LON is the master protease that protects against protein aggregation in human mitochondria through direct degradation of misfolded proteins

Maintenance of mitochondrial protein homeostasis is critical for proper cellular function. Under normal conditions resident molecular chaperones and proteases maintain protein homeostasis within the organelle. Under conditions of stress however, misfolded proteins accumulate leading to the activation of the mitochondrial unfolded protein response (UPRmt). While molecular chaperone assisted refolding of proteins in mammalian mitochondria has been well documented, the contribution of AAA+ proteases to the maintenance of protein homeostasis in this organelle remains unclear. To address this gap in knowledge we examined the contribution of human mitochondrial matrix proteases, LONM and CLPXP, to the turnover of OTC-∆, a folding incompetent mutant of ornithine transcarbamylase, known to activate UPRmt. Contrary to a model whereby CLPXP is believed to degrade misfolded proteins, we found that LONM, and not CLPXP is responsible for the turnover of OTC-∆ in human mitochondria. To analyse the conformational state of proteins that are recognised by LONM, we examined the turnover of unfolded and aggregated forms of malate dehydrogenase (MDH) and OTC. This analysis revealed that LONM specifically recognises and degrades unfolded, but not aggregated proteins. Since LONM is not upregulated by UPRmt, this pathway may preferentially act to promote chaperone mediated refolding of proteins.

Scientific RepoRts | 5:17397 | DOI: 10.1038/srep17397 proteins, organisms have also evolved a number of integrated stress response pathways such as the heat shock response, envelope stress response and unfolded protein response in the ER 3,10-12 . These stress response pathways mediate the up-regulation of molecular chaperones and proteases and in some cases attenuate translation in an attempt to re-establish protein homeostasis. In many cases PQC components also play an integral role in the sensing and signalling of the stress 2,3,10-12 .
In mammalian mitochondria the PQC network contains key chaperones such as mtHSP70 (also known as HSPA9 and mortalin) and HSP60 (also known as HSPD1) and their cofactors and five different AAA+ proteases. Two of these proteases, LONM (also known as LONP1) and CLPXP, are located in the matrix while intermembrane space facing YME1L1 (also known as i-AAA protease) and two different matrix facing m-AAA protease complexes are anchored to the inner membrane 13,14 . Consistent with an important role in maintaining mitochondrial protein homeostasis, inherited mutations in PQC components are associated with several diseases [15][16][17][18][19][20] . Apart from naturally occurring mutations, a disruption to mitochondrial protein homeostasis is achieved experimentally by altering the stoichiometry of nuclear and mitochondrial encoded protein subunits of various respiratory chain complexes [21][22][23][24] or by overexpressing a folding-defective protein within the mitochondrion 25 . In higher eukaryotes, this disturbance to protein homeostasis within the mitochondrion is sensed, and a mitochondria-to-nucleus signalling pathway is activated. This pathway, known as the mitochondrial unfolded protein response (UPR mt ), was initially identified in a monkey cell line, by over-expression of a folding-incompetent form of the mitochondrial protein, ornithine transcarbamylase (OTC) termed OTC-Δ 25 . This pathway has subsequently been described in human and mouse cells lines as well as in Drosophila melanagaster and Caenorhabditis elegans 24,[26][27][28] . Although the UPR mt signalling pathway is only poorly understood in mammals, the pathway and many of its components have been defined in C. elegans [29][30][31][32] . Recent data suggests that the subcellular distribution of ATFS-1 (activating transcription factor associated with stress-1) is a key regulatory branch-point in the pathway 32 . Under non-stress conditions ATFS-1 is directed to the mitochondrial matrix by an N-terminal mitochondrial targeting signal, where it is rapidly degraded by the AAA+ protease, Lon. Under conditions of stress however, ATFS-1 import (into mitochondria) is inhibited and it is instead directed to the nucleus by a nuclear localisation signal, resulting in the up-regulation of nearly 400 genes 32 . Significantly, the redistribution of ATFS-1 to the nucleus is dependent on the inner membrane embedded ABC-family peptide transporter Haf-1 (Haf transporter 1). Furthermore, both ClpX and ClpP from C. elegans have been experimentally implicated as direct components of the UPR mt signal transduction pathway 30,31 . Interestingly, Haynes and colleagues proposed a model for C. elegans, whereby stress within the mitochondrial matrix, is signalled across the inner membrane (via Haf-1) by a factor that is generated by the ClpXP protease 31 . Hence a ClpXP-dependent degradation product may regulate ATFS-1 trafficking.
There is growing evidence that the mitochondrial unfolded protein response is associated with several pathophysiology conditions such as inflammatory disease, infection, cancer and neurodegeneration 33,34 . It is currently a focal point in mitochondrial theory of aging and has potential to be manipulated as a strategy to reduce age-dependent cellular decline. Despite this, and the detailed understanding of this signalling pathway in worms, our understanding of the pathway in mammals remains rudimentary. In mammals the pathway is triggered by overexpression of OTC-Δ , which engages with mitochondrial PQC components such as HSP60 25 . The transient expression of OTC-Δ in mammalian cells leads to the selective transcriptional up-regulation of nuclear genes encoding mitochondrial chaperones (HSP60, HSP10 and mtDnaJ) and proteases (CLPP, YME1L1 and MPPβ ) as well as the nuclear transcription factor C/EBP homology protein (CHOP/DDIT3) 25,35 , while the expression of other prominent PQC proteins such as mtHSP70 and LONM are not effected. The mechanisms of sensing the increased load of unfolded protein and transfer of the signal across the two membranes of the mitochondrion to the cytoplasm however is not yet understood. Nevertheless, given that human mitochondrial CLPXP is known to recognise and degrade the model unfolded protein, casein and mammalian CLPP is transcriptionally upregulated upon over-expression of OTC-Δ 25,36,37 , it is reasonable to suggest that human CLPXP may play an important role in the degradation of misfolded proteins and subsequent stress signalling as reported in C. elegans.
In addition to CLPXP, C. elegans and mammalian mitochondria both harbour soluble Lon family AAA+ proteases. Mammalian mitochondrial LONM degrades proteins that are mildly damaged by oxidation, unassembled native proteins, as well as natively folded proteins for regulatory purposes [38][39][40][41][42][43] . However, it is not known if LONM and CLPXP both act to clear unfolded or aggregated proteins from the mitochondrial matrix. Here we report on a systematic investigation of the contribution of LONM and CLPXP to the processing of UPR mt -inducing protein OTC-∆ and the nature of the conformational state of the proteins that are recognised. Our data suggest that LONM, and not CLPXP, is primarily responsible for degradation of misfolded protein in the mitochondria matrix. Thus UPR mt stress signalling by the direct and promiscuous degradation of unfolded proteins by CLPXP may not be a conserved mechanism.

Results
OTC-Δ is rapidly degraded in mitochondria in an ATP-dependent process. To study the stability of OTC-∆, relative to wild-type OTC, we initially employed an import-chase proteolysis assay 38,44 . For this purpose, radiolabeled preprotein (either OTC or OTC-Δ ) was incubated with mitochondria isolated from HeLa cells. As expected, OTC preprotein was processed into an intermediate (~38 kDa) and Scientific RepoRts | 5:17397 | DOI: 10.1038/srep17397 a mature (37 kDa) form of the protein 45 when incubated with energised mitochondria (Fig. 1A, lanes 2  and 3). Importantly, these data also demonstrated that radiolabeled OTC-Δ (synthesised in rabbit reticulocyte lysate) was not only competent for in vitro import into mitochondria, but also correctly processed into the intermediate and mature forms of the protein (Fig. 1A, lanes 5 and 6).
Next, to assess the stability of OTC-Δ relative to OTC, an import-chase proteolysis assay was performed as outlined (Fig. 1B, top left panel). This analysis revealed that OTC-∆ is highly unstable in isolated mitochondria (Fig. 1B, bottom left panel) with only ~10% of the protein remaining after 120 min (Fig. 1B, right panel, open circles) compared to ~80% remaining for OTC at the equivalent time (Fig. 1B, right panel, filled circles). This in vitro analysis, which permits an assessment of protein turnover independent of other cellular processes such as mitophagy, revealed that OTC-∆ was degraded, within mitochondria, with a half-life of ~25 min (compared to a half-life > 120 min, for wild type OTC, under the same conditions).
To determine if the turnover of OTC-Δ was mediated by a mitochondrial ATP-dependent protease, we performed an import-chase proteolysis assay following the manipulation of ATP levels within mitochondria (Fig. 1C, left panels) 38,46 . In the presence of "high" levels of ATP, OTC-Δ was rapidly degraded with less than 20% remaining after the 2 h chase (Fig. 1C, right panel, open circles). In comparison, depletion of ATP from mitochondria strongly inhibited the turnover of OTC-Δ , with ~70% of the protein remaining after the 2 h chase (Fig. 1C, right panel, filled circles). Collectively, these data demonstrated that OTC-Δ is rapidly degraded in mitochondria and that its turnover is mediated by an ATP-dependent protease.
Depletion of mitochondrial matrix AAA+ proteases reveals a lack of functional overlap in the clearance of OTC-Δ. Mammalian mitochondria contain the two soluble matrix AAA+ proteases; CLPXP and LONM. The CLPXP protease consists of the peptidase CLPP, which is composed of two heptameric rings stacked back-to-back, and a hexameric unfoldase -CLPX, which can associate at either or both ends of CLPP 36,47 . The central component (CLPP) is essential for hydrolysis of peptide bonds, while CLPX uses the energy stored in ATP to drive the unfolding and translocation of proteins into the proteolytic chamber of CLPP 48,49 . To experimentally determine the relative contribution of CLPXP and LONM to the turnover of OTC-Δ , each component was depleted from mitochondria using RNA interference (RNAi) and an import-chase proteolysis assay of OTC-Δ performed (Fig. 2). Initially, the stability of OTC-Δ was monitored in mitochondria depleted of CLPX. Following import of OTC-Δ into CLPX-depleted mitochondria (CLPX↓ ), an import-chase assay was performed. Despite the successful knock down of CLPX in mitochondria ( Fig. 2A, middle panel) the half-life of OTC-Δ didn't change ( Fig. 2A). Since CLPP, and not CLPX, is transcriptionally upregulated upon activation of UPR mt we examined the possibility that CLPP could act independently of CLPX in the degradation of OTC-∆. Despite the successful knockdown of CLPP by RNAi- (Fig. 2B, middle panel) the rate and extent of OTC-Δ degradation in CLPP-depleted (CLPP↓ ) mitochondria (Fig. 2B, filled circles) was unchanged compared to control mitochondria (Fig. 2B, open circles). Collectively these data indicate that OTC-Δ is not a substrate of the mammalian CLPXP protease.
Next, to determine if the matrix AAA+ protease LONM was responsible for the turnover of OTC-∆, it was depleted from HeLa mitochondria using RNAi. The steady state level of mitochondrial LONM was highly reduced in the siRNA treated cells ( Fig. 2C middle panel, compare lanes 6-10 with lanes 1-5). Importantly, the turnover of OTC-Δ was almost completely inhibited in LONM depleted-mitochondria ( Fig. 2C top panel, lanes 6-10). Indeed, quantification of four independent import-chase proteolysis experiments, demonstrated that after the 2 h chase ~85% of the imported protein remained in LONM-depleted mitochondria (Fig. 2C, filled circles), while only ~20% of the imported protein remained in control mitochondria (Fig. 2C, open circles). Consistently, the half-life of OTC-Δ increased dramatically, from ~25 min in control mitochondria to > 120 min in LONM depleted mitochondria. Collectively these data suggest that LONM, and not CLPXP, is responsible for the turnover of OTC-Δ in mammalian mitochondria.

The LONM protease is responsible for degradation of OTC-Δ in mitochondria.
To exclude possible off-pathway effects associated with using RNAi, we sought to validate the import-chase proteolysis data using an alternative approach. To do so, we used MG132, an inhibitor of the 26S proteasome that is known to enter mitochondria where it can inhibit LONM 41,50 . Currently, the mechanism of action of MG132 on LONM is unclear, as is its effect on human CLPXP. To ensure that MG132 was indeed a specific inhibitor of human mitochondrial LONM (and not CLPXP) we monitored the degradation of α -casein in vitro, using either purified recombinant human LONM (Fig. 3A) or human CLPXP (Fig. 3B). Consistent with the findings of Orly and colleagues 41 the LONM-mediated degradation of α -casein was inhibited by MG132 in a concentration dependent manner (Fig. 3A). Importantly, the CLPXP-mediated degradation of α -casein was not inhibited by MG132 even at concentrations (up to 400 μ M) well in excess of those used to inhibit LONM. Having established that MG132 specifically inhibited LONM and not CLPXP activity, we applied these conditions to the mitochondrial import-chase proteolysis assay to determine if the turnover of OTC-Δ was mediated by LONM. Consistent with the effect of LONM-depletion on OTC-Δ stability, MG132 inhibited the turnover of OTC-Δ in a dose-dependent manner (Fig. 3C). Indeed at the highest concentration of MG132 used (50 μ M), the turnover OTC-Δ was almost completely inhibited (Fig. 3C, filled squares) with ~80% of the protein remaining after a 2 h chase, in comparison to only ~20% remaining for the untreated mitochondria (Fig. 3C, open circles). Next, to confirm that LONM was directly responsible for the turnover of OTC-Δ we performed a series of immunoprecipitation (IP) experiments with LONM. Given that the interaction between a protease and its substrate is transient we performed the IP in the presence (and absence) of MG132 to inhibit LONM proteolytic activity and hence potentially trap OTC-Δ in the degradation chamber of the protease. Following import of OTC-∆ preprotein and incubation in the absence or presence of MG132, mitochondria were lysed. Endogenous LONM was then immunoprecipitated, using anti-LONM antiserum and compared to control IPs, using pre-immune serum. Importantly, OTC-∆ co-immunoprecipitated with LONM, but only in the presence of MG132 (Fig. 3D, lane 6). These data confirm that LONM interacts directly with OTC-∆ and is responsible for the degradation of this folding impaired protein in mitochondria.
Unfolded but not aggregated proteins are degraded by LONM. Next we asked the question, which form of OTC-Δ (i.e. unfolded or aggregated) is recognised and degraded by LONM? To address this question, in vitro degradation assays were performed using either purified recombinant OTC or OTC-∆ together with a well-characterised model substrate, mitochondrial MDH 15,51-53 . Initially, we wanted to determine if LONM was capable of recognising and degrading aggregated OTC (OTC agg ). To examine this, we first developed conditions to aggregate wild type OTC ( Supplementary Fig. S1). Heat treatment of OTC at 55 °C but not at 47 °C or 50 °C, resulted in a rapid increase in turbidity (absorbance at 320 nm), reaching a maximum level within ~20 min ( Supplementary Fig. S1A, filled circles). Following removal from the treatment, the aggregates were stable for > 2 h at RT, with more than 95% of the heat-treated OTC remaining in the pellet fraction after centrifugation ( Supplementary Fig. S1B, lane 6). Consistent with published data 52 , MDH was irreversibly aggregated following a 30 min incubation at 47 °C ( Supplementary Fig. S1A, open circles and Supplementary Fig. S1B, lane 3). As such, in this study the thermal aggregation of OTC was performed at 55 °C for 30 min. Next, the stability of OTC agg was monitored in the presence of either LONM (Fig. 4A, filled squares), E. coli ClpAP (Fig. 4A, open circles) or E. coli ClpAPS (Fig. 4A, filled circles). Neither ClpAPS nor LONM was able to degrade aggregated OTC. To validate these in vitro degradation assays, we used a well-established model aggregate, thermally aggregated MDH (MDH agg ) as a control 51,52,54 . Importantly, and consistent with published data 51 , MDH agg was efficiently degraded by E. coli ClpAP but only in the presence of ClpS (Fig. 4B, filled circles). However, degradation of MDH agg was not observed in the presence of LONM (Fig. 4B, filled squares). Collectively, the data suggest that LONM is unable to degrade aggregated mitochondrial matrix proteins. These data also indicate that both aggregated proteins exhibit different properties, as ClpAPS was only able to degrade MDH agg (Fig. 4B, filled circles) and not OTC agg (Fig. 4A, filled circles).
Next, we monitored the in vitro degradation of unfolded MDH (MDH U ) by LONM. In this case, MDH was chemically denatured (with guanidine HCl) using a well-established method 53,55,56 , then extensively diluted into a degradation buffer containing LONM. These data demonstrate that unfolded MDH (MDH U ) was rapidly and completely degraded by LONM in an ATP-dependent manner (Fig. 4C, top  panel), with a half-life of ~4 min, as determined from the quantification of four independent experiments (Fig. 4C, filled circles). As an additional control the stability of native MDH (MDH N ) was also monitored in the presence of LONM (Supplementary Fig. S2A). Importantly, MDH N was completely stable for the duration of the assay (180 min). Collectively, these data indicate that LONM specifically recognises and degrades MDH U but not MDH N or MDH agg . To determine if this behaviour was also true for folding impaired OTC-∆ responsible for the induction of UPR mt in mammals, the stability of both unfolded OTC-∆ (OTC-∆ U ) and unfolded OTC (OTC U ) was monitored in the presence of LONM (Fig. 4D). Purified recombinant mature OTC-∆ (isolated from inclusion bodies) and OTC (isolated from  soluble lysate) were chemically denatured (as for MDH) and subsequently diluted into degradation buffer containing LONM. Importantly, in contrast to native OTC, which was stable in the presence of LONM and ATP ( Supplementary Fig. S2B), both OTC U and OTC-Δ U were rapidly degraded by LONM in an ATP-dependent manner (Fig. 4D). Quantification of three independent degradation assays revealed that unfolded OTC-Δ (Fig. 4D, filled triangles) was degraded much more rapidly than unfolded OTC (Fig. 4D, filled circles) with a half-life of ~2 min compared to ~15 min for OTC U (Fig. 4D). It is currently unclear why OTC-∆ U is degraded more rapidly than OTC U , but it is likely related to the nature of the degron presented to LONM in the unfolded state of the two proteins. Specifically, OTC-∆ is distinct from OTC, as it lacks 84 contiguous residues. Collectively these data demonstrate that LONM is responsible for the efficient recognition and degradation of unfolded proteins, but not of aggregated proteins. Finally, as a control we monitored the turnover of unfolded OTC-∆ by purified recombinant human CLPXP (Supplementary Fig. S3). As expected and consistent with the import-chase proteolysis assays (Fig. 2) active CLPXP was unable to degrade OTC-∆ U (Supplementary Fig. S3A).

Discussion
Maintenance of mitochondrial protein homeostasis and bioenergetics is fundamental to health and normal aging of eukaryotic organisms 14 . It is achieved through multiple layers of interrelated processes, from chaperone-and protease-mediated PQC, to activation of the mitochondrial unfolded protein response and finally the selective destruction of mitochondria by mitophagy 14 . In mammalian cell lines, the overexpression of a folding-impaired mitochondrial protein OTC-∆ has been shown to induce both UPR mt and mitophagy in energised mitochondria 25,26 . Currently however, the trigger for activation of these processes in mammals is unknown. Nevertheless, several studies in C. elegans have revealed that the mitochondrial matrix protease ClpXP contributes to the UPR mt signalling pathway and it has been proposed that peptides (or cofactors), derived from CLPXP-mediated degradation of unfolded proteins act to transmit the stress signal 30,31 . Here we have tracked the fate of the UPR mt inducing protein OTC-∆ in human mitochondria to gain insight into the contribution that the matrix proteases (CLPXP and LONM) make to protein homeostasis in human mitochondria. The apparent turnover of mitochondrial proteins can be attributed to either the selective degradation of the protein by resident proteases or the selective degradation of mitochondria by the lysosome [57][58][59][60] . By examining the fate of OTC-∆ imported into isolated mitochondria (Fig. 1) rather than in cells we have established that this aggregation-prone protein is unstable and selectively cleared from within the organelle in an ATP-dependent manner. Given that the ClpXP protease was shown to play an important role in the UPR mt signalling pathway in C. elegans we speculated that human CLPXP would be responsible for the degradation of OTC-∆. Surprisingly, and contrary to the proposed model of UPR mt signal transduction in humans, our data demonstrated that the rate of OTC-Δ degradation in both CLPX-and CLPP-depleted mitochondria was unchanged relative to control mitochondria (Fig. 2), indicating that CLPXP does not contribute to the removal of OTC-Δ in human mitochondria. In contrast, the loss of LONM function in mitochondria, either through depletion of LONM by RNAi-mediated knock-down or inactivation of LONM by MG132, almost completely stabilised OTC-Δ with ~80-85% of the protein remaining after a 2 h incubation at 37 °C (Figs 2 and 3). Collectively these data indicate that the turnover of OTC-Δ in mitochondria is largely mediated by LONM. We cannot however exclude the possibility that the m-AAA proteases or other unknown ATP-dependent proteases play a minor role in the turnover of OTC-∆.
The apparent inability of CLPXP to degrade OTC-∆ suggests that human CLPXP may not degrade folding intermediates of mitochondrial matrix proteins. These data also suggest that, UPR mt signalling in human mitochondria may not be linked to the CLPXP-mediated degradation of unfolded proteins. It could however be argued that CLPXP is responsible for the turnover of a specific mitochondrial matrix protein (or proteins) that only becomes accesible when the PQC machinery is over-burdened. However, it was recently revealed that overexpression of OTC-∆ in a human cell line induces mitophagy in polarised mitochondria via a PINK1 (PTEN-induced putative kinase 1) and PARK2 (Parkin E3 ubiquitin ligase) dependent pathway 26 . Consistent with our identification of LONM as the major protease responsible for the turnover of OTC-∆, Youle and colleagues revealed that depletion of LONM (and not CLPP) enhanced PINK1 and PARK2 recruitment (and hence mitophagy) to mitochondria exposed to unfolded protein stress. Hence CLPP does not modulate mitophagy because it does not contribute to the clearance of misfolded proteins in mitochondria, while in contrast LONM, as the central PQC protease, is a key moderator of mitophagy when matrix protein homeostasis in compromised. Interestingly, mutations in CLPP are associated with Perrault syndrome 3 (MIM 614129) which is characterised by sensorineural hearing loss and premature ovarian failure 19 and CLPP knockout mice display phenotypes consistent with Perrault syndrome as well as other distinct effects such as elevated mtDNA levels and induction of selected matrix protein quality control components and inflammatory factors 61 . Hence we speculate that human CLPXP may contribute to the temporal degradation of specific proteins for the purpose of regulating mitochondrial homeostasis or cellular processes, rather than contributing to the direct and promiscous degradation of unfolded or misfolded proteins for general PQC. However, this model requires further examination through the direct determination and comparison of the human CLPXP substrate proteome under normal and protein folding stress conditions.
In contrast to CLPXP, several substrates of human LONM have been defined. Human LONM is responsible for the turnover of mildly oxidised proteins such as aconitase as well as unassembled proteins [38][39][40]62 . LONM is also responsible for the regulated turnover of natively folded mitochondrial proteins including mammalian steroidogenic acute regulatory protein (StAR), 5-aminolevulinic acid synthase 1 (ALAS-1) and mitochondrial transcription factor A (TFAM) [41][42][43]62 . To further dissect the role of AAA+ proteases in human mitochondrial homeostasis we examined the specificity of LONM using three mitochondrial matrix proteins MDH, OTC and OTC-Δ in different conformational states (native, unfolded and aggregated). Strikingly, when all three proteins were chemically unfolded and diluted into aqueous solution containing LONM, they were rapidly degraded in an ATP-dependent manner. This is consistent with E. coli Lon (ecLon), which recognises short hydrophobic polypeptide stretches on proteins that are usually buried in the folded protein and exposed on the surface only upon unfolding 7,63 . Interestingly, the rate of degradation of unfolded OTC-Δ (t ½ ~ 4 min) was much faster than that of unfolded OTC (t ½ ~ 15 min) (Fig. 4). We speculate that this is due to the exposure of different hydrophobic residues or patches in the unfolded state of these proteins, which may be recognised with different affinities. Alternatively, the faster rate of turnover (of OTC-Δ ) may result from reduced folding kinetics of the mutant protein facilitating increased partitioning of the substrate to the protease. Consistently, ecLon exhibited different degradation rates for truncated products of β -galactosidase 63,64 . Here we provide direct evidence that LONM has the capacity to bind and degrade unfolded or misfolded states of proteins, likely due to exposure of hydrophobic residues that are normally buried in the native protein.
Since the E. coli AAA+ protease ClpAPS is known to recognise and degrade aggregated proteins 51 we questioned whether a human AAA+ protease also exhibited such an activity. To examine this, we Scientific RepoRts | 5:17397 | DOI: 10.1038/srep17397 monitored the stability of MDH agg and OTC agg in vitro in the presence of LONM (Fig. 4). Over a 3-hour incubation, in the presence of ATP regenerating system, little to no degradation of either aggregated MDH or OTC was observed. This data suggests that aggregated mitochondrial matrix proteins are either degraded very slowly by LONM or are not substrates of this protease. This is consistent with a previous study showing aggregated aconitase (formed due to severe oxidative damage) is a poor substrate of LONM 39 . Similarly, LONM was unable to degrade insoluble yeast MPPα generated by either chemical or heat denaturation 62 . Thus, LONM appears to recognise and degrade the soluble unfolded or misfolded state of proteins before they aggregate. Consistently, Lon-deficient (Δ pim1) yeast cells show an accumulation of aggregated mitochondrial polypeptides following treatment with heat, hydrogen peroxide (H 2 O 2 ) or menadione 65 , while electron microscopy images of mitochondria isolated from either Δ pim1 yeast cells or LONM-depleted human lung fibroblasts (WI-38 VA-13) revealed mitochondria with electron-dense inclusions within the organelle, which likely represent protein aggregates 66,67 . Furthermore, cells derived from patients with CODAS syndrome, a multisystem developmental disorder caused by mutations in LONP1 (coding for LONM), contain electron dense inclusions within mitochondria 68 .
Despite the presence of a CLPXP protease in human mitochondria, our data suggests that LONM is the central PQC protease in the mammalian mitochondrial matrix. Interestingly, the unfolded states of the substrates examined in this study (MDH and OTC-∆) also interact with human HSP60 both in vitro and in organello 15,25 . Therefore unfolded proteins are likely to partition between LONM and molecular chaperones such as HSP60. Consequently, any factor that slows the kinetics of protein folding (e.g. genetically inherited mutation), is also likely to reduce its half-life due to increased degradation by LONM. Since HSP60, and not LONM, is up-regulated by UPR mt 35 it seems important for mitochondria (under condition of protein folding stress) to prevent aggregation and recover folded functional proteins rather than removing them by proteolysis. In the case that the folding and degradation capacity of the mitochondrial matrix is exceeded, aggregates will accumulate. In bacteria and eukaryotes such as yeast, mechanisms exist in the cell to recover natively folded functional proteins from misfolded amorphous aggregates. This process is mediated by the AAA+ protein ClpB (Hsp78 in yeast mitochondria) together with DnaK (Ssc1p in yeast) and cofactors 52,69,70 . To date a functionally equivalent machinery has not been described in human mitochondria. Thus, in the absence of a mechanism to clear aggregates from within mitochondria (by degradation or dissaggregation and refolding) it seems the selective clearance of compromised mitochondria by mitophagy may play a critical role in maintaining mitochondrial protein homeostasis.

Methods
In vitro transcription and translation. In vitro transcription of Rattus norvegicus OTC precursor was performed using pSP65/pre-OTC 71 . To generate a plasmid for in vitro transcription of OTC-Δ , pSP65/pre-OTC was digested with Bgl II 25 , releasing a 255 bp insert and the linear plasmid ligated to generate pSP65/pre-OTC-Δ . To generate radiolabelled pre-OTC and pre-OTC-∆, in vitro transcription and translation was performed. Briefly, pSP65/pre-OTC and pSP65/pre-OTC-Δ were digested with Sma I then used as template to synthesise mRNA using SP6 RNA polymerase (Promega). The 35 S-labelled precursor proteins were synthesised from pre-OTC and pre-OTC-∆ mRNA in rabbit reticulocyte lysate (Promega), supplemented with 11 μ Ci [ 35 S]Met/Cys EXPRE 35 S 35 S protein labelling mix (specific activity > 1000 Ci/mmol; Perkin Elmer).
Cell culturing, mitochondrial isolation and protein import. HeLa cells were grown at 37 °C in the presence of 5% (v/v) CO 2 in Dulbecco's Modified Eagle Medium (Life Technologies) supplemented with 10% (v/v) fetal calf serum (Life Technologies). Transfection of cells with 20 nM siRNA (Life Technologies) was performed using Lipofectamine 2000 (Life Technologies). Silencer or Silencer select siRNA directed against CLPX (s21289) LONP1 (s17903) and CLPP (105684) and the appropriate negative control siRNA, were used. Isolation of mitochondria and in vitro import of radiolabelled protein was performed essentially as described previously 74,75 . Preparation of unfolded proteins. For protease mediated degradation assays, soluble unfolded proteins were prepared using guanidine hydrochloride. Specifically, either native MDH (8. Protein degradation assays. Turnover of substrates by LONM protease was performed essentially as described previously 36 . In brief, native, unfolded or aggregated protein substrates (86-3000 nM) were incubated together with LONM (0.6-1.8 μ M, protomer) in proteolysis buffer (50 mM Tris-HCl pH 8.0, 100 mM KCl, 20 mM MgCl 2 , 0.02% (v/v) Triton X-100, 10% (v/v) glycerol, 1 mM DTT) at 30 °C for 5 min with degradation initiated by the addition of ATP (5 mM). The degradation of aggregated substrates by E. coli ClpA (1.2 μ M), ClpP (2.8 μ M) and ClpS (1.2 μ M) was performed in proteolysis buffer, essentially as described 51 . Turnover of α -casein (3 μ M) or unfolded OTC-Δ (0.086 μ M) by human CLPX (2.4 μ M) and human CLPP (5.6 μ M) was performed in the presence of 5 mM ATP but without ATP regeneration as described 37 . Where indicated, MG132 (5 μ M-400 μ M) in ethanol or ethanol alone (control), was added to the proteolysis mix. Unless otherwise stated degradation assays longer than 90 min, were also supplemented with an ATP-regeneration mix (4 μ g/ml pyruvate kinase, 4 mM phosphoenolpyruvate). To terminate reactions, samples were mixed with SDS-PAGE loading buffer followed by heat treatment at 95 °C. To monitor protein degradation within isolated mitochondria, import-chase proteolysis assays were performed 38,44 . Radiolabelled precursor protein (OTC or OTC-Δ ) were imported into energised isolated mitochondria for 7.5 min at 37 °C. Following dissipation of the membrane potential with valinomycin (2 μ M), non-imported precursor protein was removed by trypsin treatment (5 μ g/ml) followed by inhibition of trypsin with soybean trypsin inhibitor (1 mg/ml). Prior to the chase (incubation for 120 min), mitochondria were either left untreated or treated with ATP regeneration mix (5 mM ATP, 10 mM creatine phosphate, 100 μ g/ml creatine kinase), ATP depletion mix (40 U/ml apyrase, 20 μ M oligomycin), MG132 (10 or 50 μ M) in ethanol or ethanol alone (control). For the immunoprecipitation experiments radiolabelled OTC-Δ preprotein was freshly imported in isolated mitochondria, then following trypsin treatment supplemented with either MG132 (50 μ M) in ethanol or ethanol alone (control) and incubated at 37 °C for 5 min. Mitochondria were reisolated, washed, then resuspended in IP buffer (50 mM Tris-HCl pH 7.5, 100 mM KCl, 10 mM Mg-acetate, 5% (v/v) glycerol, 0.5% (v/v) Triton X-100) and incubated on ice for 30 min, with occasional mixing. Soluble mitochondrial lysate was applied to antibodies (from either anti-LONM or pre-immune serum) cross-linked to Protein A Sepharose and incubated with gentle mixing for 1 h at 4 °C. Following wash steps, antibody-bound proteins were eluted using 50 mM glycine-HCl pH 2.5 followed by neutralisation of the sample.
Detection of proteins. Following separation by glycine-or Tricine-buffered SDS-PAGE 76 , proteins were detected either by Coomassie Brilliant Blue R250 staining, digital autoradiography or immunoblotting.
Dried radioactive gels were exposed to phosphor screens and scanned using the Typhoon TM Trio variable mode imager (GE Healthcare). Analysis and quantification of gel images was performed using the ImageQuant software (Version 5.1, GE Healthcare). For the analysis of experiments by both digital autoradiography and immunoblotting, proteins were first transferred from polyacrylamide gels to polyvinylidene difluoride (PVDF) membrane using a semi-dry transfer method, before the dried membrane was exposed to a phosphor screen for analysis. The same membrane was then analysed by immunoblotting with the appropriate antisera (directed against human LONM, CLPX, CLPP and NDUFA9) as described previously 38 . Antiserum against rat OTC, was kindly provided by Nick Hoogenraad (La Trobe University). MDH antibody-horse radish peroxidase (HRP) conjugate (GTX40570) was from GeneTex. In general, membrane bound primary antibodies were detected using HRP-coupled secondary antibodies (anti-mouse or anti-rabbit IgG, Sigma-Aldrich) incubated with enhanced chemiluminescence detection reagents (GE Healthcare) and images captured with a ChemiDoc XRS+ system (Bio-Rad) using Image Lab Software (Bio-Rad).