The central dogma states that DNA is transcribed to generate RNA and that the mRNA components are then translated to generate proteins; a simple statement that completely belies the complexities of gene expression. Post-transcriptional regulation alone has many points of control, including changes in the stability, translatability or susceptibility to degradation of RNA species, where both cis- and trans-acting elements will play a role in the outcome. The present review concentrates on just one aspect of this complicated process, which ultimately regulates the protein production in cells, or more specifically what governs RNA catabolism in a particular subcompartment of human cells: the mitochondrion.
Mitochondria are central to many different cellular processes. They are present in all nucleated cells of the human body and contain the only source of extranuclear DNA in animals. This mt (mitochondrial) DNA encodes genes that are present and expressed only in the organelle. The mtDNA is maintained in the innermost compartment of the organelle, i.e. the matrix, where replication, transcription and subsequently translation all take place. For this reason, mitochondria require their own machinery to facilitate each of these processes, which, in most cases, is quite different from the machinery in the nucleus or cytosol. The 16.569 kb mitochondrial genome encodes two mt-rRNAs, 22 mt-tRNAs and 13 ORFs (open reading frames), all of which encode polypeptides that are essential components of the multisubunit complexes that drive oxidative phosphorylation. The mtDNA also contains a short non-coding region that contains control elements, including an origin of replication and the transcriptional promoters  (detailed in Figure 1). A number of methods of DNA replication have been proposed, and it is likely that the physiological conditions will dictate which mechanism predominates at any one time. All, however, are dependent on RNA primers.
Transcription proceeds from three promoters, each of which generates a specific polycistronic RNA unit  (Figure 1). The HSP (heavy strand promoter) 2 and the LSP (light strand promoter) both generate products that essentially span the entire genome, except for the major non-coding region, whereas HSP1 transcribes at a 15–60-fold higher rate to produce abundant copies of the mt-rRNA species , although the mechanism by which this occurs is still debated . Since human mtDNA is such a compact and conservative molecule with no intronic sequences, the only processing events necessary are those that will cleave the polycistronic RNAs into their individual transcription units. This is followed by maturation of the mt-mRNAs by polyadenylation or the CCA addition to the mt-tRNAs (further details in ).
Why do we need RNases and what are their substrates?
There are a number of substrates, including the original polycistronic RNAs, that require processing as mentioned above, but a major substrate will be the mt-mRNAs. These species will undergo turnover, affecting the steady-state levels of RNA, which is maintained by a balance between transcription and degradation . Since human mitochondrial transcription generates full-length polycistronic mRNAs, an equal number of copies of all the immature mt-ORFs and mt-tRNAs are generated by default, so any variation in steady-state level that we know to occur  must be a consequence of processing, maturation or degradation, where the greatest influence is likely to be the rate of degradation. As the variation in steady-state levels of individual mRNAs containing a particular ORF cannot be specifically controlled at the level of transcription, it is strongly suggestive of a controlled decay mechanism that ensures appropriate levels of each transcript, but as yet this pathway has not been fully elucidated.
In humans, the mt-mRNAs lack any 5′-modification and bear simple phosphate groups at the 5′-terminus. The 3′-ends, in contrast, are constitutively polyadenylated  with the exception of MTND6 (mitochondrially encoded NADH dehydrogenase 6) . If degradation happens in an analogous fashion to other systems, one would expect a deadenylase to be present, in addition to 5′- and 3′-exonucleases and potentially an endonuclease as well, that would reduce the transcripts to mononucleotides ready for reassembly into new RNA species .
In addition to the turnover of RNAs mentioned above, there are other RNA species that need to be recycled. These include the primers that are required to initiate DNA synthesis in any current mode of replication [10–13]. As mentioned above, transcription generates primary RNA units from both strands covering virtually the entire genome. Since most of the coding sequence is present only on the L-strand RNA, there will be major fragments of H-strand RNA that are essentially antisense copies. These antisense copies, however, are rarely visible by standard Northern blot analysis of entire steady-state RNA. Similarly, although all H-strand RNA is derived from L-strand transcription, the only H-strand RNA that is visible is the ORF encoding NADH dehydrogenase 6 and not any of the transcripts corresponding to the antisense of any of the 12 L-strand ORFs. Although these species must be generated in amounts equal to those of the other species from the same strand, we know that they are rapidly degraded as they are rarely detectable by Northern blot analysis . The mechanism by which these fragments are recognized and so quickly and specifically destroyed remains unknown.
All of the RNA species described above would be the result of accurate transcription and processing. As with any biological system, there are errors and evidence of low-level misprocessing which generates RNAs that are aberrant at either the 5′- or 3′-termini, as documented previously . There are also likely to be low-level errors in mtDNA replication or transcription that will produce transcripts that contain ORF frameshifts or premature stop codons. There may be mRNAs that have stalled in the ribosome during elongation for other reasons, perhaps as a consequence of damage from free radicals, and all of these moieties will need to have some process to eliminate them. As yet no molecular mechanism for recognition or elimination of these unwanted transcripts has been characterized. Although there are a few candidates that may be responsible for the recognition, there are currently none that have been proposed for their elimination.
What are the identities of the generic mammalian mt-RNases?
Clearly mtRNAs are turned over, so the enzymes responsible must be present in the organelle, but it has been singularly challenging to identify and characterize the guilty parties. One reason may be that RNases are notoriously active, requiring very low levels to be present in the organelle. The outcome being that, in the many proteomic screens that have been performed, these molecules have been masked by more abundant proteins . Another possible confounding issue could be that these enzymes have a cytosolic or nuclear counterpart that shares regions of the ORF and proteomic data has not been able to distinguish the mitochondrial form (including UK114; Z.M.A. Chrzanowska-Lightowlers and R.N. Lightowlers, unpublished work). Further, RNases are often very small (less than 15 kDa) with basic isoelectric points, and so potentially will contain many arginine or lysine residues that can cause difficulties in identification by standard proteomics. The mitochondrion comprises ~1500 proteins, of which only 13 are encoded by mtDNA, requiring import of 99% of the components from the cytosol. There are sophisticated algorithms that have been generated in order to predict which proteins may be targeted to the mitochondrion, including MITOPRED, MitoProt, Target P and PSORT [17–20]. Mootha and colleagues devised Maestro, a ‘super’ algorithm by combining the predictive capacity of eight different parameters . By applying this to the known human genes and ranking their predicted products in terms of their likelihood of being mitochondrial, it is quite clear that in the top 1500 there are very few candidates for such RNases. So where are these proteins? Perhaps the targeting pre-sequences differ considerably from the current consensus and so have been pushed further down the listing [22,23].
So what is our current status of knowledge?
After a degree of controversy, it is now clear that the minimal components for the processing of the polycistronic transcripts into their components units are accounted for. These enzymes responsible are mt-RNase P, which recognizes the 5′-junction of the mt-tRNAs from the flanking sequence [24,25] and ELAC2, the human RNase Z-like enzyme that recognizes the 3′-junction .
An endonuclease, Endo G, has been characterized as primarily a DNase, but is potentially capable of cleaving RNA to generate the primers required to initiate DNA replication. This is an assumption made on sequence similarity to the ‘DNA/RNA non-specific endonuclease family’. A confounding feature here is that this protein is clearly localized to the IMS (intermembrane space), the compartment between the two membranes and not in the innermost matrix that contains the mtDNA and mtRNA species . This is also the major location for the human PNPase (polynucleotide phosphorylase) , which, like its bacterial orthologue, can act as a 3′→5′ phosphorylase. Interestingly, it also retains the ability to polymerize RNA in a template-independent manner generating heteropolymeric, but predominantly poly(A), extensions . This was difficult to reconcile with the IMS location and it was suggested that the change in mtRNA levels following its depletion was the consequence of some indirect effect . More recently, however, a function more compatible with its localization has been identified, indicating that PNPase acts as a ‘gatekeeper’ regulating import into mitochondria of RNA species, including the 5S rRNA and the RNA species that may be components of a mt-RNase P and RNase MRP (mtRNA processing) . The latter is responsible for cleavage of the RNA primer required for mtDNA replication , whereas the former cleaves 5′-junctions of the concatemeric mt-tRNAs as mentioned above. PNPase may also act on the putative miRNAs (microRNAs) that have been suggested to enter mitochondria, although these observations require further verification [31,32]. We too have identified an RNase that appears to be in the IMS: an as yet uncharacterized human mitochondrial protein encoded by the REXO2 gene. This, by similarity to the bacterial ORN (oligoribonuclease), acts on very short RNA and DNA oligomers of five nucleotides or fewer . This protein, like Endo G, appears to be present in more than one cellular compartment. Rexo2 appears to be located in both the IMS and the cytosol, and our characterization of this protein and its substrates are ongoing.
Other nucleases that are putatively mitochondrial which have not been fully characterized include EndoG-like (with ten splice variants, potentially in the inner mitochondrial membrane with a preference for single-stranded DNA and a possible role in apoptosis)  and HRSP12 (heat-responsive protein 12), also termed UK114. This protein is predicted by similarity to cleave single-stranded RNA and is consistent with results from our laboratory (A. Bobrowicz, F. Bruni, R.N. Lightowlers and Z.M.A. Chrzanowska-Lightowlers, unpublished work). The multiple variants have made it impossible to confirm exactly which form is present in human mitochondria, and neither has its precise substrate been confirmed.
This leaves as unknown the main enzymes responsible for the generic turnover of the mt-mRNAs (Figure 2). In recent years, a number of groups have tried to characterize this decay pathway for mt-mRNAs working from a starting assumption that an initial step is likely to follow the cytosolic paradigm and be degradation of the poly(A) tail [7,15,35]. This modification is known to stimulate translation in the cytosolic compartment and a similar, but not identical, function also appears to be true for human mitochondria [15,36]. PNPase is a 3′-phosphorylase and is a member of the bacterial degradosome. This complex also contains an endoribonuclease RNase E, a DEAD-box helicase RhlB and enolase also known for its role in glycolysis . It is this degradosome that is responsible for RNA turnover and, as a consequence, PNPase was for a long time considered to be the major candidate for initiating the degradation of mt-mRNAs . The observation that human PNPase is localized mainly to the IMS is clearly incompatible with a major role in mtRNA degradation in the mitochondrial matrix. The story is complicated, however, by the inability of any characterization to be able to fully negate the possibility that a low level of this enzyme may still be able to access the matrix. Therefore the potential for PNPase to be bifunctional with the majority of the protein locating to the IMS and promoting RNA import, with a subset being in the matrix and degrading mtRNA, cannot be precluded. A new kid on the block is PDE12 (phosphodiesterase 12) identified both by Poulsen et al.  and Rorbach et al. . Each laboratory identified the 2′-phosphodiesterase PDE12, which has a mitochondrial targeting sequence consistent with a matrix location, as having a mitochondrial localization using immunocytochemistry of tagged protein and by subcellular fractionation. PDE12 contains a catalytic nuclease domain that is found in the ‘endonuclease/exonuclease/phosphatase’ family (PFAM id: PF03372). It most closely aligns with the deadenylases found in this family . This enzyme only recognizes RNA and not DNA as a substrate, with significantly lower activity on homopolymeric-C or -G compared with poly(A) or poly(U). By in vitro testing of PDE12 with RNA corresponding to mitochondrial transcripts bearing or lacking poly(A) tails, it was possible to confirm high selectivity of poly(A) as the substrate . In contrast with the general pattern in the cytosol, loss of mt-mRNA polyadenylation does not have a single common effect on the transcript. Some mt-mRNAs are destabilized, others become more stable and there is no change in a third grouping, a curious observation, but one that has been observed by a number of independent research groups also examining the role of the poly(A) tail on mt-mRNA stability and expression [15,36,39,40]. Another consistent observation following loss or masking of mt-mRNA poly(A) tails is the reduction in translation [15,36,40].
The poly(A) tail is absent from mt-mRNAs in yeast. Degradation is triggered by recognition of a conserved dodecamer sequence in the 3′-UTR (untranslated region) and recruitment of Dss1 and Suv3 proteins, the former being a 3′→5′ exonuclease, whereas the latter is a helicase with ATPase activity . There is a human orthologue of Suv3p that encodes a helicase, hSuv3p , but it appears that it may partition between the nucleus and the mitochondrion [43,44]. The helicase activity requires a single-stranded 3′-terminus and, although it can unwind RNA, DNA or heteroduplexes, the preferred substrate appears to be double-stranded DNA. Loss of this protein through siRNA (small interfering RNA) depletion in HEK (human embryonic kidney)-293 cells resulted in a disruption of RNA degradation particularly for the ‘mirror’ antisense RNAs mentioned above . These are generated as a by-product of full-length transcription and processing, and are normally very rapidly degraded; however, they become stabilized following depletion of hSuv3p . Frustratingly, the RNase that presumably acts in concert with hSuv3p has not yet been identified and no orthologue of Dss1p exists, leaving another gaping hole in our understanding of human mtRNA degradation. There have been reports of RNase L degrading mtRNAs under stress conditions, but this does not appear to happen under normal circumstances .
There are clearly functions that are being carried out to degrade various normal and aberrant RNA molecules within the human mitochondrion. As yet, we have not identified all of the proteins responsible or the mechanisms by which they act. Since mitochondria from different organisms have evolved a diverse range of control points in gene expression, including the changes in codon usage  and the presence/absence or function of the poly(A) tail (reviewed in ), we suspect that elucidating the RNA-decay pathways may throw up yet more surprises.
This work was supported by the Wellcome Trust [grant number 074454/Z/04/Z] and the Biotechnology and Biological Sciences Research Council [grant number BB/F011520/1].
We thank Mr W. Casey Wilson for his help with Figure 1.
RNA UK 2012: An Independent Meeting held at The Burnside Hotel, Bowness-on-Windermere, Cumbria, U.K., 20–22 January 2012. Organized and Edited by Raymond O'Keefe and Mark Ashe (Manchester, U.K.).
Abbreviations: HSP, heavy strand promoter; IMS, intermembrane space; LSP, light strand promoter; mt, mitochondrial; ORF, open reading frame; PDE12, phosphodiesterase 12; PNPase, polynucleotide phosphorylase
- © The Authors Journal compilation © 2012 Biochemical Society