RNF14 is a regulator of mitochondrial and immune function in muscle

Background Muscle development and remodelling, mitochondrial physiology and inflammation are thought to be inter-related and to have implications for metabolism in both health and disease. However, our understanding of their molecular control is incomplete. Results In this study we have confirmed that the ring finger 14 protein (RNF14), a poorly understood transcriptional regulator, influences the expression of both mitochondrial and immune-related genes. The prediction was based on a combination of network connectivity and differential connectivity in cattle (a non-model organism) and mice data sets, with a focus on skeletal muscle. They assigned similar probability to mammalian RNF14 playing a regulatory role in mitochondrial and immune gene expression. To try and resolve this apparent ambiguity we performed a genome-wide microarray expression analysis on mouse C2C12 myoblasts transiently transfected with two Rnf14 transcript variants that encode 2 naturally occurring but different RNF14 protein isoforms. The effect of both constructs was significantly different to the control samples (untransfected cells and cells transfected with an empty vector). Cluster analyses revealed that transfection with the two Rnf14 constructs yielded discrete expression signatures from each other, but in both cases a substantial set of genes annotated as encoding proteins related to immune function were perturbed. These included cytokines and interferon regulatory factors. Additionally, transfection of the longer transcript variant 1 coordinately increased the expression of 12 (of the total 13) mitochondrial proteins encoded by the mitochondrial genome, 3 of which were significant in isolated pair-wise comparisons (Mt-coxII, Mt-nd2 and mt-nd4l). This apparent additional mitochondrial function may be attributable to the RWD protein domain that is present only in the longer RNF14 isoform. Conclusions RNF14 influences the expression of both mitochondrial and immune related genes in a skeletal muscle context, and has likely implications for the inter-relationship between bioenergetic status and inflammation.


Background
We are interested in understanding the regulation of muscle metabolism and its inter-relationships with development, exercise and ageing. A particular focus is the regulation of mitochondrial content which is reported to impact on metabolic syndrome in humans [1] and rats [2], feed efficiency in livestock [3] and mammalian ageing [4]. Mitochondrial content also contributes to athletic performance [5] and post-mortem meat quality [6] through the connection to muscle fibre type. The physiological connection between inflammation and muscle biology in the context of training, muscle remodelling and ageing is also of interest. For example, during exercise, muscle is routinely subject to various stressors, such as mechanical damage, hypoxia and pH decline that set in motion a proinflammatory cascade [7][8][9] which has implications for tissue remodelling. Previously, using a network science approach in cattle [10] we identified the product of the ring finger protein (RNF14) gene (aliases ARA54, TRIAD2), an incompletely characterised transcriptional regulator, as a factor that might influence not only mitochondrial transcription and function but also immune function. In this study we aimed to explore and validate the accuracy of this reverse-engineering under the tightly controlled experimental conditions afforded by in vitro cell culture.
A major foundation of our predicted function of the RNF14 protein was a bovine co-expression network [10]. Various metabolic and developmental processes were prioritised for further scrutiny on the basis of forming cohesive co-expression network gene sets or 'modules.' To build the network, bovine muscle sampled at different times during pre-and post-natal development, between genetically divergent breeds and following nutritional intervention, were subject to microarray analysis. By hunting in the module of interest for transcriptional regulators (DNA binding transcription factors and co-factors), or asking the related question "which transcriptional regulator has the highest absolute, average correlation to all the genes in the module?", we generated a ranked list of regulators predicted to control the processes in question.
These approaches correctly identified groups of genes already known to play a role in mammalian skeletal muscle biology, including master regulators of the cell cycle (E2F1), fast twitch muscle development (SIX1) and mitochondrial biogenesis (ESRRA) [10]. Several regulatory molecules of unassigned or poorly documented function were also implicated in some of these processes, and they became candidates for future gene function validation efforts. While most genes are clearly defined within the network, one prominent gene, Ring Zinc Finger 14 (RNF14) a transcriptional co-activator known to bind the androgen receptor [11], gave apparently ambiguous results. Namely, it was assigned likely roles in two different processes by our analysisimmune function and mitochondrial function [10]. Given that the majority of connections in the co-expression network are positive, the prediction is that an increase in the activity of the 'regulator' will increase the expression of the 'target' genes.
A separate analysis on a different data set used a differential network strategy called Regulatory Impact Factor analysis (RIF) [12] to contrast mitochondrial-rich brown fat versus mitochondrial-poor white fat in mice. This analysis independently assigned high likelihood to a causal role for Rnf14 in driving the phenotype differences between these cell types, also suggestive of a role in mitochondrial function and content [13]. Little is known of the function of the RNF14 protein, other than it is broadly expressed across tissues [14] and is a transcriptional co-activator that interacts with the androgen receptor transcription factor in pathways relating to sex steroid signalling. From a structural perspective, there are six RNF14 transcript variants in humans and three in mouse, in both cases producing two different protein isoforms.
The objective of this study was to characterise the regulatory role of two RNF14 isoforms in mouse muscle.
We achieved this via experimental upregulation followed by functional analysis of the subsequent genome-wide transcriptional readout. We performed a transient transfection of transcript variants encoding the two different isoforms in mouse C2C12 cells. The resultant gene expression perturbations in a number of chemokines, Interferon Regulatory Factors and related interferon signalling molecules support a role for Rnf14 in skeletal musclemediated immune and inflammatory function. Additionally, the longer transcript variant 1, which encodes a protein isoform containing an RWD domain, yielded a coordinate upregulation trend of all mitochondriallyencoded mitochondrial proteins present on the array platform (12 of the total 13) reinforcing the proposed link to the mitochondrion.

Expression constructs
PCR resulted in two differently sized Rnf14 amplicons from mouse muscle cDNA. These amplicons were individually cloned and sequenced. In both cases the sequences exhibited >99% sequence identity to the GenBank Rnf14 sequence. The longer sequence was 1457 bp and BLASTN aligned this sequence to the Mus musculus Rnf14 transcript variant 1. The shorter sequence we amplified was 1306 bp, and this was aligned by BLASTN to the Mus musculus Rnf14 transcript variant 3 (summarised in Figure 1).
Rnf14 transcript variant 1 contains an ORF capable of producing E3 ubiquitin protein ligase RNF14 protein isoform A and Rnf14 transcript variant 3 the ORF for E3 ubiquitin protein ligase RNF14 protein isoform B. The two isoforms are identical at the C-terminus end of the protein, while the longer isoforms has an RWD domain at the N-terminus end not present on the shorter isoform. Variant 2 does not encode a protein.

Microarray expression measurements
The array platform used to interrogate the C2C12 response to transfection measures genome-wide transcriptional changes using 18,129 probes. This platform contains three probes predicted to bind Rnf14; ILMN_2675078, ILMN_2682811, ILMN_2868579. Their proposed binding sites are illustrated on Figure 1. Probe ILMN_2675078 was elevated~3.4-fold in the variant 3 transfected cells and 1.1-fold in the variant 1 transfected cells. This 3.4-fold change made Rnf14 the 9th most differentially expressed (DE) gene found in cells transfected with variant 3 out of the 18,129 probes with detectable signals in at least one treatment. Correcting for an overall transfection efficiency of~10% implies that an individual transfected cell showed an increase in expression of 34-fold and 11-fold respectively.
The other two Rnf14 probes did not report a change in expression of Rnf14 following transfection, including ILMN_2682811, predicted to bind variants 1 and 2 but not transcript variant 3 (Table 1). It is not clear whether these 2 probes are reporting correctly. To unravel these observations and to further document the technical implentation of the Rnf14 transfections we performed qRT-PCR on the RNA prepared from the transfected C2C12 cultures using primers designed to detect the transfection construct produced mRNA and endogenous forms of Rnf14 mRNA.

qRT-PCR expression measurements
To compare and contrast the Rnf14 transcript variants in the different treatment groups, we designed a set of discriminatory qRT-PCR primers ( Table 2). The ANOVA model for the analysis of Ct values from the qPCR experiments accounted for 93.03% of the total variation. The two main effects (primer and treatment) and their interaction were all highly significant (P < 0.0001).
The normalised expression results are summarised in Table 3. All primer sets yielded unique dissociation curves indicating the presence of Rnf14 transcripts in the anticipated samples. The no template controls yielded no product in all cases. The transfection of Rnf14 variant 3 was clearly evident in the variant 3 transfected cells but not the other C2C12 cultures (P < 0.0001). The transcriptional output of this expression construct was also clearly detectable in the primer set that amplifies all Rnf14 mRNA species (i.e. endogenous and construct-based) yielding a 30-fold increase (P < 0.0001). The disparity between this result and the array fold-change is presumably attributable to the enhanced sensitivity of qRT-PCR.
Similarly, the transfection of the Rnf14 variant 1 was also clearly detectable in the variant 1 transfected cells but not the other cell cultures (P < 0.0001). The expression by the construct was not as effective as for variant 3 and was not reflected in increased amplification by the all Rnf14 species (endogenous plus construct based) primer set, presumably because it forms a much more modest proportion of all the Rnf14 mRNA species in the system. There was very little variability in native Rnf14 expression (i.e. all three endogenous variants summed) across the three treatment groups. Overall, these findings imply the global transcriptional readout in the transfected C2C12 cells described by the microarray platform can be attributed to the impact of the transfection of the Rnf14 variants.

Functional enrichment analyses
With regard to the microarray analysis, a mixed-model normalisation procedure was applied to the raw intensity values, as previously described [15]. Clustering on columns discriminates the treatments based on global considerations of the gene expression patterns. The visualisation of the clustering analysis indicated that two of the six control samples were outliers and hence were removed for the subsequent DE analysis. Overall, the global gene expression patterns in the two control groups (no transfection and empty construct transfection) could not  Assuming a transfection efficiency of~10%, the fold-changes for the transfected cells are~33-fold and 12-fold on a per cell basis (based on ILMN_2675078). The location of the probes in the different Rnf14 transcripts is given in Figure 1.
be discriminated from each other. Consequently, the gene expression values for the two control groups were combined to form a single control for the purposes of computing DE between controls and treatments. The Rnf14 variant 3 and 1 transfected cells were both clustered separately to the controls (more so) and separate to each other (less so).
We computed a list of DE genes (Tables 4 and 5) as previously described [16], performing the analysis at the probe level. To explore the genome-wide expression output for functional enrichment we determined statistically significant DE for each treatment contrast. We then submitted the DE lists versus a background list of all genes present on the array to the GOrilla webtool [17], which uses hypergeometric statistics to determine functional enrichments.
With regard to upregulation following over-expression of the shorter variant 3, "response to biotic stimulus", "chemokine activity" and "extracellular region" gave Pvalues of 8.44E -12 , 2.93E -7 and 7.4E -14 for the Process, Function and Component ontological levels respectively.
With regard to upregulation following over-expression of the longer variant 1, the top functional enrichments were "immune response," chemokine activity" and "extracellular region" with hypergeometric P-values of 1.98 E -9 , 3.83 E -6 and 2.06 E -10 .
Various extracellular region components were clearly among the most downregulated genes following transfection with both variants ( Table 5). The enrichment was highly significant in both cases, but more significant in the transcript 3 (P = 3.69 E -19 ) compared to the transcript 1 transfected cells (P = 1.32 E -14 ).
Contrasting effects of the variant 1 transfection with that of the variant 3 transfection yielded similar functional enrichments of "response to biotic stimulus" (3.94E -8 ), signal transducer activity (2.15E -4 ) and extracellular region (1.77E -5 ). The identity of the perturbed genes is further illustrated on Figure 2 and tabulated on Tables 4 and 5. The normalised mean expression results for the entire data set are in Additional file 1. Transfection with the empty construct could not be discriminated from untransfected cells, ruling out the possibility that these responses are an experimental artefact relating to the presence of the construct. The overall spread of the perturbed transcripts is greater (4-fold DE in both directions) in the variant 3 transfected cells, which may reflect the greater abundance of the variant 3 based construct.
We also computed a modified DE metric called Phenotypic Impact Factors (PIF) [12,13], a product of the average abundance of the gene and its DE. We have previously found that this accounts for the increased noise of the rarer transcripts and increases the sensitivity for detecting DE of the more abundant transcripts [12]. Figure 3 illustrates those genes either DE or awarded a high PIF score in at least one of the two transfections. Immune and mitochondrial genes are highlighted based on functional annotations performed by importing the list into the DAVID web tool [18]. Immune genes and nuclear-encoded mitochondrial genes are prominent among the DE genes but the direction of change is not consistent.
On the other hand, among the significant PIF transcripts in the cells transfected with the long transcript variant are three of the 13 mitochondrially-encoded mitochondrial proteins (Mt-coxII, Mt-nd2 and Mt-nd4l).
Moreover, a deeper exploration shows that all the mitochondrially-encoded genes represented on the array (12 of the 13) display a coherent trend of upregulation Transciption from the Actb gene was used as a control. A schematic of the known mouse Rnf14 mRNA species is presented in Figure 1. The standard error is 0.447 in all cases because we performed a system-wide normalization. 'No amplification' denotes multiple dissociation peaks i.e. non-specific amplification.
in the variant 1 transfected cells based on at least one probe ( Figure 3; Table 6; Additional file 2). The standard error bars for Figure 3 were calculated using the standard curve method [19]. A subset of these mitochondrially encoded genes that are significantly DE as assayed in pairwise comparisons are also highlighted on Figure 2. The upregulation of mitochondrially-encoded mitochondrial proteins was not observed in the Rnf14 variant 3 transfected cells (Additional file 2). The full list of DE and differentially PIF genes can be found in Additional file 3.

Motif analysis and bioinformatics
We next attempted to identify regulatory motifs that are conserved among the RNF14 responsive genes in an attempt to determine the cellular pathway linking Rnf14 upregulation to the observed mitochondrial and inflammatory output. Hunting for conserved Transcription Factor Binding Sites 1000 bp upstream of the DE genes using Whole Genome RVista for mouse, enriched for Hfh3 in both the variant 1 (P = 9.29-log 10 ) and variant 3 transfection experiments (P = 6.57-log 10 ). We also undertook a combination of bioinformatic analyses and literature mining. A protein motif analysis of RNF14 within UNIPROTKB indicated the presence of 1) an N-terminal destruction box which could act as a recognition signal for ubiquitin proteosome degradation and 2) RING type zinc finger essential for interaction with UBE2E2.

Discussion
In an attempt to infer the functional role of the RNF14 protein, previously found to be co-expressed with mitochondrial and immune genes in developing bovine longissimus muscle, we transfected each of two Rnf14 transcript variants into a mouse myoblast cell line. Interestingly, and despite low expression levels of that transcript, transfection with transcript variant 1 culminated in a significant upregulation in the transcription of three of the 13 mitochondrially-encoded mitochondrial genes (i.e. Mt-nd4L, Mt-coxII and Mt-nd2) [12]. Furthermore, while we only have microarray expression data for 12 of the 13 of these genes (Mt-coxI is missing) a deeper exploration shows that the remaining nine all display a modest but coherent trend of upregulation. The direction of this observation is consistent with the initial network connections being based on positive rather than negative co-expression values. While the fold-change is Genes may be represented by more than one probe. The controls are a combination of untransfected cells and cells transfected with an empty construct. only 1.1 to 1.4-fold, all these transcripts are very abundantly expressed which provides a favourable signal to noise ratio for reliable detection. The array does not report on the 22 mitochondrially encoded tRNAs and two ribosomal RNAs that make up the remaining transcriptional output of the mammalian mitochondrion, which encodes 37 different genes in total. The expression of a number of nuclear-encoded mitochondrial proteins was also upregulated following transfection of Rnf14 variant 1 (Figure 2). For example, Cmpk2 (alias Tyki) (2.4-fold up regulated) is a nucleoside monophosphate kinase that localises to the mitochondria and has previously been found to be tightly correlated with macrophage activation and inflammation [20]. A very recent publication has documented Rnf14 as a positive regulator of canonical Wnt signalling in human cells [21], with canonical Wnt signalling previously reported to be a potent activator of mitochondrial biogenesis [22]. This recent body of work clearly complements our findings linking Rnf14 with mitochondrial physiology.
While the broad transcriptional impact of Rnf14 variant 1 transfection on our samples was in line with our functional prediction, there were some interesting deviations. Firstly, the nuclear and mitochondrially-encoded mitochondrial proteins occupy distinct parts of the original bovine muscle co-expression network [10]. While Rnf14 sits in the nuclear-encoded portion of the in vivo bovine network, transfection with variant 1 appears to exert the most coherent transcriptional influence on the mitochondrially-encoded mitochondrial proteins. By way of contrast, Rnf14 variant 3 transfection did not lead to a detectable change on the expression of mitochondriallyencoded mitochondrial genes, despite an overabundance of the variant 3 transcript in the variant 3 transfected cells.
Both Rnf14 variants influenced expression of genes encoding proteins relating to immune function. Unlike the mitochondrially-encoded mitochondrial proteins upregulated observed in Rnf14 variant 1 transfected cells, genes belonging to inflammatory processes were both up-and downregulated. Prominent among the perturbed immune genes were chemokines (e.g. Ccl2, Ccl4, Ccl5, Ccl7, Cxcl10, Cxcl12), interferon regulatory factors and related interferon responsive and signalling genes (e.g.  Irf1, Irf7, Irf9, Isg15, Isg20, Ifit2, Ifit3, Psmb8, Usp18, Adar, Gbp2). In humans following eccentric exercise, the in vivo inflammatory response includes activation of chemokines [23]. A number of these genes also imply apoptosis, a mitochondrial phenomenon [24]. These data go some way towards resolving the question posed by our apparently ambiguous (i.e. strong co-expression to both mitochondrial and immune genes) observations from the bovine muscle co-expression network, and imply that both mitochondrial and immune predictions are supported, depending on the particular transcript variant under consideration. Both RNF14 motifs (N-terminal destruction box and RING type zinc finger) indicate some involvement in ubiquitin mediated proteolysis which ties in with apoptosis, and UBE2E2 is known to play a specific role in adaptive immunity signalling. The motif analysis shows that most of the large transcription factor motifs (Zinc fingers and RNA binding domains) of the protein reside in the C-terminus shared by both isoforms, while the missing amino acids in the shorter isoform result in the loss of an RWD domain. We hypothesise that the RWD domain accounts for the mitochondrial response observed after transfection with Rnf14 transcript variant 1. Recent work has emphasised deep functional connections between mitochondria and innate immunity in general [25,26], and mitochondria and antiviral processing in particular [27], which is clearly of interest given the very same dual roles outlined here for RNF14.
The downregulation of a set of extracellular region and extracellular matrix transcripts following transfection with both Rnf14 transcript variants was unexpected. Example downregulated molecules common to both transfections included multiple collagen isoforms (Col14a1, Col6a2, Col6a1, Col8a2, Col16a1), other matrix structural components (Dcn, Mglap) and matrix remodelers (Mmp2, Adamts2). We ascribe these observations to one of two phenomena. On the one hand, it may reinforce the transmission of the immune signals we have observed, as it has been documented that the extracellular matrix plays a crucial role in the inflammatory process [28]. Alternatively, the signal may correspond to differences in myocyte progression through proliferation and differentiation, the transition through which is known to be accompanied by various changes in matrix-mediated adhesion [29].
Interpreting the various lines of evidence linking RNF14 protein to immune and mitochondrial functions is complicated by the cross-species sources of data. The original co-expression prediction of RNF14's gene function was made mainly from bovine expression data. Disentangling the various pieces of information is challenging given cattle are a non-model organism and we have incomplete knowledge of bovine functional genomics. For example, it is not clear how many bovine RNF14 transcript variants exist in total, which clearly complicates our (co-expression) interpretation of the Agilent probe that provided the original foundation for some of the predictions.
Nevertheless, the outcome of this validation experiment supports three of our systems biology gene discovery approaches: 1) Partial Correlation and Information Theory (PCIT) [30] 2) Module-to-Regulator analysis [10] and 3) Regulatory Impact Factors [12,13]. While there is some overlap in the exact molecules present in the coexpression modules and those perturbed in this subsequent transfection experiment (Prsmb8 and Irf1 being common to both in an immune context), the overlap is very patchy. This implies that predictions based on considerations of co-expression or differential co-expression are perhaps best made in terms of broad function rather than specific molecules.

Conclusions
Rnf14, which encodes a transcriptional co-factor, acts via two differentially spliced transcripts to modulate innate immune responses, tissue energy homeostasis and the tissue matrix. We do not know under what stimuli either transcript is produced by cells of different lineages. We have shown that gene expression information generated in a non-model organism can be used to develop hypotheses that can be validated in more conventional systems. The validation of this approach enhances the intrinsic value of many existing datasets generated in a range of species as it provides a methodology for detailed analysis of fundamental biological phenomena, such as mitochondrial transcription and biogenesis. Future work could aim to tease out whether one or more of the RNF14 protein isoforms are mitochondrially colocalised in addition to the mRNA transcripts being mitochondrially co-expressed, and/or whether in vivo evidence can be detected for a role in immune function  and mitochondrial transcription or biogenesis to supplement the in vitro data we have presented.

Prioritisation of RNF14 for experimental analysis
Three separate lines of evidence were used to make a prediction of RNF14's possible role in immune versus mitochondrial function. Firstly, a co-expression network on developing bovine muscle samples -inferred using the PCIT algorithmconnected RNF14 to a dense nuclear-encoded mitochondrial module (comprising the following with correlations >0.85 MDH1, PDHA1, NDUFS2, ES1, UQCRC1, NDUFV1, MDH2, GOT2, NDUFA9, SOD2, CYCS, NDUFV2, SDHA, BRP44, ACO2, APOO) [10]. The position of RNF14 in the mitochondrial module of the co-expression network made a prediction about a role in mitochondrial function.
Thirdly, application of the Regulatory Impact Factor (RIF) algorithm to an entirely independent data set (comparing high-mitochondrial content brown fat versus low-mitochondrial content white fat) awarded high likelihood of phenotype causality to the RNF14 protein [13].

Amplification, cloning and transformation of two transcript variants of RNF14
The mouse ortholog for RNF14 (GenBank accession NM_020012.1) was identified through sequence alignment to the bovine RNF14 protein. Mouse muscle cDNA was used as template for a PCR designed to amplify the full length mRNA encoding the Rnf14 gene. To increase recognition of the first ATG by the ribosome, the forward primer incorporated the mammalian Kozak sequence (CCATGG) prior to the start of the Open Reading Frame. PCR primers also incorporated vector sequence, restriction enzyme sites used for cloning (NcoI at 5' and XhoI 3') and gene specific sequence. The forward primer for PCR was 5'-GAGATATGCCACCATGGAGT CGGCAGAAGACCTGGAAGCCCAG-3' and the reverse primer 5'-GTGATGGTGGTGCTCGAGGTCGTCGTCG TCGTCTTCATCATCGT-3'.
A conventional PCR with a 50°C annealing temperature using high fidelity Taq polymerase, amplified a band of 1457 base pairs. This was gel purified and cloned into the pTANDEM-1 vector (EMD4 Biosciences, Merck, USA) using a fusion homologous recombination method. The resultant expression plasmid contained a His 6 tag at the C terminus of the RNF14 gene for protein purification and detection by antibodies.
The two variants are referred to as Rnf14 transcript variant 1 and Rnf14 transcript variant 3. Following The homeobox genes encode a highly conserved family of transcription factors that play an important role in morphogenesis transformation into TOP10 chemically competent E.coli, clones were picked and diagnosed for the presence of the insert in the correct orientation using restriction digests (XhoI and NcoI) followed by gel electrophoresis. Clones predicted to contain the correct insert were sequenced from both ends of the expression construct using the BGH, T7 and pTandem1 Tandem DOWN1 primer and aligned to the GenBank mouse RNF14 sequence. Expression constructs containing the correct sequence were midi-prepped (Qiagen) following the manufacturer's instructions and stored at 1 μg/μl at −20°C in preparation for transfection. The Qiagen midi-prepping protocol removes LPS and provides transfection-ready reagents.
A control construct containing no insert (hereon referred to as "empty") was also produced and subsequently used for transfection. A separate control was derived from untransfected cells.

Cell culture and transfection conditions
The murine myoblast cell line, C2C12, was obtained from the American Type Culture Collection and cultured in DMEM supplemented with penicillin/streptomycin and 10% fetal bovine serum. Transfection conditions for the C2C12 cells (passage numbers 15 -20) were explored for a range of confluences and Lipofectamine 2000 (Invitrogen) transfection reagent: DNA ratios. A pEGFP.C3 vector containing Green Fluorescent Protein (GFP) was used to visualise transfection efficiency under a Zeiss AX10 fluorescence microscope fitted with an Axiocam camera. As previously reviewed [31], we found that C2C12 cells transfected with a low transfection efficiency of~10%. Optimal cell number was found to be 5 × 10 4 cells/well transfected with 5 μl of Lipofectamine 2000 and 800 ng DNA per well for 48 hours.
The experiment was run in two 24-well Geltrex-coated plates using the optimised procedure. Cells were cultured for 24 hours prior to transfection. At the time of transfection the cells were~50% confluent. Two independent controls were used, one with no transfection and the other through transfection of the empty construct.
Twelve independent wells were run for each of the four treatment groups (no transfection, empty construct, Rnf14 short transcript variant and Rnf14 long transcript variant) giving a 48 well cell culture experiment in total. At 48 hours post transfection the cells from each well were harvested for RNA, while a GFP transfection run in parallel was fixed and stained to provide an estimate of transfection efficiency.

RNA extraction and microarray hybridisation
Total RNA was independently extracted from each of the 48 cell culture wells using QIAshredder homogenates and QIAgen RNeasy columns (Qiagen), incorporating the oncolumn DNase treatment (Ambion) to remove genomic DNA contamination. RNA integrity and purity was assayed visually by gel electrophoresis and additionally by A 260/280 spectrophotometry. All the RNA samples resolved into discrete 16S and 26S ribosomal RNA bands.
Because each of the four treatment groups comprised 12 independent RNA samples, these 12 were pooled into 3 major groups of 4 replicates with each replicate contributing 0.5 μg. This process yielded 12 by 2.0 μg pooled RNA samples as follows (no transfection 1, no transfection 2, no transfection 3, empty construct 1, empty construct 2, empty construct 3, short transcript 1, short transcript 2, short transcript 3, long transcript 1, long transcript 2, long transcript 3).
The 12 RNA pools were submitted to the Gene Expression Centre at the University of Queensland's Institute for Molecular Bioscience for cDNA synthesis and hybridisation to the Illumina WG6 mouse microarray platform. This facility performed an additional set of quality checks based on possession of an RNA Integrity Number (RIN) >8 assayed on a Bioanalyzer.

Statistical analysis
The analytical procedures to normalize the data and to identify differentially expressed genes were based on methods developed by our group and published elsewhere. In particular, we followed the methodology described in [15,32,33].
The raw data from the Illumina Microarray System contained expression signals from 45,281 probes across the 12 hybridizations.

Initial normalization
We base-2 Log-transformed the expression signals to stabilize the variance, capturing only those probes with a "detectable" signal (P < 0.01) in at least one of the 12 hybridizations. This produced a "cleaned" dataset of 18,129 Illumina probes.

Chip normalization
The "cleaned" dataset was subject to a within chip normalization in which each signal was normalized by subtracting the within-chip mean and dividing by the within-chip standard deviation. The chips were therefore standardized to have a mean (μ) of 0 and a standard deviation (σ) of 1, enabling data from different chips to be combined without the risk of being influenced by differences in mean or SD. This allows subsequent linear functions to be more reasonably derived, in our case computation of differential expression. The resulting normalized dataset was subjected to hierarchical cluster analyses using the PermutMatrix software [34].