IDE (insulin-degrading enzyme) is a widely expressed zinc-metallopeptidase that has been shown to regulate both cerebral amyloid β-peptide and plasma insulin levels in vivo. Genetic linkage and allelic association have been reported between the IDE gene locus and both late-onset Alzheimer's disease and Type II diabetes mellitus, suggesting that altered IDE function may contribute to some cases of these highly prevalent disorders. Despite the potentially great importance of this peptidase to health and disease, many fundamental aspects of IDE biology remain unresolved. Here we identify a previously undescribed mitochondrial isoform of IDE generated by translation at an in-frame initiation codon 123 nucleotides upstream of the canonical translation start site, which results in the addition of a 41-amino-acid N-terminal mitochondrial targeting sequence. Fusion of this sequence to the N-terminus of green fluorescent protein directed this normally cytosolic protein to mitochondria, and full-length IDE constructs containing this sequence were also directed to mitochondria, as revealed by immuno-electron microscopy. Endogenous IDE protein was detected in purified mitochondria, where it was protected from digestion by trypsin and migrated at a size consistent with the predicted removal of the N-terminal targeting sequence upon transport into the mitochondrion. Functionally, we provide evidence that IDE can degrade cleaved mitochondrial targeting sequences. Our results identify new mechanisms regulating the subcellular localization of IDE and suggest previously unrecognized roles for IDE within mitochondria.
IDE (insulin-degrading enzyme; also known as insulysin or insulinase; EC 126.96.36.199) is a member of the M16 family of zinc-metalloendopeptidases that contain at their catalytic centre a tetrad of conserved residues [His-Xaa-Xaa-Glu-Xaa76-Glu] that are involved in zinc binding and peptide hydrolysis . Although the insulin-degrading abilities of IDE have been studied for more than 50 years, IDE is expressed in both insulin-sensitive and insulin-insensitive tissues, suggesting that there may be unrecognized functions for this highly conserved endopeptidase. Within the cell, IDE is localized predominantly to cytoplasm , but in different cell types it can be secreted into the extracellular space  or associated with the cell surface [3,4]. In addition, IDE is present in several subcellular compartments, including peroxisomes, endosomes and the nucleus (see ). The localization of IDE to peroxisomes is explained by the presence of a conserved type-1 peroxisomal-targeting sequence (Ala-Lys-Leu) at the C-terminus . However, the mechanisms responsible for the targeting of IDE to other subcellular compartments are unknown.
Renewed interest in IDE has emerged following the observation that it can degrade Aβ (amyloid β-protein), which accumulates in the brains of AD (Alzheimer's disease) patients [2,6–8]. Building on this in vitro observation, we  and others  have recently shown that genetic deletion of IDE in mice leads to significant elevations in both brain Aβ and plasma insulin levels, confirming an important role for IDE in the regulation of these peptides in vivo. Our laboratory has also shown that up-regulation of IDE in neurons prevents AD-type pathology in transgenic mice overexpressing the Aβ precursor protein , supporting a potential therapeutic role for IDE in the treatment of AD.
The finding that IDE can degrade Aβ prompted a search for possible genetic linkage and/or allelic association of late-onset AD to the IDE region on chromosome 10q, evidence for which was first reported by Tanzi and colleagues . Since this initial report, this finding has been corroborated by several independent groups studying various patient cohorts around the world [13–15]. Genetic association has also been reported between the IDE gene and DMII (Type II diabetes mellitus) [16,17]. These reports are tantalizing in light of epidemiological studies that suggest a significant degree of co-morbidity between AD and DMII [18–20] or between AD and hyperinsulinaemia [21,22]. However, despite considerable effort, no pathogenic mutations have been uncovered within the coding region of IDE, raising the likelihood that the underlying pathogenic alterations are instead manifested at the transcriptional and/or translational levels.
Additional support for the possible involvement of IDE hypofunction in DMII and AD comes from a well known model of DMII, known as the GK rat . This compelling animal model was generated by inbreeding Wistar rats and selecting for animals with glucose intolerance, resulting in the production of a genetically homogeneous strain that spontaneously develops a diabetic phenotype. The GK rat has been subjected to extensive genetic analysis to uncover the elements underlying its susceptibility to DMII, leading to the discovery of two missense mutations in IDE (His18→Arg and Ala890→Val) that were reported to decrease the ability of IDE to degrade insulin . Our group has recently confirmed these findings and shown further that these mutations also impair the degradation of Aβ . Moreover, primary neuronal cultures from these animals display defects in Aβ degradation that lead to substantial accumulation of Aβ in their conditioned medium. Although the missense mutations in IDE may only partially explain the diabetic phenotype of the GK rat, IDE knockout mice show a remarkably similar phenotype; in addition to showing defects in insulin and Aβ degradation, these mice also manifest impaired glucose tolerance . Together, these findings provide compelling support for the hypothesis that co-morbidity between AD and DMII may in some cases be attributable to impairments in IDE function.
In the present study, we identify a novel mitochondrial isoform of IDE generated from translation beginning at an in-frame initiation codon located upstream of the canonical start site. We show that this longer isoform contains an N-terminal mitochondrial targeting sequence and that this indeed targets IDE to mitochondria. In addition, we provide evidence that IDE functions to degrade cleaved mitochondrial leader peptides. Our results identify new mechanisms regulating IDE subcellular localization and suggest novel functions for IDE that may play a role in both health and disease.
MATERIALS AND METHODS
Generation of cDNA constructs
The cDNA for full-length IDE containing its authentic termination codon was amplified from human brain RNA by reverse transcription–PCR and cloned into the pEF6/V5-His-TOPO vector (Invitrogen), using the primers IDE-Fwd (5′-GGCTAATGCGGTACCGGCTAGC-3′) and IDE-TA-Rev (5′-ATCTTCAGAGTTTTGCAGCC-3′), to generate pEF6-IDE-STOP. A second cDNA clone was generated using a primer lacking the termination codon and thus permitting the in-frame addition of the C-terminal V5-6×His fusion tag encoded by the vector, to generate pEF6-IDE-V5-His. Constructs encoding IDE beginning obligatorily at Met1 or Met42 (Met1-IDE or Met42-IDE) were generated from pEF6-IDE-STOP by mutagenesis of the codons encoding Met42 or Met1 respectively to Ala using the QuikChange Site-Directed Mutagenesis Kit (Stratagene). EGFP (enhanced green fluorescent protein) fusion constructs containing the N-terminal portion of Met1-IDE or Met42-IDE were generated by subcloning BamHI–BamHI fragments of the IDE cDNA (encoding peptides comprising residues Met1 or Met42 to Pro100) into pEGFP-N1 (Clontech). An internal HA (haemagglutinin) tag was added to the wild-type, Met1-IDE or Met42-IDE constructs immediately after Arg43 using a PCR-based method. A Kozak consensus sequence was also incorporated into the Met1-HA-IDE construct by PCR using the primers ATG1/Koz-IDE-Fwd (5′-CGCGGATCCGCCACCATGCGGTACCGGCTAGCGTGGCTTCTG-3′) and IDE-TA-Rev.
Generation of stable cell lines
Because we wished to compare the relative expression levels of the different HA-tagged IDE cDNA constructs, we elected to use Invitrogen's Flp-In system, which uses homologous recombination to stably integrate a single copy of each construct into the same genomic locus of a single cell line. To this end, the HA-tagged cDNAs were subcloned into pcDNA5/FRT and transfected into a CHO (Chinese hamster ovary) cell line stably expressing the Flp-In receiving vector, pFRT/lacZeo. Stable isogenic clones were selected and characterized according to the manufacturer's recommendations.
Subcellular fractionation by differential centrifugation
Groups of cells were washed two times in PBS, collected in ice-cold buffer A (5 mM Tris/HCl, 1 mM EDTA, 0.25 M sucrose, pH 7.4) and gently disrupted using a Dounce homogenizer. Cell nuclei and unbroken cells were pelleted (P1) by centrifugation at 500 g for 10 min. Pelleted fractions enriched for various subcellular organelles were obtained from the post-nuclear supernatant by successive centrifugation at 3000 g for 15 min (P2; enriched for heavy mitochondria and plasma membrane fragments), 10000 g for 15 min (P3; enriched for light mitochondria, lysosomes, peroxisomes and Golgi) and 100000 g for 60 min (P4; containing endoplasmic reticulum vesicles, plasma membrane, endosomes, etc.), and the high-speed supernatant was retained as the soluble fraction (S1) (see ). After each centrifugation step, the resulting pellets were washed two times by resuspension in an excess of buffer A, followed by centrifugation at the appropriate relative centrifugal force (g).
Purification of mitochondria from HEK-293 cells
Confluent monolayers of cultured cells (2×108) were washed with PBS, collected by scraping, and pelleted at 1000 g for 20 min. Cells were resuspended in 6 vol. of ice-cold cell homogenization medium (150 mM NaCl, 10 mM KCl, 10 mM Tris/HCl, pH 6.7) and disrupted with a Dounce homogenizer. The cell homogenate was supplemented with 1/3 vol. of cell homogenization medium containing 1 M sucrose, and nuclei and unbroken cells were removed by centrifugation at 1000 g for 20 min. The supernatant was centrifuged at 5000 g for 20 min, and the resulting pellet was resuspended in ice-cold sucrose/Mg2+ medium (0.25 M sucrose, 150 mM MgCl2, 10 mM Tris/HCl, pH 6.7). Following repeated washing and centrifugation at 5000 g for 20 min to remove trapped peroxisomes and other organelles, the purified mitochondria were resuspended in mitochondrial suspension medium (0.25 M sucrose, 10 mM Tris/HCl, pH 7.0).
CHO cells plated on to sterile glass-bottomed cell chambers (Nalge Nunc International) were transfected with the IDENTF::EGFP cDNA constructs either alone or together with cDNAs encoding DsRed directed either to the mitochondrion or to the endoplasmic reticulum (DsRed2-mito and DsRed2-ER respectively; Clontech). After 32 days, fluorescence in the green and red channels was visualized in the living cells using a Zeiss LSM 510 confocal microscope fitted with a 63× Plan NeoFluar oil-immersion objective using the 488 nm and 568 nm laser lines of argon and krypton lasers respectively.
Western blots were performed as described in . The IDE-5 antibody was generated by vaccination of rabbits with a peptide corresponding to amino acids 28–41 of the human IDE sequence and was affinity purified using this same peptide (Research Genetics). Other antibodies used in the present study include αIDE-1 , 3F10 (reactive with the HA tag; Roche), α-hsp60 (Stressgen Biotechnologies), α-catalase (Rockland, Inc.) and anti-α-syntaxin 13 (Stressgen Biotechnologies).
Cryosections were prepared from frozen blocks of formaldehyde-fixed cultured cells infiltrated with 2.3 M sucrose in PBS and transferred to formvar-carbon-coated copper grids for gold labelling. All antibodies and Protein A–gold were diluted in 1% (w/v) BSA. Grids were floated on drops of 1% BSA for 10 min to block unspecific labelling, transferred to 5 μl drops of primary antibody and incubated for 30 min. The grids were then washed in four drops of PBS for a total of 15 min and transferred to 5 μl drops of Protein A–gold (Amersham). Contrasting/embedding of the labelled grids was carried out on ice in 0.3% (w/v) uranyl acetate in 2% (w/v) methyl cellulose for 10 min. The grids were examined in a JEOL 1200EX transmission electron microscope, and images were recorded at a primary magnification of ×20000–30000.
Peptide degradation assays
A peptide corresponding to the cleaved mitochondrial targeting sequence of the E1α subunit of human pyruvate dehydrogenase  was synthesized by Fmoc (fluoren-9-ylmethoxycarbonyl) chemistry at the Tufts University Core Facility (Boston, MA, U.S.A.). Degradation assays were performed at 37 °C using the HPLC-purified peptide dissolved in PBS to a final concentration of 100 μM and reacted with 50 nM recombinant IDE . Aliquots of the reaction mixture were removed at various time points and terminated by boiling or by the addition of 2 mM 1,10-phenanthroline. Degradation products were separated by reverse-phase HPLC using a Vydac C-18 column (250 mm×4.6 mm) with a 50-min linear gradient of 7–70% (v/v) acetonitrile/0.1% (v/v) trifluoroacetic acid at 1.5 ml/min and detected by absorbance at 226 nm.
Alternative initiation codons are a conserved feature of mammalian IDE homologues
The open reading frames of all known mammalian IDE cDNAs contain two in-frame translation initiation codons, and therefore potentially encode proteins beginning either at the first or the 42nd amino acid (designated Met1-IDE and Met42-IDE; respectively; Figure 1A). Current experimental evidence indicates that Met42-IDE is the predominant, if not exclusive, isoform expressed in tissues and cultured cells, and previous attempts to detect Met1-IDE have proved unsuccessful [28,29]. Factors influencing translation efficiency have been postulated to account for this preference, as the nucleotide sequence surrounding the second initiation codon contains a better Kozak consensus sequence for initiation of translation (Figure 1B) . Although predicted to be less efficiently translated, the Met1-IDE isoform nevertheless could account for a significant fraction of total cellular IDE. Moreover, the novel N-terminal sequence contained within the Met1-IDE protein might determine an alternative subcellular localization for this translational isoform.
The amino acid and nucleic acid sequences corresponding to the N-terminus of IDE are conserved among mammalian species
Translation beginning at Met1 encodes a mitochondrial targeting sequence
To address this issue, we performed bioinformatic analyses on the Met1-IDE protein sequence using several N-terminal sorting sequence/subcellular localization prediction programs available on the ExPASy on-line proteomics server (www.expasy.ch). Two such programs predicted the presence of a signal peptide within the N-terminus of Met1-IDE (SignalP  and PSortII ), while two others predicted a mitochondrial targeting sequence in this region (MitoProt  and TargetP ). To resolve this issue experimentally, we generated DNA constructs that encoded fusion proteins consisting of N-terminal fragments (ending at Pro100) of either Met1-IDE (with Met42 mutated to Ala) or Met42-IDE (with Met1 mutated to Ala) fused in-frame to the N-terminus of EGFP, which is normally cytosolic (Figure 2A). These fusion constructs, designated Met1-IDENTF::EGFP and Met42-IDENTF::EGFP respectively, or the unmodified EGFP construct were transfected separately into CHO cells and visualized by confocal microscopy. Like EGFP itself, the Met42-IDENTF::EGFP fusion protein was expressed overwhelmingly in the cytoplasm (Figure 2B). This result is consistent with the principally cytosolic localization reported for wild-type IDE and suggests that the Met42-IDE isoform lacks an organelle-specific targeting sequence at its N-terminus. In marked contrast, the Met1-IDENTF::EGFP fusion protein was expressed within subcellular organelles possessing a vermiform morphology (Figure 2B). To determine the identity of these organelles, the Met1-IDENTF::EGFP construct was co-transfected with cDNAs encoding the DsRed fluorescent protein directed to different subcellular compartments. Met1-IDENTF::EGFP co-localized perfectly with mitochondrion-directed DsRed (Figure 2B, lower panels), but did not overlap with endoplasmic reticulum-directed DsRed (results not shown). These results demonstrate that the N-terminal sequence present in Met1-IDE includes a mitochondrial targeting sequence that is necessary and sufficient for transport into mitochondria.
The N-terminus of the Met1-IDE isoform contains a mitochondrial targeting sequence
To investigate the subcellular localization of the full-length IDE translational isoforms, we generated stable cell lines expressing the following HA-tagged expression constructs (Figure 3A): wtHA–IDE, encoding full-length wild-type IDE (i.e. containing both initiation codons); Met42-HA–IDE, beginning obligatorily at Met42 and lacking the first initiation codon; Met1-HA–IDE, beginning obligatorily at Met1 and with Met42 mutated to an alanine; and Koz-Met1-HA–IDE, comprising the latter construct with a Kozak consensus sequence engineered to precede Met1. We generated these cell lines using a homologous recombination-based strategy (Invitrogen Flp-In System) in order to integrate each expression construct into the same genomic locus of a single parental CHO cell line. Because the expression constructs in the different cell lines are integrated at the same site(s), the relative expression levels of the different isoforms can be validly compared.
The full-length Met1-IDE isoform is expressed in a cell fraction enriched in mitochondria
Cells expressing the different constructs were homogenized in an isotonic buffer, fractionated by differential centrifugation (see the Materials and methods section), and analysed by Western blot for the presence of the HA-tagged IDE proteins. Consistent with the known subcellular localization of endogenous IDE, the wtHA–IDE construct was expressed predominantly in the high-speed supernatant fraction (cytosol), but was also found at lower levels in each of the pelleted fractions (Figure 3B). Met42-IDE was expressed in a closely similar pattern (Figure 3B), confirming previous studies showing that the majority of endogenous IDE protein represents this isoform. In striking contrast, the Koz-Met1-HA–IDE construct was present most abundantly in the post-nuclear 3000 g pellet (P2), which is enriched in mitochondria (Figure 3B), as confirmed by Western analysis of marker proteins (Figure 3D). It initially appeared that the Kozak consensus signal within Koz-Met1-HA–IDE was essential for translation beginning at Met1, since the Met1-HA–IDE protein (lacking the Kozak consensus sequence) did not appear to be present on Western blots that detected the other isoforms (Figure 3B). However, upon longer exposure, expression of Met1-HA–IDE could, in fact, be detected solely in the mitochondrion-enriched P2 fraction (Figure 3C). Taken together, the results in Figure 3 indicate that translation of Met1-IDE can occur, albeit at lower levels than that of Met42-IDE, and they show that the Met1-IDE isoform is localized to a mitochondrion-enriched subcellular fraction. Similar results (not shown) were obtained with transiently transfected COS cells.
Endogenous IDE is present within mitochondria
To attempt to confirm that an endogenous mitochondrial IDE isoform exists, we performed Western blots on extensively washed mitochondria purified from untransfected HEK-293 cells (see the Materials and methods section) and detected using our well characterized anti-IDE antibody (IDE-1). This antibody is highly specific for IDE (both isoforms), recognizing a ∼110 kDa band on Western blots of whole-cell lysates or tissue homogenates of numerous mammalian species that is absent from samples prepared from IDE knockout mice . Endogenous IDE was readily detected in purified mitochondrial preparations (far left lanes in Figures 4A and 4B), which were shown by Western analysis to be free of the peroxisomal marker catalase, but highly enriched in the mitochondrial marker hsp60 (Figure 4A). The IDE present in these fresh intact mitochondria was protected from digestion by trypsin (Figure 4B), excluding the possibility of contamination with the cytosolic Met42-IDE isoform and consistent with the predicted intramitochondrial localization of Met1-IDE based on its N-terminal targeting sequence.
Endogenous IDE is present in purified mitochondria
The mitochondrial presequence of IDE is probably removed upon import into mitochondria
The Met1-IDE and Met42-IDE isoforms are predicted to differ in size and could conceivably be discriminated on this basis. In practice, however, we could not detect a reliable size difference between the Met1-IDE and Met42-IDE HA-tagged isoforms even following separation on 4% (w/v) polyacrylamide gels, although we were readily able to detect a sizable mobility shift in constructs containing a 45-residue tag at the C-terminus, even on 8% (w/v) polyacrylamide gels (results not shown). This result can be explained if the mitochondrial targeting sequence is cleaved upon import to the mitochondrion by MPP (mitochondrial processing peptidase), as is true for most mitochondrial presequences. Consistent with this idea, the signal prediction program MitoProt predicts cleavage of Met1-IDE at the Leu31–Cys32 peptide bond (‘probability’=0.9770 ) (arrow in Figure 1A). Cleavage at this site would leave the proteolytically processed Met1-IDE isoform with only 10 additional amino acids beyond the 978 amino acids present in Met42-IDE, explaining the difficulty in discriminating between Met42-IDE and cleaved Met1-IDE. We were also unable to directly detect the non-cleaved Met1-IDE isoform using an affinity-purified polyclonal antibody (IDE-5) that was raised to a peptide corresponding to amino acids 28–40 within the predicted mitochondrial targeting sequence (Figure 1A, underlined). However, cleavage at the predicted site (Leu31–Cys32) would be expected to remove significant portions of the epitope recognized by IDE-5, and this could explain its inability to recognize the proteolytically processed variant of Met1-IDE. These results are therefore consistent with the idea that the Met1-IDE isoform represents a mitochondrial precursor whose targeting sequence is removed proteolytically by MPP upon import into the mitochondrion. In this regard, we noticed that, in purified mitochondria, the IDE-1 antibody does recognize a faint band of the predicted size of the unprocessed Met1-IDE isoform (Figure 4B, arrowhead) that is susceptible to trypsin digestion. We speculate that this band represents a small pool of Met1-IDE that has been translocated to mitochondria (e.g. by cytosolic hsc70 or other mitochondrial presequence chaperones) but that has not yet been proteolytically cleaved by MPP.
Ultrastructural evidence for the presence of IDE within mitochondria
To confirm the mitochondrial localization of IDE at the ultrastructural level, we performed immuno-electron microscopy on CHO cells expressing HA-tagged IDE constructs. In CHO cells (Figure 5A) expressing the wild-type isoform, IDE was clearly detected within mitochondria, in addition to its expected cytoplasmic localization, and similar results were obtained with COS cells (results not shown). In CHO cells expressing the Koz-Met1-HA–IDE construct, IDE was detected exclusively in mitochondria (Figure 5B).
Detection of the mitochondrial isoform of IDE by immuno-electron microscopy
Potential function of IDE within mitochondria
IDE within peroxisomes has been shown to degrade the cleaved leader sequences removed from peroxisomal precursor proteins . Moreover, likely IDE orthologues in plants have been shown to degrade N-terminal sorting sequences cleaved from both mitochondrial and chloroplast proteins [36,37]. To determine whether IDE may perform a similar function in mitochondria, we tested the ability of IDE to degrade a synthetic peptide corresponding to the cleaved mitochondrial targeting sequence for the E1α subunit of human pyruvate dehydrogenase  (Figure 6A). This peptide was indeed hydrolysed at several sites by purified recombinant IDE (Figure 6B), and the degradation was completely inhibited by 1,10-phenanthroline, a broad-spectrum zinc-metalloprotease inhibitor (results not shown). Analysis of the rate of hydrolysis of this substrate suggested that degradation of this peptide was highly efficient: based on disappearance of the intact peptide as monitored by HPLC, we estimated a kcat value of ∼200 min−1. Although this value is likely to reflect a lower boundary of the true value, it is comparable with the kcat values determined for the degradation of Aβ (∼100–200 min−1) [38,39], another avid IDE substrate.
IDE efficiently degrades a prototypical mitochondrial targeting sequence
Here we provide evidence for the existence of a previously unidentified isoform of IDE that is localized to mitochondria. This isoform is generated by translation beginning at an alternative initiation codon upstream of the canonical start site, producing a novel protein with 41 amino acids added to the N-terminus. We show that this N-terminal sequence, which we find is predicted to contain a mitochondrial targeting signal, is both necessary and sufficient to transport normally cytosolic proteins (IDE and EGFP) into mitochondria, where, as for most nuclear-encoded mitochondrial proteins, the signal appears to be removed proteolytically. Within mitochondria, the functional role of IDE may include the degradation of cleaved mitochondrial targeting sequences.
IDE appears to be unique among mammalian zinc-metalloproteases in using alternative translation initiation to regulate the expression of multiple protein isoforms with different subcellular localizations. By way of comparison, a mitochondrial variant of another metalloendopeptidase, neurolysin (EC 188.8.131.52), is regulated via an alternative promoter, the use of which generates a splice isoform that encodes a protein containing a cleavable mitochondrial targeting sequence . However, the translational regulation apparently used by IDE is not unprecedented. For example, such a mechanism has been found to regulate the expression of mitochondrial and cytosolic variants of fumarase . As compared with transcriptional regulation, which can be influenced by numerous cis- and trans-acting factors, regulation via translation would appear to be relatively inflexible, being influenced only by the primary nucleotide sequence of the mRNA. In the case of fumarase, however, RNA species were found that could suppress translation of the cytosolic variant and enhance expression of the mitochondrial one . Such trans-acting factors might also regulate the translation of Met1-IDE and Met42-IDE. In this regard, we found a larger relative proportion of endogenous mitochondrial IDE (Figure 4A) than was expected on the basis of our studies in stable cell lines expressing the HA-tagged IDE constructs (Figure 3B). It is possible that the latter constructs, which lack the complete 5′ and 3′ untranslated regions present in endogenous IDE mRNAs, may have escaped the influence of cis- or trans-acting factors regulating the endogenous transcripts. In view of the potentially important role of IDE in health and disease, factors that regulate the translation of the two IDE isoforms should be viewed as attractive targets for future genetic studies and drug development.
It is instructive to compare the present findings with a study by Hospital and colleagues , published after initiation of our work, which examined translation initiation and subcellular localization of a close homologue of IDE, NRDC (N-arginine dibasic convertase; nardilysin). Like IDE, NRDC also contains two in-frame initiation codons, the second of which is believed to be used preferentially if not exclusively. When translated beginning at the first initiation codon, the N-terminal amino acid sequence is predicted by both MitoProt and PSORT II to encode a mitochondrial targeting sequence, as we found for IDE. When Hospital and co-workers added a Kozak consensus sequence in front of the first initiation codon (but did not mutate the second), they observed a longer isoform of NRDC in addition to the usual isoform in whole-cell lysates. Intriguingly, they could detect both isoforms in the cytosol as well as on the cell surface, where biotinylation could take place. Although these investigators did not report on the possible localization of the longer isoform within mitochondria, their results nonetheless show that this N-terminal sorting signal is not absolutely required for the transport of NRDC to the cell surface (nor were C-terminal sorting signals involved, since Hospital and colleagues analysed constructs containing C-terminal tags). Given the very high degree of sequence similarity between NRDC and IDE, it seems likely that the N-terminal extension of IDE that we describe here is also not required for the targeting of wild-type IDE to the cell surface, where IDE can be biotinylated [3,4].
In the present study, we show that IDE efficiently degrades a prototypical cleaved mitochondrial targeting sequence. Such sequences are characterized by the presence of numerous positively charged residues, particularly Arg. In this regard, it is notable that several IDE homologues show a preference for substrates containing Arg residues. For example, IDE's closest homologue, NRDC, shows a preference for cleavage within the motif -Xaa↓Arg-Lys-, or less commonly at -Arg↓Arg-Xaa-, where Xaa represents any amino acid except Arg or Lys . Moreover, the cleavage of leader peptides by MPP frequently fits the motif Arg-Xaa↓Xaa, where Xaa represents any amino acid . Furthermore, we have recently found that IDE itself shows a strong preference for cleavage of peptides containing Arg residues at the +1 and +2 positions, based on active-site mapping experiments using combinatorial peptide mixtures (M. A. Leissring, B. E. Turk, L. C. Cantley and D. J. Selkoe, unpublished work). Taken together, these observations suggest that cleaved mitochondrial targeting sequences are innately good substrates for IDE, lending support to the view that their degradation represents a physiological function of mitochondrial IDE. We emphasize, however, that we cannot rule out other, more vital functions for IDE within mitochondria. Indeed, other members of the M16 metalloprotease family have evolved to play essential roles in mitochondrial functions, as exemplified by the core subunits of the mitochondrial cytochrome bc1 complex , which play an indispensable structural role in the integrity of this respiratory complex. Moreover, outside of mitochondria, IDE and its homologues have been reported to participate in diverse cellular functions, such as regulation of the proteasome , interaction with androgen and glucocorticoid receptors , and regulation of bud site selection in haploid Saccharomyces cerevisiae , the latter process being independent of its proteolytic activity.
As mentioned in the Introduction, an inbred rat model of DMII has been found to harbour two missense mutations in IDE, namely His18→Arg and Ala890→Val . It is notable that one of these mutations, His18→Arg, lies within the mitochondrial targeting sequence that we have identified in Met1-IDE. Interestingly, an Arg residue is normally present at this position in both the human and murine IDE homologues (Figure 1A). This Arg residue appears to contribute to the positive charge that is essential for the function of mitochondrial signal sequences , suggesting that this mutation may alter the subcellular localization of rat IDE. Interestingly, when Fakhrai-Rad and colleagues , who originally identified the missense mutations in IDE in the GK rat, transfected cells with cDNAs that harboured either both mutations together or the individual mutations alone, they observed defects in insulin degradation only in cells transfected with constructs containing both mutations. We recently revisited this question  and found significant differences in the catalytic efficiency of IDE within the cytosolic fraction of GK rat brains relative to wild-type controls. Because most IDE within this fraction lacks the mitochondrial targeting sequence and therefore the His18→Arg mutation, our results suggest that the Ala890→Val mutation alone is sufficient to confer significant impairments in the degradation of insulin and Aβ. Nonetheless, it remains possible that the His18→Arg mutation could contribute independently to the diabetic phenotype in the GK rat, perhaps by altering the subcellular localization of the Met1-IDE isoform. Intriguingly, mitochondria isolated from the brains of GK rats were recently reported to exhibit increased vulnerability to exposure to Aβ peptides . Our discovery that IDE is present within mitochondria should help to elucidate the mechanism(s) underlying this intriguing finding.
It is presently uncertain whether the mitochondrial IDE precursor identified in the present study plays any role in human disease. However, emerging evidence implicates mitochondrial dysfunction in the pathogenesis of DMII [51,52], and mitochondria also play important roles in neurodegeneration . Because both variants can be translated from the same mRNA, mutations or other factors affecting the relative abundance of the Met1-IDE isoform would necessarily affect the abundance of the Met42-IDE isoform, which is presumably the principal form involved in the degradation of Aβ and insulin in vivo. However, one cannot rule out the possibility of a more direct role for the Met1-IDE isoform in these diseases. While significantly more research will be required to fully uncover the physiological and pathophysiological roles of IDE within mitochondria, such efforts should be facilitated by the finding that mitochondrial IDE is derived from a unique translational isoform that can be manipulated experimentally.
We thank Dominic Walsh for providing CHO cell lines stably expressing the pFRT/lacZeo vector of the Flp-In system, Maria Ericsson for performing electron microscopy, Chittaranjan Das for assistance with HPLC, and Kostas Vekrellis and Stefan Mansourian for assistance with the generation and purification of the αIDE-5 antibody. This work is supported by National Institutes of Health grants AG12749 (to M.A.L., W.F. and D.J.S.) and K08 NS046324-01 (to W.F.).
Chinese hamster ovary
Type II diabetes mellitus
enhanced green fluorescent protein
and Met42-IDE, IDE isoforms beginning at Met1 and Met42 respectively
mitochondrial processing peptidase
N-arginine dibasic convertase