Affara et al. BMC Genomics 2013, 14:23 http://www.biomedcentral.com/1471-2164/14/23 RESEARCH ARTICLE Open Access Vasohibin-1 is identified as a master-regulator of endothelial cell apoptosis using gene network analysis Muna Affara1, Debbie Sanders1, Hiromitsu Araki2, Yoshinori Tamada3, Benjamin J Dunmore1, Sally Humphreys1, Seiya Imoto3, Christopher Savoie2, Satoru Miyano3, Satoru Kuhara4, David Jeffries5, Cristin Print6,7* and D Stephen Charnock-Jones1,8* Abstract Background: Apoptosis is a critical process in endothelial cell (EC) biology and pathology, which has been extensively studied at protein level. Numerous gene expression studies of EC apoptosis have also been performed, however few attempts have been made to use gene expression data to identify the molecular relationships and master regulators that underlie EC apoptosis. Therefore, we sought to understand these relationships by generating a Bayesian gene regulatory network (GRN) model. Results: ECs were induced to undergo apoptosis using serum withdrawal and followed over a time course in triplicate, using microarrays. When generating the GRN, this EC time course data was supplemented by a library of microarray data from EC treated with siRNAs targeting over 350 signalling molecules. The GRN model proposed Vasohibin-1 (VASH1) as one of the candidate master-regulators of EC apoptosis with numerous downstream mRNAs. To evaluate the role played by VASH1 in EC, we used siRNA to reduce the expression of VASH1. Of 10 mRNAs downstream of VASH1 in the GRN that were examined, 7 were significantly up- or down-regulated in the direction predicted by the GRN.Further supporting an important biological role of VASH1 in EC, targeted reduction of VASH1 mRNA abundance conferred resistance to serum withdrawal-induced EC death. Conclusion: We have utilised Bayesian GRN modelling to identify a novel candidate master regulator of EC apoptosis. This study demonstrates how GRN technology can complement traditional methods to hypothesise the regulatory relationships that underlie important biological processes. Keywords: Vasohibin, HUVEC, Bayesian, Gene regulatory network Background generated by modern high throughput methods such as The explosion of systems biology in recent years, facili- microarrays or RNAseq, to holistically model interac- tated by sequencing of the human genome [1,2] and the tions between molecules in cells and tissues. GRN are development of high throughput methods to rapidly usually displayed as directed graphs - nodes represent characterise and quantify biological systems [3-6], has mRNA abundance and edges represent some form of promoted understanding of complex biological and regulatory relationship between the nodes. The reverse pathological processes. Gene regulatory networks (GRN) engineering of GRN from gene expression data has been represent a systems biology approach, taking advantage used to understand molecular interactions in both bac- of the growing number of RNA abundance data sets terial and lower eukaryotic organisms, as well as in more complex mammalian systems. GRN employ simple cor- * Correspondence: [email protected]; [email protected] relation [7] or Boolean [8] methods, algorithms based on 6 Department of Molecular Medicine and Pathology, School of Medical mutual information [9,10] as well as Bayesian methods. Sciences, University of Auckland, Auckland, New Zealand 1Department of Obstetrics and Gynaecology, University of Cambridge, The Computational frameworks have been generated to sim- Rosie Hospital, Robinson Way, Cambridge CB2 0SW, U.K ultaneously perform several types of GRN analysis [11]. Full list of author information is available at the end of the article © 2013 Affara et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Affara et al. BMC Genomics 2013, 14:23 Page 2 of 12 http://www.biomedcentral.com/1471-2164/14/23 Bayesian GRN are in theory especially powerful for in- the more likely it is that accurate predictions can be ference of causal relationships between mRNAs in noisy made by GRN. Therefore, in this current study we have microarray data [12,13]. In Bayesian GRN, the probabil- expanded our previous analysis by combining triplicated ity of the abundance of each mRNA node is modelled eight time point SFD time course data with a much lar- using a function that takes as its inputs the abundance ger library of EC siRNA disruptant microarray data, of the node's parent mRNAs. The edges in a Bayesian which was generated from the knockdown of 351 differ- GRN can represent hidden protein, non-coding RNA or ent mRNA transcripts that encode proteins with a broad metabolite-based regulatory relationships [14]. There- range of functions in EC [11]. This expanded analysis fore, Bayesian GRN can in theory capture information identified numerous GRN master regulators, many of about a subset of the complex cellular regulatory cir- which were already known to play important roles in EC cuitry [15]. Many GRN developed to date have had a biology. However, we noted one major master regulator ‘scale free’ structure [10,16], in which a small number of RNA named Vasohibin-1 (VASH1) that had not at the “hub” RNAs can be identified that are connected to large time been extensively studied in EC apoptosis. There- numbers of downstream RNAs in the networks. These fore, we investigated the function of VASH1 in regulat- hub RNAs are candidate master-regulators of transcrip- ing mRNA abundance and in the process of EC tion and other cellular processes. Their identification is apoptosis. We targeted VASH1 using siRNA and then based on relationships in the data between the hub RNA used quantitative polymerase chain reaction (PCR) to and their downstream RNAs in the GRN structure, examine the abundance of 10 of the 31 mRNAs directly which are usually referred to as "children". Therefore, downstream of VASH1 in the GRN. 7 of these 10 the amount of data behind the identification of hub mRNAs were significantly up- or down-regulated in the RNAs is much greater than the amount of data behind direction predicted by the GRN when VASH1 expression the identification of individual edges, and correct identi- was reduced. We also show that VASH1 is required for fication of hubs may be easier in theory than the correct the apoptotic response in EC treated with SFD. identification of individual edges. Apoptosis is pivotal for normal EC function [17], and Methods the dysregulation of endothelial apoptosis is a key step Cell culture and siRNA transfection in the development of numerous pathologies [18], includ- Umbilical cords were collected with written informed ing cardiovascular disease [19-21] and tumourogenesis maternal consent and the approval of the Cambridge [22-25]. Understanding the regulatory events occurring (UK) Research Ethics Committee. Human Umbilical during this process in a holistic manner may provide Vein ECs (HUVECs) were isolated by collagenase diges- insight into normal vascular development and mainten- tion, as previously described [31]. Cells were cultured in ance, as well as vascular pathologies. Although there has fully supplemented media without antibiotics (basal been extensive characterisation of the EC proteins EBM-2 with a propriety mix of heparin, hydrocortisone, involved in apoptotic pathways [26-28], there have been epidermal growth factor, fibroblast growth factor, vascu- fewer investigations into regulation of the transcriptome lar endothelial growth factor, 2% foetal calf serum (FCS) in ECs undergoing apoptosis [17,29]. Lonza, Cambridge, UK), at 37°C/5% CO2. To carry out To begin to address this issue, our group have siRNA transfection, HUVEC pools consisting of 10 bio- previously used Bayesian GRN to identify molecular logical isolates (of equal contribution) were prepared interactions involved in survival factor deprivation (SFD)- using passage 3 cultured cells. The HUVEC pools were induced EC apoptosis and cell cycle arrest [18]. This plated in 6-well plates at 2.5 × 105 per well and left for previous study used micorarray data over a triplicated 24hrs until approximately 70% confluent. siRNA transfec- eight time point SFD time course. Previous studies have tion was carried out using pools of four siRNA duplexes illustrated the value of supplementing time series data from Dharmacon Inc (Lafyette, Colorado, USA) and the with gene disruption data (e.g. [30]). Since at the time we SiFectamine transfection reagent (ICVEC, London, UK) were especially interested in regulation of the cell cycle, in according to the manufacturer’s instructions. this previous work the time series data was supplemented by eight microarrays from EC cultures treated with RNA processing and microarray preparation W siRNAs targeting molecules associated with the cell cycle. RNA was extracted using TRIzol reagent (Invitrogen, This analysis identified several GRN master regulator London, UK). RNA quality was assessed using the RNAs including the γ-amino butyric acid (GABA)- Agilent 2100 bioanalyser. Biotin labelled cRNA was gen- Receptor Associated Protein (GABARAP) [18]. erated and hybridised on the CodeLink Human Uniset In theory, the greater the volume of high-quality 20K microarrays following the manufacturer’s instruc- siRNA data used to supplement time course data, and tions (Applied Microarrays, Tempe,