A total of 108 primary AML samples and 36 control samples were selected from the Gene Expression Omnibus (GEO) database [GEO accession no. GSE20452 (12), last update, March 22, 2012]. In GSE20452, blasts from patients with AML were obtained at the time of diagnosis. Two batches of experiments were performed and analysed separately. One group of specimens contained AML samples (n=38) and the other contained AML samples (n=70), CD34⁺ progenitor cells (n=21) and white blood cells (n=15) as controls. We used GEO2R (https://www.ncbi.nlm.nih.gov/geo/geo2r/) to compare two groups of samples so as to identify genes that are differentially expressed across experimental conditions. GEO2R terms with corrected P-values <0.05 were considered significant, as were logFC of GEO2R terms >1.0.

Different peak analysis was based on the fold-enrichment of peaks in various experiments. A peak was defined as different when the odds ratio (OR) between two groups was more than 2. Using the same method, genes associated with different peaks were identified and subjected to Gene Ontology (GO) enrichment analysis. We used KEGG Orthology-Based Annotation System (KOBAS) 3.0 online (14) to test the statistical enrichment of peak-related genes in Kyoto Encyclopedia of Genes and Genomes (KEGG) (15–17) pathways. GO terms with corrected P-values <0.05 were considered significantly enriched by peak-related genes.

The STRING (https://string-db.org/) database provides protein-protein interaction (PPI) information, including direct (physical) and indirect (functional) associations (18). Pathways from KEGG and the extended network were constructed for m1A regulators and related protein-coding genes signatures using Cytoscape 3.5.1.

The AML cell line THP-1 was kindly donated by Professor Ravi Bhatia (City of Hope National Medical Center, Duarte, CA, USA). The THP-1 cell line was cultured in Iscoves modified Dulbeccos medium or RPMI-1640 medium (Invitrogen; Thermo Fisher Scientific, Inc.) with 10% foetal bovine serum (FBS) at 37°C in a humidified incubator containing 5% CO₂.

Chidamide was kindly donated by Shenzhen Chipscreen Co. AML cells were treated for 24 h. For THP-1 cells, the dose of chidamide was 0.5 µM (20).

Chromatin immunoprecipitation (ChIP) experiments were performed as previously described (13,21), and ChIP-seq was based on the Illumina Technology Sequencing platform (Illumina, Inc., USA). The single/paired-end method (13,21) was used to complete the ChIP-seq sequencing analysis of the THP-1 cell line. The antibody against histone was H3K9me3 (cat. #4260, RRID: AB_10828006; Cell Signaling Technology, Inc.).

The significances of differences were calculated using the moderated t-statistic (only available when two groups of samples were defined). P-values were adjusted for multiple testing. Genes with the smallest P-values were considered the most reliable. P<0.05 after adjustment was considered statistically significant. Log2-fold change between two experimental conditions (only available when two groups of samples were defined) was carried out. Moderated F-statistic combined the t-statistics was conducted for all pair-wise comparisons into an overall test of significance for that gene (only available when more than two groups of samples are defined).

When samples were collected from the GEO database (accession no. GSE20452), the differential expression of genes between AML samples and control samples was analysed using GEO2R. There were more than 2,000 genes that showed significant differential expression. The definition of the value in log(FC) for upregulation was >1.0, and the definition of the value in log(FC) for downregulation was <-1.0. According to the definition of upregulation or downregulation together with P<0.05, there were 147 genes related to alterations in H3K9me3 showing downregulation and 170 genes related to the change in H3K9me3 showing upregulation (data not shown).

The function of genes collected from GEO were further analysed using geometric mean of semantic similarities in biological processes (BPs), molecular functions (MFs), and cellular components (CCs), that were assessed using the GOSemSim package (22) by considering the GO topological structure in a more precise and unbiased manner. In BPs, there were 20 processes that showed significant changes (Fig. 1A). The main processes were related to metabolism, including cofactor metabolic process, cellular metabolic processes, and others. In MFs, there were only three significantly altered processes: Protein binding, binding, and molecular-function (Fig. 1B). In CCs, there were 20 processes that showed significant alterations, including intracellular and extracellular processes (Fig. 1C). The relationship of GO items in the regulation of apoptosis and autophagy are shown in Fig. 2A and B. These GO items were included in BPs, MFs, and CCs.

The related pathways were predicted using KOBAS 3.0 online. One of these pathways, the longevity pathway (Fig. 3), may related to the survival of leukaemia cells, including the PI3K/AKT pathway, the AMPK pathway, the FOXO pathway, the P53 pathway, and others. During analysis, SIRT1 was also found to be related to the AMPK pathway and autophagy pathway, consistent with findings from our previous study (23).

A total of 317 regulator genes, together with lysine-specific demethylase 4A (KDM4A), a regulation protein for the H3K9me3 network containing 220 nodes and 202 edges, were obtained from the STRING online database and Cytoscape software. The PPI network showed detailed protein interactions; however, the PPI enrichment P-value was above 0.05, suggesting that the network did not have significantly more interactions than expected. Nevertheless, the modulation enzyme lysine-specific demethylase 4A (KDM4A) for H3K9me3 could have an interaction with other proteins in the network by HISTIH4C (Histone cluster 1, H4c, a modulation enzyme for histone H4). There were also many protein interactions that could help identify the mechanism of H3K9me3 modulation (Fig. 4).

The ChIP-seq of H3K9me3 was performed in AML cells in which levels of SIRT1 were inhibited by chidamide; this was conducted to identify genes that may be involved in the mechanisms of cell death. The peaks almost clustered in the gene and intergenic areas (nearly 99%); others clustered in areas such as CDS and D10K20K (Fig. 5A).

GO analysis showed that the peak of processes that had significant changes for AML cells treated by chidamide were BPs and MFs. In the BPs, the most significant peaks involved metabolic processes such as the macromolecule metabolic process and cellular macromolecule metabolic processes. In the MFs, the most significant peaks involved metal ion binding or cation binding. KEGG analysis also revealed that the richest peak involved metabolic processes (Fig. 5B).

Network analysis of GO enrichment in AML cells treated by chidamide were in BPs. Units responsive to stress in directed acyclic graph (DAG) had the most significant changes, especially for the DNA repair and cells responsive to stress. In the CCs, the ribosome unit showed the most significant change; other units also showed significant changes, including cytoplasm and intracellular ribonucleoprotein. In the MFs, phosphotransferase, nucleotide kinase, and adenylate kinase showed the most significant changes.

Our previous research (13) suggested that the level of H3K9me3 could be affected by a low dose of chidamide, thus we aimed to ascertain in the present study whether the network of GO enrichment in AML cells similar to AML patients could be affected by chidamide. This result may be evidence for further research concerning the use of chidamide in AML patients. An interesting result was that the network analysis of GO enrichment in AML cells treated with vhidamide were similar in AML patients. In GO0050789, GO0050794, GO0050896, and GO0080090, all of which are related to BPs, GO enrichment showed significant changes in AML patients compared with control groups; the same also appeared in AML cells treated with chidamide (Fig. 5E), which was different from THP-1 cells without chidamide treatment (Fig. 5C). In the CCs, GO enrichment such as GO005623, GO005622, and GO0044444 showed significant changes in AML patients as well as in the AML cells treated with chidamide. However, in MFs, only GO005488 showed significant changes in AML patients and in AML cells treated with chidamide (Fig. 5F), which was different from THP-1 cells without chidamide treatment (Fig. 5D).

The network of analysis for AML cells treated with chidamide showed several results of GO enrichment that were the same as the results of GO enrichment obtained from the patients in GSE20452. Our previous study showed that SIRT1 (located on chromosome 10) was in the domain of H3K9me3 on chromosome 10 that was downregulated by chidamide, suggesting there may be interaction between SIRT1 and the regulation of H3K9me3 (17). A total of 317 regulator genes together with SIRT1, containing 219 nodes and 209 edges, were obtained from the STRING online database and Cytoscape software. The PPI network showed detailed protein interactions; however, the PPI enrichment P-value was above 0.05, suggesting that the network did not have significantly more interactions than expected. Nevertheless, SIRT1 had a direct interaction with several proteins in the network compared with KDM4A; however, there were also many other protein interactions that were identified so as to uncover the relationship between SIRT1 and regulatory genes acting on H3K9me3 (Fig. 6A).

Our previous study (13) showed that the location of SIRT1 on chromosome 10 was in the domain of H3K9me3, suggesting there also may exist interactions between SIRT1 and KDM4A. The gene SIRT1, together with KDM4A, contained eight nodes and seven edges, obtained from STRING online database and Cytoscape software. The PPI network showed the detailed protein interactions, and the PPI enrichment P-value was below 0.05, suggesting that the network had significantly more interactions than expected. Nevertheless, there were no direct interactions between SIRT1 and KDM4A. The protein M-phase phosphoprotein 8 (MPHOSPH8) may be the bridge for SIRT1 and KDM4A, and interactions between MPHOSPH8 and KDM4A need to be demonstrated (Fig. 6B).