HDAC2‐dependent miRNA signature in acute myeloid leukemia

Acute myeloid leukemia (AML) arises from a complex sequence of biological and finely orchestrated events that are still poorly understood. Increasingly, epigenetic studies are providing exciting findings that may be exploited in promising and personalized cutting‐edge therapies. A more appropriate and broader screening of possible players in cancer could identify a master molecular mechanism in AML. Here, we build on our previously published study by evaluating a histone deacetylase (HDAC)2‐mediated miRNA regulatory network in U937 leukemic cells. Following a comparative miRNA profiling analysis in genetically and enzymatically HDAC2‐downregulated AML cells, we identified miR‐96‐5p and miR‐92a‐3p as potential regulators in AML etiopathology by targeting defined genes. Our findings support the potentially beneficial role of alternative physiopathological interventions.

Acute myeloid leukemia (AML) is a multifactorial and highly heterogeneous malignancy, whose incidence rises with age [1]. The evolution of the disease is characterized by uncontrolled progenitor cell proliferation and block of differentiation. To date, although many genetic mutations in AML have been identified, prognosis has markedly improved in recent years but still remains poor [2]. It is well known that epigenetic mechanisms regulate gene expression and consequently define pathways involved in the physiopathogenesis of AML. Among the many epigenetic regulators, histone deacetylases (HDACs) are tightly involved in AML etiology [3]. Notably, HDAC2 is highly overexpressed in solid and hematological cancers, including AML [4][5][6][7][8][9]. HDAC2 silencing by enzymatic inhibition has a substantial impact on leukemia cell proliferation and immune regulation, as described in our previous work [10], where we established an HDAC2-knockdown AML clone to better understand the role of cancer cell proliferation dynamics with and without treatment with the wellstudied HDAC inhibitors (HDACi) suberanilohydroxamic acid (SAHA) and entinostat (also known as MS-275). Several epigenetic drugs, including HDACi, are in fact actively undergoing clinical investigation as single agents or mainly in combination with consolidated chemotherapeutics [11]. Similarly, miRNAs are now recognized as epigenetic regulators of transcripts in nearly all physiological processes and human cancers, including AML [12]. The key involvement of miRNAs in crucial biological pathways hints at their functional role in complex molecular gene networks in cancer. Recently, the potential use of cellular and circulating miRNAs as biomarkers for AML diagnosis/prognosis, and as therapeutic targets has been widely explored, and many miRNAs were found to be associated with HDAC2 dysfunction in leukemia [13].
Here, we identified a cluster of common up-and downregulated miRNAs in both SAHA-treated and HDAC2-downregulated cells. By miRNA target network computational analysis, we defined an HDAC2mediated miRNA signatures in AML by genetic and enzymatic HDAC2 deficiency in a U937 leukemic cell line. We propose a crucial role of miR-96-5p and miR-92a-3p and related target genes and their relationship with HDAC2 in AML. Here, we corroborated our previous findings and strongly suggested an HDAC2mediated regulation of the immune system in AML, involving major histocompatibility complex (MHC) class II genes and specific miRNAs, via finely tuned molecular mechanisms.

Stable transfection of sh2 vector
Silencing of HDAC2 in U937 cells was performed as previously described [10].

RNA isolation and miRNA expression analysis
Total miRNA-enriched RNA was isolated and miRNA expression levels were analyzed by real-time PCR as previously reported [14].

Gene set enrichment and functional annotation analysis
The relative abundance of 'Biological Process' (BP), Pathways (by KEGG), oncogenic and immunologic signatures Gene Ontology terms in each of the selected lists was analyzed using the Molecular Signatures Database v6.2 (MSigDB) in Gene Set Enrichment Analysis (GSEA) software (http://software.broadinstitute.org/gsea/msigdb) for Annotation, Visualization and Integrated Discovery.
Gene expression microarray profiling and data analysis using the Agilent platform Gene expression profiles of U937 scramble vector (scr) and HDAC2 knockdown (shHDAC2) cells were analyzed by Whole Human Genome Two-Color Microarray (G4112F; Agilent), following the manufacturer's protocol. Microarray data are available in the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/gds) under the accession number GSE37529 [10]. Probe-level raw intensity was processed using R/BioConductor and limma package. Background correction was performed using 'normexp' limma method and data normalization was carried out in two steps: LOWESS normalization within array to correct systematic dye bias and quantile normalization between arrays to detect systematic nonbiological bias. Ratios representing the relative target mRNA intensities compared to control RNA probe signals were derived from normalized data. For each P-value, the Benjamini-Hochberg procedure was used to calculate the FDR in order to avoid the problem of multiple testing.

Results
Differentially expressed miRNA profiling in HDAC2-defective AML We built on our previous epigenetic study [10] by evaluating the impact of HDAC2 deficiency on miRNA expression in AML. miRNome analysis was performed in HDAC2-silenced (shHDAC2) and relative scramble control (scr) U937 stable clones, previously obtained and retested for mRNA and HDAC2 protein expression levels (Fig. 1A,B). In addition, we treated scr cells with the well-known HDACi SAHA for 6 h (scrSA-HA6h) to enzymatically mimic HDAC2 silencing. Volcano plots display differentially expressed miRNAs in shHDAC2 and scrSAHA6h compared to scr cells ( Fig. 1C,D). Student's t-test analysis revealed the presence of 29 and 14 differentially expressed miRNAs in shHDAC2/scr and scrSAHA6h/scr cells, respectively, by applying an FDR significance threshold < 0.05 (Tables 1 and 2). Comparative analysis showed that 11 miRNAs are commonly regulated both when HDAC2 is genetically silenced and enzymatically inhibited ( Fig. 2A). Among these, miR-801 and miR-923 were  excluded because they were removed from the miR-Base (v22) [16]; miR-801 appears to be a fragment of U11 spliceosomal RNA, while miR-923 seems to be a fragment of the 28S rRNA. The nine miRNAs altered (three downregulated and six upregulated) in each HDAC2-defective condition are shown in Fig. 2B. We speculate that this cluster of miRNAs may suggest an HDAC2-dependent miRNA signature.

miRNA target networks and enrichment analysis
To predict miRNA targets, we interrogated miRNet (https://www.mirnet.ca/) [17] and identified 1711 predicted targets of the three downregulated miRNAs (Table S1), and 1418 predicted targets of the six upregulated miRNAs (Table S2). To investigate the biological functions, regulatory mechanisms, and disease relevance of differentially expressed HDAC2-dependent miRNAs and their relative target genes, we used MSigDB (v6.2) software applying an FDR q-value significance threshold < 0.001. Figure 3A,B shows the biological processes of predicted target genes of the three downregulated and six upregulated miRNAs in shHDAC2/scr and scrSAHA6h/scr cells, respectively. HDAC2 dysfunction in AML cells after both genetic and enzymatic downregulation is in line with the biological processes in terms of cell cycle regulation, cell and protein localization, and response to organic substance (i.e., the HDACi SAHA). We also looked for the immunologic signature, consisting of gene sets representing cell types, conditions and alterations within the immune system, in predicted target genes of the three downregulated miR-NAs (Table 3) and six upregulated miRNAs ( Table 4). As shown in Tables 3 and 4, many genes involved in immunoregulatory mechanisms are perturbed, further confirming our previous finding that the immune system is affected in an HDAC2-defective AML clone (in both genetic and enzymatic conditions).

Identification of HDAC2-dependent miRNA targets
To validate target prediction analysis, we performed a very stringent intersection analysis between the commonly altered genes in shHDAC2/scr and scrSA-HA6h/scr cells (GSE37529) [10] (Table 5) and the predicted miRNA hits. Among the targets of the six upregulated miRNAs, we identified five predicted gene targets (TRIB3, SLC37A3, EMP1, SCD, IL1B) also regulated in gene expression profiles of shHDAC2/scr and scrSAHA6h/scr cells (Fig. 4A). Only one downregulated hit corresponding to TRIB3 gene displayed an according trend compared to the related regulating miRNA, mir-96-5p. Figure 4B shows TRIB3 microarray expression fold change (FC) in log2 in shHDAC2 and scrSAHA6h compared to scr cells. In contrast, distinguishing between the targets of the three downregulated miRNAs, we identified four predicted gene targets (SLC37A3, TBC1D8, SCD, FAM49A) regulated in gene expression profiles of shHDAC2/scr and scrSAHA6h/scr cells (Fig. 5A). SLC37A3, TBC1D8, and FAM49A are upregulated hits targeted by miR-92a-3p. Figure 5B shows the microarray expression FC in log2 of three upregulated target genes in both shHDAC2 and scrSAHA6h compared to scr cells. SCD was excluded as it showed a different trend in the two conditions (FC = À1.03 in scrSA-HA6h/scr; FC = 1.46 in shHDAC2/scr). These data are in line with trends in mRNA regulation, target prediction, and miRNA expression levels.

Validation of miRNAs and target genes in HDAC2-defective U937 cells
Following miRNA microarray profiling and computational prediction of miRNA target genes, we investigated the expression levels of miRNAs and their corresponding target genes. We analyzed miR-92a-3p and miR-96-5p expression levels by real-time PCR (Fig. 6A). Gene expression levels of TRIB3, a target of miR-96-5p, were analyzed in scr and shHDAC2 cells as well as in scr cells untreated or treated with SAHA for 6 h. TRIB3 relative expression was downregulated in both HDAC2-deficient and scr U937 cells treated with SAHA, suggesting a correlated response due to HDAC2 enzymatic inhibition and genetic silencing (Fig. 6B). The expression levels of SLC37A3, FAM49A, and TBC1D8, target genes of hsa-miR-92a-3p, were also analyzed (Fig. 6C). According to miRNA expression, all target genes were upregulated in shHDAC2 as well as in scrSAHA treated cells, related to scr. Notably, miRNA-mRNA regulation in HDAC2-downregulated cells was comparable after both enzymatic and pharmacological inhibition by SAHA, supporting the hypothesis that upregulated expression levels of HDAC2 indicate dysfunction of these regulators in AML.  Discussion miRNAs are known to play a critical and functional role in a broad range of key molecular processes via sophisticated regulation of distinct targets, orchestrating a molecular intracellular balance of gene expression. miRNA activity and expression are affected in cancer. Altered miRNAs in AML are involved in a variety of biological pathways [18], and a better understanding of their signatures might help unravel the complexity associated with the emergence of this disease. Since HDAC2 deregulation affects cell proliferation, apoptosis, and immune system in AML, focusing on specific miRNAs altered by this epigenetic regulator may identify potential markers for determining the best strategies in AML treatment. To date, many miR-NAs were found directly targeted by HDAC2 in several cancers such as colorectal cancer [19], hepatocellular carcinoma [20], breast cancer [21], and AML [22]. In AML, differentially expressed miRNAs have a prognostic and functional role associated with cytogenetics, molecular features, molecular markers, morphology, and clinical outcome [23]. In a previous study, we found that HDAC2 gene is considerably upregulated in AML ex vivo patient samples and cell lines. Following HDAC2 silencing and enzymatic inhibition using the epigenetic-based drug SAHA, we observed a pivotal HDAC2-dependent modulation of chromatin architecture leading to transcriptional changes promoting mainly activation of an immune response. Specifically, HDAC2 acts directly at epigenetic level by regulating the promoter regions of specific allelic forms of MHC class II genes (HLA-DRA and HLA-DPA1). Here, based also on our previous findings, we elucidated miRNA-HDAC2 crosstalk and its involvement in AML state. Interestingly, a stringent computational analysis between the transcriptome and miRNome profiles in shHDAC2/scr and scrSAHA6h/ scr cells identified a crucial role for miR-96-5p and miR-92a-3p, and defined their target gene regulation enclosing convergent pathways already identified in our previous work. We investigated upregulated miR-96-5p, which has an oncogenic role in several cancer types [24][25][26]. Low expression levels of miR-96-5p were found in a specific cohort of AML patients, suggesting that this epigenetic marker could be considered a prognostic factor for AML at diagnosis [27]. miR-96-5p upregulation acts as an antiapoptotic factor in bladder cells, as it negatively regulates specific targets such as CDKN1A, which is involved in cell cycle regulation and DNA damage pathway in bladder cancer [28]. Our data show that TRIB3 is targeted by miR-96-5p. The protein encoded by TRIB3 gene is a potential kinase that negatively regulates NF-kB and Akt1 pathways, affecting cell proliferation and apoptosis, and promoting ubiquitination-dependent degradation of several key proteins. TRIB3 is strongly expressed in AML with t(8;21) and t(15;17) translocations, as well as in M2/M3 AML subtypes [29], although its specific role in leukemogenesis is still elusive. However, evidence suggests that TRIB3 contributes to acute promyelocytic leukemia progression by PML-RARa stabilization via specific binding to SUMOylation motifs, thereby acting on PML-RARa degradation and differentiation [30]. In addition, we found and subsequently investigated the downregulation of miR-92a-3p targeting SLC37A3, TBC1D, and FAM49A genes. High expression levels of miR-92a-3p are associated with acute megakaryoblastic leukemia and affect genes controlling apoptosis and cell proliferation [31]. High expression levels of this miRNA were also found in both AML and acute lymphoblastic leukemia cells compared to normal blasts [32]. We identified SLC37A3 as one of the targets of miR-92a-3p. This transmembrane protein is localized in the endoplasmatic reticulum and is involved in sugar transport. The SLC37A3 gene is one of the four sugar-phosphate  exchanger family members, but its functional activity is not yet clear. Evidence suggests that SLC37A3 might be involved in physiopathological regulation in pancreatic but also in immune system [33]. The methylation levels of this gene affect glucose blood degree, suggesting its potential role in epigenetic modifications via a mechanism that still requires further investigation, and its involvement in obesity-related metabolism. We also identified FAM49A as a miR-92a-3ptarget by computational analysis. This target gene was detected as a downregulated protein in bladder cancer cells [34]. FAM49A is a consensus PU.1-activated target gene. PU.1 is an E26 transformation-specific family transcription factor widely involved in hematopoiesis. FAM49A is a direct functional regulator of myeloid, dendritic cell, B cell and a differentiation factor of earliest stages of T-cell and terminal erythroid cell [35]. The third hsa-miR-92a-3p target which we investigated is TBC1D8. This gene is a member of the Tre2/Bub2/ Cdc16 (TBC) domain protein family, characterized by highly conserved TBC domains [36]. TBC1D8 was found among differentially expressed genes in pre-B acute lymphocytic leukemia samples with ALL1/AF4, E2A/PBX1, and BCR/ABL molecular rearrangements, and positively controls cell proliferation [37]. Other studies identified TBC1D8 as a target of IL4 in chronic lymphocytic leukemia and normal B cells [38]. In this work, we propose the existence of a mechanistic crosstalk between miRNAs and HDAC2 in an epigenetic superstructure regulating pathogenesis and progression of AML. All our findings converge in identifying HDAC2 and miRNA interplay in specific biological processes (Fig. 6), which potentially affects regulation of gene expression, cell cycle, apoptosis, response to stress and response to organic substance. These mechanisms are robustly altered in leukemogenesis, further confirming our previous findings. We mostly speculate on the immunologic signature of predicted target genes of both downregulated and upregulated miRNAs (Tables 4 and 5) identifies immune cells such as CD4 and CD8T in a specific gene set. It is not surprising that some hits were also associated with type I diabetes, as MHC class II genes are related to this disease [39]. Since MHC class II genes regulate initiation of immune response, our data characterize a specific signature also involving endothelial, thymic, epithelial, and B cells in gene sets. The epigenetic drug SAHA was used as a therapeutic agent in this and in our previous study. This drug plays a critical role as an immunomodulating agent by enhancing cancer cell immunogenicity. We and other authors reported that HDACi make cancer cells more responsive to immunotherapy by increasing the expression levels of tumor antigens, and drive gene expression toward a proapoptotic mechanism in cancer [40,41]. Finally, taken together, our findings identified miR-96-5p and miR-92a-3p as prospective epi-regulators in AML. This HDAC2-dependent miRNA signature in AML highlights the potentially beneficial effects of treatment with epigenetic drugs alone or in combination with other therapies (including immunotherapy) acting via a targeted mechanism involving the perturbation of genes affecting cell cycle, proliferation, apoptosis, and immune system. To date, achieving greater insights into leukemogenesis has allowed us to make progress toward the prevention and treatment of this devastating disease. Given that many disease agents including those that are not strictly biological (such as smoking, obesity, exposure to certain types of radiation or other substances) are not always controllable and vary continuously throughout a person's lifetime, the need for multifaceted therapeutic approaches is imperative.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Table S1. Predicted targets of three downregulated miRNAs obtained from miRNet database (n = 1711). Table S2. Predicted targets of six upregulated miRNAs obtained from miRNet database (n = 1418).