The experimentally validated dataset of human miRNAs and their target genes was collected from the Tarbase (http://diana.imis.athena-innovation.gr/DianaTools/index.php?r=tarbase/index) (20), miRTarBase (http://mirtarbase.mbc.nctu.edu.tw/) (21) and miRecords (http://c1.accurascience.com/miRecords/) (22). The dataset was defined as set U₁.

The dataset of human TFs regulating miRNAs was acquired from TransmiR (http://www.cuilab.cn/transmir) (23) and was considered as set U₂.

The host genes of human miRNAs were extracted from miRBase (http://www.mirbase.org/) (24) and NCBI. The dataset of host genes including miRNAs was defined as set U₃.

Genes and miRNAs are divided into two classes of abnormally expressed and related by the importance of their roles in AA, while abnormally expressed elements are involved in related elements. Abnormally expressed genes and miRNAs of AA were collected from published literature. There are numerous databases of abnormally expressed genes and miRNAs such as the NCBI SNP database (http://www.ncbi.nlm.nih.gov/SNP/), KEGG (www.genome.jp/kegg/pathway.html), Cancer GeneticsWeb (http://www.cancerindex.org/geneweb/clink30.htm) and mir2Disease (http://www.mir2disease.org/). However, the associated materials of AA were not found in these databases, thus indicating that the study of AA is lacking and requires further investigation.

The datasets of abnormally expressed genes and miRNAs were separately considered as set U₄ and U₅.

All related miRNAs were found in pertinent literatures. Related genes which affect tumor growth, migration, radial therapy and clinical outcome of AA came from three data sources including published literature, GeneCards database (http://www.genecards.org/) (25) and P-match method. GeneCards is a database of human genes that provides concise genomic related information on all known and predicted human genes (25) and the first 100 genes associated with AA were selected. Furthermore, an algorithmic P-match method, which combines pattern matching and weight matrix approaches to identify transcription factor binding sites in DNA sequences (19), was used to predict TFs that were considered to be related genes. Prior to using the P-match method, 1,000 nt promoter region sequences of targets of abnormally expressed miRNAs were downloaded from UCSC database (http://genome.ucsc.edu/) (26). Subsequently, the P-match method was used to identify transcription factor binding sites (TFBSs) in 1,000 nt promoter region sequences and mapped TFBSs onto matrix library in TRANSFAC database (http://www.gene-regulation.com/pub/databases.html). Finally, the TFs of DNA sequence were acquired. The vertebrate matrix was used and restricted to high quality matrix criteria.

Datasets of related genes and miRNAs in AA were separately considered as set U₆ and U₇. Symbols of the genes in this paper were checked against the NCBI database (http://www.ncbi.nlm.nih.gov/gene).

In this study, three networks of transcription processes related to AA were constructed, including abnormally expressed, related and global networks. Sets U₁, U₂ and U₃ were combined to derive the global network, thus indicating that the global network includes all the experimentally validated relations between miRNAs and their targets, TFs regulating miRNAs and miRNAs located on host genes.

The abnormally expressed network was derived by mapping abnormally expressed genes and miRNAs in addition to their relations from the global network. If a gene and miRNA pair contained separately in sets U₄ and U₅ was found in the global network to form a certain relation, the elements and the relation were included in the abnormally expressed network.

Datasets U₆ and U₇, that separately represent related genes and miRNAs, were organized into an AA-related network using the same method of referring to the relations in the global network. The related network is also a mapping of the global network and it includes the abnormally expressed network.

Finally, the software Cytoscape (version 3.0.0; http://www.cytoscape.org/) was used to realize network visualization, which assisted in analyzing the regulatory pathways graphically.

Using the methods and datasets mentioned above, three networks were successfully constructed. The abnormally expressed network was a core network that displays the regulatory mechanism of AA. The abnormally expressed network contains molecular elements that could be used in the diagnosis and treatment of AA, in addition to numerous crucial regulatory relations between the elements. This network indicates the pathogenesis of AA in terms of the regulatory relations between abnormally expressed genes and miRNAs.

The abnormally expressed network is composed of 15 abnormally expressed miRNAs, six abnormally expressed genes and 18 host genes, including three types of relations between miRNAs targeting genes, TFs regulating miRNAs and host genes, including miRNAs. Other abnormally expressed genes and miRNAs that do not exhibit regulatory relations with others were not included in this study, though each of these may have significance. The differences in these elements are presented in Fig. 1 according to the existing experiments. To investigate the changes in genes and miRNAs in abnormally expressed network contributes much to the network analysis.

Firstly, abnormally expressed TFs and miRNAs are analyzed. There are three TFs in Fig. 1; PTEN, TP53 and EGR1. PTEN and TP53 are crucial genes in AA. In patients with AA, PTEN and p53 mutations have been significantly associated with reduced and prolonged survival, respectively (27). The three TFs collectively regulate five abnormally expressed miRNAs (miR-21, miR-25, miR-155, miR-335 and miR-23a) and are targeted by a total of five miRNAs (miR-25, miR-16-1, miR-16-2, miR-21 and miR-106b). EGR1 regulates miR-335 and miR-23a, and is not itself targeted by any known miRNA. PTEN regulates miR-21 and miR-25 while being targeted by miR-21 and miR-106b. TP53 is targeted by miR-16-1, miR-16-2 and miR-25 while regulating miR-155. Genes that are targeted by miRNAs, but do not regulate any miRNAs, may be involved in AA. For example, TNC is the target gene of miR-335 and regulates no further miRNAs as successors. TNC may be involved in astrocytoma invasion, gliomagenesis, cell motility, proliferation and the neoplastic transformation of AA, and the upregulation of TNC may indicate the presence and metastasis of AA (28). PTEN and miR-21 form feedback loops (FBLs), via PTEN regulating the expression of miR-21, with miR-21 in turn targeting PTEN. Certain genes and miRNAs constitute a multistage regulation pathway, which proceeds as follows: miR-21 or miR-106b → PTEN → miR-25 → TP53 → miR-155. Any abnormal expression of the elements in this pathway may influence the successors and impact the carcinogenesis of AA.

Furthermore, miRNAs that exclusively targetted genes have been investigated. In Fig. 1, miR-16 (miR-16-1 and miR-16-2) and miR-106b belong to this category of miRNAs, which are not regulated by TFs. miR-16-1 and miR-16-2 target TP53, H3F3B and NOTCH2, indicating that the two miRNAs have some similarity of function. miR-106b targets PTEN and indirectly influences the successors. This type of miRNA may result in the abnormal regulatory network, functioning as start nodes.

Finally, host genes are included in the abnormally expressed network when the corresponding miRNAs are abnormally expressed. In Fig. 1, a host gene could contain multiple miRNAs, with different miRNAs being produced by a single host gene. miR-196b has five host genes, including HOXA9, HOXA10, HOXA-AS4, HOXA10-HOXA9 and MIR196B. MCM7 may be a crucial host gene in AA, as it includes three miRNAs (miR-25, miR-106b and miR-93). miR-25 and miR-106b function in the PTEN regulation pathway. miR21, which contains the key miRNA hsa-miR-21, is similarly important. miR-16-1 and miR-16-2 are from different host genes, DLEU2 and SMC4, though the two miRNAs have the same targets in AA.

Furthermore, investigation of the changes of abnormal genes and miRNAs may be of clinical relevance. The genes in Fig. 1 show abnormally expressed patterns of mutation, upregulation and downregulation when compared to normal tissues. The abnormally expressed patterns of miRNAs include up regulation and down regulation. The majority of the miRNAs serve oncogenic function in the AA abnormally expressed network. Modifying the abnormal expression of genes and miRNAs may correct faulty regulation and prevent tumorigenesis. EGR1, which is upregulated in AA (28), regulates miR-335. Furthermore, the upregulation of miR-335 (29) in AA may incorrectly target TNC and lead to the upregulation of TNC. The pathway indicates potential approaches for the therapy of AA by adjusting the abnormal expression in EGR1, miR-335 and TNC to normal expression levels. Mutation of PTEN (27) may lead to the upregulation of miR-25 (2), while mutation of TP53 (30) may result in the upregulation of miR-155 (2). In addition, Guan et al indicated that elevated miR-196 expression levels were significantly associated with poor survival in patients with AA (17). These findings suggest that the mutation, upregulation or downregulation of genes and miRNAs may contribute to tumorigenesis, and may have diagnostic and prognostic value. If the regulatory network were adjusted to a normal expression pattern, AA may hypothetically be prevented and cured. Further studies are required with the aim of correcting the abnormal expression to normal expression, based on the abnormally expressed network.

The related network contains the abnormally expressed network, and also contains additional regulatory relations not present in the abnormally expressed network. Fig. 2 displays 40 genes, including 10 abnormally expressed genes and 30 related genes (not abnormally expressed), 30 miRNAs including 11 abnormally expressed miRNAs and 19 only related miRNAs. The abnormally expressed TFs are able to regulate additional related miRNAs, and abnormally expressed miRNAs could target at related genes. The additional related elements interact with abnormally expressed elements as well as related elements. In the related network, EGR1, an abnormally expressed TF, regulates two more related miRNAs, including miR-106a and miR-24-2. TP73 is a related TF that shares a high degree of similarity with TP53 in its primary sequence and functions (30), and is involved in regulating a related miRNA miR-143. Related genes such as CCND1, CDKN1A, CDKN1B and VEGFA are targetted by abundant miRNAs. The AA-related network shows increased topology relations compared with the abnormally expressed network and aids in further understand of the pathogenesis in AA.

The global network contains all of the regulatory relations between genes and miRNAs. It is an experimentally validated biological network in the human body. It is just used as a reference of the abnormally expressed network and related network.

The predicted TFs were selected using the P-match method mentioned above. They are considered to be related genes and may be involved in the transcriptional regulation of AA. Fig. 3 displays regulatory relations between seven predicted TFs, seven abnormally expressed miRNAs and four abnormally expressed target genes in AA. A total of five predicted TFs including REL, RELA, NFKB1, E2F1 and E2F3 have the possibility to co-regulate seven abnormally expressed miRNAs, and these miRNAs target abnormally expressed genes and predicted TFs. Among the targets, PTEN, NOTCH2, H3F3B and TP53 are abnormally expressed genes and can be indirectly regulated by the predicted TFs. E2F1 and E2F3 show features in common; both regulate miR-16, miR-106b and miR-15b while being targeted by miR-106b. Furthermore, E2F1 regulates miR-25 and miR-93 and is targeted by miR-21 and miR-93 more strongly than E2F3. E2F1 and miR-106b, E2F1 and miR-93, E2F3 and miR-106b form FBLs. miR-21 is regulated by REL, RELA and NFKB1 and targets at E2F1, E2F2 and PTEN. RELA and NFKB1 co-regulate miR-155 and miR-21. Interactions between TFs and abnormally expressed miRNAs and targets may be involved in the regulatory mechanism underlying AA. Furthermore, the validated changes of abnormally expressed miRNAs and targets may aid in the investigation of predicted TFs in future studies.

The abnormally expressed genes are crucial in pathogenesis of AA. Nodes were classified according to the regulatory relation of adjacent nodes in three level networks, thus comparing and analyzing the regulatory pathway of each abnormally expressed gene. A total of 43 abnormally expressed genes were classified into five classes. The first class of abnormally expressed genes has six types of adjacent nodes (three successors and three precursor nodes), including TP53 and PTEN. Between 50 and 60% of AAs manifested TP53 mutations (31), which has significant interactions between the TP53 and miRNAs in three networks.

The predecessors and successors of TP53 are presented in three levels in Table I. It shows that miR-16 and miR-25 target at TP53 and TP53 regulates miR-155 in the abnormally expressed network. In the related network, TP53 additionally regulates miR-143 and miR-34a. Experimentally validated miRNAs in global network support these regulation relations. The predecessors of TP53 may influence the successors by affecting TP53. Furthermore, TP53 is able to indirectly influence more elements downstream via successors. Similarly, TP53 can be controlled by other elements upstream through the predecessors. For example, PTEN has indirect influence on TP53 by regulating miR-25 in the abnormally expressed network.

The second class of abnormally expressed genes has three types of adjacent nodes (three successors), including EGR1, a TF in the abnormally expressed network.

The third class has three types of predecessor, including H3F3B, NOTCH2 and TNC. They only act as target genes in the abnormally expressed network.

The fourth class has two types of predecessor, including CDK4, MMP9, NOTCH1 and PLK2. They are only targets and show up in the related network.

The majority of abnormally expressed genes have no adjacent nodes. These isolated genes are not discussed in the present study.

The regulatory pathway of each abnormally expressed miRNA is analyzed using the same method. The 12 abnormally expressed miRNAs can be classified into seven types. The first class of abnormally expressed miRNAs has six adjacent nodes (three predecessors and three successors), including miR-25, miR-21 and miR-335. It has been well documented that miR-21 is overexpressed in a number of types of human cancer, and can function as an anti-apoptotic factor by downregulating the expression of tumor suppressor genes, such as PTEN (32).

In Table II, miR-21 is regulated by PTEN and targets PTEN in the abnormally expressed network. Notably, hsa-miR-21 and PTEN form FBLs. miR-21 is regulated by seven genes while targeting six genes in the AA-related network. Two pairs of FBLs are added; miR-21 and EGFR, miR-21 and STAT3. In the global network, miR-21 has 19 predecessors and 87 successors, which are necessary for identifying the regulatory relations. The second class of abnormally expressed miRNAs has three predecessors and two successors, including miR-155. The third class of abnormally expressed miRNAs has five adjacent nodes (three successors and two predecessors), including miR-106b and miR-16. The fourth class has two successors and two predecessors, including miR-15b and miR-93. The fifth class, miR-23a, has three predecessors and no gene targets. The sixth class has two predecessors; miR-196a and miR-218.

miR-196b has no gene interactions and is an isolated miRNA.

Finally, the pathway of each predicted TF was analyzed at three levels using the same method. Among the predicted TFs, E2F1 has the most abundant adjacent nodes.

In Table III, E2F1 is targeted by three miRNAs, while regulating five miRNAs in the abnormally expressed network. Six related miRNAs target E2F1 and seven related miRNAs are regulated by E2F1. miRNAs in the global network validated by experiments provide theoretical foundation for the research. In the comparison table, miR-106b and miR-93 form FBLs with E2F1 in the abnormally expressed network, while hsa-miR-106a is added to form FBLs with E2F1 in the related network.