Nuclear Factor Kappa B Activation Occurs in the Amnion Prior to Labour Onset and Modulates the Expression of Numerous Labour Associated Genes

Background Prior to the onset of human labour there is an increase in the synthesis of prostaglandins, cytokines and chemokines in the fetal membranes, particular the amnion. This is associated with activation of the transcription factor nuclear factor kappa B (NFκB). In this study we characterised the level of NFκB activity in amnion epithelial cells as a measure of amnion activation in samples collected from women undergoing caesarean section at 39 weeks gestation prior to the onset of labour. Methodology/Principal Findings We found that a proportion of women exhibit low or moderate NFκB activity while other women exhibit high levels of NFκB activity (n = 12). This activation process does not appear to involve classical pathways of NFκB activation but rather is correlated with an increase in nuclear p65-Rel-B dimers. To identify the full range of genes upregulated in association with amnion activation, microarray analysis was performed on carefully characterised non-activated amnion (n = 3) samples and compared to activated samples (n = 3). A total of 919 genes were upregulated in response to amnion activation including numerous inflammatory genes such cyclooxygenase-2 (COX-2, 44-fold), interleukin 8 (IL-8, 6-fold), IL-1 receptor accessory protein (IL-1RAP, 4.5-fold), thrombospondin 1 (TSP-1, 3-fold) and, unexpectedly, oxytocin receptor (OTR, 24-fold). Ingenuity Pathway Analysis of the microarray data reveal the two main gene networks activated concurrently with amnion activation are i) cell death, cancer and morphology and ii) cell cycle, embryonic development and tissue development. Conclusions/Significance Our results indicate that assessment of amnion NFκB activation is critical for accurate sample classification and subsequent interpretation of data. Collectively, our data suggest amnion activation is largely an inflammatory event that occurs in the amnion epithelial layer as a prelude to the onset of labour.


Introduction
Amnion epithelial cells contain large stores of the prostaglandin precursor arachidonic acid (AA) [1,2] and synthesis of its metabolites, especially prostaglandin (PG) E2, increases dramatically at the onset of labour [3]. This increase is associated with a reduction in prostaglandin dehydrogenase activity in the chorion [4] that facilitates PG-modulated cervical ripening and uterine contractility. At labour, PG synthesis in amnion is principally via the inducible cyclo-oxygenase enzyme (COX-2) [5,6,7,8].
The amnion is an important source of pro-inflammatory chemokines and cytokines (eg. IL-8 and IL-1b), the levels of which increase in amnion epithelial cells with the onset of labour [9]. IL-8 acts by attracting neutrophils into the uterine cervix and myometrium [10,11] that subsequently contribute to fetal membrane remodelling and cervical ripening by release of metalloproteinases, such as MMP-8 (neutrophil elastase) [12]. IL-1b elicits a 'positive feed-forward' response to further increase IL-8 synthesis and PG synthesis through upregulation of COX-2. Both COX-2 and IL-8 are regulated in amnion by the transcription factor nuclear factor kappa B (NFkB) [13,14] and IL-1b is NFkB-regulated in a range of cell types [15,16,17].
NF-kB is an inducible transcription factor consisting of DNA binding dimers of various subunit combinations that determine functionality. The NF-kB subunits are derived from transcripts of five genes: NF-kB1 (encoding p50 and its precursor p105), NF-kB2 (encoding p52 and its precursor p100), RelA (p65), c-Rel, and Rel-B. Rel-B, p65, and c-Rel contain C-terminal non-homologous transactivation domains (TADs) that facilitate kinase-induced transcriptional activation. Proteolytic processing of p105 and p100 via a ubiquitin-proteasome pathway [18] leads to the formation of p50 and p52 subunits that lack TADs and are thus considered inhibitory [19]. In cells with basal or no NFkB activity, NF-kB dimers are retained in the cytoplasm in an inactive form by binding to IkB proteins, IkBa, IkBb, and IkBe. NF-kB can also be sequestered in the cytoplasm by the p105 and p100 precursor proteins where they act as IkB proteins. Classically, NFkB activation is provoked through the activation of cell surface receptors such as IL-1, TNF or Toll-like receptors, inducing a signalling cascade which converges on and activates the IkB kinase (IKK) complex, consisting of the regulatory scaffold protein NF-kB essential modulator (NEMO) and IKKa and IKKb kinases [20]. Activation of IKK leads to phosphorylation of the NFkB inhibitor IKBa and its subsequent ubiquitination and degradation [21]. The degradation of ubiquitinated IkB releases NF-kB dimers from the cytoplasmic IkB/NF-kB complex, allowing nuclear translocation of NF-kB to specific recognition elements in target gene promoters to drive gene expression [22,23,24].
In earlier studies conducted at a time (1999)(2000) when elective caesarean sections were routinely performed two weeks prior to the Figure 1. Characterisation of amnion NFkB activity by western blotting. Primary, pre-labour amnion epithelial cells were derived from 12 women undergoing caesarean section. Protein levels of activated NFkB were examined in each sample by immunoblotting for p65 in nuclear extracts, which were then normalised to b-actin (A). Following densitometric analysis, samples were categorised into three groups based upon their level of NFkB activation -low, medium and high (B). The low NFkB activation group consisted of three samples, the medium of seven samples and the high activation group consisted of two samples. Gel to gel variation was accounted for by loading equal volume of a control sample (X) on each gel. ANOVA revealed significant differences between protein levels of nuclear p65 in the three classified groups (C). Levels of nuclear p65 were also shown to correlate with levels of nuclear phosphorylated p65 in a linear fashion (R 2 = 0.6205). doi:10.1371/journal.pone.0034707.g001 estimated date of delivery, we found that amnion epithelial cells obtained following spontaneous labour and vaginal delivery exhibited consistent, high levels of nuclear NFkB-DNA binding and transcriptional activity which could not be further stimulated by incubation with IL-1b [13]. In contrast, amnion cells from cultured placentas collected following elective caesarean section before the onset of labour typically displayed low level nuclear NFkB-DNA binding and transcriptional activity that could be stimulated with IL-1b to levels similar to that seen in post-labour samples [13,17]. This pattern parallels data concerning arachidonic acid metabolism in pre-labour amnion cells in which PG synthesis is low but can be stimulated, whilst in post-labour cells PG synthesis is high and cannot be stimulated significantly further [25].
Following changes to national guidelines, elective caesarean section in the UK is now routinely performed after 39 weeks, closer to the likely time of the onset of labour. We now find that a subset of amnion samples taken from women following pre-labour elective caesarean section show levels of activation similar to those seen following the onset of labour (presumably because they are biochemically closer to the onset of clinical labour). This provides the opportunity to determine the effects of amnion activation at term, independently of any effects that labour and delivery might have upon the amnion. In this study levels of NFkB activity was used as a measurement of amnion activation. To examine which genes are regulated in association with amnion activation, samples were carefully categorised into two groups-i) low activation or ii) Figure 2. Correlations between key components of the canonical, non-canonical and atypical NFkB signalling pathways. Protein levels of activated NFkB were examined in each sample by immunoblotting as previously described. Levels of both nuclear p65 and nuclear phospho-p65 were shown to correlate highly with nuclear levels with Rel-B (A and B, R 2 = 0.8145 and R 2 = 0.6288 respectively). No correlation was detected between nuclear levels of p65 and p50 (C, R 2 = 0.06856), nuclear Rel-B and p52 (D, R 2 = 0.00008) or nuclear p65 and IkBa (E, R 2 = 0.0077). Collectively these results suggest that neither the canonical, non-canonical nor the atypical signalling pathways are responsible for the observed differences in NFkB activation. doi:10.1371/journal.pone.0034707.g002 high activation, and gene differences between the two determined using microarray analysis.

Results
NFkB plays a critical role in fetal membrane activation and the regulation of labour associated proteins such as COX-2 and IL-8. We hypothesised that the activation of NFkB should therefore occur just prior to the onset of labour. To test this, western blot analysis of primary cell cultures derived from 12 individual, term, pre-labour samples was performed to examine nuclear protein levels of p65 in these samples. Nuclear p65 levels varied markedly between samples ( Figure 1A). Three distinct groups could be distinguished which were subsequently divided upon their relative expression levels of nuclear p65 as follows; i) low nuclear p65 concentrations representing low-level NFkB activation and nonactivated amnion (,0.6 a.u. nuclear p65: b-actin), ii) medium NFkB activation (.0.6 a.u. and ,1.4 a.u. nuclear p65: b-actin) and iii) high NFkB activation (.1.4 a.u. nuclear p65:b-actin; Figures 1B and 1C). Similarly, immunoblotting for nuclear phosphorylated p65 showed variable levels in the 12 samples examined. Levels of nuclear p65 were highly correlated to levels of nuclear phosphorylated p65 indicative of the phosphorylationdependant translocation of p65 to the nucleus ( Figure 1D).
The high correlation observed between nuclear p65 and pp65 was suggestive of canonical activation of NFkB, yet high levels of nuclear Rel-B were also detected in the samples (Figure 2A). High concentrations of nuclear Rel-B were strongly correlated with high levels of both nuclear p65 and phosphorylated p65, whereas there was a poor correlation between nuclear p65 and nuclear p50, indicative of non-canonical activation of NFkB. (Figures 2B and  2C). Levels of p52, typically dimerized to Rel-B during noncanonical NFkB activation, were not correlated with nuclear p65 or nuclear Rel-B levels (Fig. 2D). No positive correlation was observed between cytoplasmic IkBa with either nuclear p65 or phosphorylated p65 (Fig. 2E).
Since a strong correlation of nuclear Rel-B with nuclear p65 and phosporylated p65 was consistently observed, the possibility that p65 and Rel-B subunits may themselves physically interact was investigated. Immunoprecipitation studies using anti-p65 antibody were performed in both non-stimulated and IL-1bstimulated pre-labour, primary cultured amnion epithelial cells ( Figure 3A). Immunoblotting using anti-pp65 revealed the presence of complexes containing pp65 in non-stimulated amnion cells. Upon stimulation, concentrations of complexes containing pp65 increased maximally at 30 min and then gradually reduced to 24 h. Dimers of p65-Rel-B were consistently high in stimulated cells through the time course of the experiment. Binding of Rel-B to the NFkB consensus sequence as assessed using a nonradioactive DNA binding assay was increased after 30 min before dropping slightly after 1 h. Peak binding was reached at 4 h before dipping again at 6 h ( Figure 3B).
Prior to microarray analysis, three samples containing high levels of NFkB activation were further characterised by measuring COX-2 expression and COX-2 protein levels. Three samples with low levels of NFkB activation and low COX-2 expression were classified as non-activated along with 3 additional highly-activated samples for microarray analysis (Figure 4). Whole genome analysis using Affymetrix U133 arrays was then performed on nonactivated (n = 3) and activated amnion samples (n = 3). Unsupervised multivariate statistical analyses approaches in the form of PCA and hierarchical clustering were performed on the gene data to examine underlying variance and correlation in the gene expression data ( Figure 5A and 5B). Both approaches revealed clear clustering of activated and non-activated amnion samples indicating that the pre-array molecular characterisation of the samples was representative of the underlying gene expression differences between the sample groups.
Of the 19198 genes examined using the microarray analysis, a total of 919 genes where found to be significantly (P,0.05) increased in activated amnion samples compared to non-activated samples. A list of the top 20 most up-regulated genes present in activated amnion was determined (Table 1) with the highest fold changes detected in COX-2 (644.4), OTR (624.1), chromosome 10 open reading frame (619.6), integrin A2 (617.7) and dimethlyarginine dimethlyaminohydrolase 1 (616). A directed search for inflammatory response genes yielded COX-2, IL-8 (66.2), IL1-RAP (64.5), thrombospondin (62.7), monoglyceride lipase (2.0), nuclear transcription factor X box binding (1.9), CD40 TNF receptor superfamily member 5 (61.7) and macrophage migration inhibition factor (61.7; Table 2). As a further validation of the microarray data, mRNA levels of OTR and IL-8 measured using real-time PCR were found to be significantly increased in the highly activated amnion samples ( Figure 5C and 5D).
To identify pathways in which the identified gene expression changes may be implicated, canonical pathway and gene ontology analysis were performed using IPA. The two main networks identified were i) cell death, cancer and morphology (25 focus genes- Figure 6) and ii) cell cycle, embryonic development and tissue development (17 focus genes- Figure 7).

Discussion
We have previously reported that the degree of amnion activation, as defined by levels of nuclear NFkB-p65, in prelabour amnion cells is highly variable [13]. Following changes to national guidelines, elective caesarean section in the UK is now routinely performed after 39 weeks, closer to the likely time of the onset of labour. In this study we examined amnion activation in cells derived from pre-labour samples collected close to term. A subset of these women exhibited low or moderate levels of NFkB activity whilst others had high activity ( Figure 1). Consistent with our previous findings, these results suggest that activation of NFkB occurs in the amnion epithelial layer as a prelude to the onset of labour where it can drive the upregulation of pro-labour genes such as COX-2 (PGHS-2) and IL-8. This pre-labour activation of NFkB appears to be persistent since it is maintained in cells in primary culture for up to 48 hours and contrasts with uterine myocytes in primary culture in which no activity of NFkB is seen without stimulation (eg by IL-1b) irrespective of whether the cells were collected before or during labour [13]. Persistent or committed activation of NFkB in amnion is logical since the amnion, as part of the fetal membranes, is expelled following delivery with the placenta. In contrast, persistent activation of NFkB in myometrium would be detrimental since it would presumably lead to post delivery myometritis.
Activation of inducible NFkB pathways occurs typically through one of three pathways; canonical, non-canonical or atypical activation, yet the mechanism governing the NFkB activation pathways in the amnion prior to labour is not clearly elucidated. We examined protein levels of numerous key modulators of NFkB activation using immunoblotting of nuclear protein extracts of amnion samples. Consistent with canonical activation of NFkB levels of nuclear p65 in the pre-labour amnion cells were highly correlated to levels of nuclear phosphorylated p65 indicative of the phosphorylation-dependant translocation of p65 to the nucleus ( Figure 1D). However, high correlation between nuclear p65 (both non-phosphorylated and phosphorylated) and nuclear Rel-B was also observed and this is consistent with non-canonical activation of NFkB (Figure 2A). However, if non-canonical activation was responsible for NFkB amnion activation we would expect to see highly correlated levels of nuclear p52 (dimerized to Rel-B during non-canonical activation) however no correlation between nuclear Rel-B and p52 was detected ( Figure 2D). No positive correlation was observed between cytoplasmic IkBa with either nuclear p65 or phosphorylated p65 meaning that NFkB activation could not be attributed to atypical signalling pathways.
Although the most commonly characterised NFkB activation pathways do not seem to be responsible for pre-labour amnion activation, we consistently observed a striking correlation between nuclear p65 and pp65 with Rel-B (R 2 = 0.8157 and R 2 = 0.6288). Using immunoprecipation, we explored the possibility that these subunits may interact physically through protein/protein interactions in both basal and activated amnion cells. For the first time, we have shown that Rel-B and p65 interact in the nucleus of amnion epithelial cells ( Figure 3A). This complex degrades quickly in the first 30 min of IL-1b stimulation before gradually increasing and peaking at 4 h. Similarly, binding of nuclear Rel-B to the NFkB consensus binding sequence using a non-radioactive DNA binding assay kit (TRANSAM) was maximal at 4 h. Rel-B contains a transcriptional activation domain but has the capacity to act as both a positive promoter of NFkB-dependent gene expression as well as a repressor of NFkB activity [26]. Few Rel-B target genes have been reported but its positive transcriptional role has been described through the use of the Rel-B 2/2 mouse. These mice lack the thymic medulla and a class of dendrytic cells suggesting a critical role for Rel-B in the development of secondary lymphatic organs [27]. Interestingly, while wild type fibroblasts lack TNF-a activity, fibroblasts isolated from Rel-B 2/2 mice do [28]. This suggests that one role of Rel-B is to epigenetically silence the TNF-a gene. Consistent with a role in mediating inflammation, Rel-B appears to play a role in mediating early innate immune responses to that of prolonged adaptive innate immune responses by down regulating acute inflammation and activating the maturation of dendritic cells necessary for antigen presentation and T-cell activation [29]. The transcriptional role of Rel-B/p65 complexes in pre-labour amnion epithelial cells remains to be elucidated, their presence in the nucleus and interaction with the NFkB consensus binding sequence along with their ability to modulate transcription warrants future investigation.
To garner a better understanding of the gene regulation underpinning amnion activation prior to labour onset, we then undertook a series of studies in which amnion epithelial cells were collected from a range of women undergoing elective caesarean section before the onset of labour. These cells were established in primary culture and their degree of NFkB activation was assessed directly and as a function of a downstream measure of NFkB activation; COX-2 mRNA and protein expression. COX-2 is an established NFkB dependent gene and marker of amnion activation. Prostaglandins produced via COX-2 facilitate cervical ripening and its increase in activated amnion samples is consistent with it being a critical player in the pathways leading to labour onset. Two distinct groups of samples were identified and classified into either low or high NFkB activity was ( Figure 4). Interestingly, we did not observe a strong correlation between COX-2 mRNA and proteins levels in a number of samples. For example, high levels of COX-2 mRNA detected in sample ''high 2'' were not translated into high protein levels in this sample. In contrast, comparatively low levels of COX-2 mRNA were observed in sample ''high 3'' yet high levels of COX-2 protein were detected in this sample. This may reflect differences in post transcriptional stability between samples, however it is more likely that high mRNA and low COX-2 levels within samples represents increased processing of COX-2 toward prostaglandin production. cDNA microarray analysis was then performed to identify the cohort of genes upregulated in association with amnion activation. The resulting gene expression data was examined using PCA and hierarchical clustering to confirm that the underlying variance in the data was sufficient to classify the samples into two distinct groups ( Figure 5). This step essentially validated the classification A table of the top 20 genes significantly upregulated with amnion activation (Table 1) as well as a list of inflammatory genes upregulated with amnion activation ( Table 2) were generated. We found a high increase in the expression of OTR (24-fold increase) in the fetal membranes upon activation. This is consistent with the concept that OTR is regulated through NFkB and is an important labour-associated gene in amnion [30] as in myometrium [31,32]. The role of oxytocin and its receptor in the amnion is not as well established as in the myometrium since the amnion is noncontractile, however, oxytocin has been shown to stimulate prostaglandin production in rabbit amnion where OTR expression increases 200-fold at the end of pregnancy [33]. Moreover, myometrial contractions can be indirectly modulated by prostaglandin production by the deciduas [34], uterine endometrium and amnion cells [35]. This has therapeutic implications in that future tocolytics such as oxytocin receptor antagonists would likely require, in addition to myometrial contractility suppression, the ability to reach the amnion, chorion and decidua in order to inhibit oxytocin induced prostaglandin production and fetal membrane activation. Small molecule OTR antagonists might therefore have an advantage over OTR peptide analogues.
Ingenuity Pathway Analysis of the microarray data revealed two major networks associated with amnion activation. The first of these networks-cell death, cancer and morphology-is consistent with marked increase in apoptosis observed with the preparation of the amnion for rupture [36,37] via the delamination and apoptosis of amnion epithelial cells ( Figure 6). However, this observation contrasts with the well described role for NFkB as an anti-apoptotic transcription factor [38]. There exists known communication between the apoptotic pathway and NFkB through TNF receptor 1-associated protein TRADD [39] and in cancer tumours, NFkB induced upregulation of COX-2 drives the synthesis of prostaglandins that in turn stimulate prostaglandin receptors and activates cell proliferation and angiogenesis [40]. Consistent with a role in angiogenesis, we observed upregulation of CTGF (connective tissue growth factor, 14.9-fold increase) and CYR61 (cysteine-rich angiogenic inducer, 6.9-fold increase) in activated amnion samples. Similarly, the cell motility modulating genes THBS1 (thrombospondin, 2.7-fold increase), TUBB3 (tubulin beta 3, 3.2-fold increase) and TUBB2A (tubulin beta 2A, 4.5-fold increase) were found to be upregulated with amnion activation. Collectively these results suggest that amnion activation involves NFkB mediated apoptotic and angiogenic pathways which may act to maintain the integrity of the amnion during the hypoxic stress of labour and protect the fetus prior to delivery. The second major gene network implemented in amnion activation was cell-to-cell signalling and interaction, DNA replication, recombination, repair and cellular development (17 focus genes- Figure 7). Much cross talk between cell-to-cell signalling and inflammatory response pathways exist and a number of additional genes involved in mediating inflammation were also observed in this study (Table 2). A 7.5-fold increase in transforming growth factor alpha (TGFa), which positively regulates epidermal growth factor (EGF) receptor activity [41,42] was detected in activated amnion samples. Similarly, we identified a 4.6-fold increase in EGF epiregulin. EGF has previously been reported to upregulate COX-2 in amnion WISH cells [43]. Thus it reasonable to suggest the powerful positive feedback loops of pro-inflammatory cytokines of NFkB and vice versa, EGF on COX-2 and TGFa on EGF may account for the massive upregulation of COX-2 in the amnion activated samples.
Our results describe a complex network of signalling pathways involved in amnion activation, many of which are regulated by NFkB. We have presented a list of genes associated with amnion activation that will serve as an important database for future work and may provide a map for rational therapeutic design to prevent preterm labour and premature preterm rupture of the membranes.

Patient cohort and Ethics Statement
Approval for the study was obtained from the research and ethics committees of the Imperial College Healthcare NHS Trust and the Imperial College and all clinical investigation was conducted according to the principles expressed in the Declaration of Helsinki. Following informed written consent, intact fetal membranes were obtained from women undergoing elective Caesarean section at term prior to the onset of labour (n = 12) or following labour (n = 8). We took prelabour samples only from women whose indication for caesarean section was breech presentation, previous caesarean section or maternal request at between 39 weeks and term in whom there were no contractions. We took post-labour samples only from women who had established in spontaneous labour and had not been given oxytocin or prostaglandin to induce or augment labour. None of these patients had pre-eclampsia, gestational diabetes or any other complication of pregnancy.

Amnion cell preparation
Amion cells were prepared from tissue as we have previously described [44]. Briefly, amnion was separated from the placenta and chorion, rinsed in PBS then cut into strips before incubating in 0.5 mM EDTA in PBS for 15 min. Strips were rinsed in PBS then digested with 2.5 mg/ml Dispase (Life Technologies, Paisley, UK) for 40 min at 37uC. Amnion epithelial cells, were dissociated by shaking vigorously in DMEM supplemented with 10% fetal calf serum (FCS, Sigma, Poole, UK). The cell suspension was centrifuged and the resulting pellet cultured in DMEM supplemented with 10% FCS, 1% L-glutamine and 1% penicillinstreptomycin at 5% CO 2 . For IL-1b treatments, cells were first serum starved for 16 h before 1 ng/ml IL-1b was added and incubated for 15, 30, 60 or 120 min. Amnion samples were characterised as low, medium and high 'activation' based upon the level of nuclear localisation of NFkB p65 as assessed by western analysis. For microarray analysis of changes in gene expression samples were further refined into low and high activation groups based upon both nuclear localisation of NFkB p65 and expression of COX-2 measured by qRT-PCR and western analysis.   in PBS-T and then incubated for 1 h with the appropriate secondary antibody at room temperature. Immunoreactive protein was detected using enhanced chemiluminescence (GE Healthcare) and film exposure. Membranes were stripped (2% SDS, 62.5 mM Tris-HCl, pH 6.7 and 100 mM 2-mercaptoethanol) for 30 min at 50uC before being washed and reprobed for b-actin as a loading control.

Coimmunoprecipitation
Non-labouring amnion epithelial cells were cultured to 90% confluence, serum starved overnight and stimulated with IL-1b for 30 min, 1 h, 4 h or 24 h. Cells were then extracted in a Nonidet P-40 (NP40) based buffer (150 mm NaCl, 50 mm Tris-HCl, 5 mm EDTA, 0.5% NP40). Co-immunoprecipitation was carried out by incubating the 300 mg whole cell extract with NFkB p65 conjugated A/G sepharose beads overnight. As a control, samples were also incubated in the presence of non-specific IgG preparations. Samples were then washed three times in lysis buffer before adding 30 mL of LDS loading buffer and boiling for 10 min. The samples were then loaded onto a 6% polyacrylamide gel, transferred to nitrocellulose as previously described and probed with antibodies raised against phospho-p65 and Rel-B.

RNA extraction, Real-time PCR and microarray studies
Total RNA (1 mg) was isolated and used as a template for reverse transcription. Expression levels of each gene was determined by real-time PCR using an ABI PRISM 7700 sequence detection system according to manufacturer's instructions (PE Applied Biosystems, Forster City, CA). Specific primers were designed for Taqman with the primer express program (PE Applied Biosystems). Acquired data were analysed using Sequence Detector version 1.7 (PE Applied Biosystems) and were normalised to ribosomal L-19.
For microarray analysis RNA was extracted using the Qiagen RNeasyH Mini kit (Qiagen, Crawley, UK) according to the manufacturer's instructions and the concentration and purity determined by measurement of the OD260 and OD280 on a spectrometer. Samples were then sent to Almac Diagnostics (Craigavon, UK) for whole genome analysis using Affymetrix U133 arrays (High Wycombe, UK). For this, 100 ng of total RNA was used for cDNA in the first synthesis using the GeneChipH Expression 39-Amplification Two-Cycle cDNA synthesis kit in conjunction with the GeneChipH Eukaryotic PolyA RNA Control kit (Affymetrix). Cleanup of the double-stranded cDNA was performed using the GeneChipH Sample Cleanup Module, which was followed by amplification and labelling using the GeneChipH Expression 39-Amplification IVT Labelling Kit. Labelled cRNA was fragmented before 15 mg was hybridised for 16 h at 45uC. Finally, the array was washed and stained on a GeneChipH fluidics station 450 and subsequently scanned using a GeneChipH Scanner 3000.

Statistical Analyses
Unless otherwise stated, results are presented as mean 6 SEM. Results were log-transformed as the raw data was not normally distributed and were analysed by repeated measures analysis of variance with a post-hoc Bonferroni correction. P-values,0.05 were considered to be significant.
For microarray data, interpretation of results was carried out using log of ratio. Data was subject to a series of quality assurance filters including a gene expression level filter based on the cross gene error model, which leverages the observed variability of many expressed genes to estimate measurement precision (set to a minimum value of 15.47), a fold-change (minimum value of 1.5) and a confidence filter (P,0.05). The fold-change value of 1.5 was specifically selected to ensure minimise loss of potentially important and relevant biologically changes [45]. The expected proportion of incorrectly rejected null hypotheses (type 1 errors) borne out of multiple comparisons was corrected using the Bejamini and Hochberg False Discovery Rate. A total of 919 genes passed this filter and constituted the stringent gene list.
Principal Components Analysis (PCA) was utilised to examine inherent variation in the sample groups. In PCA, the first principal component represents of the largest amount of correlated variation in the data set. The second principal component (representing the second largest amount of correlated variation in the data set) is placed orthogonally to this and the resulting score plot was examined for any clustering trends. Hierarchical clustering using the average linkage method was used to further demonstrate relationships between gene expression levels in the samples. Software used for data QC was Genespring v 7.0 (Agilent Technologies Inc., CA, USA).
To identify canonical pathways and gene ontology groups for the gene expression data, Ingenuity Pathway Analysis (IPA) softward (Ingenuity Systems, Redwood City, CA) was utilised.