Advertisement
  • Loading metrics

How MicroRNA and Transcription Factor Co-regulatory Networks Affect Osteosarcoma Cell Proliferation

  • Kathrin Poos,

    Affiliation: Institute of Bioinformatics, University of Münster, Münster, Germany

  • Jan Smida,

    Affiliations: Clinical Cooperation Group Osteosarcoma, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany, Children's Cancer Research Center and Department of Pediatrics, Klinikum rechts der Isar, Technische Universität München, Munich, Germany

  • Michaela Nathrath,

    Affiliations: Clinical Cooperation Group Osteosarcoma, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany, Children's Cancer Research Center and Department of Pediatrics, Klinikum rechts der Isar, Technische Universität München, Munich, Germany

  • Doris Maugg,

    Affiliations: Clinical Cooperation Group Osteosarcoma, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany, Children's Cancer Research Center and Department of Pediatrics, Klinikum rechts der Isar, Technische Universität München, Munich, Germany

  • Daniel Baumhoer,

    Affiliations: Clinical Cooperation Group Osteosarcoma, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany, Bone Tumor Reference Center at the Institute of Pathology, University Hospital Basel, Basel, Switzerland

  • Eberhard Korsching

    korschi@uni-muenster.de

    Affiliation: Institute of Bioinformatics, University of Münster, Münster, Germany

How MicroRNA and Transcription Factor Co-regulatory Networks Affect Osteosarcoma Cell Proliferation

  • Kathrin Poos, 
  • Jan Smida, 
  • Michaela Nathrath, 
  • Doris Maugg, 
  • Daniel Baumhoer, 
  • Eberhard Korsching
PLOS
x

Abstract

Osteosarcomas (OS) are complex bone tumors with various genomic alterations. These alterations affect the expression and function of several genes due to drastic changes in the underlying gene regulatory network. However, we know little about critical gene regulators and their functional consequences on the pathogenesis of OS. Therefore, we aimed to determine microRNA and transcription factor (TF) co-regulatory networks in OS cell proliferation. Cell proliferation is an essential part in the pathogenesis of OS and deeper understanding of its regulation might help to identify potential therapeutic targets. Based on expression data of OS cell lines divided according to their proliferative activity, we obtained 12 proliferation-related microRNAs and corresponding target genes. Therewith, microRNA and TF co-regulatory networks were generated and analyzed regarding their structure and functional influence. We identified key co-regulators comprising the microRNAs miR-9-5p, miR-138, and miR-214 and the TFs SP1 and MYC in the derived networks. These regulators are implicated in NFKB- and RB1-signaling and focal adhesion processes based on their common or interacting target genes (e.g., CDK6, CTNNB1, E2F4, HES1, ITGA6, NFKB1, NOTCH1, and SIN3A). Thus, we proposed a model of OS cell proliferation which is primarily co-regulated through the interactions of the mentioned microRNA and TF combinations. This study illustrates the benefit of systems biological approaches in the analysis of complex diseases. We integrated experimental data with publicly available information to unravel the coordinated (post)-transcriptional control of microRNAs and TFs to identify potential therapeutic targets in OS. The resulting microRNA and TF co-regulatory networks are publicly available for further exploration to generate or evaluate own hypotheses of the pathogenesis of OS (http://www.complex-systems.uni-muenster.de/co_networks.html).

Author Summary

Osteosarcomas (OS) are bone tumors most frequently affecting children and young adolescents. We do not know much about its molecular pathogenesis hampering personalized therapies to more effectively cure patients. To this day, almost all patients receive adjuvant chemotherapies independent of its necessity. Hence, we need to gain a comprehensive understanding of the molecular pathogenesis of OS to uncover molecular components that can be used as therapeutic targets. MicroRNAs and transcription factors (TFs) are master regulators of the cellular system. They control the amount of genes expressed in a cell at a specific time point that ultimately results in a distinct cellular phenotype. Here, we investigated microRNA and TF co-regulatory networks in OS cell growth as one hallmark in cancer. We uncovered key microRNA and TF regulators that cooperatively control growth-related pathways in OS and proposed potential therapeutic targets. This study illustrates the benefit of analyzing complex diseases from a network perspective because no molecular component functions isolated from the underlying cellular system.

Introduction

Osteosarcoma (OS) is a rare type of cancer frequently occurring in children and young adolescents [1]. It is a complex tumor typically accompanied by severe genomic instability and extensive mutations hampering the identification of a genetic root [2][4]. These genomic alterations affect several genes to a varying extent depending on patient and OS subtype. For instance, frequent mutations and deletions of the tumor suppressor genes TP53, RB1, and CDKN2A and mutations and amplification of the MYC locus [5], [6]. However, their interactions in the molecular pathogenesis and the underlying cellular network of OS are poorly characterized.

Recently, attention has been focused on the impact of microRNAs in OS. Besides transcription factors (TFs) that transcriptionally regulate gene expression, microRNAs are a class of small, conserved, non-coding RNA molecules generally acting on the post-transcriptional level. They are mono- or polycistronically transcribed, processed to mature molecules and subsequently incorporated into the RNA Induced Silencing Complex (RISC). Once integrated in RISC, microRNAs are able to select their target genes via binding to partially complementary sequences in the 3′-UTRs of mRNAs that lead to mRNA degradation or translational inhibition. Computational prediction methods revealed that individual microRNAs regulate hundreds of target genes and one target gene might be regulated by several microRNAs [7]. According to Friedman et al. [8] around 60% of human genes are predicted to be regulated by multiple microRNAs in a cooperative manner. This huge number of target genes suggests a widespread control of biological processes including differentiation, proliferation, migration, and apoptosis [9].

In cancer, microRNAs might serve as onco- and/or tumor suppressor-microRNAs. Amplification or over-expression of oncogenic microRNAs can down-regulate tumor suppressor proteins. Likewise, deletion or under-expression of tumor suppressor microRNAs might lead to the up-regulation of oncogenes [10]. In addition, more than 50% of microRNA genes are located within fragile sites in the genome and are frequently subjected to chromosomal alterations [11]. In this manner, they can affect cancer development and progression.

MicroRNAs share several regulatory concepts with TFs, e.g. they simultaneously regulate many target genes and cooperate with other regulators. However, TFs activate or repress their target gene expression whereas microRNAs regulate their targets primarily through repression to fine-tune cell-specific gene regulatory programs [12]. Because the expression of microRNAs often depends on TF regulation and vice versa, it is not surprising that both families of regulators are tightly related to each other in gene regulatory networks. The coordinated transcriptional regulation of microRNAs and their target genes by TFs is a recurrent network motif. The two types of gene regulators frequently form 3-node feedforward loops (FFLs) with common target genes [13]. Recently, Sun et al. [14] extended this regulatory motif to 4-node FFLs by integrating additional TF target genes. The extension of 3-node to 4-node motifs illustrated a more detailed model of the oncogenesis of glioblastoma by recruiting additional disease genes not directly targeted by microRNAs.

Several studies have shown an involvement of microRNAs in the pathogenesis of OS. They demonstrated down-regulation of miR-143 in OS progression [15], up-regulation of the oncogenic miR-17∼92 cluster in OS cells [16], and regulatory functions for miR-199a-3p [17], miR-21 [18], and miR-125b [19] in OS cell proliferation and migration. Additional genome-wide microRNA analyses suggested sets of microRNAs to discriminate OS from osteoblasts and bone tissue [20][23]. All studies proposed the use of microRNAs as biomarkers in OS that might correlate with clinico-pathological parameters. However, those studies lack a comprehensive analysis of the functional consequences of aberrant microRNA expression in OS. Analyzing microRNAs in the context of their microRNA and TF co-regulatory networks might provide a deeper understanding of the pathogenesis of OS.

In this study, we joined different data sources to analyze the contribution of microRNA and TF co-regulatory 3-node and 4-node motifs to the proliferative activity of OS cells. First, we divided seven OS cell lines into high and low proliferation groups by performing proliferation assays. Expression analysis based on these groups yielded differentially expressed (DE) microRNAs and mRNAs. Second, high efficacy microRNA target genes were obtained by integrating computational predicted targets with DE mRNAs. Only microRNAs with significantly enriched target genes were considered in the analysis. Third, microRNA target genes were clustered according to their functional similarity to explore their distinct biological processes. Fourth, transcription factor binding site (TFBS) information was added to assemble 3-node and 4-node motifs of non-random microRNA and TF co-regulator pairs. Fifth, the resulting 3-node and 4-node motifs were merged to form microRNA and TF co-regulatory networks to examine the coordinated regulation of microRNAs and TFs (Figure 1).

thumbnail
Figure 1. Workflow.

The figure illustrates the procedure of the present study determining microRNA and TF co-regulation in OS cell proliferation.

http://dx.doi.org/10.1371/journal.pcbi.1003210.g001

Here, we present the first study analyzing microRNA and TF co-regulation in OS and uncover critical microRNA players of the functional processes implicated in OS cell proliferation.

Results

OS cell proliferation and differential microRNA expression

In order to investigate the deregulated microRNA and TF co-regulatory networks of proliferative OS cells, we used seven authenticated OS cell lines. The cell lines were divided according to their proliferative activity by performing a proliferation assay. Four OS cell lines exhibited a high proliferative activity with a doubling time <10 hours while three showed less proliferation (Table 1). MNNG/HOS, U2-OS, and SJSA-1 showed additional extensive migratory capabilities. The expression analysis of the microRNAs was based on these two proliferation groups. The analysis yielded nine down-regulated and eight up-regulated microRNAs that passed the differential expression criteria (False discovery rate (FDR) <0.05 & log2 Fold change (FC)≥|1|, Table 2). The derived DE microRNAs have been reported in association with neoplastic disease either due to oncogenic or tumor suppressor properties. Hierarchical clustering of them clearly separated the OS cell samples according to their proliferative activity (Figure 2). Hence, we selected the DE microRNAs as candidates that might affect OS cell proliferation for further analysis.

thumbnail
Figure 2. Clustering of differentially expressed microRNAs.

The heatmap illustrates the expression profiles of DE microRNAs (log2 FC≥|1| & FDR<0.05, y-axis) among all OS cell samples (x-axis). High (dark blue) and low (light blue) proliferative OS samples are clearly separated. The red/green color code corresponds to the expression deviation from the average expression among all samples. Complete-linkage clustering was performed with the Pearson correlation as distance metric.

http://dx.doi.org/10.1371/journal.pcbi.1003210.g002

thumbnail
Table 2. Differentially expressed microRNAs dependent on OS cell proliferation.

http://dx.doi.org/10.1371/journal.pcbi.1003210.t002

OS proliferation-related microRNA target genes

To explore the functional consequences of DE microRNAs on OS cell proliferation, we determined their target genes by integrating gene expression profiles with computational predicted target genes.

First, the expression analysis of mRNAs resulted in a total of 666 up-regulated and 610 down-regulated mRNAs. We applied loose filter criteria for DE mRNAs without correcting for multiple tests (p-value<0.05 & log2 FC≥|0.7|) because microRNA regulation might lead to subtle changes in gene expression. Next, we superimposed the DE genes with predicted microRNA targets to obtain target genes affecting OS cell proliferation. We assumed that microRNAs exhibit an inverse regulatory relationship to their functional target genes, i.e. microRNA expression is inversely correlated to its target gene expression. Hence, down-regulated targets were assigned to up-regulated microRNAs in high proliferative OS cells and vice versa.

To exclude DE microRNAs with random association to OS cell proliferation, we tested for microRNA target gene enrichment within the list of DE genes. Among the 17 DE microRNAs, 12 are significantly enriched due to their targets (FDR<0.05, Table S1). To account for different numbers of targets that might influence the enrichment analysis, we also computed the target gene enrichment of 1,000 permuted samples. The permutation procedure confirmed previous results (Figure S1). Consequently, we excluded 5 microRNAs from further analyses (miR-92b, let-7f, miR-9-3p, miR-151-5p, and miR-100). The remaining 12 OS proliferation-related microRNAs are implicated in the regulation of 474 target genes. Hierarchical clustering of the target genes resulted in a distinct separation of the high and low proliferative OS cell samples (Figure 3A).

thumbnail
Figure 3. Functional implications of microRNA target genes in OS proliferation.

(A) Differentially expressed microRNA target genes. The heatmap shows the expression of DE microRNA target genes (log2 FC≥|0.7| & p-value<0.05, y-axis) among all OS cell samples (x-axis). High (dark blue) and low (light blue) proliferative OS samples are grouped into distinct clusters. The red/green color code corresponds to the expression deviation from the average expression among all samples. Complete-linkage clustering was applied with the Pearson correlation as distance metric. (B) Principal component analysis based on GO functional similarity. The figure draws the first (x-axis) and second (y-axis) principal components based on the GO functional similarity of microRNA target genes. The target genes are color coded according to their cluster membership determined by FCM. The radius of each gene represents the strength of association to its cluster. (C) Comparison of the GO enrichment analyses of the distinct clusters. Each row corresponds to a specific biological process category. The number of genes included in each category is indicated in the round brackets. Circle sizes correspond to the percentage of row-genes located in a cluster. Significantly enriched categories are color coded in red and random categories in blue.

http://dx.doi.org/10.1371/journal.pcbi.1003210.g003

Functional clustering of microRNA targets

We further investigated the underlying biological processes of OS proliferation-related microRNAs. We classified microRNA target genes according to their functional similarities of their gene ontology (GO) biological process terms using fuzzy c-means clustering (FCM). After determining FCM parameters (Figure S2), we obtained two clusters. Principal Component Analysis (PCA) supported the results. The first two components separate the determined clusters (Figure 3B). Cluster C1 consists of 172 members and cluster C2 contains 212 members. The remaining 90 microRNA targets could not be annotated with a GO biological process term and were excluded from further analysis. The clustering suggested that the microRNA targets can be classified into two broad functional classes.

GO enrichment analyses revealed that members of C1 are mainly involved in metabolic processes like protein modification, nucleic acid metabolism, and carbohydrate metabolism, whereas members of C2 are implicated in signal transduction pathways leading to proliferation, differentiation, apoptosis, and migration. Both clusters demonstrate that cancer cells adapt metabolic processes for cell proliferation and survival [24]. A comparison between the five most informative GO terms (FDR<0.05) illustrating the specific biological aspects of each cluster is shown in Figure 3C.

Identifying microRNA and TF co-regulated target genes

Transcriptional regulation of TFs is tightly coupled with the post-transcriptional regulation of microRNAs. We utilized their 3-node and 4-node co-regulatory motifs to study DE microRNA and TF co-regulation in OS cell proliferation for each functional cluster.

Every possible 3-node and 4-node FFL motif was determined to assess significant microRNA and TF combinations (FDR<0.2) by using the hypergeometric test (Table S2 and S3). For the 3-node FFL, we obtained non-random microRNA and TF pairs with common target genes (Figure 4A). For 4-node FFL motifs, we assessed non-random microRNA and TF pairs that regulate gene neighbors in the protein interaction network (Figure 4B). The individual 3-node and 4-node FFL motifs are listed in Table S4 and S5, respectively.

thumbnail
Figure 4. MicroRNA and TF co-regulatory motifs.

Schematic illustration of (A) 3-node and (B) 4-node FFL motifs. A 3-node motif contains a microRNA, a TF, and a commonly regulated target gene. In contrast, 4-node FFLs comprise a microRNA, a TF, a microRNA target gene (primary target), and a TF target gene (secondary target) that interacts with the primary target. MicroRNAs are indicated with diamonds, TF with triangles, primary targets with rectangles, and secondary targets with circles. (C) Coexpression of microRNA and TF target genes. The plot shows the observed and random cumulative distribution functions (CDFs, y-axis) of the Pearson correlation coefficients (x-axis) between any gene pair regulated by a specific microRNA and TF 3-node and 4-node motif. The observed and random CDFs are compared using the KS test. The p-values are indicated within the plots. Green and orange color codes correspond to the distinct clusters C1 and C2.

http://dx.doi.org/10.1371/journal.pcbi.1003210.g004

Subsequently, we analyzed the co-regulated target genes of significant microRNA and TF combinations. The results are summarized in Table 3. Noticeably, the microRNA and TF duo with the highest number of co-regulated target genes in both functional clusters is miR-9-5p and SP1 (Figure 5) indicating a prominent role in OS cell proliferation.

thumbnail
Figure 5. miR-9-5p and SP1 co-regulatory motifs.

The co-regulatory motifs of miR-9-5p and SP1 are illustrated as graphs with nodes and edges for (A) C1 and (B) C2. MicroRNAs are marked with diamonds, TFs with triangles, primary targets with rectangle, and secondary targets with a circle. Yellow edges tag protein-DNA interactions, blue edges microRNA-target interactions, and dashed edges protein interactions. The red/green color code presents the corresponding log2 FC.

http://dx.doi.org/10.1371/journal.pcbi.1003210.g005

Further, we examined the co-expression of genes co-regulated by the same microRNA and TF pairs. We computed the Pearson correlation coefficients between co-regulated gene pairs as a measure of their co-expression. The distribution of the resulting correlation values was compared to the correlation distribution of random genes by the Kolmogorov-Smirnov (KS) test. The co-expression of co-regulated gene pairs tends to be significantly higher than for random genes (p-value<2.2×10−16, Figure 4C). This result supports the hypothesis of non-random microRNA and TF co-regulation within the list of their common or interacting target genes and suggests a similar functional context for their targets.

Generating microRNA and TF co-regulatory networks

Subsequently, we constructed the microRNA and TF co-regulatory networks that highlighted the combinatorial regulation patterns and regulated biological processes of microRNAs and TFs. The networks of C1 and C2 were generated by joining all significant co-regulatory relationships of microRNAs and TFs (Table 3). The resultant microRNA and TF co-regulatory networks are provided for full exploration on our website (http://www.complex-systems.uni-muenster.de/co_networks.html).

To assess the contribution of individual nodes in the co-regulatory networks on the networks' stability and robustness, we calculated the node degree and betweenness centrality parameters. The node degree distributions are highly right skewed. A large fraction of nodes shows a low degree and only few nodes have high degrees (Figure S3). Almost all microRNAs and TFs are located at higher node degrees as indicated by their average node degrees (C1: microRNAs 19 and TFs 19, C2: microRNAs 25 and TFs 49). We expected that finding as microRNA and TF co-regulation is the main subject of the present study.

Each network contains three types of nodes, namely microRNAs, TFs, and target genes. We ranked the nodes according to their node degrees and node type. The top 25% of microRNAs and TFs and the top 5% of target genes were considered as hubs in the C1 and C2 networks (Table S6). We detected the hub microRNA miR-214 and the hub TFs CREB1, SP1, and ZIC2 in both networks suggesting a central function in OS cell proliferation. Strikingly, around 50% of microRNA and TF target gene hubs in the two networks are TFs themselves. The microRNA and TF co-regulatory network derived from C1 contains ATF6, GTF2A1, HIVEP2, KLF5, LMO3, NFKB1, and TBPL1 and the C2 network comprises BCL6, BCL6B, E2F4, HES1, JUN, LMO4, RARA, REST, SIN3A, TCF7L2, and ZBTB16. Some of these TFs (ATF6, E2F4, JUN, RARA, and REST) are implicated in building 3-node and 4-node FFL motifs in one or two networks. The remaining TFs were either not existent in the UCSC conserved TFBS track or do not produce any significant FFL. Additionally, we found epigenetic modulators and genes involved in protein modification processes, like protein ubiquitination and phosphorylation. As already mentioned, the microRNA targets in C2 are associated with signal transduction for maintaining OS cell proliferation. Among the hub genes in the network derived from C2, 30% of target gene hubs (AMOT, ARF6, CACNA1A, CTNNB1, GRB2, NOTCH1, PDGFRB, PIK3R1, SMAD7, and TGFBR2) are related to signaling pathways that participate in cell proliferation, survival, and migration.

Moreover, we assessed over-represented functional pathways derived from the KEGG database [25]. The enriched categories (FDR<0.05) are shown in Table S7. We detected an enrichment of genes involved in the cell cycle and cancer related pathways in both networks. We expected to observe these functional categories as we analyzed the proliferative potential of OS tumor cell lines. Further, we observed a similar functional trend between the C1 and C2 networks as for their corresponding functional clusters. Within the co-regulatory network of C2, signaling pathways are significantly over-represented such as the MAPK-, TGFB-, and WNT-signaling pathways. In contrast, the network of C1 comprises a significant number of genes required for the basal transcription machinery.

Tightly connected microRNA and TF co-regulated subnetworks

After examining the global co-regulation patterns of microRNAs and TFs in both networks, we were interested in sets of microRNAs and TFs that co-regulate densely connected network modules. To investigate the local structure of the OS proliferation-related co-regulatory networks, the walktrap algorithm was applied [26].

The algorithm obtained six modules within the metabolic network of C1 (Figure S4) and six modules in the signaling network of C2 (Figure S5). The size and node types within each module are indicated in Table 4. Strikingly, miR-9-5p is located in the largest module and is regulated through the TFs ATF2, BACH1, CREB1, and SP1 in both networks. As mentioned before, miR-9-5p and SP1 co-regulate the largest number of target genes and thus indicate a prominent function in OS cell proliferation.

Further, we run the Functional Annotation clustering Tool of the DAVID database [27] to classify the distinct network modules according to their GO biological process and molecular function terms. We annotated each module with the biological aspect of its maximum enrichment score (ES). Among the 12 modules, five are mainly involved in transcriptional regulation processes, which is in accordance with previous studies that illustrated that microRNAs function via TFs to regulate various biological processes like cell proliferation [28], [29]. Despite the top scored biological associations, one module (C2.1) is related to negative regulation of differentiation (ES>3.9), particularly to osteoblast differentiation due to the genes CDK6, MEN1, SKI, SMAD3, and SOX2, which might provide a link to the pathogenesis of OS (Figure S5A). Within this module, the TF MYC (node degree 83) co-regulates several targets with miR-138 (node degree 25). The top ranked target gene in this module is SIN3A (node degree 57).

Discussion

OS is a complex tumor with varying degrees of genomic alterations and affected disease genes. This genomic complexity makes it difficult to identify a genetic cause and deregulated pathways in the pathogenesis of OS. Systems biological approaches provide tools to investigate the interactions between candidate genes by integrating different data sources on the network level. Thereof, cooperative or compensative effects between candidate genes might be observed. Hence, network-based approaches seemed to be ideally suited to study the implication of functional pathways in OS. This study represents the first attempt to investigate microRNA and TF co-regulatory networks in the pathogenesis of OS. In the course of the study, several data sources were integrated, namely microRNA and mRNA expression profiles, TFBS information, and protein interaction data. Therewith, we aimed to unravel possible candidate genes and their interplay that ultimately result in a high proliferative phenotype of OS cell lines. The derived microRNA and TF networks are publicly available (http://www.complex-systems.uni-muenster.de/co_networks.html).

OS proliferation-related microRNAs are associated with a proliferative phenotype

The microRNA and TF co-regulatory networks modeling OS cell proliferation are based on 12 proliferation-related microRNAs. Among these microRNAs, 11 were previously mentioned in OS [20], [22], [23], [18], whereas miR-138 was exclusively obtained in this study. Previous studies focused on global microRNA alterations in OS with respect to osteoblast cells and bone tissue. However, microRNA expression was partially inconsistent between different studies. Namløs et al. [23] hypothesized that contradictory microRNA regulation in different genome-wide studies might be explained due to distinct differentiation stages of OS progenitor cells. In this study, the DE microRNAs were assessed between high and low proliferative OS cells. The varying proliferation activity between these cell lines might reflect distinct differentiation stages of their progenitor cells. This might explain the detection of proliferation-related microRNAs in global OS studies.

In different biological contexts the down-regulated microRNAs have been associated with decreased proliferation [30][33]. In turn, up-regulated microRNAs have been related to an increased proliferation potential [34][37]. These studies emphasize the proposed influence of the derived microRNAs on OS cell proliferation.

miR-9-5p and SP1 co-regulated targets are highly involved in OS cell proliferation

The assembly of non-random microRNA and TF co-regulators revealed several interesting combinations. Based on the huge number of co-regulated target genes, the most notably co-regulation duo is the miR-9-5p and SP1 pair. Depending on the derived cluster, these co-regulators seem to affect distinct functional pathways due to their target genes.

The C1 derived miR-9-5p and SP1 co-regulated target genes seem to participate in NFKB-signaling. Commonly regulated target genes include the TFs NFKB1, NFKB2, RELA, RELB, and BCL3 and the inhibitors NKRF, NFKBIA, and TNIP2 that cooperatively activate or block target gene expression of NFKB, respectively [38]. This pathway is implicated in OS cell proliferation [39], and NFKB1 is an experimentally validated target gene of miR-9-5p [40]. Furthermore, a regulatory circuit including SP1/NFKB1/HDAC and miR-29b is known to induce leukemic growth [41]. Thus, miR-9-5p might function in a similar context in OS than miR-29b in leukemia.

On the other hand, the C2 derived miR-9-5p and SP1 co-regulated genes might be involved in focal adhesion. We observed cadherins (CDH1, CDH2, DSC2, and DSG2), further cell adhesion molecules (e.g. FN1, ITGA6, ITGB1, JUP, PKP3, and VCL), and calcium signaling receptors (e.g. CACNA1A and CALR). According to StringDB [42], almost all commonly regulated genes interact with each other indicating a functional association (Figure S6). A possible pathway of miR-9-5p and SP1 co-regulation could implicate CDH1 and CTNNB1 that modulate cell proliferation [43]. CTNNB1 is not a commonly regulated target gene of miR-9-5p and SP1. However, we observed some of its binding partners (e.g. CDH1, CDH3, CTNNBPI1, RUNX2, and SMAD7). Additionally, CTNNB1 is a hub gene in the C2 co-regulatory network.

Moreover, increased expression of miR-9-5p resulted in down-regulation of the NFKB-SNAIL pathway and simultaneously to up-regulation of CDH1 in melanoma cells [44]. All these observations suggest a central role of miR-9-5p and SP1 co-regulation in OS cell proliferation.

Functional implications of the miR-138-MYC-SIN3A module in the pathogenesis of OS

Merging significant microRNA and TF co-regulators resulted in co-regulatory networks of C1 and C2. The networks provided a global view on microRNA and TF co-regulation in OS cell proliferation. To analyze the local co-regulation patterns within the networks, we specifically extracted densely connected modules. One of these modules, namely the C2.1 module, is implicated in negative regulation of differentiation, particularly osteoblast differentiation. It contains the hubs miR-138, MYC, and SIN3A. We hypothesize a function for members of C2.1 that might be specific for the pathogenesis of OS. According to StringDB [42], the genes in the module form a densely connected network illustrating a tight functional relationship (Figure S7). The module comprises members of the cell cycle (CCND1, CCND3, CDKN1A, CDKN2C, and CDK6), all involved in the RB1-pathway [45]. The complex of the module members SIN3A, NCOR1, SKI, and HDAC can bind to RB1 and repress E2F target genes [46]. Therefrom, we assumed a connection between module members and RB1, which has been reported to be frequently deregulated in OS [5].

Further, SIN3A is an experimentally validated target gene of miR-138 [47]. In the global co-regulatory network of C2 as well as in C2.1 module, it depicts a hub gene. Its role in cancer is contradictory, on the one hand it shows tumor suppressor functions [48]. On the other hand, it acts in tumor growth [49]. Taken together with the fact that SIN3A can interact with RB1, we suggest a possible role for SIN3A in the pathogenesis of OS.

Model of OS cell proliferation

After examining and discussing the structural and functional aspects of the co-regulatory networks, we integrated the main results of the present study into a potential model of OS cell proliferation (Figure 6). The focus of the model is on microRNA and TF co-regulation of the microRNAs miR-9-5p, miR-138, and miR-214 and the TFs SP1 and MYC.

thumbnail
Figure 6. Model of OS cell proliferation.

The picture illustrates the proposed model of OS cell proliferation co-regulated by miR-9-5p, miR-138, and miR-214 and the TFs SP1 and MYC. These regulators control mainly components of the NFKB- and RB1-pathway and of focal adhesion processes. Primary microRNA targets are rectangular and secondary targets are ellipse shaped. Red marks up-regulated genes in proliferative active OS cell lines and green indicates down-regulated genes. White nodes represent genes not DE in OS cell proliferation. Solid and dashed arrows illustrate direct and indirect functional associations, respectively. RB1 is shaded in red to indicate that it is no member of the global networks.

http://dx.doi.org/10.1371/journal.pcbi.1003210.g006

In proliferative active OS cell lines, miR-9-5p, miR-138, and miR-214 are significantly down-regulated leading to the up-regulation of their direct target genes CDK6, E2F4, HES1, ITGA6, NFKB1, NOTCH1, and SIN3A. CDK6 phosphorylates RB1 and therewith the RB1/SIN3A/SKI/NCOR1/HDAC complex cannot repress E2F4 target gene expression [45], [46]. Activated NOTCH1 induces HES1 and sustains NFKB-signaling through NFKB1, NFKB2, RELA, RELB [50]. CTNNB1 stabilizes cell-cell adhesions in complex with CDH1. Unbound CTNNB1 can translocate to the nucleus [43]. All these signals end up in the nucleus where they induce expression of proliferation promoting genes like microRNAs up-regulated in high proliferative OS cells, CCND1, and FN1. In turn, they repress genes implicated in growth-arrest.

Conclusion

The resulting microRNA and TF co-regulatory networks display a detailed picture of the regulation of OS cell proliferation. In the present study, we concentrated on distinct functional aspects unraveled from the networks. The outcome suggests that down-regulation of miR-9-5p, miR-138, and miR-214 results in a strong proliferative phenotype of OS cells due to their impact on NFKB- and RB1-signaling and on focal adhesion.

Our study provides potential therapeutic targets in OS and proposes concepts for further research. In addition, we demonstrated how systems biological approaches support the analyses of complex diseases.

Materials and Methods

Data sets

We used microRNA and mRNA expression data of seven authenticated OS cell lines, six from our previously published study [16] and one additional (ZK-58). Prior to microarray analyses, RNA was isolated and further processed as described in [16]. MicroRNA and mRNA expression profiles were determined on Exiqon's miRCURY LNA and Affymetrix's Human Gene 1ST arrays, respectively.

Conserved TFBSs (hg19) were downloaded from the UCSC Table Browser [51]. The track contained predicted TFBSs conserved in the human/mouse/rat alignment that were determined by using the Transfac Matrix Database 7.0 [52]. Protein interaction data were obtained from BioGRID release 3.1.92 [53].

Proliferation assay to group OS cell lines

The OS cell lines were evaluated regarding their proliferative, migrative, and invasive potential by using in vitro-assays (BD Biosciences). Cells utilized in the assays showed 60 to 80% confluence growth. Prior to the assays cell lines were synchronized to ensure a uniform cell growth.

To analyze OS cell proliferation, duplicates of each cell line (1×105 cells) were seeded in 25 cm2 cell culture flasks. Cells were harvested at 24, 48, 72, 96, and 168 hours of growth. The cell number was determined by an automated cell counter (Beckman Coulter). For each cell line and time point, the mean cell number was calculated to estimate the growth rate and subsequently the doubling time.

Further, the Biocoat Matrigel Invasion Assay and a migration assay (BD Biosciences) were performed for each cell line in duplicate with matrigel-coated and uncoated inserts, respectively. Experiments were performed according to the manufacturer's instructions. Evaluation of invaded and migrated cells was done after 24 and 48 hours by light-microscopic analysis. Ten visual fields (magnification 10×) were analyzed by counting stained cells on the membranes.

Expression analysis of microRNAs and mRNAs

The expression data sets were analyzed using the Bioconductor package limma [54]. DE genes between high and low proliferative OS cells were determined using eBayes [54].

MicroRNA expression data were annotated with miRBase release 18 [55], background corrected by normexp+offset 10 [56], and normalized with printtiploess followed by RQuant [57]. In the differential expression analysis we considered the top 75% of microRNA probes that showed largest variation over all samples. Multiple test correction was performed using Benjamini and Hochberg's FDR approach [58].

The mRNA expression data were preprocessed by the Bioconductor package affy [59]. The Affymetrix Human Gene 1 ST array contains probes mapping among the whole transcript. Therefore, we filtered probes that mapped to exons present in at least 80% of a gene's transcripts to get one stable expression value per gene. Transcripts were derived from Ensembl release 63 [60]. The raw probe intensities were background corrected, normalized, and summarized to the gene-level by applying the robust multi-array average algorithm (rma) [59].

MicroRNA and TF target prediction

Predicted microRNA targets were obtained by running the local perl scripts targetscan_60.pl and targetscan_61_context_scores.pl that were online available at the TargetScan website (http://www.targetscan.org/) [61]. Mature microRNA sequences were downloaded from miRBase release 18 [55]. To derive high efficacy targets, we filtered target predictions with a context score≤−0.1 [61].

To determine TF target genes, we downloaded the transcriptional start sites (TSSs) of all genes included in the mRNA expression data set from the hg19 assembly of the UCSC Table Browser [51] and the TSSs from our proliferation related microRNA genes from miRStart [62]. Further, we defined the promoter region to −/+ 2000 nucleotides around the TSSs. Genes having a TFBS that completely overlapped their promoter regions were considered as TF targets. We just considered human TFs that were expressed in at least one proliferation group of our OS cell samples (log2 intensity≥8).

Enrichment of microRNA target genes

To assess the enrichment of microRNA target genes in the list of DE genes, a hypergeometric test was performed. Multiple test correction was done by determining the FDR according to Benjamini and Hochberg [58]. To account for the different number of target genes of the DE microRNAs, a permutation procedure was applied. We randomly selected the number of DE genes out of the genes in our mRNA expression data set and counted the number of random microRNA target genes. For each permutation an enrichment score (ES, -log10 p-value) was calculated. The permutation procedure was repeated 1,000 times. The resulting permutation p-values were obtained by counting the number of permuted ESs exceeding the observed one. This was done for all DE microRNAs separately.

Fuzzy c-means clustering of DE microRNA targets

To classify DE microRNA targets according to their functional similarity, their GO semantic similarity scores based on biological process terms were computed using Resnik's information content approach of the GOSim package [63]. The resulting functional similarity scores for any target gene pair were listed in a similarity matrix, which was further utilized as distance matrix for clustering. We applied FCM [64] to classify microRNA target genes according to their functional similarity using the function fanny from the R cluster package [65]. The fuzziness parameter was estimated by Dunn's coefficient [66] among a range of 1.1 to 1.5 and the cluster number was estimated over a range of 2 to 15 using Dunn's index [67]. The Dunn coefficient (range 0–1) indicates the fuzziness of a cluster [66]. The Dunn index compared the between cluster variation to the within cluster variation and measures the cluster separation. A Dunn index >1 indicates a satisfying clustering [67].

The derived functional clusters were evaluated by running a GO enrichment analysis with the Bioconductor package GOStats [68]. Multiple test correction was performed by using Benjamini and Hochberg's FDR approach [58].

Testing and evaluating microRNA and TF co-regulation

We tested for non-random microRNA and TF 3-node and 4-node motifs by using the hypergeometric test adapted from Sun et al. [14]. In contrast to them, we applied a different null model to derive p-values specific for the underlying microRNA and TF co-regulation pairs.

For the 3-node motifs we tested if a microRNA and TF pair had significantly more commonly DE target genes than computationally predicted target genes. In turn, co-regulation of microRNA and TF pairs in 4-node motifs was tested based on commonly regulated secondary target genes and compared to the genes with corresponding TFBS in the whole 1st-neighbor protein interaction network. The 1st-neighbor protein interaction network was determined by extracting all 1st-neighbors of microRNA target genes from the protein interaction data. Benjamini and Hochberg's FDR was used to adjust for multiple testing [58].

Furthermore, evaluation of significant pairs of microRNAs and TFs was performed by assessing the coexpression of genes targeted by the same microRNA and TF pair. The Pearson correlation was used as a measure for coexpression. Statistical significance was determined by a permutation procedure. We randomly chose the same number of genes targeted by 3-node and 4-node FFLs out of all genes annotated in the mRNA expression data and computed their correlation coefficients. The permutation procedure was repeated 1,000 times. Finally, we tested if the coexpression of the genes in the FFLs was significantly greater than in randomly selected gene pairs using the KS test.

MicroRNA and TF co-regulatory network generation and analyses

The microRNA and TF co-regulatory networks were constructed by merging all 3-node and 4-node FFL motifs. The networks were modeled as graphs with nodes and edges. Nodes corresponded to microRNAs, TFs, or target genes and edges corresponded to microRNA-target regulation, TF-target regulation, or protein interactions. To identify crucial network players, we computed network centralities, namely node degree and betweenness, using the R package igraph [26]. The node degree is defined as the number of direct neighbors of a node in a network. Nodes having a high number of direct neighbors are thought to be important regulatory hubs inside the network. In contrast, a node's betweenness is a measure of the number of shortest paths between any pair of nodes that run through it [69].

To detect tightly connected groups of nodes in the network, we run the walktrap algorithm [70]. This algorithm finds modules in connected graphs. It is based on random walks and assumes that the random walker is trapped in dense parts of a network [26].

For further network evaluation we used the Functional Annotation Tool of the DAVID database [27]. The networks were visualized with Cytoscape 2.8.3 [71] and Cytoscape Web 1.0.2 [72].

Supporting Information

Figure S1.

Enrichment of proliferation-related microRNAs. The barplot of enrichment scores (ESs) of observed microRNA target genes (cyan) and randomly selected targets (grey). The ES of randomly selected microRNA targets is illustrated as mean±stdev. Per microRNA we computed 1,000 random ESs. P-values between observed and random ESs were obtained by counting the number of random ESs exceeding the observed one.

doi:10.1371/journal.pcbi.1003210.s001

(PDF)

Figure S2.

Distribution of Dunn coefficients and indices determined by FCM clustering. (A) Assessing the optimal fuzziness parameter. The plot illustrates the Dunn coefficients (y-axis) among a range of fuzziness parameters (x-axis) for different cluster numbers. The fuzziness was set to 1.1, where the Dunn coefficient distribution exceeds 0.5 for all cluster numbers. (B) Determining the optimal cluster number. The plot shows the Dunn indices (y-axis) among a range of cluster numbers (x-axis) for different fuzziness parameters. The optimal cluster number was set to 2, where the Dunn index reached its maximum value.

doi:10.1371/journal.pcbi.1003210.s002

(PDF)

Figure S3.

Node degree distribution of the microRNA and TF co-regulatory networks. The plots show the fraction of proteins (y-axis) among all node degrees (x-axis) from the microRNA and TF co-regulatory networks (grey) of (A) C1 and (B) C2. Different colors indicate distinct degree distributions of different node types. Horizontal lines mark the average node degree of individual node types. The values of the average node degrees are listed in the plots' legends.

doi:10.1371/journal.pcbi.1003210.s003

(PDF)

Figure S4.

MicroRNA and TF co-regulatory network modules derived from C1. The figure shows network modules defined by the walktrap algorithm. The modules C1.1 to C2.6 are labeled from (A) to (F). Node shapes correspond to the distinct node types: microRNAs (diamond), TFs (triangle), primary target (rectangle), and secondary target (ellipse). Yellow edges mark TF-DNA interactions, blue edges microRNA-target interactions, and dashed grey edges protein interactions. The red/green color code indicates the log2 FC.

doi:10.1371/journal.pcbi.1003210.s004

(PDF)

Figure S5.

MicroRNA and TF co-regulatory network modules derived from C2. The figure shows network modules defined by the walktrap algorithm. The modules C2.1 to C2.6 are labeled from (A) to (F). Node shapes correspond to the distinct node types: microRNAs (diamond), TFs (triangle), primary target (rectangle), and secondary target (ellipse). Yellow edges mark TF-DNA interactions, blue edges microRNA-target interactions, and dashed grey edges protein interactions. The red/green color code indicates the log2 FC. The C2.1 module implicated in negative regulation of differentiation of osteoblast cells is tagged with a red frame.

doi:10.1371/journal.pcbi.1003210.s005

(PDF)

Figure S6.

miR-9-5p and SP1 target gene associations. The network is derived from the STRING 9.0 database [42]. It illustrates experimental and literature-mined functional associations between miR-9-5p and SP1 target genes.

doi:10.1371/journal.pcbi.1003210.s006

(TIF)

Figure S7.

Module C2.1 target gene associations. The network is derived from the STRING 9.0 database [42]. It illustrates experimental and literature-mined functional associations between genes within the C2.1 network module.

doi:10.1371/journal.pcbi.1003210.s007

(TIF)

Table S1.

Enrichment of proliferation-related microRNA target genes. Results of the hypergeometric test to examine significantly enriched microRNA target genes within the list of DE genes. The table marks the total number of predicted target genes, the number of DE target genes, and the corresponding FDR.

doi:10.1371/journal.pcbi.1003210.s008

(XLS)

Table S2.

Significant microRNA and TF co-regulatory 3-node motifs. The table summarizes the number of common target genes of each non-random microRNA and TF co-regulatory pair with corresponding statistics.

doi:10.1371/journal.pcbi.1003210.s009

(XLS)

Table S3.

Significant microRNA and TF co-regulatory 4-node motifs. The table summarizes the number of interacting target genes of each non-random microRNA and TF co-regulatory pair with corresponding statistics.

doi:10.1371/journal.pcbi.1003210.s010

(XLS)

Table S4.

Individual microRNA and TF co-regulatory 3-node and 4-node motifs of C1.

doi:10.1371/journal.pcbi.1003210.s011

(XLS)

Table S5.

Individual microRNA and TF co-regulatory 3-node and 4-node motifs of C2.

doi:10.1371/journal.pcbi.1003210.s012

(XLS)

Table S6.

MicroRNA and TF co-regulatory network hubs. The table holds the node degree and betweenness parameters of hub genes within the co-regulatory network of C1 and C2 for each node type.

doi:10.1371/journal.pcbi.1003210.s013

(XLS)

Table S7.

Over-represented KEGG pathways.

doi:10.1371/journal.pcbi.1003210.s014

(XLS)

Author Contributions

Conceived and designed the experiments: KP JS EK. Performed the experiments: KP JS. Analyzed the data: KP. Contributed reagents/materials/analysis tools: JS MN DM DB EK. Wrote the paper: KP JS EK.

References

  1. 1. Picci P (2007) Osteosarcoma (osteogenic sarcoma). Orphanet J Rare Dis 2: 6. doi: 10.1186/1750-1172-2-6
  2. 2. Man T-K, Chintagumpala M, Visvanathan J, Shen J, Perlaky L, et al. (2005) Expression profiles of osteosarcoma that can predict response to chemotherapy. Cancer Res 65: 8142–8150. doi: 10.1158/0008-5472.can-05-0985
  3. 3. Ozaki T, Schaefer K-L, Wai D, Buerger H, Flege S, et al. (2002) Genetic imbalances revealed by comparative genomic hybridization in osteosarcomas. Int J Cancer 102: 355–365. doi: 10.1002/ijc.10709
  4. 4. Smida J, Baumhoer D, Rosemann M, Walch A, Bielack S, et al. (2010) Genomic alterations and allelic imbalances are strong prognostic predictors in osteosarcoma. Clin Cancer Res 16: 4256–4267. doi: 10.1158/1078-0432.ccr-10-0284
  5. 5. Scholz RB, Kabisch H, Weber B, Röser K, Delling G, et al. (1992) Studies of the RB1 gene and the p53 gene in human osteosarcomas. Pediatr Hemat Oncol 9: 125–137. doi: 10.3109/08880019209018328
  6. 6. Ueda T, Healey JH, Huvos AG, Ladanyi M (1997) Amplification of the MYC Gene in Osteosarcoma Secondary to Paget's Disease of Bone. Sarcoma 1: 131–134. doi: 10.1080/13577149778209
  7. 7. Selbach M, Schwanhäusser B, Thierfelder N, Fang Z, Khanin R, et al. (2008) Widespread changes in protein synthesis induced by microRNAs. Nature 455: 58–63. doi: 10.1038/nature07228
  8. 8. Friedman RC, Farh KK-H, Burge CB, Bartel DP (2009) Most mammalian mRNAs are conserved targets of microRNAs. Genome Res 19: 92–105. doi: 10.1101/gr.082701.108
  9. 9. Davis BN, Hata A (2009) Regulation of MicroRNA Biogenesis: A miRiad of mechanisms. Cell Commun Signal 7: 18. doi: 10.1186/1478-811x-7-18
  10. 10. Esquela-Kerscher A, Slack FJ (2006) Oncomirs - microRNAs with a role in cancer. Nat Rev Cancer 6: 259–269. doi: 10.1038/nrc1840
  11. 11. Calin GA, Sevignani C, Dumitru CD, Hyslop T, Noch E, et al. (2004) Human microRNA genes are frequently located at fragile sites and genomic regions involved in cancers. P Natl Acad Sci USA 101: 2999–3004. doi: 10.1073/pnas.0307323101
  12. 12. Hobert O (2008) Gene regulation by transcription factors and microRNAs. Science 319: 1785–1786. doi: 10.1126/science.1151651
  13. 13. Shalgi R, Lieber D, Oren M, Pilpel Y (2007) Global and local architecture of the mammalian microRNA-transcription factor regulatory network. PLoS Comput Biol 3: e131. doi: 10.1371/journal.pcbi.0030131
  14. 14. Sun J, Gong X, Purow B, Zhao Z (2012) Uncovering MicroRNA and Transcription Factor Mediated Regulatory Networks in Glioblastoma. PLoS Comput Biol 8: e1002488. doi: 10.1371/journal.pcbi.1002488
  15. 15. Zhang H, Cai X, Wang Y, Tang H, Tong D, et al. (2010) microRNA-143, down-regulated in osteosarcoma, promotes apoptosis and suppresses tumorigenicity by targeting Bcl-2. Oncol Rep 24: 1363–1369. doi: 10.3892/or_00000994
  16. 16. Baumhoer D, Zillmer S, Unger K, Rosemann M, Atkinson MJ, et al. (2012) MicroRNA profiling with correlation to gene expression revealed the oncogenic miR-17-92 cluster to be up-regulated in Osteosarcoma. Cancer Genet 205: 212–219. doi: 10.1016/j.cancergen.2012.03.001
  17. 17. Duan Z, Choy E, Harmon D, Liu X, Susa M, et al. (2011) MicroRNA-199a-3p is downregulated in human Osteosarcoma and regulates cell proliferation and migration. Mol Cancer Ther 10: 1337–1345. doi: 10.1158/1535-7163.mct-11-0096
  18. 18. Ziyan W, Shuhua Y, Xiufang W, Xiaoyun L (2011) MicroRNA-21 is involved in osteosarcoma cell invasion and migration. Med Oncol 28: 1469–1474. doi: 10.1007/s12032-010-9563-7
  19. 19. Liu L, Li H, Li J, Zhong H, Zhang H, et al. (2011) miR-125b suppresses the proliferation and migration of osteosarcoma cells through down-regulation of STAT3. Biochem Bioph Res Co 416: 31–38. doi: 10.1016/j.bbrc.2011.10.117
  20. 20. Jones KB, Salah Z, Del Mare S, Galasso M, Gaudio E, et al. (2012) miRNA signatures associate with pathogenesis and progression of osteosarcoma. Cancer Res 72: 1865–1877. doi: 10.1158/0008-5472.can-11-2663
  21. 21. Lulla RR, Costa FF, Bischof JM, Chou PM, F Bonaldo M de, et al. (2011) Identification of Differentially Expressed MicroRNAs in Osteosarcoma. Sarcoma 2011: 732690. doi: 10.1155/2011/732690
  22. 22. Maire G, Martin JW, Yoshimoto M, Chilton-MacNeill S, Zielenska M, et al. (2011) Analysis of miRNA-gene expression-genomic profiles reveals complex mechanisms of microRNA deregulation in osteosarcoma. Cancer Genet 204: 138–146. doi: 10.1016/j.cancergen.2010.12.012
  23. 23. Namløs HM, Meza-Zepeda LA, Barøy T, Ostensen IHG, Kresse SH, et al. (2012) Modulation of the Osteosarcoma Expression Phenotype by MicroRNAs. PLoS One 7: e48086. doi: 10.1371/journal.pone.0048086
  24. 24. Fritz V, Fajas L (2010) Metabolism and proliferation share common regulatory pathways in cancer cells. Oncogene 29: 4369–4377. doi: 10.1038/onc.2010.182
  25. 25. Kanehisa M, Goto S (2000) KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res 28: 27–30. doi: 10.1093/nar/28.1.27
  26. 26. Csardi G, Nepusz T (2006) The igraph software package for complex network research. Inter Journal Complex Systems 1695.
  27. 27. Huang DW, Sherman BT, Tan Q, Collins JR, Alvord WG, et al. (2007) The DAVID Gene Functional Classification Tool: a novel biological module-centric algorithm to functionally analyze large gene lists. Genome Biol 8: R183. doi: 10.1186/gb-2007-8-9-r183
  28. 28. Huang Y, Chen H-C, Chiang C-W, Yeh C-T, Chen S-J, et al. (2012) Identification of a two-layer regulatory network of proliferation-related microRNAs in hepatoma cells. Nucleic Acids Res 40: 10478–10493. doi: 10.1093/nar/gks789
  29. 29. John B, Enright AJ, Aravin A, Tuschl T, Sander C, et al. (2004) Human MicroRNA targets. PLoS Biol 2: e363. doi: 10.1371/journal.pbio.0020363
  30. 30. Fei J, Li Y, Zhu X, Luo X (2012) miR-181a post-transcriptionally downregulates oncogenic RalA and contributes to growth inhibition and apoptosis in chronic myelogenous leukemia (CML). PLoS One 7: e32834. doi: 10.1371/journal.pone.0032834
  31. 31. Peng R-Q, Wan H-Y, Li H-F, Liu M, Li X, et al. (2012) MicroRNA-214 suppresses growth and invasiveness of cervical cancer cells by targeting UDP-N-acetyl-α-D-galactosamine:polypeptide N-acetylgalactosaminyltransferase 7. J Biol Chem 287: 14301–14309. doi: 10.1074/jbc.m111.337642
  32. 32. Selcuklu SD, Donoghue MTA, Rehmet K, Souza Gomes Mde, Fort A, et al. (2012) MicroRNA-9 inhibition of cell proliferation and identification of novel miR-9 targets by transcriptome profiling in breast cancer cells. J Biol Chem 287: 29516–29528. doi: 10.1074/jbc.m111.335943
  33. 33. Wang W, Zhao L-J, Tan Y-X, Ren H, Qi Z-T (2012) MiR-138 induces cell cycle arrest by targeting cyclin D3 in hepatocellular carcinoma. Carcinogenesis 33: 1113–1120. doi: 10.1093/carcin/bgs113
  34. 34. Galardi S, Mercatelli N, Giorda E, Massalini S, Frajese GV, et al. (2007) miR-221 and miR-222 expression affects the proliferation potential of human prostate carcinoma cell lines by targeting p27Kip1. J Biol Chem 282: 23716–23724. doi: 10.1074/jbc.m701805200
  35. 35. Wu W-H, Hu C-P, Chen X-P, Zhang W-F, Li X-W, et al. (2011) MicroRNA-130a mediates proliferation of vascular smooth muscle cells in hypertension. Am J Hypertens 24: 1087–1093.
  36. 36. Xie Q, Chen X, Lu F, Zhang T, Hao M, et al. (2012) Aberrant expression of microRNA 155 may accelerate cell proliferation by targeting sex-determining region Y box 6 in hepatocellular carcinoma. Cancer 118: 2431–2442. doi: 10.1002/cncr.26566
  37. 37. Zhong Z, Dong Z, Yang L, Gong Z (2012) miR-21 induces cell cycle at S phase and modulates cell proliferation by down-regulating hMSH2 in lung cancer. J Cancer Res Clin Oncol 138: 1781–1788. doi: 10.1007/s00432-012-1287-y
  38. 38. Gilmore TD (2006) Introduction to NF-kappaB: players, pathways, perspectives. Oncogene 25: 6680–6684. doi: 10.1038/sj.onc.1209954
  39. 39. Eliseev RA, Zuscik MJ, Schwarz EM, O'Keefe RJ, Drissi H, et al. (2005) Increased radiation-induced apoptosis of Saos2 cells via inhibition of NFkappaB: a role for c-Jun N-terminal kinase. J Cell Biochem 96: 1262–1273. doi: 10.1002/jcb.20607
  40. 40. Guo L-M, Pu Y, Han Z, Liu T, Li Y-X, et al. (2009) MicroRNA-9 inhibits ovarian cancer cell growth through regulation of NF-kappaB1. FEBS J 276: 5537–5546. doi: 10.1111/j.1742-4658.2009.07237.x
  41. 41. Liu S, Wu L-C, Pang J, Santhanam R, Schwind S, et al. (2010) Sp1/NFkappaB/HDAC/miR-29b regulatory network in KIT-driven myeloid leukemia. Cancer Cell 17: 333–347. doi: 10.1016/j.ccr.2010.03.008
  42. 42. Szklarczyk D, Franceschini A, Kuhn M, Simonovic M, Roth A, et al. (2011) The STRING database in 2011: functional interaction networks of proteins, globally integrated and scored. Nucleic Acids Res 39: D561–D568. doi: 10.1093/nar/gkq973
  43. 43. Stockinger A, Eger A, Wolf J, Beug H, Foisner R (2001) E-cadherin regulates cell growth by modulating proliferation-dependent beta-catenin transcriptional activity. J Cell Biol 154: 1185–1196. doi: 10.1083/jcb.200104036
  44. 44. Liu S, Kumar SM, Lu H, Liu A, Yang R, et al. (2012) MicroRNA-9 up-regulates E-cadherin through inhibition of NF-κB1-Snail1 pathway in melanoma. J Pathol 226: 61–72. doi: 10.1002/path.2964
  45. 45. Knudsen ES, Wang JYJ (2010) Targeting the RB-pathway in cancer therapy. Clin Cancer Res 16: 1094–1099. doi: 10.1158/1078-0432.ccr-09-0787
  46. 46. Tokitou F, Nomura T, Khan MM, Kaul SC, Wadhwa R, et al. (1999) Viral ski inhibits retinoblastoma protein (Rb)-mediated transcriptional repression in a dominant negative fashion. J Biol Chem 274: 4485–4488. doi: 10.1074/jbc.274.8.4485
  47. 47. Ramachandran S, Karp PH, Jiang P, Ostedgaard LS, Walz AE, et al. (2012) A microRNA network regulates expression and biosynthesis of wild-type and DeltaF508 mutant cystic fibrosis transmembrane conductance regulator. P Natl Acad Sci USA 109: 13362–13367. doi: 10.1073/pnas.1210906109
  48. 48. Suzuki H, Ouchida M, Yamamoto H, Yano M, Toyooka S, et al. (2008) Decreased expression of the SIN3A gene, a candidate tumor suppressor located at the prevalent allelic loss region 15q23 in non-small cell lung cancer. Lung Cancer 59: 24–31. doi: 10.1016/j.lungcan.2007.08.002
  49. 49. Dannenberg J-H, David G, Zhong S, Torre J van der, Wong WH, et al. (2005) mSin3A corepressor regulates diverse transcriptional networks governing normal and neoplastic growth and survival. Genes Dev 19: 1581–1595. doi: 10.1101/gad.1286905
  50. 50. Espinosa L, Cathelin S, D'Altri T, Trimarchi T, Statnikov A, et al. (2010) The Notch/Hes1 pathway sustains NF-κB activation through CYLD repression in T cell leukemia. Cancer Cell 18: 268–281. doi: 10.1016/j.ccr.2010.08.006
  51. 51. Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, et al. (2002) The human genome browser at UCSC. Genome Res 12: 996–1006. doi: 10.1101/gr.229102
  52. 52. Matys V, Kel-Margoulis OV, Fricke E, Liebich I, Land S, et al. (2006) TRANSFAC and its module TRANSCompel: transcriptional gene regulation in eukaryotes. Nucleic Acids Res 34: D108–D110. doi: 10.1093/nar/gkj143
  53. 53. Breitkreutz B-J, Stark C, Tyers M (2003) The GRID: The General Repository for Interaction Datasets. Genome Biol 3: R23. doi: 10.1186/gb-2003-4-3-r23
  54. 54. Smyth GK (2004) Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat Appl Genet Mol 3: Article3. doi: 10.2202/1544-6115.1027
  55. 55. Griffiths-Jones S, Grocock RJ, Dongen S van, Bateman A, Enright AJ (2006) miRBase: microRNA sequences, targets and gene nomenclature. Nucleic Acids Res 34: D140–D144. doi: 10.1093/nar/gkj112
  56. 56. Ritchie ME, Silver J, Oshlack A, Holmes M, Diyagama D, et al. (2007) A comparison of background correction methods for two-colour microarrays. Bioinformatics 23: 2700–2707. doi: 10.1093/bioinformatics/btm412
  57. 57. Yang YH, Dudoit S, Luu P, Lin DM, Peng V, et al. (2002) Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation. Nucleic Acids Res 30: e15. doi: 10.1093/nar/30.4.e15
  58. 58. Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc 57: 289–300.
  59. 59. Gautier L, Cope L, Bolstad BM, Irizarry RA (2004) affy–analysis of Affymetrix GeneChip data at the probe level. Bioinformatics 20: 307–315. doi: 10.1093/bioinformatics/btg405
  60. 60. Flicek P, Amode MR, Barrell D, Beal K, Brent S, et al. (2011) Ensembl 2011. Nucleic Acids Res 39: D800–D806. doi: 10.1093/nar/gkq1064
  61. 61. Grimson A, Farh KK-H, Johnston WK, Garrett-Engele P, Lim LP, et al. (2007) MicroRNA Targeting Specificity in Mammals: Determinants beyond Seed Pairing. Mol Cell 27: 91–105. doi: 10.1016/j.molcel.2007.06.017
  62. 62. Chien C-H, Sun Y-M, Chang W-C, Chiang-Hsieh P-Y, Lee T-Y, et al. (2011) Identifying transcriptional start sites of human microRNAs based on high-throughput sequencing data. Nucleic Acids Res 39: 9345–9356. doi: 10.1093/nar/gkr604
  63. 63. Fröhlich H, Speer N, Poustka A, Beissbarth T (2007) GOSim–an R-package for computation of information theoretic GO similarities between terms and gene products. BMC Bioinformatics 8: 166. doi: 10.1186/1471-2105-8-166
  64. 64. Kaufman L, Rousseeuw PJ (1990) Finding Groups in Data: An Introduction to Cluster Analysis. Wiley.
  65. 65. Maechler M, Rousseeuw P, Struyf A, Hubert M, Hornik K (2012) cluster: Cluster Analysis Basics and Extensions, R package version 1.14.3.
  66. 66. Trauwaert E (1988) On the meaning of Dunn's partition coefficient for fuzzy clusters. Fuzzy Set Syst 25: 217–242. doi: 10.1016/0165-0114(88)90189-3
  67. 67. Haldiki M, Batistakis Y, Vazirgiannis M (2002) Cluster validity methods. SIGMOD 31: 40–45. doi: 10.1145/565117.565124
  68. 68. Falcon S, Gentleman R (2007) Using GOstats to test gene lists for GO term association. Bioinformatics 23: 257–258. doi: 10.1093/bioinformatics/btl567
  69. 69. He X, Zhang J (2006) Why do hubs tend to be essential in protein networks? PLoS Genet 2: e88. doi: 10.1371/journal.pgen.0020088.eor
  70. 70. Pons P, Latapy M (2006) Computing communities in large networks using random walks. JGAA 10: 191–218. doi: 10.7155/jgaa.00124
  71. 71. Smoot ME, Ono K, Ruscheinski J, Wang P-L, Ideker T (2011) Cytoscape 2.8: new features for data integration and network visualization. Bioinformatics 27: 431–432. doi: 10.1093/bioinformatics/btq675
  72. 72. Lopes CT, Franz M, Kazi F, Donaldson SL, Morris Q, et al. (2010) Cytoscape Web: an interactive web-based network browser. Bioinformatics 26: 2347–2348. doi: 10.1093/bioinformatics/btq430
  73. 73. Boyerinas B, Park S-M, Hau A, Murmann AE, Peter ME (2010) The role of let-7 in cell differentiation and cancer. Endocr Relat Cancer 17: F19–F36. doi: 10.1677/erc-09-0184
  74. 74. Haug BH, Henriksen JR, Buechner J, Geerts D, Tømte E, et al. (2011) MYCN-regulated miRNA-92 inhibits secretion of the tumor suppressor DICKKOPF-3 (DKK3) in neuroblastoma. Carcinogenesis 32: 1005–1012. doi: 10.1093/carcin/bgr073
  75. 75. Wang X-C, Tian L-L, Wu H-L, Jiang X-Y, Du L-Q, et al. (2010) Expression of miRNA-130a in nonsmall cell lung cancer. Am J Med Sci 340: 385–388. doi: 10.1097/maj.0b013e3181e892a0
  76. 76. Jiang S, Zhang H-W, Lu M-H, He X-H, Li Y, et al. (2010) MicroRNA-155 functions as an OncomiR in breast cancer by targeting the suppressor of cytokine signaling 1 gene. Cancer Res 70: 3119–3127. doi: 10.1158/0008-5472.can-09-4250
  77. 77. Chun-Zhi Z, Lei H, An-Ling Z, Yan-Chao F, Xiao Y, et al. (2010) MicroRNA-221 and microRNA-222 regulate gastric carcinoma cell proliferation and radioresistance by targeting PTEN. BMC Cancer 10: 367. doi: 10.1186/1471-2407-10-367
  78. 78. Luedde T (2010) MicroRNA-151 and its hosting gene FAK (focal adhesion kinase) regulate tumor cell migration and spreading of hepatocellular carcinoma. Hepatology 52: 1164–1166. doi: 10.1002/hep.23854