Plant cells harbor two types of endosymbiotic organelle: mitochondria and chloroplasts. As a consequence of endosymbiotic gene transfer, the majority of their proteins are encoded in the nucleus and post-translationally ‘re’-imported into the respective target organelle. The corresponding transport signals are usually selective for a single organelle, but several proteins are transported into both the mitochondria and chloroplasts.
To estimate the number of proteins with such dual targeting properties in Arabidopsis, we classified the proteins encoded by nuclear genes of endosymbiotic origin according to the respective targeting specificity of their N-terminal transport signals as predicted by the TargetP software package. Selected examples of the resulting protein classes were subsequently analyzed by transient transformation assays as well as by in organello protein transport experiments.
It was found that most proteins with high prediction values for both organelles show dual targeting with both experimental approaches. Unexpectedly, however, dual targeting was even found among those proteins that are predicted to be localized solely in one of the two endosymbiotic organelles. In total, among the 16 candidate proteins analyzed, we identified 10 proteins with dual targeting properties.
This unexpectedly high proportion suggests that such transport properties are much more abundant than anticipated.
Mitochondria and chloroplasts are the result of two consecutive endosymbiotic events in which two prokaryotic organisms, namely an α-proteobacterium and a cyanobacterium, are successively taken up by a host cell and incorporated as organelles (Giovannoni et al., 1988; Douglas, 1998; Gray, 1999). The single processing steps in the evolution of the eukaryotic cell have not been elucidated and are a subject of ongoing debate (SET, serial endosymbiont theory; symbiogenesis hypothesis; PhaT, phagocytosing archaeon theory) (Embley & Martin, 2006; Lang & Burger, 2012; Martijn & Ettema, 2013). However, the establishment of organelles is accompanied by a massive transfer of genetic information, predominantly from the endosymbiont to the host nucleus (Martin & Herrmann, 1998; Bock & Timmis, 2008). As a consequence, many proteins encoded by such genes must be transported ‘back’ into the respective organelle after synthesis. For this purpose, they are synthesized in the cytosol as precursor polypeptides carrying N-terminal extensions called presequences or transit peptides that comprise the entire information for organelle targeting and import. In most instances, such transport is monospecific, i.e., a given protein is targeted specifically into either mitochondria or chloroplasts. However, in recent years, a number of proteins have been identified that exhibit dual targeting properties, i.e., they are imported into both endosymbiotic organelles. In some cases, such dual targeting results from transit peptides comprising two independent transport signals in tandem. As a result of differential transcription, splicing and/or translation processes, either of the two signals can be exposed at the N-terminus of the precursor protein, where it determines the target organelle. In other instances, however, dual targeting is mediated by ambiguous transit peptides, which are able to interact with the protein transport machineries of both endosymbiotic organelles (summarized in Peeters & Small, 2001; Karniely & Pines, 2005).
Many features of such ambiguous transit peptides are similar to those of the corresponding monospecific transport signals, such as low content of negatively charged residues, over-representation of hydroxylated residues and an overall positive charge (von Heijne et al., 1989). It has been suggested that dual targeting transit peptides are intermediate in character between mitochondrial and chloroplastic transit peptides (Peeters & Small, 2001), although with a slight preference for hydrophobic residues (Berglund et al., 2009). In line with this, most established software packages predicting the subcellular localization of proteins, including PSORT (Nakai & Kanehisa, 1992), TargetP (Emanuelsson et al., 2000) and Predotar (Small et al., 2004), do not take dual targeting into both mitochondria and chloroplasts into account. Only the recently developed ‘ambiguous targeting predictor’ (ATP) program (Mitschke et al., 2009) aims at specifically predicting proteins targeted to both mitochondria and chloroplasts. Consequently, among the c. 100 proteins from different organisms carrying such ambiguous transit peptides that have been described in the literature (e.g. Silva-Filho, 2003; Berglund et al., 2009; Carrie & Small, 2013), most were identified by accident (e.g. Creissen et al., 1995; Rödiger et al., 2011).
In this study, we have attempted to estimate the number of proteins with dual targeting properties by a combination of extended in silico analyses and protein transport experiments. We screened the nuclear genome of Arabidopsis thaliana for genes of endosymbiotic origin encoding proteins with transit peptides, which were then classified into different groups according to the difference in their targeting prediction values for either mitochondria or chloroplasts. Several candidates from these groups were then analyzed with respect to their organelle transport specificity using both in vivo and in organello protein transport experiments.
Materials and Methods
Selection of candidate proteins and prediction of their subcellular localization
The databases of the non-redundant protein-encoding genes from A. thaliana, Synechocystis sp. PCC 6803 (Yang & McFadden, 1994; Kaneko et al., 1996, 2003), Nostoc sp. PCC 7120 (Kaneko et al., 2001), Thermosynechococcus elongatus BP-1 (Nakamura et al., 2002), Rickettsia prowazekii strain Madrid E (Andersson et al., 1998), Caulobacter crescentus strain CB15 (Nierman et al., 2001) and Rhizobium meliloti strain 1021 (Barnett et al., 2001; Capela et al., 2001; Finan et al., 2001; Galibert et al., 2001) were taken from the European Bioinformatics Institute server (http://www.ebi.ac.uk/integr8). Protein sequences deduced from the A. thaliana genome were aligned with those of the selected representatives of both cyanobacteria and α-proteobacteria using the BlastP algorithm (Altschul et al., 1990, 1997) (algorithm parameter: expect value ≤ 10−10). Approximately 2000 proteins passed both alignments and were further analyzed concerning their putative subcellular localization. The prediction values for the likelihood of organellar transport into mitochondria (M*) and chloroplasts (C*) of the candidate proteins were obtained using the software package TargetP V1.0 (Emanuelsson et al., 2000) in the ‘plant’ mode without cutoff restrictions. Protein sequences passing the arbitrary threshold level M* + C* ≥ 0.8 (c. 700 in total) were ranked into classes according to the difference in the TargetP prediction values for the likelihood of transport into mitochondria or chloroplasts (ΔM*C*). From several of these classes, candidate proteins were selected in a random manner for further analysis.
Full-size cDNA clones encoding the authentic candidate proteins from A. thaliana were obtained from the Arabidopsis Biological Resource Center (Columbus, OH, USA) and the Institut National de la Recherche Agronomique (Paris, France). The reporter constructions encoding the fusion proteins were generated by PCR amplification, as described in Rödiger et al. (2011). All chimeric constructions as well as the authentic cDNAs were cloned into the two plasmid vectors pBAT (Annweiler et al., 1991) and pRT100 Ω/Not/Asc (Ueberlacker & Werr, 1996), which are suitable for in vitro transcription and transient gene expression in plants, respectively.
Protein transport experiments
In organello experiments were performed as described in Rödiger et al. (2010). Biolistic transformation of epidermal cells of pea (Pisum sativum var. Feltham First) leaves followed the protocol published in Rödiger et al. (2011). Confocal laser scanning microscopy after transient expression was performed according to Baudisch & Klösgen (2012).
Gel electrophoresis of proteins under denaturing conditions was carried out according to Laemmli (1970). The gels were exposed to phosphorimaging screens and analyzed with a Fujifilm FLA-3000 (Fujifilm, Düsseldorf, Germany) using the software packages BAS Reader (version 3.14) and AIDA (version 3.25) (Raytest, Straubenhardt, Germany). Protein concentration was determined according to Bradford (1976) and chlorophyll concentration according to Arnon (1949). All other methods followed published protocols (Sambrook & Russell, 2001).
Selection of candidate proteins
In order to identify candidate proteins for dual targeting into mitochondria and chloroplasts, we screened the nuclear genome of A. thaliana for protein-encoding genes with homology to α-proteobacterial and cyanobacterial genes because we assumed that, within this group, the number of proteins that are transported ‘back’ into either of the two endosymbiotic organelles would be particularly high. As representatives of mitochondrial ancestors, Rickettsia prowazekii and Rhizobium meliloti (both living intracellularly within human and plant cells, respectively), as well as Caulobacter crescentus (as an example of a free-living prokaryote), were used. With respect to chloroplasts, the cyanobacteria Synechocystis PCC 6803 (unicellular), Anabaena PCC 7120 (filamentous) and Thermosynechococcus elongatus BP-1 (unicellular, thermophilic) were selected. From this analysis, c. 4100 non-redundant protein-encoding genes with homology to the ancestors of plastids were identified, which is in good agreement with literature data suggesting 4500 protein-encoding genes of cyanobacterial origin (Martin et al., 2002). Homology to the ancestors of mitochondria was found for c. 3000 protein-encoding genes. In addition, c. 2000 genes showed homology to both α-proteobacterial and cyanobacterial genes (Supporting Information Table S1).
To obtain a first hint of the presumed subcellular localization of the corresponding 2000 proteins, the N-terminal 150 amino acid residues of the deduced polypeptide sequences of the respective genes were analyzed with the software package TargetP, which is commonly used to predict the likelihood of organellar targeting of a given protein sequence (Emanuelsson et al., 2000). We chose TargetP for this purpose because its training sets also included mitochondrial sequences from plants, in contrast with most other prediction programs. By applying an arbitrary threshold level of 0.8 for the probability of targeting to endosymbiotic organelles in general (see the 'Materials and Methods' section), c. 700 of the 2000 input protein sequences were extracted (Supporting Information Table S2). Most are clearly predicted to be targeted to only one of the two endosymbiotic organelles. However, in an unexpectedly large number of cases, the prediction values for the two organelles were remarkably similar, i.e., the software calculated comparable likelihood of targeting into each, mitochondria and chloroplasts. Taking the difference of the prediction values for transport into mitochondria (M*) and chloroplasts (C*) as a distinctive feature, we arranged the candidates into classes, starting with proteins with an almost identical prediction for the two organelles (class A, i.e. a high probability for dual targeting is expected) and ending with proteins for which an almost exclusive localization within either mitochondria or chloroplasts is predicted (class Z, i.e. a strictly monospecific localization into one organelle is expected) (Table 1). From each of the classes A (ΔM*C* < 0.1), B (ΔM*C* = 0.1 < 0.2) and C (ΔM*C* = 0.2 < 0.3), four candidates were randomly selected for further analyses (Table 2). Among the proteins of class A was α-MPP2, a subunit of the mitochondrial processing peptidase that has been shown recently to exhibit dual targeting properties in in vivo and in organello protein transport experiments (Baudisch & Klösgen, 2012). For comparison, four candidates from class Z (ΔM*C* ≥ 0.8) were further analyzed, two each with prediction for mitochondria and chloroplasts, respectively (Table 2).
Table 1. Classification of precursor proteins from Arabidopsis thaliana according to the difference in TargetP prediction values for the likelihood of transport into mitochondria and chloroplasts (ΔM*C*)
Number of proteins
0.1 < 0.2
0.2 < 0.3
Table 2. Selected candidate proteins from Arabidopsis thaliana identified by the in silico approach
C, chloroplasts; M, mitochondria; M*(C*), likelihood of transport into mitochondria (chloroplasts); ΔM*C*, difference in TargetP prediction values for transport into mitochondria and chloroplasts.
Almost all transit peptides of candidates from classes A and B show dual targeting in vivo
As the first step in the investigation of the organelle targeting specificity of the candidate proteins of classes A and B, in vivo transport experiments were performed. For this purpose, epidermal cells of pea leaves were transiently transformed by biolistic transformation with artificial genes encoding chimeric reporter proteins under the control of the 35S promoter of Cauliflower Mosaic Virus. Each chimeric reporter construction consists of the N-terminal 100 residues of a candidate precursor protein, which presumably comprise the complete information for the targeting and transport of the protein into its target organelle (e.g. Berglund et al., 2009), fused to enhanced yellow fluorescent protein (EYFP) (Fig. 1a). The subcellular localization of EYFP within the transformed cells was analyzed by confocal laser scanning microscopy. It was found that all reporter proteins based on candidates of classes A and B co-localize with the mitochondria of the transformed cells, i.e., they are detectable as highly mobile punctate structures with a size of c. 0.5 μm (Fig. 1b–h). Remarkably, with one exception, they also show EYFP fluorescence in plastids, as indicated by the co-localization of the EYFP signal with the autofluorescence signal of chlorophyll (Fig. 1b–g). This demonstrates that the N-terminal 100 residues of seven of the eight candidate proteins belonging to classes A and B comprise transport signals that are capable of recognizing both endosymbiotic organelles as targets. The only exception is ATPS (At5g08680) for which transport into chloroplasts, and thus dual targeting, cannot be observed (Fig. 1h).
This result was confirmed by reporter constructions consisting of the full-length candidate proteins fused to EYFP. Biolistic transformation of pea leaves with these chimeric genes led principally to the same subcellular localization as the constructs comprising the N-terminal 100 residues (e.g. GCS, EF-Tu, GrpE, see Supporting Information Fig. S3a–c). However, in some instances, it resulted in cytosolic aggregates (Fig. S3d–f), which is the reason why we restricted our analysis to the 100-residue constructions in the subsequent experiments.
The efficiency of transport into either mitochondria or plastids, as indicated by the relative intensity of EYFP fluorescence accumulating in either of the organelles, differs between the candidates. Preferential accumulation of the reporter in only one type of organelle is observed, for example, with GCS(1–100)/EYFP (stronger EYFP signal in mitochondria, Fig. 1g) or PDF(1–100)/EYFP (stronger EYFP signal in chloroplasts, Fig. 1f). Such variability in organelle-linked EYFP fluorescence, depending on the chimeric reporter protein, suggests different affinities of the transit peptides for the respective transport receptors, although it cannot be ruled out that the accumulation of EYFP within the organelles might also have been influenced by parameters such as the expression rate or protein stability.
Most candidate proteins of classes A and B also show dual targeting in in organello assays
A complementary approach to study protein transport involves in organello experiments, in which radiolabeled authentic precursor proteins obtained by in vitro translation in cell-free translation systems are incubated with freshly isolated intact mitochondria or chloroplasts. By analyzing the precursors of the eight candidate proteins of classes A and B with such in organello import experiments, it was found, again, that all proteins except ATPS (At5g08680) show dual targeting and import (Fig. 2). Within the organelles, the precursor proteins are processed to products of lower molecular weight, which are resistant to protease added externally to the organelles after the import reaction, proving that they are the result of complete internalization of the candidate proteins into the respective organelles. The size of the processing products accumulating in either mitochondria or chloroplasts is, in some cases, identical, suggesting cleavage at close proximity within the precursor polypeptide (Fig. 2b,d–f), as was also found with Thr-tRNA synthetase (Berglund et al., 2009). In other instances, however, the processing products of the candidate proteins within the two organelles show different mobility on sodium dodecylsulfate-polyacrylamide gel electrophoresis (SDS-PAGE) (Fig. 2a,c), in line with the accumulation of divergent processing products within mitochondria and chloroplasts described, for example, for Phe-tRNA synthetase (Pujol et al., 2007) and cytochrome c1 (Rödiger et al., 2011). In most cases, the precursor proteins are processed to single products in each organelle. However, in some instances, multiple processing can be observed (Fig. 2a,b,e), possibly as a result of stepwise processing events, such as those reported earlier (Branda et al., 1999; Richter & Lamppa, 2003). Remarkably, the organelle preference observed for the candidate proteins on in organello import is similar to that observed in the corresponding transient expression studies described above (e.g. Figs 1g, 2f and 1f, 2e).
Congruent results between the in vivo and in organello experiments were also observed for ATPS (At5g08680), the only candidate exhibiting monospecific targeting characteristics in the transient transformation assays. This protein is efficiently imported into isolated mitochondria and cleaved to a single product with an apparent size of 50 kDa which is resistant to protease added externally to the organelles after import (Fig. 2g). Incubation of the precursor with chloroplasts, however, leads merely to the binding of the protein to the surface of the organelles, and neither specific processing products nor protease resistance can be observed (Fig. 2g).
It should be mentioned that we did not detect any unspecific import of a plastid protein into isolated pea mitochondria in our experiments, in contrast with published data (Cleary et al., 2002). We performed numerous control experiments to reconfirm the protein selectivity of our isolated organelles (see, for example, Fig. S2) and can only speculate that the observed differences in protein import selectivity of the various isolated organelles result from differences in the experimental conditions (for a detailed discussion, see also Rödiger et al., 2010).
Organelle specificity of class C candidate proteins
Next, we analyzed the organelle targeting specificity of four proteins that were grouped into class C based on the slightly larger difference in the prediction values for mitochondrial and chloroplast targeting (ΔM*C* = 0.2 < 0.3). Again, in organello and in vivo experiments, like those described above, were performed. It was found that two of the four candidates, notably DhoDH (At5g23300) and Gtred (At3g15660), show dual targeting in both assays. The precursor proteins are imported into isolated mitochondria, as well as chloroplasts, where they are processed to single or multiple products, respectively (Fig. 3b,d). Likewise, confocal laser scanning microscopy of epidermal leaf cells transiently expressing either of the two reporter constructions (DhoDH(1–100)/EYFP, Gtred(1–100)/EYFP) demonstrates localization of the fluorescent reporter within both organelles (Fig. 3a,c).
By contrast, GgpS (At3g29430) shows monospecific targeting to chloroplasts in both assay systems. Incubation of the radiolabeled precursor with isolated organelles results in the import of the protein into chloroplasts, where it is cleaved to a processing product of higher mobility, whereas neither specific processing nor protease resistance can be observed in the mitochondrial assay (Fig. 3f). In line with this, transiently transformed cells expressing GgpS(1–100)/EYFP show EYFP fluorescence solely in plastids, as demonstrated by the co-localization of the EYFP signal with chlorophyll autofluorescence (Fig. 3e).
Such co-localization of EYFP and chlorophyll is also detectable in cells expressing GAPDH(1–100)/EYFP, the fourth member of class C candidates (Fig. 3g). Again, labeling of mitochondria cannot be observed, which suggests that the N-terminal 100 residues of GAPDH (At1g42970) are able to target the passenger protein in vivo exclusively into plastids. However, the authentic precursor protein of GAPDH (At1g42970) shows a deviating targeting behavior in in organello experiments (Fig. 3h). Within both organelle fractions, a processing product of higher mobility is observed that is protected against externally added protease, which suggests that GAPDH (At1g42970) exhibits dual targeting properties in organello. Thus, this is the first example in our study in which the results obtained with the two assay systems deviate from each other.
Even candidate proteins of class Z can show dual targeting
For comparison, we also analyzed a set of candidate proteins that were predicted to be strictly monospecific for either mitochondria or chloroplasts (Table 1). Unexpectedly, however, none of these candidates show unambiguous monospecific targeting in our experiments. Instead, both candidates for mitochondrial localization, fumarate hydratase (FumH, At2g47510) and NADH (At1g79010) (Table 2), show dual localization within mitochondria and plastids after transient expression of the corresponding chimeric reporter proteins (FumH(1–100)/EYFP, NADH(1–100)/EYFP) (Fig. 4a,c). In the case of FumH (At2g47510), such dual targeting is also observed in the complementary in organello assays (Fig. 4b). Although import into chloroplasts in this case is considerably weaker than into mitochondria, it confirms that, even among monospecific candidate proteins, dual targeting properties can be found. NADH (At1g79010), however, shows in organello import only into mitochondria, but not into chloroplasts (Fig. 4d). Thus, NADH (At1g79010) shows deviating targeting characteristics in the in organello and in vivo experiments.
Exactly opposite results, notably dual targeting with in organello assays but monospecific targeting after transient expression, are found for the two candidate proteins predicted to show chloroplast localization. With in organello assays, the proteins SHBP (At3g55800) and DXPS (At4g15560) show processing products of similar mobility on SDS-PAGE within both organelle fractions (Fig. 4f,h). By contrast, confocal laser scanning microscopy of epidermal leaf cells expressing either SHBP(1–100)/EYFP or DXPS(1–100)/EYFP demonstrates, in both cases, targeting solely into plastids, as visualized by co-localization of the EYFP signal with chlorophyll autofluorescence (Fig. 4e,g). No EYFP signal associated with mitochondria is observed after transient expression of the two chimeric reporter constructions. Thus, these two peptides show the same targeting behavior as GAPDH (At1g42970) (Fig. 3g,h), namely monospecific targeting in vivo and dual targeting in organello.
It was the goal of this study to assess the frequency, and thus relevance, of dual targeting of proteins to both mitochondria and chloroplasts within plant cells. Using a combination of in silico analysis and protein transport experiments, we attempted to obtain a picture of the number of proteins in A. thaliana that potentially possess such dual targeting properties.
Suitability of the experimental approach
In the nuclear genome of A. thaliana, c. 2000 non-redundant protein-encoding genes with homology to genes of both α-proteobacteria and cyanobacteria are found (Table S1). Approximately 30% of these genes encode proteins predicted to carry transit peptides mediating import into mitochondria and/or chloroplasts (Table S2). These transit peptides were classified according to the presumed target organelle using the software package TargetP (Emanuelsson et al., 2000) (Table 1), and candidate proteins from various classes were analyzed by a combination of in vivo and in organello protein transport experiments (Table 2). All selected candidate proteins, with both experimental systems, show transport into at least one of the two endosymbiotic organelles, demonstrating that TargetP is principally suitable to predict such transit peptides. Furthermore, from the 16 candidate proteins analyzed, 12 show identical organelle specificity in both experimental systems, whereas, in four cases, the two experimental systems lead to deviating results (Table 3). The reason for such deviation is not yet fully understood, but might be the result of the specific experimental conditions. For example, in the in organello experiments, targeting and transport of an authentic precursor protein are analyzed with isolated organelles, i.e., the assays are devoid of potentially important cytosolic targeting factors and contain instead components of the heterologous in vitro translation mixture. However, in the in vivo experiments, chimeric reporter constructions rather than authentic precursor proteins are analyzed, and it has been known for a long time that reporter proteins, such as EYFP, have a strong tendency to aggregate after overexpression (Garcia-Mata et al., 1999; Vogel et al., 2007), which eventually can prevent membrane transport. Taken together, our results demonstrate that, for a reliable conclusion on the targeting properties of a protein, two independent and complementary experimental systems are required. A single assay system might be sufficient for a reasonable guess, but the result should not be taken as final proof.
Table 3. Experimentally determined subcellular localization of the candidate proteins from Arabidopsis thaliana
C, chloroplasts; Dual, dual targeting; M, mitochondria.
For a few of the candidate proteins analyzed here, complementary data are available in the literature or databases which support our findings. For example, GCS (At2g35370), which shows dual localization in our in vivo and in organello protein transport experiments (Figs 1g, 2f), was likewise detected in both mitochondria and plastids by MS/MS analyses (see Table S2). In addition, the exclusive plastidic localization of DXPS(1–100)/EYFP in transiently transformed epidermal leaf cells of pea (Fig. 4g) corresponds well with the data of Araki et al. (2000), who transiently expressed a similar construct (DXPS(1–58)/GFP; GFP, green fluorescent protein) in cultured cells of tobacco. Unfortunately, the authors did not perform any complementary experiments, so that it remains unsolved whether the dual targeting properties of the authentic DXPS precursor protein, which we observed in our in organello assays (Fig. 4h), could likewise be confirmed.
In three instances, however, our findings are contradictory to the literature data at first glance. Although FumH (At2g47510), OhmT (At2g46110) and Gtred (At3g15660) show dual targeting with both experimental systems used here (Figs 4a,b, 1c, 2b, and 3c,d, respectively), an exclusive mitochondrial localization has been described for the three proteins in the literature (Chew et al., 2003; Ottenhof et al., 2004; Pracharoenwattana et al., 2010; respectively). However, in the case of FumH (At2g47510), the authors did not actually look for a plastidic localization, and both the chimeric reporter construct (GFP localized in the middle of the precursor protein) and the methodical details (e.g. transformation of cell cultures) (Pracharoenwattana et al., 2010) are so different from those of our study that the results are hardly comparable. This is not the case for OhmT (At2g46110) for which a full-length reporter construction with GFP was analyzed in vivo basically in the same manner as described here (Ottenhof et al., 2004). Unfortunately, the corresponding full-length fusion with EYFP showed cytosolic aggregation in our assays (Supporting Information Fig. S3e) and complementary in organello data were not provided by the authors, so that the reason for the divergent results remains unresolved. The same also holds true for Gtred (At3g15660), although, in this case, in organello experiments were performed (Chew et al., 2003). However, the epifluorescence microscopy data provided are difficult to interpret and, again, the comparable full-length reporter construction shows cytosolic localization and aggregation in our experiments (Fig. S3f). Thus, despite the fact that, for several other enzymes of the ascorbate–glutathione cycle dual targeting has been demonstrated (Chew et al., 2003), the organelle specificity of the Gtred precursor remains enigmatic.
The number of proteins with dual targeting properties is unexpectedly high
The classification of the candidate proteins according to the difference in the TargetP prediction values for mitochondria and chloroplasts was found to be suitable for the selection of proteins with dual targeting properties. In classes A and B, which comprise proteins with only minor differences in these prediction values, a high probability for dual targeting into both organelles is found. Seven of the eight candidate proteins are transported into both mitochondria and chloroplasts irrespective of the experimental system used. Moreover, even 50% of the tested candidate proteins of class C, which exhibit a slightly larger difference in the prediction values, show dual targeting in vivo as well as in organello. If we assume that the behavior of the analyzed candidate proteins is exemplary for all proteins of classes A, B and C, which comprise 38, 27 and 29 proteins, respectively (Table 1), these three classes should contain c. 70 proteins with dual targeting properties. Considering that proteins with ΔM*C* = 0.3 < 0.8 have been excluded from the analysis and that, even in class Z, which comprises proteins with strictly monospecific prediction (ΔM*C* ≥ 0.8), dual targeting can be found (e.g. FumH (At2g47510), Fig. 4a), it can be concluded that the actual number of proteins with dual targeting properties is substantially higher. The following points support this conclusion. First, the restriction of our analysis to genes with homology to both α-proteobacteria and cyanobacteria is a considerable limitation of the potential number of candidate genes. Hence, within the nuclear genome of A. thaliana, numerous further genes encoding proteins with dual targeting transit peptides can be assumed. Second, the high arbitrary threshold level of 0.8 that was used for the definition as a transit peptide (M* + C* ≥ 0.8; see the 'Materials and Methods' section) further reduced the number of potential candidate proteins. Indeed, almost one-half of the dual targeted proteins identified by MS (summarized in Carrie et al., 2009a), which are listed in the SUBA database (Heazlewood et al., 2007), missed the criteria described above and were thus excluded from our analysis. Third, our set of c. 700 candidate proteins (M* + C* ≥ 0.8; see Table S2) comprises only 20 of the c. 80 proteins of A. thaliana with dual targeting properties that have been identified to date (summarized in Berglund et al., 2009; Carrie et al., 2009b; Morgante et al., 2009; Carrie & Small, 2013). Finally, most estimates regarding the number of nuclear-encoded proteins of mitochondria and chloroplasts reach > 4000 in total (Leister, 2003; Millar et al., 2006), which significantly exceeds the number of 700 candidate proteins identified here. Taking all these points into consideration, it appears reasonable to assume that the actual number of proteins with dual targeting properties in plants is in fact in the range of several hundred. Remarkably, Mitschke et al. (2009) came to a similar conclusion by a completely different approach which was mainly based on protein sequence predictions.
Why so much dual targeting?
Intracellular sorting of proteins is usually considered to be a highly specific and selective process achieved by the specific recognition of a transport signal by receptor compounds on the surface of the target organelle. An early example of proteins targeted to both mitochondria and chloroplasts involves aminoacyl-tRNA synthetases, which are essential for translation (Duchêne et al., 2005) and thus provide a function in both target organelles. Likewise, the dually targeted ribosomal protein S16 of Medicago truncatula and Populus alba (Ueda et al., 2008) was assumed to specifically compensate for the loss of the respective gene, and thus protein function, in the organelles. Consequently, for these proteins, a positive selection pressure for the establishment and/or maintenance of dual targeting appears likely. However, the functionality of a given protein in both organelles may not be a prerequisite for dual targeting. For example, within plastids, a function of FumH (At2g47510), an enzyme of the citrate cycle showing dual targeting properties (Fig. 4a,b), cannot easily be attributed. Likewise, cytochrome c1, a component of the respiratory electron transport chain, shows dual targeting, although a function within chloroplasts is not likely (Rödiger et al., 2011). This suggests that protein import into mitochondria and chloroplasts is not as strictly selective as often assumed. Instead, dual targeting might be acceptable for the cell as long as the degree of ‘mistargeting’ into the ‘wrong’ organelle is not sufficiently harmful to cause negative selective pressure. However, dual targeting might even be the result of positive selection pressure because it could facilitate the transfer of complete metabolic pathways across organelle borders (Martin, 2010). An interesting alternative hypothesis was recently provided by Carrie & Small (2013), who speculated that dual targeting might be one of the processes limiting genetic variability as a result of the double selection pressure on proteins with functions in two different environments. In any case, dual targeting of nuclear-encoded organelle proteins probably represents an evolutionary remnant and a direct consequence of the two consecutive endosymbiotic events (Staiger et al., 2009).
We thank Daniela Ditfe, Swanhild Lohse and Stefanie Max for their help with the protein localization studies. This work was supported by the Martin Luther University Halle-Wittenberg.