used to screen independent prognostic factors to establish the prediction model. Finally, age, race, risk strati cation of molecular mutations, induction type, transplant or not, WBC count, and methylation level of SYNJ2 (cg10977795) were signi cantly associated with OS ( Table 2). The hazard ratio (HR) of methylation of SYNJ2 was 2.130 (95%CI 1.010-4.492) and the p value was 0.047, indicating hypermethylation of SYNJ2 is an independent unfavorable prognostic factor.
In the following subgroup analysis, the samples were classi ed as "HMA group" and "chemotherapy group" according to the determined induction type. The clinical features of the HMA and chemotherapy groups were compared and shown in Table 1. The patients receiving HMA had a signi cantly higher median age, lower ratio of AML-M3, more mutations, inferior risk strati cation, lower transplant ratio and lower WBC count than the chemotherapy group.
For the HMA group, univariable/multiple analysis was shown in Table 3. Mutation count, risk strati cation of molecular mutations and cytogenetics, expression and methylation level of EHD1 and methylation level of CORO1A were included in multivariable analysis, based on the results of univariable analysis. The Cox hazards model identi ed methylation of CORO1A as the only signi cant favorable factor (p value = 0.012, HR = 0.004), associating with longer OS. For the chemotherapy group, univariable analysis revealed that age, race, FAB subtype (M3 or non-M3), risk strati cation of molecular mutations and cytogenetics, WBC, expression and methylation level of CORO1A/MPO/SYNJ2, methylation level of EHD1/SLC11A1/GAS2L1 were signi cant prognostic factors and included in multivariable analysis. Notably, the expression of CORO1A was reported to be downregulated in CEBPA-mutated AML by Federzoni et al (26). Therefore, we downloaded corresponding CEBPA mutation data from the TCGA database and used CEBPA mutation status as a stratifying factor in the following multivariable Cox hazards analysis. The age (p value = 0.008, HR = 1.004), risk strati cation of molecular mutations (p < 0.05), transplant (p value = 0.012, HR = 0.512), WBC count (p value = 0.011, HR = 1.006), and SYNJ2 methylation (p value = 0.041, HR = 2.544) were assessed. Notably, the methylation of CORO1A was not a signi cant predictor for chemotherapy patients.
Moreover, we classi ed the samples as "CORO1A hypermethylated group" and "CORO1A hypomethylated group" inside the HMA and chemotherapy groups, in comparison with median beta value of each group. Kaplan-Meier plot displayed the opposite impact by methylation of CORO1A in the HMA and chemotherapy groups ( Figure 4A&B). Therefore, CORO1A methylation level may serve as a biomarker for HMA treatment. The AML patients harboring hypermethylated CORO1A may get superior response and survival bene t in HMA treatment instead of chemotherapy.
Furthermore, to explore the association of CORO1A and HMA, GSE40871 dataset was employed, including expression and methylation data of primary AML samples with decitabine or mock treatment. The transcription of CORO1A in decitabine treated samples was upregulated compared with that in the control group (p value = 0.0484, log2foldchange = 0.668), while the expression in cytarabine treated samples didn't vary compared with that in control group, indicating CORO1A may be a substantial target demethylated by HMA. While the methylation analysis found no methylation changes among target genes by comparing decitabine and mock treated group. Lacking direct methylome evidence, the relation of HMA and CORO1A still need validation.
Association between genome-wide methylation pro le and SYNJ2 methylation 74 differentially methylated regions (DMRs) were identi ed to be associated with SYNJ2 methylation signi cantly (FDR-adjusted p value < 0.05, the list of DMRs are shown in Supplementary Table 3). The DMRs included the following vital genes: 1) HOXA9, an oncogenic transcription factor (27); 2) CALR, the mutation of which is often reported in myeloid neoplasms (MPN), but the methylation of which was rarely studied for AML; 3) ABI1, identi ed as a new partner gene for MLL in an AML patient(28) ;4) STAT3, involving in leukemogenesis and immune evasion in AML (29). All abovementioned genes were hypomethylated in SYNJ2 hypermethylated group. The DMRs are associated with SYNJ2 were enriched in Chemokine signaling pathway, Hematopoietic cell lineage, Graft-versus-host disease, etc.
26 upregulated differentially expressed microRNAs (DEmiRs) and 75 downregulated DEmiRs were identi ed as signi cantly associated with SYNJ2 methylations (adjusted p value < 0.05, |log2foldchange| > 1). We selected the AML-related DEmiRs for further exploration, according to previously validated evidence by experiment in vitro or vivo. The low level of miR-217 expression was reported to be associated with unfavorable prognosis in AML (30). The miR-485-3p was reported to regulated expression of DNA topoisomerase and induce chemoresistance via targeting nuclear factor YB, in leukemic cells (31). The miR-889-3p may inhibit myeloid cell nuclear differentiation antigen expression, acting as potentially targets in AML (32). The potential target mRNAs were predicted by miRWalk 2.0 online tools.
An integrated network of DEmiRs, target genes and DEGs was displayed in Figure 8.

Discussion
According to data from The Cancer Genome Atlas Network (2013), the count of genomic mutations tends to be less in AML than in other cancers (33). Moreover, a similar distribution of mutations was observed before and after progression, but methylation pro le shifts and have a functional impact on gene regulation (34), suggesting the increasing signi cance of understanding epigenetic changes in AML. Global methylation patterns are associated with the prognosis of AML, and several studies have explored the prognostic value of gene methylation signatures. Deneberg et al reported that a high level of global methylation predicted a poorer complete remission rate and that p15 methylation was associated with better OS and disease-free survival (DFS) (35). Sun GK et al reported that methylation of the DOK6 promoter predicts longer survival (36). These studies suggested that the methylation level of genes can be valuable prognostic biomarkers. However, systemic functional analysis of the association between gene methylation signatures and disease outcome remains rare.
Via a combination of GEP and GMP, we uncovered the 198 genes shared between DEGs and DMRs, which were inferred as silenced by aberrant methylation in AML. Sequential enrichment analysis revealed shared genes concentrated in potentially related pathways, such as endocytosis etc. In previous studies, endocytosis is a potential target in AML (37,38).
Then, 6 out of 198 genes were identi ed to be signi cantly associated with AML survival by univariable analysis. The important clinical demographics and epigenetic/transcription information were integrated into multivariable Cox hazards model, which made our results more comprehensive and the prognostic model more robust. Age, race, induction type, risk strati cation by molecular mutations, transplant or not and WBC count were traditional prognostic factors in AML included in nal multivariable results. Notably, the mutation count was signi cant in univariable analysis, but was ruled out for prognostic model in multivariable analysis, suggesting genomic alteration might not be the one and only disease-driving factor. Our results implicated that the methylation of SYNJ2 was an independent prognostic factor in the analysis of whole cohort and the chemotherapy group, included in the prognostic model derived by the multivariable Cox hazards model. SYNJ2 is an inositol 5-phosphatase, which may take part in membrane tra cking and endocytosis. The dysregulation of SYNJ2 is described in multiple cancers (39)(40)(41)(42). However, the impact of SYNJ2 has not been fully elucidated in AML. The following functional analysis using TCGA LAML cohort, revealed that hypermethylation of SYNJ2 was associated with the aberrant methylation of some crucial molecular mutations, such as HOXA9/CALR/ABI1/STAT3, etc. HOXA9 is reported as a homeodomain-containing transcription factor, generally upregulated in AML (27) and associated with an unfavorable prognosis (43). In the present analysis, HOXA9 was found to be hypomethylated and overexpressed in SYNJ2 hypermethylated group, consistent with previous studies. Intriguingly, the transcription of another HOX family member, HOXA13, was found to be upregulated in SYNJ2 hypermethylated group signi cantly (log2foldchange = 1.019, adjusted p value = 0.0148). HOXA13 plays a role in promoting metastasis and associating with poor survival in solid tumors. (44)(45)(46). In addition, sequential genome-wide miRNA differential analysis revealed that miR-217 (log2foldchange = 2.171, adjusted p value = 0.000012)/ miR-485-3p (log2foldchange = 1.1704, adjusted p value = 0.0366)/ miR-889-3p (log2foldchange = 2.135, adjusted p value = 0.00017), was underexpressed in SYNJ2 hypermethylated group. All these microRNAs potentially inhibit the expression of HOXA13 based on miRWalk 2.0 online tools, the downregulation of which lead to upregulated HOXA13 in SYNJ2 hypermethylated group. Therefore, direct or indirect upregulation of HOXA family members via epigenetic/microRNA alterations, may account for unfavorable prognosis in SYNJ2 hypermethylated group. CALR was previously extensively studied in MPN for its genomic mutations, but the signi cance of epigenetic and expression alterations in AML has not been well elucidated. ABI1 fuses with MLL protein in AML with speci c cytogenetics, which is also differentially upregulated in SYNJ2 hypermethylated group. STAT3 is a crucial transcription factor activating target genes which promote cell proliferation and prevent cell apoptosis. STAT3 is aberrantly upregulated in many cancers (47), including AML (48). The inhibition of STAT3 lead to not only suppression of tumor growth, but improvement of immune suppressive microenvironment (49)(50)(51). The impact of above genes for AML, explained why the survival is inferior in SYNJ2 hypermethylated group.
The DEGs associated with SYNJ2 methylation were uncovered to be involved in crucial pathways, including cell proliferation and immune evasion. Although only 3% AML harbors JAK2 mutation (52), the phosphorylation of JAK2 and STAT family members were implicated in the majority of AML patients with the prognostic signi cance (53,54). The e cacy and safety of ruxolitinib, a JAK inhibitor, have been validated in preliminary clinical studies (55,56), suggesting that JAK-STAT signaling plays a role in leukemogenesis and prognosis. Constitutive NF-kappaB signaling is discovered in 40% of AML patients, which promotes cell proliferation and resists apoptosis (57,58), thus contributing to leukemogenesis (59). PI3K-Akt signaling pathway was activated frequently in AML (60), and corresponding target drug is a promising therapy (61). In addition to these pathways mainly involving cell proliferation/apoptosis, the PD1-expression and PD-1 checkpoint pathway was also implicated to be activated in SYNJ2 hypermethylated group. PD1 is expressed on human AML cells, which is elevated at disease relapse, and correlated with prognosis (62). A growing body of evidence have shown the e cacy of PD-1 blockage in AML (63,64). The activation of abovementioned signaling pathways may be potential targets of therapy in SYNJ2 hypermethylated AML patients.
In the HMA group, the included patients were older, rarely received transplants and had more unfavorable risk strati cation than the chemotherapy group. Notably, the methylation of CORO1A independently predicted favorable AML survival in multivariable analysis, while signi cant association of CORO1A methylation and AML survival was not found. Furthermore, Kaplan-Meier plotter revealed the opposite impact of CORO1A hypermethylation in the HMA versus chemotherapy group, suggesting that patients with this epigenetic marker may have a greater possibility of gaining bene ts from HMA instead of traditional chemotherapy. CORO1A is a component of the cell cytoskeleton and is also involved in protrusions formation of the plasma membrane (65). The results from the analysis of GSE40871 also suggest that HMA treatment leads to upregulation of CORO1A, which may explain why hypermethylated patients have longer OS in HMA group derived from TCGA data. The above results suggest that CORO1A may act as a potential TSG in AML. However, the methylation part of GSE40871 failed to prove that the methylation level of CORO1A was modi ed in the decitabine group than that in the control group. This is merely indirect evidence explaining the relationship between HMA and CORO1A, further low-throughput validation was needed.
In addition, MPO/EHD1/GAS2L1/SLC11A1 were aberrantly methylated and underexpressed in AML. The under-expression of MPO is reported to be an unfavorable marker for AML, associating with the inferior complete remission rate (66) and worse disease-free survival (67). However, the results were controversial, a non-signi cant association even opposing results were reported. EHD1 regulates membrane reorganization and tubulation in an ATP hydrolysis dependent manner. In vitro studies, EHD1 was involved in vesiculation during endocytosis (68)(69)(70)(71). GAS2L1 regulates crosslinking of microtubes and micro laments and is involved in microtubule dynamics and stability (72,73). The expression of GAS2L1 was differentially expressed in the decitabine treated group and the control group in KG1 cell line, and its expression is diminished in AML patients across all karyotypes (74). SLC11A1 is a divalent transition metal (iron and manganese) transporter. In the present study, hypermethylation of MPO/EHD1/GAS2L1/SLC11A1/GAS2L1 predicted worse OS in the univariable analysis, suggesting that they may be potential biomarker of AML prognosis. The correlation between EHD1/MPO/GAS2L1/SLC11A1 methylation/expression pro le and AML has not been explored previously.
Due to the nature of bioinformatic analysis, our results and conclusions still require variation with a lowthroughput method and clinical evidence. Moreover, other regulatory mechanisms of transcription, such as noncoding RNA, copy number variation etc, were not considered, and the growing data and knowledge will enhance the understanding of aberrantly regulated gene pro les in the future.

Conclusion
In conclusion, we identi ed the aberrantly methylated and silenced genes of AML. Survival analysis facilitated the identi cation of crucial genes, methylation of which are associated with clinical outcome of AML. SYNJ2 was identi ed as an independent prognostic factor. Notably, subgroup analysis found that methylation of CORO1A acted as a potential biomarker for HMA treatment. These ndings added insights into the pathogenic mechanisms of AML and provided potential biomarkers.