Bacterial Proteins Predisposed for Targeting to Mitochondria
http://www.100md.com
分子生物学进展 2004年第4期
Russell Grimwade School of Biochemistry and Molecular Biology, University of Melbourne, Australia
E-mail: t.lithgow@unimelb.edu.au.
Abstract
Mitochondria evolved from an endosymbiotic proteobacterium in a process that required the transfer of genes from the bacterium to the host cell nucleus, and the translocation of proteins thereby made in the host cell cytosol into the internal compartments of the organelle. According to current models for this evolution, two highly improbable events are required to occur simultaneously: creation of a protein translocation machinery to import proteins back into the endosymbiont and creation of targeting sequences on the protein substrates themselves. Using a combination of two independent prediction methods, validated through tests on simulated genomes, we show that at least 5% of proteins encoded by an extant proteobacterium are predisposed for targeting to mitochondria, and propose we that mitochondrial targeting information was preexisting for many proteins of the endosymbiont. We analyzed a family of proteins whose members exist both in bacteria and in mitochondria of eukaryotes and show that the amino-terminal extensions occasionally found in bacterial family members can function as a crude import sequence when the protein is presented to isolated mitochondria. This activity leaves the development of a primitive translocation channel in the outer membrane of the endosymbiont as a single hurdle to initiating the evolution of mitochondria.
Key Words: endosymbiont ? mitochondria ? targeting sequence ? protein import
Introduction
A defining moment in the evolution of eukaryotic cells was the acquisition of mitochondria, and this appears to have occurred only once in the course of evolution (Martin and Müller 1998; Gray, Burger, and Lang 1999, 2001; Lang, Gray, and Burger 1999; Kurland and Andersson 2000; Emelyanov 2001). Mitochondria were derived from an endosymbiotic relationship with internalized proteobacteria, via a progressive transfer of genetic material. Copies of genes from the symbiont found their way to the host cell nucleus to be expressed as proteins that might then be imported back into the symbiont (Andersson and Kurland 1999; Lang, Gray, and Burger 1999; Cavalier-Smith 2002). Surrender of the bacterial genome provided the means for the eukaryote to control the metabolism of the endosymbiont. The relocation of the first gene from the endosymbiont to the nucleus and its deletion from the symbiont genome is a crucial step, after which the symbiont loses its status as an independent organism. This crucial step also creates an extreme topological problem: how to get proteins, now made on the host cell ribosomes, back into the appropriate compartment of the endosymbiont.
In modern eukaryotes the solution to this problem, protein targeting into mitochondria, is an elaborate and now well-understood process. Each of the several hundred proteins carry discrete targeting sequences to be selected for import into the organelle and are recognized for import by a complex protein import machinery (Neupert 1997; Voos et al. 1999). Two highly improbable events, creation of mitochondrial targeting sequences and evolution of a protein import machinery in the outer membrane of the endosymbiont were required to initiate the conversion of the -proteobacter endosymbiont to mitochondria.
Some components of the import machinery we see now in mitochondria are derived from the protein secretion apparatus of bacteria. Peptidases and elements of the translocation complexes in the mitochondrial inner membrane were probably already present in the endosymbiont, having homologs in present-day bacteria (Hermann 2003). However, the creation and evolution of the translocase in the outer mitochondrial membrane (TOM complex) remains something we understand almost nothing about. It seems likely that the genes encoding subunits of the TOM complex, like a large number of other mitochondrial proteins, were derived from the host cell's genome (Karlberg et al. 2000). The TOM complex consists of receptors that bind mitochondrial targeting sequences (Tom20 and Tom22 [Endo and Kohda 2002]), a translocation channel (Tom40) and attendant subunits (Tom6 and Tom7), and three or more other subunits that seem more species specific (Neupert 1997; Schatz 1997; Voos et al. 1999; Gabriel, Buchanan, and Lithgow 2001; Pfanner and Chacinska 2002). (The proteins are named according to their size in kilodaltons; for example, Tom20 is the 20-kDa subunit of the TOM complex [Pfanner et al. 1996]). The need for a codependent evolution of the TOM complex and the targeting sequences for each of the imported protein substrates is a major weakness in the current model explaining the evolution of mitochondria.
Comparative genomics makes a strong argument for the later stages of the evolution of mitochondria (Brennicke et al. 1993; Kurland and Andersson 2000; Gray et al. 2001): (1) escape of the DNA (or an RNA copy) of the gene from the endosymbiont, (2) transfer of the gene fragment to the nucleus and integration in the genome, (3) adaptive rearrangements to allow gene expression, (4) exon shuffling, and other means to create a mitochondrial targeting sequence, (5) transcription, then translation, of the mitochondrial protein in the cytosol, and (6) gene inactivation of the redundant copy(s) of the gene in the mitochondria (Kurland and Andersson 2000; Gray et al. 2001). However, in its current form, this model assumes the preexistence of a TOM complex. Before the TOM complex, no pressure existed to select for the sequences on a protein substrate that could promote targeting back to the early endosymbiont.
Here we show, using two independent predictive strategies, that many proteins present in extant prokaryotes carry rudimentary features for mitochondrial targeting. Application of two predictors, Mitoprot II and Predotar, and computer simulations of feasible scenarios suggest that up to 5% of proteins encoded by the extant proteobacterium Escherichia coli have features that would serve as mitochondrial targeting information, with basic and amphipathic extensions at their amino-termini and "meso-hydrophobic" character throughout the proteins. One of the bacterial proteins, YhaR, serves as an instructional case in point. It is a member of a large family of highly conserved proteins found in at least eight diverse phyla of present-day bacteria. Multiple sequence alignments of the bacterial versions of these proteins show ragged amino-termini, with YhaR from E. coli having one of the longer amino-terminal extensions. Ectopic expression of bacterial YhaR results in targeting of the protein to yeast mitochondria, suggesting that in some cases, during the course of evolution, this preadaptation meant little or no mutagenesis of upstream regions in bacterial genes to render the proteins they encode competent for import into "protomitochondria." The presence of this preexisting "targeting" information would have provided selective pressure to initiate evolution of the TOM complex.
Materials and Methods
Comparative Sequence Analysis
Iterative Blast analyses were undertaken as previously described (Macasev et al. 2000). Multiple sequence alignments were constructed with ClustalW on the European Bioinformatics Institute server (http://www.ebi.ac.uk/clustalw/).
Prediction of Mitochondrial Targeting Peptides
MitProtII was obtained from ftp://ftp.ens.fr/pub/molbio, and Predotar was obtained from http://www.inra.fr/predotar/. The precompiled Solaris executable of MitoProtII was used in calculations. For the processing of multiple sequences and analysis of results, a set of wrapper programs was developed in Perl.
Computer Simulations of Mitochondrial Predictors
For computer simulations, we assumed two predictors with the characteristics of Predotar (i.e., sensitivity = 0.86 and specificity = 0.50 [Emanuelsson and von Heijne 2001]) whose errors are uncorrelated. This is expected to be a reasonable assumption because Mitoprot II and Predotar use an entirely different approach for the prediction of mitochondrial targeting features. Simulations were carried out on genomes containing a total of 4,096 proteins, with the number of mitochondrial proteins ranging from 2% to 34%.
Given the set of assumed mitochondrial proteins, the predictions of two predictors were simulated, and the intersection of positive predictions was calculated. Given the percentage of mitochondrial proteins, the calculation was repeated 20 times, and for each run, the ratio of false negatives to false positives was evaluated. From these data, the mean and standard deviation of the ratio of false negatives to false positives was calculated. The mean, thus, represents our best estimate of this ratio, and the standard deviation gives an estimate of the variability between repeated simulations. Optimization experiments suggest the data are insensitive to small changes in the neighborhood of chosen properties of predictors.
Yeast Strains and Mitochondrial Isolation
A URA3 gene cassette was used to disrupt the coding sequence in the HMF1 gene after the codon corresponding to Gly34. The hmf1::URA3 fragment was transformed into W303a/ yeast cells to generate the strain YRG27 (Mata/, leu2/leu2, his3/his3, ade2/ade2, ura3/ura3, trp1/trp1, can1/can1, and HMF1/hmf1::URA3). YRG27 cells were sporulated and tetrads dissected to obtain haploid null mutants (Mata, leu2, his3, ade2, ura3, trp1, can1, and hmf1::URA3). Yeast mitochondria were prepared as described in Gabriel, Egan, and Lithgow (2003). Where indicated, mitochondria were exposed to osmotic shock conditions (10mM HEPES pH 7.4) or to 0.1% Triton X-100.
Antibody Production
The open reading frame encoding Mmf1 was amplified by PCR using purified yeast genomic DNA as a template and cloned into pQE30 (QIAGEN) behind a 6x Histidine coding sequence. Hexahistidine-tagged Mmf1 was expressed in Escherichia coli strain M15 (Qiagen) by induction with 0.5 mM IPTG. Denatured protein was purified using Ni-NTA resin (Qiagen) according to manufacturer's directions and used to immunize rabbits for the production of antibodies.
Mitochondrial Protein Import Assays
In vitro translation of 35S-labeled YhaR in rabbit reticulocyte lysates and import into isolated mitochondria was as previously described (Gabriel, Egan, and Lithgow 2003).
Results
The initial impetus for this study came from comparative sequence analyses of the YER057c/yjgF/UK114 family of highly conserved acid-soluble proteins (Volz 1999). With at least 50 members of the family now known, multiple sequence alignments show a striking level of sequence conservation across species from prokaryotic and eukaryotic organisms (data not shown). Just as striking was the raggedness of the amino-terminal sequence from many family members in multiple sequence alignments informed from crystal structures of two proteins (fig. 1); YjgF was crystallized from E. coli (Volz 1999) and YabJ was crystallized from Bacillus subtilis (Sinha et al. 1999), showing a compact single domain, with the amino-terminal segment of YabJ forming a four residue strand ("?1") with a loop that folds back onto the first ?-strand element of the core ("?2"). Extensions were predicted to be present in the proteins from the cyanobacterium Nostoc, the gram-positive bacterium Enterococcus faecalis, and YhaR from the proteobacterium E. coli. Among the diverse prokaryotes we surveyed, the longest extension was found in the -proteobacterium Rhodospirillum rubrum. Proteins from the eukaryotes Saccharomyces cerevisiae, Schizosaccharomyces pombe, Magnaporthe grisea, Pleurotus ostreatus, and Ustilago maydis also have amino-terminal extensions (fig. 1).
FIG. 1. Ragged amino-termini in the YER057c/yjgF/UK114 family of proteins. Iterative Blast analyses were used to compile sequences of family members. All are proteins of approximately 150 residues; the number of the residues shown is indicated. For the eukaryotes, only the fungal sequences are included (homologs were also found present in the genomes of animals and plants) aligned against representative proteins from eight diverse phyla of Bacteria. Where the proteins have been studied, names are shown in parentheses. From the structure of YjgF and YabJ, residues corresponding to those of the first ?-strand (?1) and the start of the second ?-strand (?2) are shaded gray. The experimentally verified presequence of Mmf1 is shown underlined in bold type. It is unknown whether the extensions on the bacterial proteins are functionally important. In many cases, a second AUG codon is found internally, in some cases having a good Shine-Dalgarno context for translation initiation (Lithgow, unpublished observations)
Extensions on the Eukaryotic Proteins Represent Mitochondrial Targeting Sequences
Sequence analysis of all the YER057c/yjgF/UK114 family members from fungi suggested the amino-terminal extensions represent mitochondrial targeting sequences. MitoProt II, which directly calculates a probability for mitochondrial targeting based on various physicochemical properties of mitochondrial proteins, including a "mesohydrophobicity" score that correlates well with import competence of known mitochondrial proteins (Claros et al. 1995; Claros and Vincens 1996), predicts the extended proteins as highly probable mitochondrial proteins (M. grisea, P = 0.939; S. pombe, P = 0.998; P. ostreatus, P = 0.941; U. maydis; P = 0.9844, and S. cerevisiae, P = 0.994), whereas the proteins without extensions predict strongly as nonmitochondrial proteins. The mitochondrial targeting sequences of hundreds of proteins are now known and prove to be discrete segments rich in positively charged residues and with tendencies to form two to three turns of a helix with amphipathic character (von Heijne 1986). Predotar, a neural network predictor (http://www.inra.fr/predotar/) also predicts the extensions of the fungal proteins in figure 1 as mitochondrial targeting sequences.
Pmf1 is now known to be mitochondrial in S. pombe, and epitope-tagged Mmf1 has been located in mitochondria of S. cerevisiae (Oxelmark et al. 2000; Kim, Yoshikawa, and Shirahige 2001; Marchini et al. 2002). An antibody raised to purified, recombinant protein found Mmf1 exclusively in the mitochondria (data not shown) and further localized Mmf1 within the mitochondrial matrix (fig. 2), demonstrating Mmf1 is not simply attached to the mitochondrial surface but has been imported across the mitochondrial membranes. The shorter protein from S. cerevisiae, Hmf1, is found only in the cytosol (data not shown). To verify that the amino-terminal extension of Mmf1 is a cleavable mitochondrial targeting sequence, a preparation of matrix proteins was isolated from yeast mitochondria, separated by Tris-tricine SDS-PAGE (Sch?gger and von Jagow 1991) and blotted to PVDF membrane for amino-terminal sequencing. Mmf1 is processed between amino acids Gly17 and Ile18, truncating the mature protein close to the point conforming to the size of the shorter family members (see figure 1).
FIG. 2. Mmf1 is processed to be a soluble protein in the mitochondrial matrix. (A) Mitochondria (100 μg protein) isolated from wild-type yeast cells and treated with 10 μg proteinase K either in iso-osmotic buffer to protect the integrity of the mitochondrial membranes (lane 1), in hypo-osmotic buffer (10 mM HEPES pH 7.4) to rupture the outer membrane and allow proteinase K access to proteins such as cytochrome b2 (Cyb2) in the intermembrane space (lane 2), or, in the presence of 0.1% Triton X-100 (lane 3), to solubilize both the outer and inner membrane and allow proteinase K access to proteins in the matrix such as -ketoglutarate dehydrogenase (Kgd1). After separation on SDS-PAGE, immunoblots were analyzed with antisera raised to the purified proteins. (B) Mitochondria (100 μg protein) were extracted with cold 100 mM Na2CO3 (pH 10.0), incubated on ice for 30 minutes, centrifuged (30 min) at 75,000 rpm in a TL100.2 rotor (Beckman) to yield a membrane pellet (P) that includes the integral membrane protein Por1 and a soluble extract (S) containing the matrix-located chaperonin Cpn10 and Cyb2 from the intermembrane space
Some Bacterial Family Members, Including Yhar, Are Predisposed for Targeting to Mitochondria
Several of the bacterial family members also have putative amino-terminal extensions. Might the bacterial proteins fortuitously carry sequences suitable for mitochondrial targeting? The open-reading frame corresponding to YhaR from E. coli was cloned into a vector for in vitro transcription and translation, and the [35S]methionine-labeled protein incubated together with mitochondria isolated from S. cerevisiae. The 17-kDa protein associates with mitochondria (fig. 3, lane 1) and can interact with the TOM complex to initiate translocation across the outer membrane. If proteinase K is added to mitochondria, the protein is degraded to a fragment of approximately 9 kDa (lane 2). The protein fragment is within the intermembrane space because rupture of the outer membrane by osmotic shock renders the fragment accessible to proteinase K (lane 3). Quantitative recovery of the [35S]methionine-labeled fragment suggests that it represents the amino-terminal two-thirds of YhaR, given the placement of methionine residues (at positions 1, 6, 22, 47, and 95) in the 150 residue protein.
FIG. 3. The bacterial protein YhaR is partially translocated into isolated mitochondria. Mitochondria (625μg protein/ml) in import buffer (0.6 M sorbitol, 50 mM HEPES pH 7.4, 2mM potassium phosphate pH 7.4, 25 mM KCl, 10 mM MgCl2, 0.5 mM EDTA, 25 mg/ml BSA, 1 mM DTT, 1 mM ATP, and 5 mM NADH) were aliquoted (100 μl), incubated with 10 μl translation mix containing [35S]methionine-labeled YhaR and incubated for 20 minutes at 25°C. Mitochondria were then collected by centrifugation and resuspended in import buffer or hypo-osmotic buffer (10 mM HEPES pH 7.4, "osmotic shock") or in import buffer containing 0.1% Triton X-100 to solubilize the inner membrane. Where indicated (+), 10 μg Proteinase K was added for 15 minutes on ice before mitochondria were reisolated and prepared for SDS-PAGE and fluorography
To determine whether the mitochondrial targeting propensity of YhaR is unique, we scanned a conceptual translation of the E. coli genome (4,289 protein sequences) with the programs MitoProt (Claros and Vincens 1996) and Predotar. A total of 487 E. coli proteins (11.4% of the total conceptual proteome) are predicted with greater than 0.7 probability to be mitochondrial MitoProtII. The program Predotar reports "mitochondrial scores," and the Predotar server (http://www.inra.fr/predotar/) predicts a protein as being mitochondrial if the score is greater than 0.5. To increase the stringency of predictions, we have used a Predotar score of greater than 0.7 as a high-confidence prediction of mitochondrial location. Of the 4,289 E. coli proteins, 613 (14.3 %) satisfy this criterion. Figure 4A shows the plot of Predotar scores versus MitoProtII probabilities for all protein sequences derived from the E. coli genome.
FIG. 4. Mitochondrial targeting predictions. (A) A conceptual translation of the E. coli genome yields 4,289 protein sequences, each of which was analyzed by Predotar and by MitoProt II. Each cross represents the values of the Predotar mitochondrial score versus the MitoProt II probability for a single protein. The squared area encloses 210 proteins predicted by both MitoProt II and Predotar to carry mitochondrial targeting information. The identities of these 210 sequences and the predictor scores are shown in the Supplementary Material online. (B) Computer simulations were used to calculate the ratio of false negatives to false positives in combined predictions from two independent predictors with properties equivalent to Predotar, depending on the percentage of mitochondrial proteins encoded in a genome. For each data point, the mean (diamond) and the standard deviation (error bar) of this ratio was calculated from 20 simulations. The dashed line shows the ratio of 1:1, for which the prediction matches the true number mitochondrial proteins. For the assumed properties of two predictors, the ratio of false negatives to false positives in the combined prediction was greater than 1 when the fraction of mitochondrial proteins was less than 25%, implying that the combined prediction underestimates the actual number of proteins that carry mitochondrial targeting sequences
Reliability of the Mitochondrial Targeting Prediction Methods
Despite considerable success in predicting mitochondrial target sequences, current prediction methods are imperfect. In a previous study, a set of 2,738 nonplant proteins of known subcellular location was compiled and screened with Predotar: 76.3% were correctly assigned (Emanuelsson and von Heijne 2001). Predotar prediction contained a large number of false positives (50% of the total positive assignments), which greatly increases the total number of sequences estimated to carry mitochondrial target signal. False negatives offset this to a certain extent, but in any calculation where false positives greatly exceed false negatives, the final estimate is unrealistically high. Our goal was to estimate the total number of proteins from the E. coli genome predisposed for targeting to mitochondria. Considering the sensitivity of current predictors (Emanuelsson and von Heijne 2001), the predictions of MitoProtII and Predotar (11.4% and 17.9%) cannot be taken as reliable.
The two programs rely on different methods for the prediction of mitochondrial targeting peptides, and, in the first approximation, one could assume that errors in the predictions made by MitoProtII and Predotar are uncorrelated. We investigated this scenario by computer simulations to determine how reliable a combined prediction of mitochondrial targeting sequences might be. The simulations assumed two predictors with properties similar to that of Predotar (i.e., with sensitivity of 0.86 and specificity of 0.50 [Emanuelsson and von Heijne 2001]). Given the number of "true" mitochondrial proteins, the data from the two predictors were simulated, and the number of proteins predicted to be mitochondrial by both (henceforth referred as the "combined prediction") were analyzed. We were specifically interested in the ratio of false negatives to false positives in the combined prediction. When this ratio is greater than 1, the predictor underestimates the total number mitochondrial proteins; when this ratio is less than 1 the predictor overestimates the number of mitochondrial proteins (one can calculate this ratio to be 0.16 when Predotar alone is used).
The properties of the combined prediction depend on the fraction of "true" mitochondrial proteins. Thus, we simulated genomes where the proportion of mitochondrial proteins has been set at values in the range 2% to 34% (see Materials and Methods). Computer simulations show that a combined prediction drastically reduces the number of false positives relative to each predictor, especially when the fraction of mitochondrial proteins is small. At the same time, the combined prediction roughly halves the number of total positives and also doubles the number of false negatives, independently of the fraction of mitochondrial proteins. The ratio of false negatives to false positives in the combined prediction is shown in figure 4B. For genomes with less than approximately 25% of mitochondrial proteins and the assumed predictors, the combined prediction always underestimates the number of true mitochondrial proteins.
Of the proteins encoded in the E. coli genome, 210 sequences (4.9%) have both a MitoProt probability greater than 0.7 and a Predotar score greater than 0.7 (fig. 4A). Based on the computer simulations, we suggest that this number represents an underestimate of the actual fraction of E. coli proteins predisposed for targeting to mitochondria. When checking through the identities of these proteins many, like YhaR, have functions conserved in mitochondria today, including targeting sequence processing, phospholipid biosynthesis, membrane transporters (for metal ions and small molecules), and ribosomal proteins. Furthermore, like YhaR, the ribosomal proteins L18 and S20 and the phosphoserine decarboxylases, show raggedness in their amino-terminal sequences in multiple sequence alignments. We conclude that at least 5% of proteins from E. coli are predisposed for targeting to mitochondria.
Discussion
The conversion of an intracellular, endosymbiotic bacteria to an organelle depended on transfer of bacterial genes to the host cell nucleus (Andersson and Kurland 1999; Lang, Gray, and Burger 1999; Gray, Burger, and Lang 2001; Cavalier-Smith 2002). The gene products translated in the cytosol then needed to be recognized for translocation into protomitochondria. Although seemingly problematic, it is now clear that physicochemical properties of a sizeable number of bacterial proteins like YhaR, present in diverse phyla of extant bacteria and therefore likely to have been inherent in proteins of the ancestral proteobacter, were available as a preadaptation to be used as the basis to specify mitochondrial targeting.
Plastids, the other organelle derived from endosymbiosis, have a protein translocation machinery built of components that are also found in cyanobacteria. These similarities suggest some form of primitive translocation machinery preexisted in the early cyanobacter-type endosymbiont (Bolter et al. 1998; Reumann and Keegstra 1999). Although the mitochondrial TOM complex does not have obvious homologs in proteobacter species, the central component of the translocation channel is a membrane-embedded ?-barrel protein called Tom40. Membrane-embedded ?-barrels are found only in the outer membrane of bacteria and in outer membranes of mitochondria and plastids. It is widely accepted that Tom40 evolved from a bacterial ?-barrel porin (Mannella, Neuwald, and Lawrence 1996; Bains and Lithgow 1999; Gabriel, Buchanan, and Lithgow. 2001; Hermann 2003), perhaps one that could fortuitously bind proteins displaying the preadaptive characteristics found in YhaR.
Once a simple protein translocation system was established for proteins such as YhaR, the selective pressure would be established for the genetic rearrangement of exon sequences to produce basic, amphipathic peptides to those proteins of the endosymbiont that did not already carry such characteristics. Previous work (Baker and Schatz 1987; Lemire et al. 1989) has shown a significant proportion of DNA fragments generated from random sequences can function as a mitochondrial targeting sequence if cloned into a cassette upstream from the mitochondrial protein CoxIV (engineered to have lost its own mitochondrial targeting sequence). Exon duplication and exon shuffling, as often required to propagate this process, is readily seen in the case of the relatively recent transfer of the rps11 gene from the mitochondrial genome of the plant Oryza sativa. Two copies of rps11 are now present in the nuclear genome of the plant: rps11a, transcribed to incorporate an exon duplicated from the sequence encoding the mitochondrial presequence of the ATPase ?-subunit and rps11b, transcribed with a presequence-encoding exon from the CoxIV gene (Kadowaki et al. 1996).
We suggest that the first steps in the evolution of protein import into protomitochondria involved a very primitive set-up: a ?-barrel protein in the outer membrane of the internalized bacteria and a few substrates predisposed for targeting to mitochondria (fig. 5). Once these few substrates could be acquired from the cytosol of the earliest eukaryotic cells, a scenario would have been set to allow for the inactivation and loss of the first copies of genes from within the bacterial endosymbiont (Kurland and Andersson 2000). With this as impetus, the evolution of receptor subunits for the TOM complex would have been selected for codependently as the number of substrate proteins for import increased and the diversity of their targeting sequences expanded.
FIG. 5. A model for the primitive stages in the evolution of eukaryotic cells. Continued random escape of DNA from the internalized bacterium would allow integration of some early genes into the nucleus and translation of these proteins in the cytosol. If one of the existing ?-barrel "porins" in the outer membrane of the endosymbiont could, even inefficiently, recognize the protein for translocation, the protein would function within the endosymbiont. Continued gene escape would increase the number and diversity of proteins to be recognized by the ?-barrel translocator. Once the genes encoding any one of the essential proteins predisposed for import were deleted from the symbiont's genome, the process of "mitochondrification" would be at a point of no return
Acknowledgements
We thank Jeff Schatz and Sabine Rospert for antibodies, Rosemary Condron for protein sequencing, Geoff McFadden, Lena Burri, Marcio Castro-Filho, Kip Gabriel, and Peter Walsh for comments on the manuscript. This work was supported by a grant from the Australian Research Council.
Literature Cited
Andersson, S.G., and C.G. Kurland. 1999. Origins of mitochondria and hydrogenosomes. Curr. Opin. Microbiol. 2:535-544.
Bains, G., and T. Lithgow. 1999. The Tom channel in the mitochondrial outer membrane: alive and kicking. Bioessays 21:1-4.
Baker, A., and G. Schatz. 1987. Sequences from a prokaryotic genome or the mouse dihydrofolate reductase gene can restore the import of a truncated precursor protein into yeast mitochondria. Proc. Natl. Acad. Sci. USA 84:3117-3121.
Bolter, B., J. Soll, A. Schulz, S. Hinnah, and R. Wagner. 1998. Origin of a chloroplast protein importer. Proc. Natl. Acad. Sci. USA 95:15831-15836.
Brennicke, A. L., L. Grohmann, R. Heisel, V. Knoop, and W. Schuster. 1993. The mitochondrial genome on its way to the nucleus: different stages of gene transfer in higher plants. FEBS Lett. 325:140-145.
Cavalier-Smith, T. 2002. The phagotrophic origin of eukaryotes and phylogenetic classification of Protozoa. Int. J. Syst. Evol. Microbiol. 52:297-354.
Claros M. G., and P. Vincens. 1996. Computational method to predict mitochondrially imported proteins and their targeting sequences. Eur. J. Biochem. 241:770-786.
Claros M. G., et al. 1995. Limitations to in vivo import of hydrophobic proteins into yeast mitochondria: the case of a cytoplasmically synthesized apocytochrome b. Eur. J. Biochem. 228:762-767.
Emanuelsson, O., and G. von Heijne. 2001. Prediction of organellar targeting signals. Biochim. Biophys. Acta 1541:114-119.
Emelyanov, V. V. 2001. Evolutionary relationship of Rickettsiae and mitochondria. FEBS Lett. 501:11-18.
Endo, T., and D. Kohda. 2002. Functions of outer membrane receptors in mitochondrial protein import. Biochim. Biophys. Acta 1592:3-14.
Gabriel, K., S. K. Buchanan, and T. Lithgow. 2001. The alpha and the beta: protein translocation across mitochondrial and plastid outer membranes. Trends Biochem Sci. 26:36-40.
Gabriel, K., B. Egan, and T. Lithgow. 2003. Tom40, the import channel of the mitochondrial outer membrane, plays an active role in sorting imported proteins. EMBO J. 22:2380-2386.
Gray, M. W., G. Burger, and B. F. Lang. 1999. Mitochondrial evolution. Science 283:1476-1481.
Gray, M. W., G. Burger, and B. F. Lang. 2001. The origin and early evolution of mitochondria. Genome Biol. 2:REVIEWS1018.
Hermann, J. M. 2003. Converting bacteria to organelles: evolution of mitochondrial protein sorting. Trends Microbiol. 11:74-79.
Kadowaki, K., N. Kubo, K. Ozawa, and A. Hirai. 1996. Targeting presequence acquisition after mitochondrial gene transfer to the nucleus occurs by duplication of existing targeting signals. EMBO J. 15:6652-6661.
Karlberg, O., B. Canback, C. G. Kurland, and S. G. Andersson. 2000. The dual origin of the yeast mitochondrial proteome. Yeast 17:170-187.
Kim, J. M., H. Yoshikawa, and K. Shirahige. 2001. A member of the YER057c/yjgf/Uk114 family links isoleucine biosynthesis and intact mitochondria maintenance in Saccharomyces cerevisiae. Genes Cells 6:507-517.
Kurland, C. G., and S. G. Andersson. 2000. Origin and evolution of the mitochondrial proteome. Microbiol. Mol. Biol. Rev. 64:786-820.
Lang, B. F., M. W. Gray, and G. Burger. 1999. Mitochondrial genome evolution and the origin of eukaryotes. Annu. Rev. Genet. 33:351-397.
Lemire, B. D., C. Frankhauser, A. Baker, and G. Schatz. 1989. The mitochondrial targeting function of randomly generated peptide sequences correlates with predicted helical amphiphilicity. J. Biol. Chem. 264:20206-20215.
Macasev, D., E. Newbigin, J. Whelan, and T. Lithgow. 2000. How do plant mitochondria avoid importing chloroplast proteins: Tom20 and Tom22 from Arabidopsis thaliana differ from their fungal counterparts. Plant Physiol. 123:1-6.
Mannella, C. A., A. F. Neuwald, and C. E. Lawrence. 1996. Detection of likely transmembrane beta strand regions in sequences of mitochondrial pore proteins using the Gibbs sampler. J. Bioenerg. Biomembr. 28:163-169.
Marchini, A., R. Accardi, and I. Malanchi, et al. (9 co-authors). 2002. Schizosaccharomyces pombe Pmf1p is structurally and functionally related to Mmf1p of Saccharomyces cerevisiae. Yeast 19:703-711.
Martin, W., and M. Müller. 1998. The hydrogen hypothesis for the first eukaryote. Nature 392:37-44.
Neupert, W. 1997. Protein import into mitochondria. Annu. Rev. Biochem. 66:863-917.
Oxelmark, E., A. Marchini, and I. Malanchi, et al. (9 co-authors). 2000. Mmf1p, a novel yeast mitochondrial protein conserved throughout evolution and involved in maintenance of the mitochondrial genome. Mol. Cell. Biol. 20:7784-7797.
Pfanner, N., and A. Chacinska. 2002. The mitochondrial import machinery: preprotein-conducting channels with binding sites for presequences. Biochim. Biophys. Acta. 1592:15-24.
Pfanner, N., M. G. Douglas, and T. Endo, et al. (10 co-authors). 1996. Uniform nomenclature for the protein transport machinery of the mitochondrial membranes. Trends Biochem. 21:51-52.
Reumann, S., and K. Keegstra. 1999. The endosymbiotic origin of the protein import machinery of chloroplastic envelope membranes. Trends Plant Sci. 4:302-307.
Sch?gger, H., and G. von Jagow. 1991. Blue native electrophoresis for isolation of membrane protein complexes in enzymatically active form. Anal. Biochem. 199:223-231.
Schatz, G. 1997. Just follow the acid chain. Nature 388:121-122.
Sinha S., et al. 1999. Crystal structure of Bacillus subtilis YabJ, a purine regulatory protein and member of the highly conserved YjgF family. Proc. Natl. Acad. Sci. USA 96:13074-13079.
Volz, K. 1999. A test case for structure-based functional assignment: the 1.2 ? crystal structure of the yjgF gene product from Escherichia coli. Protein Sci. 8:2428-2437.
von Heijne, G. 1986. Mitochondrial targeting sequences may form amphiphilic helices. EMBO J. 5:1335-1342.
Voos, W., et al. 1999. Mechanisms of protein translocation into mitochondria. Biochim. Biophys. Acta 1422:235-254.(Rebecca Lucattini, Vladim)
E-mail: t.lithgow@unimelb.edu.au.
Abstract
Mitochondria evolved from an endosymbiotic proteobacterium in a process that required the transfer of genes from the bacterium to the host cell nucleus, and the translocation of proteins thereby made in the host cell cytosol into the internal compartments of the organelle. According to current models for this evolution, two highly improbable events are required to occur simultaneously: creation of a protein translocation machinery to import proteins back into the endosymbiont and creation of targeting sequences on the protein substrates themselves. Using a combination of two independent prediction methods, validated through tests on simulated genomes, we show that at least 5% of proteins encoded by an extant proteobacterium are predisposed for targeting to mitochondria, and propose we that mitochondrial targeting information was preexisting for many proteins of the endosymbiont. We analyzed a family of proteins whose members exist both in bacteria and in mitochondria of eukaryotes and show that the amino-terminal extensions occasionally found in bacterial family members can function as a crude import sequence when the protein is presented to isolated mitochondria. This activity leaves the development of a primitive translocation channel in the outer membrane of the endosymbiont as a single hurdle to initiating the evolution of mitochondria.
Key Words: endosymbiont ? mitochondria ? targeting sequence ? protein import
Introduction
A defining moment in the evolution of eukaryotic cells was the acquisition of mitochondria, and this appears to have occurred only once in the course of evolution (Martin and Müller 1998; Gray, Burger, and Lang 1999, 2001; Lang, Gray, and Burger 1999; Kurland and Andersson 2000; Emelyanov 2001). Mitochondria were derived from an endosymbiotic relationship with internalized proteobacteria, via a progressive transfer of genetic material. Copies of genes from the symbiont found their way to the host cell nucleus to be expressed as proteins that might then be imported back into the symbiont (Andersson and Kurland 1999; Lang, Gray, and Burger 1999; Cavalier-Smith 2002). Surrender of the bacterial genome provided the means for the eukaryote to control the metabolism of the endosymbiont. The relocation of the first gene from the endosymbiont to the nucleus and its deletion from the symbiont genome is a crucial step, after which the symbiont loses its status as an independent organism. This crucial step also creates an extreme topological problem: how to get proteins, now made on the host cell ribosomes, back into the appropriate compartment of the endosymbiont.
In modern eukaryotes the solution to this problem, protein targeting into mitochondria, is an elaborate and now well-understood process. Each of the several hundred proteins carry discrete targeting sequences to be selected for import into the organelle and are recognized for import by a complex protein import machinery (Neupert 1997; Voos et al. 1999). Two highly improbable events, creation of mitochondrial targeting sequences and evolution of a protein import machinery in the outer membrane of the endosymbiont were required to initiate the conversion of the -proteobacter endosymbiont to mitochondria.
Some components of the import machinery we see now in mitochondria are derived from the protein secretion apparatus of bacteria. Peptidases and elements of the translocation complexes in the mitochondrial inner membrane were probably already present in the endosymbiont, having homologs in present-day bacteria (Hermann 2003). However, the creation and evolution of the translocase in the outer mitochondrial membrane (TOM complex) remains something we understand almost nothing about. It seems likely that the genes encoding subunits of the TOM complex, like a large number of other mitochondrial proteins, were derived from the host cell's genome (Karlberg et al. 2000). The TOM complex consists of receptors that bind mitochondrial targeting sequences (Tom20 and Tom22 [Endo and Kohda 2002]), a translocation channel (Tom40) and attendant subunits (Tom6 and Tom7), and three or more other subunits that seem more species specific (Neupert 1997; Schatz 1997; Voos et al. 1999; Gabriel, Buchanan, and Lithgow 2001; Pfanner and Chacinska 2002). (The proteins are named according to their size in kilodaltons; for example, Tom20 is the 20-kDa subunit of the TOM complex [Pfanner et al. 1996]). The need for a codependent evolution of the TOM complex and the targeting sequences for each of the imported protein substrates is a major weakness in the current model explaining the evolution of mitochondria.
Comparative genomics makes a strong argument for the later stages of the evolution of mitochondria (Brennicke et al. 1993; Kurland and Andersson 2000; Gray et al. 2001): (1) escape of the DNA (or an RNA copy) of the gene from the endosymbiont, (2) transfer of the gene fragment to the nucleus and integration in the genome, (3) adaptive rearrangements to allow gene expression, (4) exon shuffling, and other means to create a mitochondrial targeting sequence, (5) transcription, then translation, of the mitochondrial protein in the cytosol, and (6) gene inactivation of the redundant copy(s) of the gene in the mitochondria (Kurland and Andersson 2000; Gray et al. 2001). However, in its current form, this model assumes the preexistence of a TOM complex. Before the TOM complex, no pressure existed to select for the sequences on a protein substrate that could promote targeting back to the early endosymbiont.
Here we show, using two independent predictive strategies, that many proteins present in extant prokaryotes carry rudimentary features for mitochondrial targeting. Application of two predictors, Mitoprot II and Predotar, and computer simulations of feasible scenarios suggest that up to 5% of proteins encoded by the extant proteobacterium Escherichia coli have features that would serve as mitochondrial targeting information, with basic and amphipathic extensions at their amino-termini and "meso-hydrophobic" character throughout the proteins. One of the bacterial proteins, YhaR, serves as an instructional case in point. It is a member of a large family of highly conserved proteins found in at least eight diverse phyla of present-day bacteria. Multiple sequence alignments of the bacterial versions of these proteins show ragged amino-termini, with YhaR from E. coli having one of the longer amino-terminal extensions. Ectopic expression of bacterial YhaR results in targeting of the protein to yeast mitochondria, suggesting that in some cases, during the course of evolution, this preadaptation meant little or no mutagenesis of upstream regions in bacterial genes to render the proteins they encode competent for import into "protomitochondria." The presence of this preexisting "targeting" information would have provided selective pressure to initiate evolution of the TOM complex.
Materials and Methods
Comparative Sequence Analysis
Iterative Blast analyses were undertaken as previously described (Macasev et al. 2000). Multiple sequence alignments were constructed with ClustalW on the European Bioinformatics Institute server (http://www.ebi.ac.uk/clustalw/).
Prediction of Mitochondrial Targeting Peptides
MitProtII was obtained from ftp://ftp.ens.fr/pub/molbio, and Predotar was obtained from http://www.inra.fr/predotar/. The precompiled Solaris executable of MitoProtII was used in calculations. For the processing of multiple sequences and analysis of results, a set of wrapper programs was developed in Perl.
Computer Simulations of Mitochondrial Predictors
For computer simulations, we assumed two predictors with the characteristics of Predotar (i.e., sensitivity = 0.86 and specificity = 0.50 [Emanuelsson and von Heijne 2001]) whose errors are uncorrelated. This is expected to be a reasonable assumption because Mitoprot II and Predotar use an entirely different approach for the prediction of mitochondrial targeting features. Simulations were carried out on genomes containing a total of 4,096 proteins, with the number of mitochondrial proteins ranging from 2% to 34%.
Given the set of assumed mitochondrial proteins, the predictions of two predictors were simulated, and the intersection of positive predictions was calculated. Given the percentage of mitochondrial proteins, the calculation was repeated 20 times, and for each run, the ratio of false negatives to false positives was evaluated. From these data, the mean and standard deviation of the ratio of false negatives to false positives was calculated. The mean, thus, represents our best estimate of this ratio, and the standard deviation gives an estimate of the variability between repeated simulations. Optimization experiments suggest the data are insensitive to small changes in the neighborhood of chosen properties of predictors.
Yeast Strains and Mitochondrial Isolation
A URA3 gene cassette was used to disrupt the coding sequence in the HMF1 gene after the codon corresponding to Gly34. The hmf1::URA3 fragment was transformed into W303a/ yeast cells to generate the strain YRG27 (Mata/, leu2/leu2, his3/his3, ade2/ade2, ura3/ura3, trp1/trp1, can1/can1, and HMF1/hmf1::URA3). YRG27 cells were sporulated and tetrads dissected to obtain haploid null mutants (Mata, leu2, his3, ade2, ura3, trp1, can1, and hmf1::URA3). Yeast mitochondria were prepared as described in Gabriel, Egan, and Lithgow (2003). Where indicated, mitochondria were exposed to osmotic shock conditions (10mM HEPES pH 7.4) or to 0.1% Triton X-100.
Antibody Production
The open reading frame encoding Mmf1 was amplified by PCR using purified yeast genomic DNA as a template and cloned into pQE30 (QIAGEN) behind a 6x Histidine coding sequence. Hexahistidine-tagged Mmf1 was expressed in Escherichia coli strain M15 (Qiagen) by induction with 0.5 mM IPTG. Denatured protein was purified using Ni-NTA resin (Qiagen) according to manufacturer's directions and used to immunize rabbits for the production of antibodies.
Mitochondrial Protein Import Assays
In vitro translation of 35S-labeled YhaR in rabbit reticulocyte lysates and import into isolated mitochondria was as previously described (Gabriel, Egan, and Lithgow 2003).
Results
The initial impetus for this study came from comparative sequence analyses of the YER057c/yjgF/UK114 family of highly conserved acid-soluble proteins (Volz 1999). With at least 50 members of the family now known, multiple sequence alignments show a striking level of sequence conservation across species from prokaryotic and eukaryotic organisms (data not shown). Just as striking was the raggedness of the amino-terminal sequence from many family members in multiple sequence alignments informed from crystal structures of two proteins (fig. 1); YjgF was crystallized from E. coli (Volz 1999) and YabJ was crystallized from Bacillus subtilis (Sinha et al. 1999), showing a compact single domain, with the amino-terminal segment of YabJ forming a four residue strand ("?1") with a loop that folds back onto the first ?-strand element of the core ("?2"). Extensions were predicted to be present in the proteins from the cyanobacterium Nostoc, the gram-positive bacterium Enterococcus faecalis, and YhaR from the proteobacterium E. coli. Among the diverse prokaryotes we surveyed, the longest extension was found in the -proteobacterium Rhodospirillum rubrum. Proteins from the eukaryotes Saccharomyces cerevisiae, Schizosaccharomyces pombe, Magnaporthe grisea, Pleurotus ostreatus, and Ustilago maydis also have amino-terminal extensions (fig. 1).
FIG. 1. Ragged amino-termini in the YER057c/yjgF/UK114 family of proteins. Iterative Blast analyses were used to compile sequences of family members. All are proteins of approximately 150 residues; the number of the residues shown is indicated. For the eukaryotes, only the fungal sequences are included (homologs were also found present in the genomes of animals and plants) aligned against representative proteins from eight diverse phyla of Bacteria. Where the proteins have been studied, names are shown in parentheses. From the structure of YjgF and YabJ, residues corresponding to those of the first ?-strand (?1) and the start of the second ?-strand (?2) are shaded gray. The experimentally verified presequence of Mmf1 is shown underlined in bold type. It is unknown whether the extensions on the bacterial proteins are functionally important. In many cases, a second AUG codon is found internally, in some cases having a good Shine-Dalgarno context for translation initiation (Lithgow, unpublished observations)
Extensions on the Eukaryotic Proteins Represent Mitochondrial Targeting Sequences
Sequence analysis of all the YER057c/yjgF/UK114 family members from fungi suggested the amino-terminal extensions represent mitochondrial targeting sequences. MitoProt II, which directly calculates a probability for mitochondrial targeting based on various physicochemical properties of mitochondrial proteins, including a "mesohydrophobicity" score that correlates well with import competence of known mitochondrial proteins (Claros et al. 1995; Claros and Vincens 1996), predicts the extended proteins as highly probable mitochondrial proteins (M. grisea, P = 0.939; S. pombe, P = 0.998; P. ostreatus, P = 0.941; U. maydis; P = 0.9844, and S. cerevisiae, P = 0.994), whereas the proteins without extensions predict strongly as nonmitochondrial proteins. The mitochondrial targeting sequences of hundreds of proteins are now known and prove to be discrete segments rich in positively charged residues and with tendencies to form two to three turns of a helix with amphipathic character (von Heijne 1986). Predotar, a neural network predictor (http://www.inra.fr/predotar/) also predicts the extensions of the fungal proteins in figure 1 as mitochondrial targeting sequences.
Pmf1 is now known to be mitochondrial in S. pombe, and epitope-tagged Mmf1 has been located in mitochondria of S. cerevisiae (Oxelmark et al. 2000; Kim, Yoshikawa, and Shirahige 2001; Marchini et al. 2002). An antibody raised to purified, recombinant protein found Mmf1 exclusively in the mitochondria (data not shown) and further localized Mmf1 within the mitochondrial matrix (fig. 2), demonstrating Mmf1 is not simply attached to the mitochondrial surface but has been imported across the mitochondrial membranes. The shorter protein from S. cerevisiae, Hmf1, is found only in the cytosol (data not shown). To verify that the amino-terminal extension of Mmf1 is a cleavable mitochondrial targeting sequence, a preparation of matrix proteins was isolated from yeast mitochondria, separated by Tris-tricine SDS-PAGE (Sch?gger and von Jagow 1991) and blotted to PVDF membrane for amino-terminal sequencing. Mmf1 is processed between amino acids Gly17 and Ile18, truncating the mature protein close to the point conforming to the size of the shorter family members (see figure 1).
FIG. 2. Mmf1 is processed to be a soluble protein in the mitochondrial matrix. (A) Mitochondria (100 μg protein) isolated from wild-type yeast cells and treated with 10 μg proteinase K either in iso-osmotic buffer to protect the integrity of the mitochondrial membranes (lane 1), in hypo-osmotic buffer (10 mM HEPES pH 7.4) to rupture the outer membrane and allow proteinase K access to proteins such as cytochrome b2 (Cyb2) in the intermembrane space (lane 2), or, in the presence of 0.1% Triton X-100 (lane 3), to solubilize both the outer and inner membrane and allow proteinase K access to proteins in the matrix such as -ketoglutarate dehydrogenase (Kgd1). After separation on SDS-PAGE, immunoblots were analyzed with antisera raised to the purified proteins. (B) Mitochondria (100 μg protein) were extracted with cold 100 mM Na2CO3 (pH 10.0), incubated on ice for 30 minutes, centrifuged (30 min) at 75,000 rpm in a TL100.2 rotor (Beckman) to yield a membrane pellet (P) that includes the integral membrane protein Por1 and a soluble extract (S) containing the matrix-located chaperonin Cpn10 and Cyb2 from the intermembrane space
Some Bacterial Family Members, Including Yhar, Are Predisposed for Targeting to Mitochondria
Several of the bacterial family members also have putative amino-terminal extensions. Might the bacterial proteins fortuitously carry sequences suitable for mitochondrial targeting? The open-reading frame corresponding to YhaR from E. coli was cloned into a vector for in vitro transcription and translation, and the [35S]methionine-labeled protein incubated together with mitochondria isolated from S. cerevisiae. The 17-kDa protein associates with mitochondria (fig. 3, lane 1) and can interact with the TOM complex to initiate translocation across the outer membrane. If proteinase K is added to mitochondria, the protein is degraded to a fragment of approximately 9 kDa (lane 2). The protein fragment is within the intermembrane space because rupture of the outer membrane by osmotic shock renders the fragment accessible to proteinase K (lane 3). Quantitative recovery of the [35S]methionine-labeled fragment suggests that it represents the amino-terminal two-thirds of YhaR, given the placement of methionine residues (at positions 1, 6, 22, 47, and 95) in the 150 residue protein.
FIG. 3. The bacterial protein YhaR is partially translocated into isolated mitochondria. Mitochondria (625μg protein/ml) in import buffer (0.6 M sorbitol, 50 mM HEPES pH 7.4, 2mM potassium phosphate pH 7.4, 25 mM KCl, 10 mM MgCl2, 0.5 mM EDTA, 25 mg/ml BSA, 1 mM DTT, 1 mM ATP, and 5 mM NADH) were aliquoted (100 μl), incubated with 10 μl translation mix containing [35S]methionine-labeled YhaR and incubated for 20 minutes at 25°C. Mitochondria were then collected by centrifugation and resuspended in import buffer or hypo-osmotic buffer (10 mM HEPES pH 7.4, "osmotic shock") or in import buffer containing 0.1% Triton X-100 to solubilize the inner membrane. Where indicated (+), 10 μg Proteinase K was added for 15 minutes on ice before mitochondria were reisolated and prepared for SDS-PAGE and fluorography
To determine whether the mitochondrial targeting propensity of YhaR is unique, we scanned a conceptual translation of the E. coli genome (4,289 protein sequences) with the programs MitoProt (Claros and Vincens 1996) and Predotar. A total of 487 E. coli proteins (11.4% of the total conceptual proteome) are predicted with greater than 0.7 probability to be mitochondrial MitoProtII. The program Predotar reports "mitochondrial scores," and the Predotar server (http://www.inra.fr/predotar/) predicts a protein as being mitochondrial if the score is greater than 0.5. To increase the stringency of predictions, we have used a Predotar score of greater than 0.7 as a high-confidence prediction of mitochondrial location. Of the 4,289 E. coli proteins, 613 (14.3 %) satisfy this criterion. Figure 4A shows the plot of Predotar scores versus MitoProtII probabilities for all protein sequences derived from the E. coli genome.
FIG. 4. Mitochondrial targeting predictions. (A) A conceptual translation of the E. coli genome yields 4,289 protein sequences, each of which was analyzed by Predotar and by MitoProt II. Each cross represents the values of the Predotar mitochondrial score versus the MitoProt II probability for a single protein. The squared area encloses 210 proteins predicted by both MitoProt II and Predotar to carry mitochondrial targeting information. The identities of these 210 sequences and the predictor scores are shown in the Supplementary Material online. (B) Computer simulations were used to calculate the ratio of false negatives to false positives in combined predictions from two independent predictors with properties equivalent to Predotar, depending on the percentage of mitochondrial proteins encoded in a genome. For each data point, the mean (diamond) and the standard deviation (error bar) of this ratio was calculated from 20 simulations. The dashed line shows the ratio of 1:1, for which the prediction matches the true number mitochondrial proteins. For the assumed properties of two predictors, the ratio of false negatives to false positives in the combined prediction was greater than 1 when the fraction of mitochondrial proteins was less than 25%, implying that the combined prediction underestimates the actual number of proteins that carry mitochondrial targeting sequences
Reliability of the Mitochondrial Targeting Prediction Methods
Despite considerable success in predicting mitochondrial target sequences, current prediction methods are imperfect. In a previous study, a set of 2,738 nonplant proteins of known subcellular location was compiled and screened with Predotar: 76.3% were correctly assigned (Emanuelsson and von Heijne 2001). Predotar prediction contained a large number of false positives (50% of the total positive assignments), which greatly increases the total number of sequences estimated to carry mitochondrial target signal. False negatives offset this to a certain extent, but in any calculation where false positives greatly exceed false negatives, the final estimate is unrealistically high. Our goal was to estimate the total number of proteins from the E. coli genome predisposed for targeting to mitochondria. Considering the sensitivity of current predictors (Emanuelsson and von Heijne 2001), the predictions of MitoProtII and Predotar (11.4% and 17.9%) cannot be taken as reliable.
The two programs rely on different methods for the prediction of mitochondrial targeting peptides, and, in the first approximation, one could assume that errors in the predictions made by MitoProtII and Predotar are uncorrelated. We investigated this scenario by computer simulations to determine how reliable a combined prediction of mitochondrial targeting sequences might be. The simulations assumed two predictors with properties similar to that of Predotar (i.e., with sensitivity of 0.86 and specificity of 0.50 [Emanuelsson and von Heijne 2001]). Given the number of "true" mitochondrial proteins, the data from the two predictors were simulated, and the number of proteins predicted to be mitochondrial by both (henceforth referred as the "combined prediction") were analyzed. We were specifically interested in the ratio of false negatives to false positives in the combined prediction. When this ratio is greater than 1, the predictor underestimates the total number mitochondrial proteins; when this ratio is less than 1 the predictor overestimates the number of mitochondrial proteins (one can calculate this ratio to be 0.16 when Predotar alone is used).
The properties of the combined prediction depend on the fraction of "true" mitochondrial proteins. Thus, we simulated genomes where the proportion of mitochondrial proteins has been set at values in the range 2% to 34% (see Materials and Methods). Computer simulations show that a combined prediction drastically reduces the number of false positives relative to each predictor, especially when the fraction of mitochondrial proteins is small. At the same time, the combined prediction roughly halves the number of total positives and also doubles the number of false negatives, independently of the fraction of mitochondrial proteins. The ratio of false negatives to false positives in the combined prediction is shown in figure 4B. For genomes with less than approximately 25% of mitochondrial proteins and the assumed predictors, the combined prediction always underestimates the number of true mitochondrial proteins.
Of the proteins encoded in the E. coli genome, 210 sequences (4.9%) have both a MitoProt probability greater than 0.7 and a Predotar score greater than 0.7 (fig. 4A). Based on the computer simulations, we suggest that this number represents an underestimate of the actual fraction of E. coli proteins predisposed for targeting to mitochondria. When checking through the identities of these proteins many, like YhaR, have functions conserved in mitochondria today, including targeting sequence processing, phospholipid biosynthesis, membrane transporters (for metal ions and small molecules), and ribosomal proteins. Furthermore, like YhaR, the ribosomal proteins L18 and S20 and the phosphoserine decarboxylases, show raggedness in their amino-terminal sequences in multiple sequence alignments. We conclude that at least 5% of proteins from E. coli are predisposed for targeting to mitochondria.
Discussion
The conversion of an intracellular, endosymbiotic bacteria to an organelle depended on transfer of bacterial genes to the host cell nucleus (Andersson and Kurland 1999; Lang, Gray, and Burger 1999; Gray, Burger, and Lang 2001; Cavalier-Smith 2002). The gene products translated in the cytosol then needed to be recognized for translocation into protomitochondria. Although seemingly problematic, it is now clear that physicochemical properties of a sizeable number of bacterial proteins like YhaR, present in diverse phyla of extant bacteria and therefore likely to have been inherent in proteins of the ancestral proteobacter, were available as a preadaptation to be used as the basis to specify mitochondrial targeting.
Plastids, the other organelle derived from endosymbiosis, have a protein translocation machinery built of components that are also found in cyanobacteria. These similarities suggest some form of primitive translocation machinery preexisted in the early cyanobacter-type endosymbiont (Bolter et al. 1998; Reumann and Keegstra 1999). Although the mitochondrial TOM complex does not have obvious homologs in proteobacter species, the central component of the translocation channel is a membrane-embedded ?-barrel protein called Tom40. Membrane-embedded ?-barrels are found only in the outer membrane of bacteria and in outer membranes of mitochondria and plastids. It is widely accepted that Tom40 evolved from a bacterial ?-barrel porin (Mannella, Neuwald, and Lawrence 1996; Bains and Lithgow 1999; Gabriel, Buchanan, and Lithgow. 2001; Hermann 2003), perhaps one that could fortuitously bind proteins displaying the preadaptive characteristics found in YhaR.
Once a simple protein translocation system was established for proteins such as YhaR, the selective pressure would be established for the genetic rearrangement of exon sequences to produce basic, amphipathic peptides to those proteins of the endosymbiont that did not already carry such characteristics. Previous work (Baker and Schatz 1987; Lemire et al. 1989) has shown a significant proportion of DNA fragments generated from random sequences can function as a mitochondrial targeting sequence if cloned into a cassette upstream from the mitochondrial protein CoxIV (engineered to have lost its own mitochondrial targeting sequence). Exon duplication and exon shuffling, as often required to propagate this process, is readily seen in the case of the relatively recent transfer of the rps11 gene from the mitochondrial genome of the plant Oryza sativa. Two copies of rps11 are now present in the nuclear genome of the plant: rps11a, transcribed to incorporate an exon duplicated from the sequence encoding the mitochondrial presequence of the ATPase ?-subunit and rps11b, transcribed with a presequence-encoding exon from the CoxIV gene (Kadowaki et al. 1996).
We suggest that the first steps in the evolution of protein import into protomitochondria involved a very primitive set-up: a ?-barrel protein in the outer membrane of the internalized bacteria and a few substrates predisposed for targeting to mitochondria (fig. 5). Once these few substrates could be acquired from the cytosol of the earliest eukaryotic cells, a scenario would have been set to allow for the inactivation and loss of the first copies of genes from within the bacterial endosymbiont (Kurland and Andersson 2000). With this as impetus, the evolution of receptor subunits for the TOM complex would have been selected for codependently as the number of substrate proteins for import increased and the diversity of their targeting sequences expanded.
FIG. 5. A model for the primitive stages in the evolution of eukaryotic cells. Continued random escape of DNA from the internalized bacterium would allow integration of some early genes into the nucleus and translation of these proteins in the cytosol. If one of the existing ?-barrel "porins" in the outer membrane of the endosymbiont could, even inefficiently, recognize the protein for translocation, the protein would function within the endosymbiont. Continued gene escape would increase the number and diversity of proteins to be recognized by the ?-barrel translocator. Once the genes encoding any one of the essential proteins predisposed for import were deleted from the symbiont's genome, the process of "mitochondrification" would be at a point of no return
Acknowledgements
We thank Jeff Schatz and Sabine Rospert for antibodies, Rosemary Condron for protein sequencing, Geoff McFadden, Lena Burri, Marcio Castro-Filho, Kip Gabriel, and Peter Walsh for comments on the manuscript. This work was supported by a grant from the Australian Research Council.
Literature Cited
Andersson, S.G., and C.G. Kurland. 1999. Origins of mitochondria and hydrogenosomes. Curr. Opin. Microbiol. 2:535-544.
Bains, G., and T. Lithgow. 1999. The Tom channel in the mitochondrial outer membrane: alive and kicking. Bioessays 21:1-4.
Baker, A., and G. Schatz. 1987. Sequences from a prokaryotic genome or the mouse dihydrofolate reductase gene can restore the import of a truncated precursor protein into yeast mitochondria. Proc. Natl. Acad. Sci. USA 84:3117-3121.
Bolter, B., J. Soll, A. Schulz, S. Hinnah, and R. Wagner. 1998. Origin of a chloroplast protein importer. Proc. Natl. Acad. Sci. USA 95:15831-15836.
Brennicke, A. L., L. Grohmann, R. Heisel, V. Knoop, and W. Schuster. 1993. The mitochondrial genome on its way to the nucleus: different stages of gene transfer in higher plants. FEBS Lett. 325:140-145.
Cavalier-Smith, T. 2002. The phagotrophic origin of eukaryotes and phylogenetic classification of Protozoa. Int. J. Syst. Evol. Microbiol. 52:297-354.
Claros M. G., and P. Vincens. 1996. Computational method to predict mitochondrially imported proteins and their targeting sequences. Eur. J. Biochem. 241:770-786.
Claros M. G., et al. 1995. Limitations to in vivo import of hydrophobic proteins into yeast mitochondria: the case of a cytoplasmically synthesized apocytochrome b. Eur. J. Biochem. 228:762-767.
Emanuelsson, O., and G. von Heijne. 2001. Prediction of organellar targeting signals. Biochim. Biophys. Acta 1541:114-119.
Emelyanov, V. V. 2001. Evolutionary relationship of Rickettsiae and mitochondria. FEBS Lett. 501:11-18.
Endo, T., and D. Kohda. 2002. Functions of outer membrane receptors in mitochondrial protein import. Biochim. Biophys. Acta 1592:3-14.
Gabriel, K., S. K. Buchanan, and T. Lithgow. 2001. The alpha and the beta: protein translocation across mitochondrial and plastid outer membranes. Trends Biochem Sci. 26:36-40.
Gabriel, K., B. Egan, and T. Lithgow. 2003. Tom40, the import channel of the mitochondrial outer membrane, plays an active role in sorting imported proteins. EMBO J. 22:2380-2386.
Gray, M. W., G. Burger, and B. F. Lang. 1999. Mitochondrial evolution. Science 283:1476-1481.
Gray, M. W., G. Burger, and B. F. Lang. 2001. The origin and early evolution of mitochondria. Genome Biol. 2:REVIEWS1018.
Hermann, J. M. 2003. Converting bacteria to organelles: evolution of mitochondrial protein sorting. Trends Microbiol. 11:74-79.
Kadowaki, K., N. Kubo, K. Ozawa, and A. Hirai. 1996. Targeting presequence acquisition after mitochondrial gene transfer to the nucleus occurs by duplication of existing targeting signals. EMBO J. 15:6652-6661.
Karlberg, O., B. Canback, C. G. Kurland, and S. G. Andersson. 2000. The dual origin of the yeast mitochondrial proteome. Yeast 17:170-187.
Kim, J. M., H. Yoshikawa, and K. Shirahige. 2001. A member of the YER057c/yjgf/Uk114 family links isoleucine biosynthesis and intact mitochondria maintenance in Saccharomyces cerevisiae. Genes Cells 6:507-517.
Kurland, C. G., and S. G. Andersson. 2000. Origin and evolution of the mitochondrial proteome. Microbiol. Mol. Biol. Rev. 64:786-820.
Lang, B. F., M. W. Gray, and G. Burger. 1999. Mitochondrial genome evolution and the origin of eukaryotes. Annu. Rev. Genet. 33:351-397.
Lemire, B. D., C. Frankhauser, A. Baker, and G. Schatz. 1989. The mitochondrial targeting function of randomly generated peptide sequences correlates with predicted helical amphiphilicity. J. Biol. Chem. 264:20206-20215.
Macasev, D., E. Newbigin, J. Whelan, and T. Lithgow. 2000. How do plant mitochondria avoid importing chloroplast proteins: Tom20 and Tom22 from Arabidopsis thaliana differ from their fungal counterparts. Plant Physiol. 123:1-6.
Mannella, C. A., A. F. Neuwald, and C. E. Lawrence. 1996. Detection of likely transmembrane beta strand regions in sequences of mitochondrial pore proteins using the Gibbs sampler. J. Bioenerg. Biomembr. 28:163-169.
Marchini, A., R. Accardi, and I. Malanchi, et al. (9 co-authors). 2002. Schizosaccharomyces pombe Pmf1p is structurally and functionally related to Mmf1p of Saccharomyces cerevisiae. Yeast 19:703-711.
Martin, W., and M. Müller. 1998. The hydrogen hypothesis for the first eukaryote. Nature 392:37-44.
Neupert, W. 1997. Protein import into mitochondria. Annu. Rev. Biochem. 66:863-917.
Oxelmark, E., A. Marchini, and I. Malanchi, et al. (9 co-authors). 2000. Mmf1p, a novel yeast mitochondrial protein conserved throughout evolution and involved in maintenance of the mitochondrial genome. Mol. Cell. Biol. 20:7784-7797.
Pfanner, N., and A. Chacinska. 2002. The mitochondrial import machinery: preprotein-conducting channels with binding sites for presequences. Biochim. Biophys. Acta. 1592:15-24.
Pfanner, N., M. G. Douglas, and T. Endo, et al. (10 co-authors). 1996. Uniform nomenclature for the protein transport machinery of the mitochondrial membranes. Trends Biochem. 21:51-52.
Reumann, S., and K. Keegstra. 1999. The endosymbiotic origin of the protein import machinery of chloroplastic envelope membranes. Trends Plant Sci. 4:302-307.
Sch?gger, H., and G. von Jagow. 1991. Blue native electrophoresis for isolation of membrane protein complexes in enzymatically active form. Anal. Biochem. 199:223-231.
Schatz, G. 1997. Just follow the acid chain. Nature 388:121-122.
Sinha S., et al. 1999. Crystal structure of Bacillus subtilis YabJ, a purine regulatory protein and member of the highly conserved YjgF family. Proc. Natl. Acad. Sci. USA 96:13074-13079.
Volz, K. 1999. A test case for structure-based functional assignment: the 1.2 ? crystal structure of the yjgF gene product from Escherichia coli. Protein Sci. 8:2428-2437.
von Heijne, G. 1986. Mitochondrial targeting sequences may form amphiphilic helices. EMBO J. 5:1335-1342.
Voos, W., et al. 1999. Mechanisms of protein translocation into mitochondria. Biochim. Biophys. Acta 1422:235-254.(Rebecca Lucattini, Vladim)