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) . 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 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 . A very recent publication has documented Rnf14 as a positive regulator of canonical Wnt signalling in human cells , with canonical Wnt signalling previously reported to be a potent activator of mitochondrial biogenesis . 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 . 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 mitochondrially-encoded 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 . A number of these genes also imply apoptosis, a mitochondrial phenomenon . 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 , 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 . 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 .
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)  2) Module-to-Regulator analysis  and 3) Regulatory Impact Factors [12, 13]. While there is some overlap in the exact molecules present in the co-expression 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.