- Research
- Open Access
Parameter optimization for constructing competing endogenous RNA regulatory network in glioblastoma multiforme and other cancers
- Yu-Chiao Chiu†1, 2,
- Tzu-Hung Hsiao†2, 3,
- Yidong Chen2, 4Email author and
- Eric Y Chuang1, 5Email author
https://doi.org/10.1186/1471-2164-16-S4-S1
© Chiu et al; licensee BioMed Central Ltd. 2015
- Published: 21 April 2015
Abstract
Background
In addition to direct targeting and repressing mRNAs, recent studies reported that microRNAs (miRNAs) can bridge up an alternative layer of post-transcriptional gene regulatory networks. The competing endogenous RNA (ceRNA) regulation depicts the scenario where pairs of genes (ceRNAs) sharing, fully or partially, common binding miRNAs (miRNA program) can establish coexpression through competition for a limited pool of the miRNA program. While the dynamics of ceRNA regulation among cellular conditions have been verified based on in silico and in vitro experiments, comprehensive investigation into the strength of ceRNA regulation in human datasets remains largely unexplored. Furthermore, pan-cancer analysis of ceRNA regulation, to our knowledge, has not been systematically investigated.
Results
In the present study we explored optimal conditions for ceRNA regulation, investigated functions governed by ceRNA regulation, and evaluated pan-cancer effects. We started by investigating how essential factors, such as the size of miRNA programs, the number of miRNA program binding sites, and expression levels of miRNA programs and ceRNAs affect the ceRNA regulation capacity in tumors derived from glioblastoma multiforme patients captured by The Cancer Genome Atlas (TCGA). We demonstrated that increased numbers of common targeting miRNAs as well as the abundance of binding sites enhance ceRNA regulation and strengthen coexpression of ceRNA pairs. Also, our investigation revealed that the strength of ceRNA regulation is dependent on expression levels of both miRNA programs and ceRNAs. Through functional annotation analysis, our results indicated that ceRNA regulation is highly associated with essential cellular functions and diseases including cancer. Furthermore, the highly intertwined ceRNA regulatory relationship enables constitutive and effective intra-function regulation of genes in diverse types of cancer.
Conclusions
Using gene and microRNA expression datasets from TCGA, we successfully quantified the optimal conditions for ceRNA regulation, which hinge on four essential parameters of ceRNAs. Our analysis suggests optimized ceRNA regulation is related to disease pathways and essential cellular functions. Furthermore, although the strength of ceRNA regulation is dynamic among cancers, its governing functions are stably maintained. The findings of this report contribute to better understanding of ceRNA dynamics and its crucial roles in cancers.
Keywords
- Acute Myeloid Leukemia
- Lung Squamous Cell Carcinoma
- Essential Cellular Function
- Essential Biological Function
- ceRNA Network
Background
A group of short single-stranded RNAs, namely microRNAs (miRNAs), has been widely investigated in this decade. With an average length of 22 nucleotides only, miRNAs are not protein coding transcripts. Instead, they fulfill the role of regulators of gene expression by complementarily binding to 3' untranslated regions (3' UTRs) of target mRNA transcripts [1, 2]. According to existing biological evidence, the binding of miRNAs on mRNA can cause mRNA degradation or suppression of translation, and may affect expression of up to one third of the protein coding genes in humans [2]. In cancers, the dysregulation of miRNAs has been proven to be involved in oncogenesis (reviewed in [3]), tumor progression [4, 5], and clinical outcomes, such as patient survival [6, 7]. With advances in next-generation sequencing, a great number of novel miRNAs have been identified and deposited in the public database miRBase [8], increasing the complexity of miRNA regulation.
Recently, reports postulated and experimentally validated that miRNAs can serve as an alternative layer of post-transcriptional gene-gene regulation, namely the competing endogenous RNAs (ceRNAs) [9–11]. Pairs of genes (ceRNAs) fully or partially sharing common binding miRNAs can establish crosstalk with each other through competition for a limited pool of the common miRNAs (miRNA programs; abbreviated as miRP). When expression level of one ceRNA rises (or decreases) in cells, it attracts (or releases) the targeting miRNAs away from (or toward) the other ceRNAs, and in turn has protective (or degradative) effects on expression of the other ceRNA partners. In other words, this postulation provides the scenario that genes can, facilitated by miRNAs, regulate each other without direct interaction. Through bioinformatic analysis and in vitro experiments on the tumor suppressor gene PTEN, previous studies suggested that ceRNAs of PTEN, e.g., VAPA and ZEB2, can possess tumor-suppressive properties by modulating (i.e. coexpressing with) PTEN expression levels in a miRNA-dependent while protein-coding independent manner [9–11]. Through in silico analysis of glioblastoma gene expression datasets, our recent study further demonstrated that ceRNA regulation, while only accounting for a small portion of global gene regulation, plays an essential role in transient cellular responses to dynamic inter-cellular signals [12]. Taken together, these observations have revealed that ceRNA regulation provides an alternative mechanism of gene regulation in essential cellular processes and functions. To address the optimal cellular conditions for ceRNA regulation, several recent studies used the mathematical mass-action simulation [13, 14] and cell line experiments [13] to demonstrate the dependency of ceRNA regulation on the dosages (i.e., cellular concentration) of both ceRNAs and miRPs, and number of miRNA response elements, suggesting the dynamic and condition-specific properties of ceRNA regulation in vitro.
Realizing that biological processes typically involve more complex mechanisms in vivo than in vitro, in the first part of this study we investigate the optimal conditions of ceRNA regulation in expression datasets derived from clinical samples. The optimal conditions may depend on the following essential factors: 1) Size of miRNA programs, 2) Number of miRNA program binding sites, 3) Expression level of miRNA programs, and 4) Expression level of ceRNAs. Here we developed an analytic scheme for determining whether these factors affect strength of ceRNA regulation. By integrating four factors' effects, the biological functions governed by optimal ceRNA regulation can be elucidated. On the other hand, while pan-cancer genomic analysis has been widely utilized to reveal tumor-specific and distinct molecular signature to better understand cancer heterogeneity [15, 16], pan-cancer analysis of ceRNA regulation, to our knowledge, has not been systematically explored. Collectively, the present study provides a systematic investigation of optimal conditions for ceRNA regulation, explores associated biological functions, and conducts pan-cancer analysis of ceRNAs in four cancer types.
Results
Model overview and data preparation
Analysis flowchart of this study. The present study is aimed to systematically explore optimal conditions and related biological functions of ceRNA regulation in GBM, and confer cancer type specific and independent effects. (A) First we defined 47,451,423 putative ceRNA pairs as pairs of genes (i.e., pairs of ceRNAs) sharing any number of predicted targeting miRNAs in the TargetScan database. Pairwise correlation coefficients of putative ceRNA pairs were computed in 481-sample TCGA GBM datasets. (B) Correlation coefficients were partitioned into groups based on the states of each essential factor of putative ceRNA pairs, followed by inter-group goodness-of-fit tests (K-S tests) that pinpointed the essential factors and optimal conditions for ceRNA regulation. (C) 551,175 pairs of ceRNAs fulfilling all of the identified optimal conditions were defined as optimal ceRNA pairs. In order to address differential and constitutive functions, we then included TCGA OV, LUSC, and LAML datasets and performed pan-cancer analysis.
Increased size of miRNA program and number of miRNA program binding sites intensify ceRNA regulation in GBM
Effects of size of miRNA programs and number of miRNA program binding sites on ceRNA regulation. (A, B) Density functions and cumulative distribution functions of correlation coefficients of putative ceRNA pairs. The putative ceRNA pairs were divided into four groups by the quartiles of miRNA program sizes. (C, D) Density functions and cumulative distribution functions of correlation coefficients of putative ceRNA pairs, which were partitioned based on number of miRNA program binding sites.
As defined in the Methods section, the number of miRNA program binding sites (#miRPBS) was determined by summing up the total number of interacting sites of miRPs on the corresponding pairs of ceRNAs. Among all the putative ceRNA pairs the #miRPBS fell in the range of 2-1,859 (histogram in Additional file 2: Figure S2). We grouped the putative ceRNA pairs based on the #miRPBS with identical criteria as used in analyzing size of miRNA programs, resulting in 4 groups of 10,829,459, 12,443,877, 12,296,702, and 11,881,385 putative ceRNA pairs. With the K-S tests, significant p-values between any two groups (p-value ~0) indicate that the number of miRP binding sites is positively associated with ceRNA coexpression (Figure 2C-D). Taken together, our data demonstrate that increased numbers of common targeting miRNAs as well as the abundance of binding sites intensify the strength of ceRNA regulation.
Strength of ceRNA regulation is dependent on expression levels of miRNA programs and ceRNAs in GBM
Effects of expression levels of miRNA programs and ceRNAs on ceRNA regulation. (A, B) Density functions and cumulative distribution functions of correlation coefficients of putative ceRNA pairs. The putative ceRNA pairs were split into groups by quartiles of miRNA programs expression levels. (C) Density functions of correlation coefficients of putative ceRNA pairs. Here the putative ceRNA pairs were categorized based on expression states (i.e., H, M, and L) of their composed ceRNAs. (D) Cumulative distribution functions of (C), focused on comparison of ceRNA pairs composed of two highly expressed genes (H-H) to other ceRNA pairs and non-ceRNA pairs.
Number of ceRNA pairs in groups categorized based on expression states of their composed ceRNAs.
High-expression genes (H) | Medium-expression genes (M) | Low-expression genes (L) | |
---|---|---|---|
High-expression genes (H) | 4551383 (9.59%) | ||
Medium-expression genes (M) | 8262660 (17.41%) | 3786914 (7.98%) | |
Low-expression genes (L) | 10134010 (21.36%) | 9302868 (19.61%) | 5654651 (11.92%) |
Intertwined signaling among optimal ceRNAs is associated with essential cellular functions and disease pathways
The optimal ceRNA regulatory network. (A) Cumulative distribution functions of correlation coefficients of optimal ceRNA pairs satisfying four optimal conditions, other ceRNA pairs, and non-ceRNA pairs. (B) The optimal ceRNA regulatory network. The network is constructed by merging the identified 551,175 optimal ceRNA pairs comprising 2,405 ceRNAs. Nodes and edges denote ceRNAs and optimal regulatory relationship, respectively. (C) The subnetwork of intracellular transport (GO:0046907), generated by extracting 181 genes related to the function and corresponding ceRNA regulatory pairs from (B). Node size is proportional to the number of first-order neighbors and nodes accounting for more than 1% of all intra-function ceRNA regulating pairs are labeled with gene symbols. (D) The subnetwork of protein localization (GO:0008104).
Top 20 hub genes of the optimal ceRNA network.
Hub genes | No. of first-order neighbors | Percentage of total optimal ceRNA pairs | Entrez gene name | Typea | Disease/Functiona |
---|---|---|---|---|---|
CDS2 | 1480 | 0.269% | CDP-diacylglycerol synthase (phosphatidate cytidylyltransferase) 2 | enzyme | |
PARVA | 1470 | 0.267% | parvin, alpha | other | cancer; cell death and survival |
SLC1A2 | 1466 | 0.266% | solute carrier family 1 (glial high affinity glutamate transporter), member 2 | transporter | neurological disease; hereditary disorder |
NFIB | 1435 | 0.260% | nuclear factor I/B | transcription regulator | |
SSR1 | 1421 | 0.258% | signal sequence receptor, alpha | other | |
GTF2H5 | 1407 | 0.255% | general transcription factor IIH, polypeptide 5 | other | neurological disease; hereditary disorder |
SAR1B | 1398 | 0.254% | SAR1 homolog B (S. cerevisiae) | enzyme | hereditary disorder |
GSK3B | 1392 | 0.253% | glycogen synthase kinase 3 beta | kinase | cancer; cell death and survival; neurological disease; hereditary disorder |
HEG1 | 1390 | 0.252% | heart development protein with EGF-like domains 1 | other | cancer |
ZNF148 | 1390 | 0.252% | zinc finger protein 148 | transcription regulator | cell death and survival |
EIF5 | 1387 | 0.252% | eukaryotic translation initiation factor 5 | translation regulator | cancer |
SMAD4 | 1382 | 0.251% | SMAD family member 4 | transcription regulator | cancer; cell death and survival; prognosis biomarker; neurological disease; hereditary disorder |
TCF4 | 1374 | 0.249% | transcription factor 4 | transcription regulator | cell death and survival; neurological disease; hereditary disorder |
QKI | 1366 | 0.248% | QKI, KH domain containing, RNA binding | other | cancer; neurological disease; hereditary disorder |
LSAMP | 1364 | 0.247% | limbic system-associated membrane protein | other | |
ATXN1 | 1354 | 0.246% | ataxin 1 | transcription regulator | cancer; cell death and survival; neurological disease; hereditary disorder |
DCP2 | 1354 | 0.246% | decapping mRNA 2 | enzyme | |
PSD3 | 1346 | 0.244% | pleckstrin and Sec7 domain containing 3 | other | cancer; neurological disease |
SLC38A1 | 1340 | 0.243% | solute carrier family 38, member 1 | transporter | |
VAPB | 1340 | 0.243% | VAMP (vesicle-associated membrane protein)- associated protein B and C | other | cell death and survival; neurological disease; hereditary disorder |
Top 5 clusters of Gene Ontology terms enriched in the 2,405 optimal ceRNAs
GO Term | No. of genes | Bonferroni adjusted P-value | Total No. of optimal ceRNA pairs/ceRNAs | GBM corea | OV corea | LUSC corea | LAML corea | CV among cancersb |
---|---|---|---|---|---|---|---|---|
Cluster 1 (Enrichment Score: 17.99) | ||||||||
GO:0046907~intracellular transport | 184 | 9.26E-18 | ||||||
GO:0008104~protein localization | 225 | 7.88E-17 | ||||||
GO:0015031~protein transport | 200 | 4.67E-16 | ||||||
GO:0045184~establishment of protein localization | 201 | 6.18E-16 | 8229/261 | 7152/261 | 6235/258 | 3961/254 | 2755/247 | 34.85%/2.06% |
GO:0070727~cellular macromolecule localization | 126 | 2.75E-14 | ||||||
GO:0034613~cellular protein localization | 124 | 1.09E-13 | ||||||
GO:0006886~intracellular protein transport | 111 | 3.17E-11 | ||||||
Cluster 2 (Enrichment Score: 8.27) | ||||||||
GO:0009057~macromolecule catabolic process | 180 | 2.95E-08 | ||||||
GO:0044265~cellular macromolecule catabolic process | 168 | 1.04E-07 | ||||||
GO:0030163~protein catabolic process | 143 | 6.75E-06 | ||||||
GO:0043632~modification-dependent macromolecule catabolic process | 134 | 8.41E-06 | ||||||
GO:0019941~modification-dependent protein catabolic process | 134 | 8.41E-06 | 4838/200 | 4418/200 | 3836/197 | 2379/193 | 2049/191 | 31.08%/1.79% |
GO:0051603~proteolysis involved in cellular protein catabolic process | 138 | 1.31E-05 | ||||||
GO:0044257~cellular protein catabolic process | 138 | 1.85E-05 | ||||||
GO:0006511~ubiquitin-dependent protein catabolic process | 68 | 8.96E-05 | ||||||
GO:0006508~proteolysis | 177 | 1 | ||||||
Cluster 3 (Enrichment Score: 6.01) | ||||||||
GO:0016192~vesicle-mediated transport | 141 | 8.98E-08 | ||||||
GO:0016044~membraneorganization | 93 | 3.70E-04 | 3462/157 | 2882/155 | 2519/156 | 1607/149 | 1208/147 | 32.82%/2.53% |
GO:0010324~membrane invagination | 50 | 0.935246 | ||||||
GO:0006897~endocytosis | 50 | 0.935246 | ||||||
Cluster 4 (Enrichment Score: 5.67) | ||||||||
GO:0032446~protein modification by small protein conjugation | 45 | 6.85E-05 | ||||||
GO:0070647~protein modification by small protein conjugation or removal | 50 | 2.36E-04 | ||||||
GO:0016567~protein ubiquitination | 40 | 7.30E-04 | ||||||
GO:0019787~small conjugating protein ligase activity | 45 | 0.012039 | 607/67 | 555/67 | 488/66 | 335/65 | 266/62 | 28.13%/2.88% |
GO:0016881~acid-amino acid ligase activity | 51 | 0.020482 | ||||||
GO:0004842~ubiquitin-protein ligase activity | 39 | 0.084550 | ||||||
GO:0016879~ligase activity, forming carbon-nitrogen bonds | 53 | 0.213866 | ||||||
Cluster 5 (Enrichment Score: 5.10) | ||||||||
GO:0010608~posttranscription al regulation of gene expression | 59 | 0.001017 | ||||||
GO:0006417~regulation of translation | 40 | 0.036298 | 1312/105 | 1132/105 | 961/101 | 643/99 | 480/100 | 31.90%/2.25% |
GO:0032268~regulation of cellular protein metabolic process | 96 | 0.576893 |
Pan-cancer analysis revealed dynamic ceRNA regulation among constitutive ceRNAs
Venn diagrams of core ceRNAs among four cancer types. (A) Comparison of core ceRNAs in four cancer datasets. Core ceRNAs are genes that comprise the core ceRNA pairs with significant positive correlation coefficients in a cancer dataset. (B) Comparison of top hub core ceRNAs which collectively account for 10% of the total core ceRNA regulating pairs in corresponding cancer datasets.
Discussion
Besides the well-studied role of miRNAs in directly regulating gene expression, emerging evidence postulates that ceRNA regulation is an alternative mechanism through which miRNAs participate in gene regulation. Regulation of ceRNAs has been proved to govern essential biological functions in human development and diseases including cancer (reviewed in [24, 25]). While recent studies have proved the dynamicity of ceRNA regulation among cellular conditions based on in silico and in vitro experiments [13, 14], comprehensive investigation into the strength of in vivo ceRNA regulation remains largely unexplored. Addressing this, in the present study we started by characterizing crucial factors in determination of optimal conditions using gene expression profiles derived from tumor specimens from TCGA. Our analyses indicated the dose effect of miRNA programs; i.e., increased size of miRNA programs as well as increased number of miRNA program binding sites enhance the competing relationship among genes and thus elevate inter-ceRNA coexpression. Furthermore, the expressional levels of both miRNA programs and ceRNAs affect ceRNA regulation and lead to statistically significant differences in distributions of correlation coefficients, suggestive of the existence of optimal molecular conditions in which ceRNA regulation prevails. Intermediate expression levels of miRNA programs allow efficient and effective competition, thus further optimize the power of ceRNA regulation; while varied expression levels of ceRNAs exhibited divergent effects on ceRNA regulation. Taken together, our analyses clearly demonstrated that ceRNA regulation highly depends on states of the essential factors and thus may involve complex and dynamic processes in vivo. Incorporating the optimal conditions of ceRNA regulation, we identified the optimal ceRNA pairs and revealed the biological functions significantly associated with protein transport, protein catabolic processes, and regulation of translation. These functions are all of essential significance in regular cellular routines, indicative of the indispensable involvement of ceRNA regulation in vivo. Recently, Denzler et al. assessed the ceRNA effect in hepatocytes and liver using quantitative biological experiments [26]. Our findings agree with their paper where it was concluded that ceRNA regulation is more likely to occur when both ceRNAs are highly expressed or miRNA binding sites are sufficient. Interestingly, analyzing the unusually highly expressed miRNA miR-122, they showed that coexpression of miR-122 target genes was achieved specifically at extremely high target site abundance. Our data further showed the dependence of ceRNA regulation on the essential factors and cancer types. Taken together, we elucidated that ceRNA regulation is a complex and sophisticated mechanism in vivo, thus difficult to be observed under some cellular conditions. Future biological studies may investigate it in detail and carry out essential clues to complex ceRNA regulation.
With the growing volume of DNA microarray and next-generation sequencing samples, pan-cancer analysis has unveiled both common and unique characteristics of genomic aberrations [15], expression profiles [27], oncogenic microRNAs [16], and secretome [28] across cancer types. This emerging research domain illuminates the tumor type-specific and independent molecular properties, and further contributes to enhanced understanding of tumorigenesis and progression. In this pioneer report, for better characterizing ceRNA in cancers, we extended the optimal ceRNA pairs identified from GBM to large datasets deposited in TCGA, including ovarian serous cystadenocarcinoma, lung squamous cell carcinoma, and acute myeloid leukemia datasets. Our results demonstrated that ceRNA regulatory networks are massively rewired across cancer types. Acute myeloid leukemia exhibited the most distinctive ceRNA pattern from other solid tumors. Remarkably, although the strength and wiring of ceRNA regulation changed immensely, the recruitment of genes into the ceRNA regulatory network is highly stable among cancer types; i.e., the highly intertwined ceRNA regulatory relationship enables genes to be effectively regulated by some of their ceRNA partners regardless of perturbations to cellular conditions. This property of ceRNA regulation stabilizes intra-function regulation and thus facilitates maintenance of essential biological functions in cells. With the increasing number of molecular profiles of cancers, future analysis may extend the analysis to more cancer types and provide universal landscape of ceRNA regulation.
In the present study, for each of four factors we attempted with the quartiles and specific percentiles to partition the putative ceRNAs into groups and compare inter-group distribution of correlation coefficients. For inferring more subtle changes in distribution of correlation coefficients, future work may use other methods that are capable of revealing local fluctuations of distribution functions. For defining putative ceRNA pairs, we employed prediction data from the miRNA-target gene prediction algorithm of TargetScan. TargetScan is a widely used prediction algorithm that takes into consideration both sequence complementarity (especially the seed regions of miRNAs) and conservativity of binding sites. There are still a handful of prediction algorithms, such as PicTar, based on genome-wide sequence alignment [29] and mirBridge utilizing gene set enrichment analysis [30]. While different prediction methods, as well as species-specific targeting, define dissimilar miRNA-target gene pairs, since our present report was aimed to investigate the systematic view of ceRNA regulation and its optimal conditions, out of simplicity we only employed the TargetScan algorithm. Indeed, ceRNAs with larger number of targeting miRNAs are expected to have more putative ceRNA partners, and thus account for a higher proportion of the 47 million putative ceRNA pairs. However, in the analysis of optimal conditions for each parameter, the calculation of correlation coefficients was based on "ceRNA pairs" instead of "ceRNAs". Furthermore, since TargetScan is one of the algorithms with the highest prediction precision (reviewed in [31]), in this study the putative ceRNA pairs were defined with high confidence, regardless of the numbers of their ceRNA partners. Thus, we reason that differences in the number of ceRNA partners among genes would not cause major systematic biases to our analysis.
Besides, here we adopted the biologically straightforward Pearson correlation as a measure of gene-gene coexpression, other methods such as mutual information and polynomial regression may provide alternatives for modeling non-linear properties of miRP-modulated coexpression of ceRNAs. In the report, out of simplicity we focused on "pairwise" relationship between ceRNA pairs. However, the competition for a set of miRNAs may not be exclusively limited to pairs of ceRNAs since a handful of ceRNAs can compete, fully or partially, for common targeting miRNAs. Realizing that one miRNA can target up to hundreds of mRNAs in the genome-wide scale, taking all these factors into account will exponentially complicate the problem and thus require more complex mathematical models.
Conclusions
Here we carried out a comprehensive investigation into optimal conditions for competing endogenous RNA regulation, associated biological functions, and pan-cancer effects of ceRNA regulation. Using TCGA GBM microarray datasets, we demonstrated that regulation between ceRNAs is dynamic, however the optimal conditions are quantifiable. The obtained optimal ceRNA regulatory network is associated with diseases pathways and essential cellular functions. Pan-cancer analysis revealed that while strength of ceRNA regulation is dynamic across cancer types, the highly intertwined ceRNA signaling stably maintains the essential functions it governs. Therefore, we expect the study presented here brings biological insights into the dynamicity and essential roles of ceRNA regulation.
Methods
Microarray expression datasets
Microarray datasets of glioblastoma multiforme (GBM) [17], ovarian serous cystadenocarcinoma (OV) [19], lung squamous cell carcinoma (LUSC) [20], and acute myeloid leukemia (LAML) [21] patients were downloaded from The Cancer Genome Atlas (TCGA) database. We extracted 481 GBM samples with paired miRNA and mRNA expression profiles from the datasets of 557-sample Affymetrix Human Genome U133A Arrays and 505-sample Agilent 8 × 15K Human miRNA-specific microarrays. The OV, LUSC, and LAML datasets were composed of 585-sample Affymetrix Human Genome U133A Arrays, 133-sample Affymetrix Human Genome U133A Arrays, and 197-sample Affymetrix Human Genome U133 Plus 2.0 Arrays, respectively. We utilized TCGA Level 3 data, which were previously normalized and merged into gene- or miRNA-level expression values by TCGA, for consequent analysis in this study.
MiRNA targeting genes
where T' is the transpose of T. For each putative ceRNA pair, we calculated the number of miRNA program binding sites (#miRPBS) by summing the numbers of binding sites of all miRNAs belonging to the miRP on the pair of ceRNAs.
Statistical analysis
The Pearson correlation coefficient was utilized to measure degree of coexpression between expression profiles of pair of ceRNAs. In order to identify the essential factors that affect ceRNA regulation, we compared distributions of correlation coefficients obtained from different sets of putative ceRNA pairs. Each set of ceRNA pairs was derived by varying the size of miRPs, number of miRNA binding sites, etc., all essential factors for determination of ceRNA pairs. The goodness-of-fit between two cumulative distribution functions (CDFs) was evaluated with the two-sample Kolmogorov-Smirnov test (K-S test). By measuring the K-S statistic as the maximum vertical distance of two CDFs, the K-S test statistically infers whether two sets of samples are drawn from the same distribution and thus provides a nonparametric measure for comparing two CDFs. The putative ceRNA pairs satisfying the optimal states of essential factors in the GBM dataset were selected and defined as the "optimal ceRNA pairs".
Construction and visualization of ceRNA networks
To confer higher-order signaling properties among ceRNA pairs, we first identified optimal ceRNA pairs and then constructed the ceRNA networks for various cancer types. Nodes and edges in the network denote ceRNAs and co-expressions between ceRNA pairs, respectively. We utilized the open source software Cytoscape [32] for visualization of the network.
Notes
Declarations
Acknowledgements
The study is supported partly by the Ministry of Science and Technology, Taiwan (grant IDs 103-2917-I-002-166 and 103-2320-B-002-065-MY3). The authors wish to thank Center of Genomic Medicine, National Taiwan University for financial support and computing servers. The authors also greatly appreciate the brilliant and constructive inputs into this study from researchers during the 2013 IEEE International Workshop on Genomic Signal Processing and Statistics.
Declarations
The publication costs for this article were funded by the Ministry of Science and Technology, Taiwan (MOST 103-2320-B-002-065-MY3).
This article has been published as part of BMC Genomics Volume 16 Supplement 4, 2015: Selected articles from the IEEE International Workshop on Genomic Signal Processing and Statistics (GENSIPS) 2013. The full contents of the supplement are available online at http://www.biomedcentral.com/bmcgenomics/supplements/16/S4.
Authors’ Affiliations
References
- Bartel DP, Chen CZ: Micromanagers of gene expression: the potentially widespread influence of metazoan microRNAs. Nat Rev Genet. 2004, 5 (5): 396-400.View ArticlePubMedGoogle Scholar
- Lewis BP, Burge CB, Bartel DP: Conserved seed pairing, often flanked by adenosines, indicates that thousands of human genes are microRNA targets. Cell. 2005, 120 (1): 15-20. 10.1016/j.cell.2004.12.035.View ArticlePubMedGoogle Scholar
- Croce CM: Causes and consequences of microRNA dysregulation in cancer. Nat Rev Genet. 2009, 10 (10): 704-714. 10.1038/nrg2634.PubMed CentralView ArticlePubMedGoogle Scholar
- Valastyan S, Reinhardt F, Benaich N, Calogrias D, Szasz AM, Wang ZC, Brock JE, Richardson AL, Weinberg RA: A pleiotropically acting microRNA, miR-31, inhibits breast cancer metastasis. Cell. 2009, 137 (6): 1032-1046. 10.1016/j.cell.2009.03.047.PubMed CentralView ArticlePubMedGoogle Scholar
- Oneyama C, Morii E, Okuzaki D, Takahashi Y, Ikeda J, Wakabayashi N, Akamatsu H, Tsujimoto M, Nishida T, Aozasa K, et al: MicroRNA-mediated upregulation of integrin-linked kinase promotes Src-induced tumor progression. Oncogene. 2012, 31 (13): 1623-1635. 10.1038/onc.2011.367.View ArticlePubMedGoogle Scholar
- Hu Z, Chen X, Zhao Y, Tian T, Jin G, Shu Y, Chen Y, Xu L, Zen K, Zhang C, et al: Serum microRNA signatures identified in a genome-wide serum microRNA expression profiling predict survival of non-small-cell lung cancer. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2010, 28 (10): 1721-1726. 10.1200/JCO.2009.24.9342.View ArticleGoogle Scholar
- Boeri M, Verri C, Conte D, Roz L, Modena P, Facchinetti F, Calabro E, Croce CM, Pastorino U, Sozzi G: MicroRNA signatures in tissues and plasma predict development and prognosis of computed tomography detected lung cancer. Proc Natl Acad Sci USA. 2011, 108 (9): 3713-3718. 10.1073/pnas.1100048108.PubMed CentralView ArticlePubMedGoogle Scholar
- Kozomara A, Griffiths-Jones S: miRBase: integrating microRNA annotation and deep-sequencing data. Nucleic Acids Res. 2011, 39 (Database): D152-D157. 10.1093/nar/gkq1027.PubMed CentralView ArticlePubMedGoogle Scholar
- Karreth FA, Tay Y, Perna D, Ala U, Tan SM, Rust AG, DeNicola G, Webster KA, Weiss D, Perez-Mancera PA, et al: In vivo identification of tumor-suppressive PTEN ceRNAs in an oncogenic BRAF-induced mouse model of melanoma. Cell. 2011, 147 (2): 382-395. 10.1016/j.cell.2011.09.032.PubMed CentralView ArticlePubMedGoogle Scholar
- Sumazin P, Yang X, Chiu HS, Chung WJ, Iyer A, Llobet-Navas D, Rajbhandari P, Bansal M, Guarnieri P, Silva J, et al: An extensive microRNA-mediated network of RNA-RNA interactions regulates established oncogenic pathways in glioblastoma. Cell. 2011, 147 (2): 370-381. 10.1016/j.cell.2011.09.041.PubMed CentralView ArticlePubMedGoogle Scholar
- Tay Y, Kats L, Salmena L, Weiss D, Tan SM, Ala U, Karreth F, Poliseno L, Provero P, Di Cunto F, et al: Coding-independent regulation of the tumor suppressor PTEN by competing endogenous mRNAs. Cell. 2011, 147 (2): 344-357. 10.1016/j.cell.2011.09.029.PubMed CentralView ArticlePubMedGoogle Scholar
- Chiu Y-C, Chuang EY, Hsiao T-H, Chen Y: Modeling competing endogenous RNA regulatory networks in glioblastoma multiforme. Bioinformatics and Biomedicine (BIBM), 2013 IEEE International Conference on: 18-21 Dec. 2013. 2013, 201-204.View ArticleGoogle Scholar
- Ala U, Karreth FA, Bosia C, Pagnani A, Taulli R, Leopold V, Tay Y, Provero P, Zecchina R, Pandolfi PP: Integrated transcriptional and competitive endogenous RNA networks are cross-regulated in permissive molecular environments. Proc Natl Acad Sci USA. 2013, 110 (18): 7154-7159. 10.1073/pnas.1222509110.PubMed CentralView ArticlePubMedGoogle Scholar
- Bosia C, Pagnani A, Zecchina R: Modelling Competing Endogenous RNA Networks. PLoS One. 2013, 8 (6): e66609-10.1371/journal.pone.0066609.PubMed CentralView ArticlePubMedGoogle Scholar
- Kandoth C, McLellan MD, Vandin F, Ye K, Niu B, Lu C, Xie M, Zhang Q, McMichael JF, Wyczalkowski MA, et al: Mutational landscape and significance across 12 major cancer types. Nature. 2013, 502 (7471): 333-339. 10.1038/nature12634.PubMed CentralView ArticlePubMedGoogle Scholar
- Hamilton MP, Rajapakshe K, Hartig SM, Reva B, McLellan MD, Kandoth C, Ding L, Zack TI, Gunaratne PH, Wheeler DA, et al: Identification of a pan-cancer oncogenic microRNA superfamily anchored by a central core seed motif. Nature communications. 2013, 4: 2730-PubMed CentralView ArticlePubMedGoogle Scholar
- Cancer Genome Atlas Research N: Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature. 2008, 455 (7216): 1061-1068. 10.1038/nature07385.View ArticleGoogle Scholar
- Garcia DM, Baek D, Shin C, Bell GW, Grimson A, Bartel DP: Weak seed-pairing stability and high target-site abundance decrease the proficiency of lsy-6 and other microRNAs. Nature structural & molecular biology. 2011, 18 (10): 1139-1146. 10.1038/nsmb.2115.View ArticleGoogle Scholar
- Cancer Genome Atlas Research N: Integrated genomic analyses of ovarian carcinoma. Nature. 2011, 474 (7353): 609-615. 10.1038/nature10166.View ArticleGoogle Scholar
- Cancer Genome Atlas Research N: Comprehensive genomic characterization of squamous cell lung cancers. Nature. 2012, 489 (7417): 519-525. 10.1038/nature11404.View ArticleGoogle Scholar
- Cancer Genome Atlas Research N: Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia. N Engl J Med. 2013, 368 (22): 2059-2074.View ArticleGoogle Scholar
- Dennis G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki RA: DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome biology. 2003, 4 (5): P3-10.1186/gb-2003-4-5-p3.View ArticlePubMedGoogle Scholar
- Huang D, Sherman B, Lempicki R: Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. 2009, 4: 44-57.View ArticleGoogle Scholar
- de Giorgio A, Krell J, Harding V, Stebbing J, Castellano L: Emerging roles of competing endogenous RNAs in cancer: insights from the regulation of PTEN. Mol Cell Biol. 2013, 33 (20): 3976-3982. 10.1128/MCB.00683-13.PubMed CentralView ArticlePubMedGoogle Scholar
- Tay Y, Rinn J, Pandolfi PP: The multilayered complexity of ceRNA crosstalk and competition. Nature. 2014, 505 (7483): 344-352. 10.1038/nature12986.PubMed CentralView ArticlePubMedGoogle Scholar
- Denzler R, Agarwal V, Stefano J, Bartel DP, Stoffel M: Assessing the ceRNA hypothesis with quantitative measurements of miRNA and target abundance. Mol Cell. 2014, 54 (5): 766-776. 10.1016/j.molcel.2014.03.045.PubMed CentralView ArticlePubMedGoogle Scholar
- Parikh N, Hilsenbeck S, Creighton CJ, Dayaram T, Shuck R, Shinbrot E, Xi L, Gibbs RA, Wheeler DA, Donehower LA: Effects of TP53 mutational status on gene expression patterns across 10 human cancer types. J Pathol. 2014, 232 (5): 522-533. 10.1002/path.4321.PubMed CentralView ArticlePubMedGoogle Scholar
- Wu CC, Hsu CW, Chen CD, Yu CJ, Chang KP, Tai DI, Liu HP, Su WH, Chang YS, Yu JS: Candidate serological biomarkers for cancer identified from the secretomes of 23 cancer cell lines and the human protein atlas. Molecular & cellular proteomics : MCP. 2010, 9 (6): 1100-1117. 10.1074/mcp.M900398-MCP200.PubMed CentralView ArticlePubMedGoogle Scholar
- Krek A, Grun D, Poy MN, Wolf R, Rosenberg L, Epstein EJ, MacMenamin P, da Piedade I, Gunsalus KC, Stoffel M, et al: Combinatorial microRNA target predictions. Nat Genet. 2005, 37 (5): 495-500. 10.1038/ng1536.View ArticlePubMedGoogle Scholar
- Tsang JS, Ebert MS, van Oudenaarden A: Genome-wide dissection of microRNA functions and cotargeting networks using gene set signatures. Mol Cell. 2010, 38 (1): 140-153. 10.1016/j.molcel.2010.03.007.PubMed CentralView ArticlePubMedGoogle Scholar
- Witkos TM, Koscianska E, Krzyzosiak WJ: Practical Aspects of microRNA Target Prediction. Current molecular medicine. 2011, 11 (2): 93-109. 10.2174/156652411794859250.PubMed CentralView ArticlePubMedGoogle Scholar
- Smoot ME, Ono K, Ruscheinski J, Wang PL, Ideker T: Cytoscape 2.8: new features for data integration and network visualization. Bioinformatics. 2011, 27 (3): 431-432. 10.1093/bioinformatics/btq675.PubMed CentralView ArticlePubMedGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.