The mitochondrial proteome is estimated to contain ∼1100 proteins, the vast majority of which are nuclear-encoded, with only 13 proteins encoded by the mitochondrial genome. The import of these nuclear-encoded proteins into mitochondria was widely believed to be unidirectional, but recent discoveries have revealed that many these ‘mitochondrial’ proteins are exported, and have extra-mitochondrial activities divergent from their mitochondrial function. Surprisingly, three of the exported proteins discovered thus far are mitochondrially encoded and have significantly different extra-mitochondrial roles than those performed within the mitochondrion. In this review, we will detail the wide variety of proteins once thought to only reside within mitochondria, but now known to ‘emigrate’ from mitochondria in order to attain ‘dual citizenship’, present both within mitochondria and elsewhere.
Historically, mitochondria have been perceived almost exclusively as the cell's energy generator, principally responsible for cellular ATP (adenosine triphosphate) production, as well as an evolutionary curiosity due to their endosymbiotic nature and α-proteobacterial ancestry. However over the past two decades, it has become apparent that mitochondria are far more dynamic than previously thought and have critical roles to play in apoptosis , Fe–S cluster generation , and the production of reactive oxygen species . Interestingly, it appears that there is substantial cross-talk between the organelle and the outside cytosolic world [4,5], and a growing number of mitochondrial proteins, both mitochondrially and nuclear-encoded, have been identified at extra-mitochondrial sites, the roles for only some of which are known (Figure 1 and Table 1). Many recent studies have sought to catalogue the human mitochondrial proteome, notably the databases MitoCarta 2.0  and MitoCarta+ , which list 1158 and 1166 candidate proteins respectively. Another subcellular mapping study of the human proteome identified 1074 mitochondrial proteins, of which 475 displayed dual subcellular localisation, observed both in mitochondria and at least one other non-mitochondrial site .
Location, function, and MDV presence of dual localised proteins.
Protein translocation in mitochondria has long been believed to be solely one-way — inward. The mechanism governing mitochondrial protein import has been substantially characterised, with eight translocases known to recognise and import more than 1100 proteins into the mitochondria . However, recently an intracellular transport pathway from mitochondria to other organelles has been discovered — the first such description of export machinery for mitochondrial proteins . These mitochondrially derived vesicles (MDVs) are selective for their cargo and have been shown to transport specific proteins to either peroxisomes  or lysosomes . Interestingly, a vesicular export system has already been characterised in the mitochondrion's extended family: bacteria have a vesicular secretion pathway critical for a wide variety of functions . Another mitochondrial protein export route via the outer mitochondrial membrane import channel Tom40 has recently been described . Thus, with the recent discovery of these mechanisms, as well as the ever increasing list of proteins shown to be exported from mitochondria to discrete extra-mitochondrial locations, it seemed an appropriate time to review our current understanding with regard to these dual localised proteins. Note that the proteins we describe are all listed in MitoCarta+ and comprise a specific subset of dual residency mitochondrial proteins. The cohorts we describe are either mitochondrially encoded or are nuclear-encoded proteins that have known extra-mitochondrial functions, but require processing in the mitochondria to attain their active state. Thereby, all 10 proteins we describe have been resident at one time in the mitochondria, but are subsequently transported to extra-mitochondrial locations.
Pyruvate dehydrogenase complex
The pyruvate dehydrogenase complex (PDC) converts pyruvate to acetyl CoA, the critical metabolic hub linking the citric acid cycle and oxidative phosphorylation, as well as a crucial component of both lipid and carbohydrate metabolism  and histone acetylation . Until recently, the PDC was believed to be imported into, and subsequently remain in, the mitochondrial matrix, a location consistent with its central role in ATP generation. However, it has recently been revealed that mitochondrial PDC is translocated out of the mitochondria to the nucleus and has many functions including producing acetyl CoA in situ for histone acetylation  and as a co-activator of gene expression . Recently, the actions of both nuclear and mitochondrial PDC have been implicated in prostate cancer tumorigenesis . The mature form of PDC is generated in the mitochondria post cleavage of the N-terminal mitochondrial-targeting sequence (MTS) of the constituent PDC subunits. This process is solely undertaken in the mitochondria, and functional PDC can only be formed once the MTSs have been removed from the subunits. Only the mature forms of the PDC subunits E1 and E2 were detected in the nucleus, consistent with an initial mitochondrial processing stage and subsequent nuclear translocation [17,18]. Additionally, treatment with the ribosomal translation inhibitor cycloheximide did not reduce the quantity of nuclear PDC detected, indicating that de novo protein translation is not involved .
The PDC is a very large complex ( ∼9 MDa), and it appears that in order to be exported from mitochondria it may first be fragmented into its constituent subunits. Both the E2 and E3 bp subunits, but not E1, have been detected in MDVs post addition of the respiratory chain inhibitor Antimycin A [12,21]. It should be noted that oxidative stress was the trigger for the export of these PDC components, whose ultimate fate was degradation in the lysosome and not transportation to the nucleus , but it is the first demonstration of a pathway by which these polypeptides can exit the mitochondria. Additionally, it has been demonstrated that the intact and functional PDC is observed at the outer mitochondrial membrane (OMM) , but whether it is assembled or translocated there from the matrix remains an open question. Further analysis on the feasibility of the MDV mechanism for PDC export specifically, as well as potential alternative export routes, has been reviewed previously [23,24].
Mortalin (also known as mthsp70, PBP74, and Grp75) is a ubiquitously expressed protein, with multiple subcellular localisations and functions. The variety of names afforded mortalin reflects its multiplicity of functions, and it has been implicated in control of cell proliferation , de novo protein folding , and intracellular trafficking . It is predominantly localised to the mitochondrial matrix , where it is an essential component of the mitochondrial protein import pathway . Newly synthesised mortalin has an N-terminal presequence that is cleaved off once the protein is transported into the mitochondria. Hep1 has been shown to be exclusively localised to the mitochondrial matrix [30,31] and is required for subsequent folding of mortalin [31,32]. This suggests that mortalin can only be exported from the mitochondrial matrix once its correct confirmation has been attained. Consistent with its multiple functions, mortalin has been observed at several extra-mitochondrial sites, including the plasma membrane [33,34], centrosome , and endoplasmic reticulum . Many studies have illustrated the role of mortalin in carcinogenesis , as well as its involvement in neurodegeneration and Parkinson's disease . Recently, it has been reported as a potential biomarker for Parkinson's disease  as well as pancreatic , colorectal , and breast cancer .
Known classically as a molecular chaperone, TRAP-1 (TNF receptor-associated protein 1, also known as HSP75), a member of the HSP90 family, is involved in protein folding in the mitochondria , but more recently its anti-apoptotic role in protecting the cell from oxidative stress has been determined [44–46]. Specifically, TRAP-1 has been reported to indirectly inhibit reactive oxygen species-dependent opening of the mitochondrial permeability transition pore , a critical player in mitochondrially directed apoptosis . A recent study outlines the potential role of TRAP-1 in Parkinson's disease and describes how TRAP-1 interacts with HtrA2, another Parkinsonism-associated mitochondrial protein . Additionally, it has been identified as a substrate for PINK1 (PTEN-induced putative kinase 1), a kinase complicit in the progression of Parkinson's disease . TRAP-1 has also been associated with various functional disorders  and appears to play a role in oncogenesis, although it has been suggested as both tumorigenic [52,53] and tumour-suppressive .
TRAP-1 has an N-terminal MTS and is localised to the mitochondrial matrix , but has also been detected at discrete extra-mitochondrial locations, including pancreatic zymogen granules, insulin secretory granules, the cell surface of blood vessel endothelial cells, and the nuclei of both pancreatic and heart cells . Recently, it has been demonstrated that TRAP-1 appears to be exported from the mitochondria of both MN9D dopaminergic neuronal cells and primary cultures of dopaminergic neurons upon exposure to 6-hydroxydopamine (6-OHDA), an apoptotic initiator . Another heat shock protein previously shown to be exported from mitochondria, mortalin, was also detected in the cytosol after 6-OHDA treatment, but many other mitochondrial proteins were not. Interestingly, neither staurosporine nor etoposide, two other cell death initiators, led to a similar increase in these cytosolic heat shock protein levels, suggesting that this export mechanism is specific to 6-OHDA, rather than as merely a by-product of mitochondrially directed apoptosis. Amoroso et al. have also recently shown a novel extra-mitochondrial location and function for TRAP-1 — it appears to interact with TBP7 in the endoplasmic reticulum in order to facilitate cellular ubiquitination and protein quality control [58,59].
Cytochrome c is a critical component of the respiratory chain, facilitating the transport of electrons from complex III to complex IV at the inner mitochondrial membrane [60,61]. However, it has also been implicated in the intrinsic apoptotic pathway [1,62,63], and its release from mitochondria is a prerequisite to activation of this caspase-9–caspase-3-mediated apoptotic cascade [64,65]. Additionally, this cytochrome c/caspase-3 pathway has recently been identified as critical for synaptic long-term depression and AMPA receptor internalisation in hippocampal neurons . The route by which cytochrome c exits the mitochondria is still a matter of some debate, with the permeability transition pore, osmotic rupture of the OMM, or Bak/Bax OMM pore formation all suggested as possible export routes [60,63]. Although cytochrome c is predominantly localised to mitochondria, significant quantities have been reported at secretory granules in both the pancreas and anterior pituitary under normal physiological conditions . Apocytochrome c does not have an MTS, but is believed to be transported into the mitochondrial intermembrane space (IMS) via the TOM (tranlocases of the outer mitochondrial membrane) complex . The subsequent addition of a haem moiety to apocytochrome c in the IMS results in the formation of functional cytochrome c , but also prevents its release back into the cytosol by direct translocation .
Another mitochondrial protein, Diablo (also known as Smac), has a pro-apoptotic role in the caspase-3 pathway, binding to IAPs (inhibitor of apoptotic proteins) in order to facilitate caspase activity [71,72], and is detected in the cytosol [72,73]. The cytosolically produced Diablo precursor has an N-terminal MTS, and it is rapidly translocated into mitochondria post translation (or co-translationally)  where it is processed by the inner membrane peptidase (IMP) complex, located on the inner mitochondrial membrane . Mitochondrial import, and subsequent processing of the Diablo precursor, is a prerequisite for facilitation of apoptosis, as only mature Diablo, but not the precursor, demonstrates pro-apoptotic capabilities . Thus, as is the case for cytochrome c, Diablo, while produced in the cytosol, must first be imported, processed, and subsequently exported from the mitochondria in order to carry out its apoptotic function.
The mitochondrial serine protease HtrA2 (also known as Omi) is another IAP-binding protein (IBP) that has demonstrated some pro-apoptotic effects when released into the cytosol [76,77]. In a similar fashion to Diablo, it has an N-terminal MTS and adopts its mature form only upon processing within the mitochondrion, exposing its IBP site . It is localised to the mitochondrial intermembrane space and is required for mitochondrial homeostasis . Many other important mitochondrial functions for HtrA2 have recently been discovered, including critical roles in OPA1 proteolysis and mitochondrial fission/fusion  as well as Parkinson's disease pathogenesis via interaction with TRAP-1 and PINK1 [49,81]. The mature HtrA2, once prompted by various apoptotic stimuli, is released from the mitochondria and can interact with IAPs, allowing the caspase-dependent apoptotic cascade [77,78]. Again, only the mature but not the apo-form of HtrA2 can act as an IBP and elicit a pro-apoptotic cascade , suggesting that the mitochondrial import, processing, and subsequent export of the protease is critical for this interaction. There is some debate on the criticality of this process for caspase-dependent apoptosis however .
Mitochondrial aspartate aminotransferase
There are two isoforms of aspartate aminotransferase, cytosolic (cAspAT also known as Got1) and mitochondrial (mAspAT also known as Got2), both of which catalyse the interconversion of oxaloacetate to aspartate. mAspAT is nuclear encoded, but is translocated to the mitochondria [82,83], where it transfers amino groups between glutamate and aspartate at the inner mitochondrial membrane [82,84]. However, mAspAT appears to have many additional functions and aliases. It is also known as FABPPM (fatty acid-binding protein), the first reported long-chain fatty acid transporter [85–87], and has recently been identified as a kynurenine transaminase (also known as KAT IV) [88,89]. In the case of the former, localisation at the plasma membrane is fundamental to facilitate fatty acid uptake by the cell, a function inconsistent with a solely mitochondrial distribution. Accordingly, mAspAT has been detected at the plasma membrane of multiple cell types, including cardiomyocytes, myocytes, lymphocytes, hepatocytes, jejunal enterocytes, adipocytes, and vascular endothelial cells [56,90,91]. mAspAT has an N-terminal MTS, which directs it to the mitochondria, and only behaves as a functional FABP once this MTS has been cleaved , a process that the enzyme only undergoes post entry into the mitochondria [87,92,93]. This suggests that mitochondrial import, processing, and subsequent export are critical for mAspAT/FABPPM functionality.
NADH dehydrogenase subunit 2
NADH dehydrogenase subunit 2 (ND2) is a mitochondrially encoded subunit of the inner mitochondrial membrane enzyme NADH:ubiquinone oxidoreductase (complex 1) . Complex 1 plays a critical role in cellular energy production and is composed of ∼44 subunits [95,96], seven of which are mitochondrially encoded . ND2 alone does not exhibit enzymatic activity but has recently been implicated in quinone binding . Over the past few years, an extra-mitochondrial role for ND2 has been identified as a critical component of the N-methyl-d-aspartate receptor (NMDAR) complex, anchoring the protein tyrosine kinase Src to NMDARs at post synaptic densities in the hippocampus . Src is a key regulator of NMDAR activity and thereby synaptic transmission [100,101]. Inhibition of the Src–ND2 interaction has also lead to suppression of Src-mediated NMDAR activity  and thereby prevents both neuropathic and inflammatory pain behaviours without affecting basal synaptic transmission . Recently, ND2 has been shown to bind directly to the GluN1 subunit of the NMDAR complex via transmembrane helices of the two proteins, and uncoupling ND2–Src from NMDARs suppresses the Src-mediated up-regulation of NMDAR currents in cultured neurons . This is the first example of a mitochondrially encoded protein having a key modulatory role to play in a non-mitochondrial process outside of this organelle.
Cytochrome c oxidase I and II
Cytochrome c oxidase (complex IV) is the terminal enzyme of the mitochondrial electron transport chain and is located on the inner mitochondrial membrane. It consists of 13 subunits, three of which, cytochrome oxidase I, II, and III (MT-CO1, 2, and 3), form the functional core and are mitochondrially encoded [104,105]. Two of the subunits, MT-CO1 and MT-CO2, are sufficient for efficient respiration-dependent proton translocation, and the other 11 subunits are believed to have principally regulatory functions [106–108]. MT-CO1 and MT-CO2 have been detected at both rat pituitary growth hormone granules and pancreatic zymogen granules . Their function at these sites has not been elucidated, but interestingly, cytochrome c, the substrate for cytochrome c oxidase, has also been detected at these same locations . MT-CO1 containing MDVs have recently been observed  and can be generated in a human osteosarcoma U2OS cell line upon Antimycin A treatment . These findings illustrate that the mitochondrially encoded protein MT-CO1 has a mechanism by which to exit mitochondria, as well as many potential destinations once it is exported. One possible destination may be the lysosome, the ultimate destination of the PDC components transported by MDVs as described earlier.
In addition to its presence at granule cells in both the anterior pituitary and pancreas in rats, MT-CO2 has been observed in other species at many extra-mitochondrial sites. In bivalves, it has been observed in both the inner and outer mitochondrial membrane in sperm, as well as the cytoplasm and the nucleoplasm, and is believed to play a role in gamete maturation and fertilisation . Subsequently, MT-CO2 was observed at both the plasma membrane and cytoplasm of eggs from the same species . This observation is consistent with the unique method of mitochondrial DNA inheritance described in bivalves, known as doubly uniparental inheritance (DUI), where both the male and female mitochondria are known to be involved in inheritance . Similarly, this suggested extra-mitochondrial role for MT-CO2 may be unique to DUI in bivalves.
Mitochondrially derived peptides
Mitochondria-to-nucleus retrograde signalling has recently been recognised as a critical factor in regulating neuronal function,  stress signalling, and the maintenance of mitochondrial protein homeostasis. One of the principal signalling mechanisms is via mitochondrially derived peptides (MDPs) . These peptides are derived from two main sources — short open reading frames encoded in mtDNA or by proteolysis of mitochondrial matrix proteins . The latter have been reported as crucial for the maintenance of mitochondrial proteostasis, serving as retrograde signalling molecules upon translocation from the mitochondrial matrix to the nucleus  as well as functioning as antigens at the cell surface . These small peptides, 6–20 amino acids in length, can be readily released from mitochondria via ABC (ATP-binding cassette) family transporters [114,116]. This export mechanism has been shown to facilitate the mitochondrial unfolded protein response pathway , an important factor in both longevity and mitochondrial proteostasis .
Two MDPs encoded within the mitochondrial genome have been reported to date — Humanin [119,120] and MOTS-c (mitochondrial open reading frame of the 12S rRNA-c) . Humanin has been shown to have neuroprotective effects as well as acting as a mitokine , and MOTS-c mediates metabolic homeostasis and insulin sensitivity upon export from the mitochondria . However, caution should be exercised in misattribution of these peptides as solely encoded and produced in the mitochondrion. Although MOTS-c is mitochondrially encoded, it is unlikely that this is also its site of translation, as the mitochondrial codon usage specifications would lead to the peptide being prematurely truncated . Therefore, the mRNA transcript must instead be exported to the cytosol prior to translation. Additionally, there is debate over whether humanin is exclusively produced in the mitochondria and subsequently exported. Humanin is also encoded for by nuclear-encoded DNA cytosolic translation machinery , and this product is biologically active , considerably weakening its claim as a solely mitochondrial derived peptide. To date, there have been no MDPs shown conclusively to be encoded, produced, and exported from the mitochondria.
Our understanding of how the mitochondrion interacts with the cellular environment, particularly with regard to its roles in biosynthesis and signalling, is constantly evolving . Mitochondria are critical for the generation of Fe–S clusters , reactive oxygen species , steroid hormones , and of course, ATP . Thus, it appears that we are only beginning to discover the contribution of mitochondria with regard to the synthesis and export of a diverse set of critical physiological and pathophysiological effectors, a role far beyond their classical textbook description as a mere ‘cellular powerhouse’. To date, studies aimed at identifying the mechanisms by which proteins are exported from the mitochondria have, for the most part, identified this export as part of the organelle's quality control system. However, given that this is an emerging field, there remains much yet to be discovered, and it has been speculated that the vesicular export pathway specifically may have a role beyond the regulation of misfolded proteins .
The mitochondrial import, processing, and export pathway for cytosolically expressed proteins can be viewed as a method of carefully controlling what proteins are rapidly available to the cell at a given time point — control of which is crucial when dealing with pro-apoptotic proteins such as cytochrome c or Diablo. This allows a readily available pool of correctly folded proteins to be sequestered and subsequently released from the mitochondria at short notice. Alternatively, this mitochondrial detour hypothesis could be interpreted as needlessly over-complicated, and many alternative routes by which dual localisation of mitochondrial proteins may occur have been suggested [130,131]. For example, incomplete translocation into the mitochondria (as is the case for fumarase and mitochondrial acitonase ), or the existence of as yet undetected cytoplasmically expressed homologues of mitochondrial processing proteins such as Hep1 or the IMP complex, could provide alternative mechanisms to explain dual localisation of these mitochondrial proteins. Indeed, many proteins previously suggested to require mitochondrial entry such as P32  or Hsp60 [134,135] have since been shown to have alternative methods for getting to their cytosolic destinations, without the need for translocation to the mitochondria [136,137]. Other methods for dual subcellular distribution include the presence of an ambiguous signal sequence or of two targeting signals on one polypeptide . Another dual localised protein, CLK-1, known to have both mitochondrial and extra-mitochondrial roles retains its MTS within the nucleus, suggesting that mitochondrial entry (which would lead to proteolytic cleavage of the MTS) is not required prior to nuclear entry . Other proteins such as protein ligase III, RNaseH1, and FEN1 are known to generate different MTS lacking isoforms that facilitate mitochondrial entry by alternative translation initiation [139–141]. Therefore, in our assessment of this relatively novel concept of mitochondrial protein export, it is important that we consider alternative explanations as to how dual localisation may occur.
However, it is critical to note that many proteins detailed here are mitochondrially encoded, and as such confound some of the potential rebuttals that could apply to their nuclear expressed compatriots. Additionally, all of the nuclear expressed proteins described in this review require processing within the mitochondrion to attain their active state, in addition to having a known extra-mitochondrial function. Thus, both cohorts must be mitochondrially resident for a period before subsequent export to discrete extra-mitochondrial sites. While protein export from mitochondria is not the only explanation for dual localisation, it is clear that proteins both can and do exit the mitochondria. This is particularly evident given the recent advances in our understanding of the MDV and protein retro-translocation export pathways, coupled with the growing cohort of both nuclearly and mitochondrially encoded proteins at extra-mitochondrial locations (Table 1).
Note that one must be prudent when attributing localisation of a particular protein to a specific subcellular location. Owing to the technical difficulties associated with obtaining ‘pure’ subcellular fractions coupled with the potential for detection of these minor contaminants using mass spectrometry, there is potential for the misattribution of proteins resident at specific subcellular sites. However, the extra-mitochondrial roles described for each of the nuclear-encoded proteins included in this review suggest that their detection at these sites is consistent with biological activity at these sites rather than a technical overdetection issue, although this possibility cannot be excluded.
Mapping the subcellular proteome is an ongoing and dynamic endeavour — three different recent studies estimate the number of mitochondrial proteins to be 1158 , 1166 , and 1074 . MitoCarta 1.0 estimated the number of proteins dual localised to both the mitochondrion and at least one other subcellular site to be 85 , but more recently this number has been revised upwards to 495 . Indeed, even this latter extensive study likely underestimates the total number of proteins present in the mitochondrion, as 8 of the 13 mitochondrially encoded proteins were not detected. Therefore, as there is not currently a single definitive database for the mitochondrial proteome, neither can there be a definitive list of all dual localised mitochondrial proteins. As far as we can ascertain, the 10 mitochondrial proteins described in this review are the only proteins for which there is evidence of mitochondrial export, but for the reasons described above they are unlikely to comprise a comprehensive list of all mitochondrially exported proteins.
It is intriguing to speculate how export mechanisms such as the MDV  and Tom40  routes may have evolved. The endosymbiont theory of mitochondrial origin — incorporation of an alphaproteobacterial ancestor into a proto-eukaryotic cell — is now almost universally acknowledged. Since this initial incorporation event, evolution of mitochondria within different species has been highly divergent, with modern mitochondrial genomes ranging in size from 6 to 11 000 kb . In animals, the mitochondrial genome has substantially reduced in size compared with that of its bacterial ancestor. Endosymbiotic gene transfer has resulted in transfer of the vast majority of these ancestral alphaproteobacterial genes to the mammalian cell nucleus, with the human genome estimated to contain over 800 genes that display alphaproteobacterial ancestry. However, only 200 of these genes code for proteins that have designated roles within the mitochondrion . Thus, the eukaryotic cell appears to have hijacked and repurposed these genes [termed NumtS (nuclear mitochondrial sequences) [144,145]] for use both within and without the mitochondria. This relationship between proto-mitochondria and host cell has evolved beyond mere symbiosis, and in the modern context is more akin to biological socialism, where ownership and utility of the genes and proteins in question are shared between the cell and the organelle. We view the export of mitochondrial proteins as a possible extension of this grand evolutionary rearrangement, with the retargeting of these ‘mitochondrial’ proteins an alternative to the gene acquisition approach. How and why each of these proteins exit their mitochondrial residence are intriguing questions yet to be answered in this emerging field.
fatty acid-binding protein
inhibitor of apoptotic proteins
inner membrane peptidase
mitochondrial aspartate aminotransferase
mitochondrially derived peptide
mitochondrially derived vesicle
mitochondrial open reading frame of the 12S rRNA-c
- MT-CO1, 2, and 3
mitochondrially encoded cytochrome oxidase I, II, and III
NADH dehydrogenase subunit 2
outer mitochondrial membrane
pyruvate dehydrogenase complex
PTEN-induced putative kinase 1
TNF receptor-associated protein 1
This work was supported by grants from the Canadian Institutes of Health Research to M.W.S. [MT-12682 and FDN-154336] and from the Ontario Research Fund research excellence programme to M.W.S.
We thank Helen McNeill and Ameet Sengar for their useful comments and suggestions.
DPS has no competing interests associated with the manuscript. MWS is an inventor on patents for inhibitors of the Src-ND2 interaction which are licensed into NoNO Inc., a privately held corporation of which he is a founder, officer and board member.