Heterogeneity of the Mitochondrial Proteome for Photosynthetic and Non-photosynthetic Arabidopsis Metabolism*S

Heterogeneity of the mitochondrial proteome in plants underlies fundamental differences in the roles of these organelles in different tissues. We quantitatively compared the mitochondrial proteome isolated from a non-photosynthetic cell culture model with more specialized mitochondria isolated from photosynthetic shoots. Differences in intact mitochondrial respiratory rates with various substrates and activities of specific enzymes provided a backdrop of the functional variation between these mitochondrial populations. Proteomics comparisons provided a deep insight into the different steady-state abundances of specific mitochondrial proteins. Combined these data showed the elevated level of the photorespiratory apparatus and its complex interplay with glycolate, cysteine, formate, and one-carbon metabolism as well as the decrease of selected parts of the tricarboxylic acid cycle, alterations in amino acid metabolism focused on 2-oxoglutarate generation, and degradation of branched chain amino acids. Comparisons with microarray analysis of these tissue types showed a positive, mild correlation between mRNA and mitochondrial protein abundance, a tighter correlation for specific biochemical pathways, but over 78% concordance in direction between changes in protein and transcript abundance in the two tissues. Overall these results indicated that the majority of the variation in the plant mitochondrial proteome occurred in the matrix, highlighted the constitutive nature of the respiratory apparatus, and showed the differences in substrate choice and/or availability during photosynthetic and non-photosynthetic metabolism.

Plant mitochondria have been traditionally purified from total cell extracts of storage organs, etiolated tissues, or cell cultures by differential centrifugation and density gradient centrifugation using sucrose or Percoll (1)(2)(3)(4). These organelles are derived from tissues where sugars or fats are used as energy sources and mitochondria have a primary role in ATP production for cellular activity. The combination of Percoll and PVP 1 gradients has also enabled purification of mitochondria away from chloroplasts in extracts of photosynthetic shoot or leaf samples from plants (3). These organelles are known to have a range of additional roles including carbon skeleton provision for nitrogen assimilation (5) and the recycling of carbon in the photorespiratory cycle (6,7).
Early reports using 1D SDS-PAGE documented apparent differences between mitochondria isolated from leaves, petioles, and roots from spinach (8); roots, leaves, and flowers from sugar beet (9); and shoot, scutella, endosperm, and cob from developing maize (10). Remy et al. (11) performed the first comparative 2D IEF/SDS-PAGE study of pea mitochondrial protein profiles isolated from different organs and quantified differences in the abundance of glycine decarboxylase complex subunits using protein spot intensities. Similar studies to compare mitochondrial proteomes were also performed on potato (12) and wheat (13). Using a combination of 2D PAGE and immunodetection, Colas des Francs-Small et al. (12) were able to compare the abundance of ATP synthase ␤ subunit in potato tuber and callus. More recently pea mitochondrial proteomes from leaves, roots, and seeds have been separated by 2D PAGE, hundreds of protein spots were detected, and a number appeared to vary in abundance between mitochondria isolated from different tissue types (14). However, despite these reports, relatively little data are available on the quantification of these differences or the identity of the proteins themselves.
The expression of the genes encoding mitochondrial respiratory components has been shown to be co-regulated in various vegetative and reproductive organs (15)(16)(17)(18). Promoter studies suggest that the existence of site II motifs in the proximal promoter regions of genes for mitochondrial components may play an important role in displaying organ-specific, metabolic, environmental, and developmental responses (17,19).
In functional studies, the specialized role of mitochondria in leaves during photosynthesis has also been investigated using a variety of methods. The regulation of mitochondrial pyruvate dehydrogenase complex by phosphorylation and inactivation in the light has suggested a down-regulation of tricarboxylic acid cycle function in the light, and the rapid rates of glycine-dependent respiration by leaf mitochondria suggest a metabolism geared to photorespiratory glycine decarboxylation (20,21). ATP provision by mitochondria in the light has been shown to be essential for efficient photosynthesis (22), and complex I function and uncoupling protein are also required to maintain photosynthetic efficiency (23,24), whereas down-regulation of mitochondrial malate dehydrogenase appears to enhance photosynthetic performance (25), and modification of mitochondrial acetyl-CoA levels inhibit both growth and photosynthesis (26). Although transcription of the mitochondrial genome cycles with diurnal rhythm, the steady-state transcript pool is found to be homoeostatic throughout the daily cycle (27). In contrast, the nuclearly encoded components of mitochondria, such as nda1 and nda2 (28), aox (29), shm1 (30), and gdcP (31), show rapid transcriptional response to light/dark transition and large changes in diurnal transcript pool sizes.
Recent proteome studies of plant mitochondria have centered on Arabidopsis as a model and have so far identified over 500 nuclearly encoded mitochondrial proteins (32). However, there is still very little data available on which of these Arabidopsis proteins are constitutively present in mitochondria and which are selectively expressed and accumulated for specialized roles in the integration of mitochondrial metabolism into different cellular environments. To further study changes in Arabidopsis mitochondrial proteomes in specific organs and to deepen our understanding of the extent of the differences between mitochondria from photosynthetic and non-photosynthetic tissues, we analyzed mitochondria from the model cell culture used in most mitochondrial proteomics investigations compared with mitochondria isolated from shoots of Arabidopsis seedlings.

Maintenance of Arabidopsis Cell Culture and Hydroponic Seedling
Culture-Arabidopsis cell suspension was cultured in growth medium (1ϫ Murashige and Skoog medium without vitamins, 3% sucrose, 0.5 mg/liter naphthaleneacetic acid, 0.05 mg/liter kinetin, pH 5.8) at 22°C under a 16/8-h day/night regime and light intensity of 90 mol m Ϫ2 s Ϫ1 with orbital shaking at 120 rpm. Cultures were maintained in 250-ml Erlenmeyer flasks by the inoculation of 25 ml of 7-10-day-old cells into 100 ml of fresh growth medium. Dark-grown cells used for mitochondrial isolation were subcultured from 7-day-old light-grown cultures and grown under the same conditions as described for light-grown cell cultures.
Conditions for the hydroponic culture were adapted from Schlesier et al. (33) with slight modifications. Approximately 50 -100 wild type Arabidopsis thaliana seeds (ecotype Columbia) were surface-sterilized in 70% ethanol for 2 min followed by 15-min incubation in 5% bleach, 0.1% Tween 20, inverting five times every 5 min. Seeds were then washed five times with sterilized water and were carefully dispensed on a stainless steel wire mesh platform (mesh size, 1 mm; 4 ϫ 4 ϫ 2 cm) layered previously with 1% sterilized agarose in a round plastic vessel (diameter, 68 mm; height, 76 mm) containing ϳ60 ml of liquid medium (( 1 ⁄4 strength Murashige and Skoog medium without vitamins, 1 ⁄4 strength Gamborg B 5 vitamin solution, 2 mM MES, 1% (w/v) sucrose, pH 5.8). Arabidopsis plants were grown under a 16/8-h light/dark period with light intensity of 100 -125 mol m Ϫ2 s Ϫ1 at 20°C over 3 weeks. Liquid medium was regularly replaced with freshly made liquid medium every 7 days.
Purification of Mitochondria from Plant Materials-Isolation of mitochondria from cell culture was carried out using the method modified from Millar et al. (34) and Sakamoto et al. (35). Cells (250 g) were incubated at 25°C for 3 h in ϳ800 ml of enzyme buffer (0.4 M mannitol, 36 mM MES, 0.4% (w/v) cellulase ("Onozuka," Yakult Pharmaceutical, Tokyo, Japan), 0.05% (w/v) pectolyase ("Y-23," Kyowa Chemical Products, Osaka, Japan), pH 5.7). Protoplasts were harvested by washing twice in enzyme buffer without cellulase and pectolyase and centrifugation at 800 ϫ g for 5 min. Cells were disrupted in extraction buffer (0.45 M mannitol, 50 mM tetrasodium pyrophosphate, 0.5% (w/v) PVP, 0.5% (w/v) BSA, 2 mM EGTA, 20 mM cysteine, pH 8.0) by five strokes in a Potter-Elvehjem homogenizer. The homogenate was centrifuged at 1500 ϫ g for 5 min, and the resulting supernatant was then centrifuged at 18,600 ϫ g for 20 min. The pellet of crude organelles was carefully resuspended in mannitol wash medium (0.3 M mannitol, 0.1% (w/v) BSA, and 10 mM TES, pH 7.5). Following one stroke in a Potter-Elvehjem homogenizer, the crude organellar fraction was gently layered over a 35-ml discontinuous Percoll density gradient consisting of 18% (5 ml) over 23% (25 ml) and 40% (5 ml) Percoll solution in mannitol wash medium. The gradient was then centrifuged at 40,000 ϫ g for 45 min. The mitochondrial fraction was seen as an off-white band near the 23-40% (v/v) Percoll interface. The upper layers of the density gradient were removed, and the mitochondrial band was collected. The transferred mitochondrial band was diluted ϳ5-fold with sucrose wash medium (0.3 M sucrose, 0.1% (w/v) BSA, 10 mM TES, pH 7.5) and centrifuged at 24,000 ϫ g for 10 min. The mitochondria-enriched homogenate was collected, and Percoll density centrifugation was repeated once as described for the first gradient. Prior to FFE, the enriched mitochondrial fraction was washed three to four times with FFE separation medium (10 mM acetic acid, 10 mM triethanolamine, 1 mM EDTA, 280 mM sucrose, pH 7.4).
Shoot mitochondria were isolated from 3-week-old hydroponically grown Arabidopsis using a method adapted from Day et al. (3) with slight modification. Approximately 100 g of shoot materials were homogenized with a Polytron blender (Kinematica, Kriens, Switzerland) in 300 ml of cold grinding medium (0.3 M sucrose, 25 mM tetrasodium pyrophosphate, 1% (w/v) PVP-40, 2 mM EDTA, 10 mM KH 2 PO 4 , 1% (w/v) BSA, 20 mM ascorbic acid, pH 7.5) for 10 s twice with 5-10-s intervals between bursts. The homogenate was filtered through four layers of Miracloth and centrifuged at 1500 ϫ g for 5 min, and the resulting supernatant was then centrifuged at 24,000 ϫ g for 15 min. The organelle pellet was washed by repeating the 1500 and 24,000 ϫ g centrifugation steps twice in sucrose wash medium. The resulting pellet of crude organelles was carefully resuspended in sucrose wash medium and gently layered over a 35-ml continuous 28% Percoll density gradient consisting of 0 -4.4% PVP-40. The gradient was then centrifuged at 40,000 ϫ g for 45 min. The mitochondrial band was seen as a yellow-brownish band near the bottom of the tube. The upper layers of the density gradient were removed, and the mitochondrial band was collected. The mitochondrial fraction was diluted ϳ5-fold with sucrose wash buffer and centrifuged at 24,000 ϫ g for 10 min. The washed fraction was further purified with Percoll density centrifugation as described for the first gradient. The mitochondrial band was collected and washed three to four times with FFE separation medium. Zone electrophoresis-FFE was conducted using the BD free flow electrophoresis unit (BD Biosciences) as described previously by Eubel et al. (36).
Gel Electrophoresis and Immunoblotting-For Western blotting with 1D SDS-PAGE, Bio-Rad Criterion precast gels (10 -20% (w/v) acrylamide, Tris-HCl, 1 mm, 18-comb) were used. Gel electrophoresis was performed at 25 mA/gel for ϳ3-4 h. Polyacrylamide gels were then incubated in transfer solution for 1 h. Protein transfer onto a Hybond TM -C extra nitrocellulose membrane (GE Healthcare) was performed using a Hoefer SemiPhor (GE Healthcare) instrument according to the manufacturer's instruction. Transferred proteins were probed with primary antibodies specifically targeted to mtHSP70, AOX, porin (all from Dr. Tom Elthon, Lincoln, NE), RbsS (from Dr. Spencer Whitney, Australian National University, Canberra, Australia), acetyl-CoA acyltransferase (Kat2) (37), and fructose bisphosphatase (commercial antibody from Agrisera, Vä nnä s, Sweden). A chemiluminescence detection linked to horseradish peroxidase was used as a secondary antibody, and quantitative light emission was recorded using a luminescent image analyzer (LAS 1000, Fuji, Tokyo). Differential (DIGE) 2D IEF/SDS-PAGE was performed according to Eubel et al. (36) with four independent experiments using DeCyder TM software. DIGE 2D blue native/SDS-PAGE was performed as described in Perales et al. (38) with three independent experiments. Fluorescent protein spots were visualized on a Typhoon TM laser scanner (GE Healthcare), and image comparison was done using the DeCyder software package (version 6.5, GE Healthcare). Sets of gels were first analyzed using the differential in-gel analysis mode of the DeCyder (GE Healthcare) software package prior to a comprehensive biological variance analysis including all three or four gel sets with the same software. Gel spots were filtered according to their presence (in nine of nine or 12 of 12 gel images), average ratio in abundance change greater than 2, and Student's t test value (p Ͻ 0.05). Gel pictures were electronically overlaid using the Image Quant TL TM software (GE Healthcare). Following scanning for fluorescent signal, gels were stained and visualized by colloidal Coomassie (G-250) staining for mass spectrometry analysis.
Trypsin Digestion of Gel Plugs and Mass Spectrometric Analysis-In-gel digestion of the selected gel plugs was performed according to Taylor et al. (39). Samples were resuspended in 5% (v/v) acetonitrile and 0.01% (v/v) formic acid. Peptides were loaded onto self-packed Microsorb (Varian Inc.) C 18 (5-m, 100-Å) reverse phase columns (0.5 ϫ 50 mm) using an Agilent Technologies 1100 series capillary liquid chromatography system and eluted into an XCT Ultra IonTrap mass spectrometer with an ESI source equipped with a low flow nebulizer in positive mode and controlled by Chemstation (Agilent Technologies) and MSD Trap Control version 6.0 (Build 38.15) software (Bruker Daltonics GmbH). Peptides were eluted from the C 18 reverse phase column at 10 l/min using a 9-min acetonitrile gradient (5-80% (v/v) in 0.1% (v/v) formic acid) at a regulated temperature of 50°C. The method used for initial ion detection utilized a mass range of 200 -1400 m/z with scan mode set to standard (8100 m/z/s), ion charge control conditions set at 250,000, and three averages taken per scan. Smart mode parameter settings were used with a target of 800 m/z, a compound stability factor of 90%, a trap drive level of 80%, and "Optimize" set to normal. Ions were selected for MS/MS after reaching an intensity of 80,000 counts/s, and two precursor ions were selected from the initial MS scan. MS/MS conditions used SmartFrag for ion fragmentation, a scan range of 70 -2200 m/z using an average of three scans, the exclusion of singly charged ions option, and ion charge control conditions set to 200,000 in ultra scan mode (26,000 m/z/s). Resulting MS/MS spectra were exported from the DataAnalysis for LC/MSD Trap version 3.3 (Build 149) software package (Bruker Daltonics GmbH) using default parameters for Au-toMS(n) and compound export.
Results were queried against an in-house Arabidopsis database comprising ATH1.pep (release 7) from The Arabidopsis Information Resource (TAIR) and the Arabidopsis mitochondrial and plastid protein sets (the combined database contained a total of 30,700 protein sequences with 12,656,682 residues) using the Mascot search engine version 2.1.04 and utilizing error tolerances of Ϯ1.2 Da for MS and Ϯ0.6 Da for MS/MS; "Max Missed Cleavages" set to 1; variable modifications of oxidation (Met), carboxymethyl (Cys), and H-loss (Cys); instrument set to ESI-TRAP, and peptide charge set at 2ϩ and 3ϩ. ATH1.pep is a non-redundant database with systematically named protein sequences based on Arabidopsis genome sequencing and annotation. Results were filtered using "Standard scoring," "Max. number of hits" set to "AUTO," and "Ions score cut-off" set at 27. A protein match was automatically validated only when at least two unique peptides both showing an ion score higher than 38 (Mascotdefined significance threshold, p Յ 0.05) were present. For proteins identified by a significant peptide having a score above the significance threshold, only the spectrum of the significant peptide was thoroughly inspected to fulfill the following criteria before accepting as a match. (i) Each peak corresponding to a fragmented ion was clearly above base-line background noise, (ii) a series of at least four continuous y or b ions were observed, and (iii) peptides did not match to any sequences in trypsin or any commonly known contaminants. For proteins identified only by multiple peptides with each ion scored above the homology threshold (usually between 27 and 37), every single MS/MS spectra was thoroughly checked. When all the criteria were met, the final protein score must exceed 37, or the match would be rejected. To estimate the false-positive rate of our protein identification strategy, a single concatenated Mascot generic format (.mgf) file generated by MASCAT (Agilent Technologies) that comprised all the MS/MS output data was then used to search against TAIR7 (target), reversed (decoy), and randomized TAIR7 (decoy) versions of the Arabidopsis database using the above search strategy. The falsepositive rate in target-decoy searches was found to be 3-4% for peptides with ion scores Ͼ27 (supplemental Data 3) as calculated using the equation described previously (40).
When peptides were matched to multiple members of a protein family encoded by different Arabidopsis genes, each protein match was manually inspected to identify the peptide(s) that was uniquely assigned to one gene product but not to the others. Protein isoforms that were identified by the same set of peptides were both assigned as protein matches (see Table II). When proteins of different families were identified in a gel spot, a reference map of the Arabidopsis mitochondrial proteome was used to identify the most probable match, taking into account the number of peptides with ion score Ͼ38 and the quality of delta mass for each peptide. For protein matches with only one unique peptide, the peptide sequence was searched against the non-redundant protein database in National Center for Biotechnology Information (NCBI) BLASTP (taxonomy was limited to Arabidopsis) to ensure that no other proteins share exactly the same peptide sequence.
Oxygen Electrode and Spectrophotometric Measurements of Isolated Mitochondria-Oxygen consumption by plant extracts and crude and purified mitochondria was measured in a computer-con-trolled Clark-type O 2 electrode unit (Hansatech-Instruments, Pentney, UK). Calibration of the electrode was carried out by the addition of sodium dithionite to remove all oxygen in the electrode chamber. The air-saturated oxygen concentration was assumed to be 240 M. All reactions were carried out using 1 ml of mitochondrial reaction medium (0.3 M sucrose, 5 mM K 2 H 2 PO 4 , 10 mM TES, 10 mM NaCl, 4 mM MgSO 4 , 0.1% (w/v) BSA, pH 7.2) and 100 g of mitochondrial sample. Cytochrome c oxidase activity was determined by the rate of oxygen consumption according to Neuburger et al. (2). Pyruvate (10 mM), succinate (5 mM), glutamate (10 mM), malate (10 mM), glycine (10 mM), formate (10 mM), NADH (1 mM), NAD (2 mM), CoA (12 M), thiamine pyrophosphate (0.2 mM), and KCN (0.5 mM) were added as appropriate to modulate the oxygen consumption rates of mitochondria.
Pyruvate dehydrogenase complex (PDC) and ␣-ketoglutarate dehydrogenase complex activity were measured according to Taylor et al. (41). Formate dehydrogenase (FDH) activity was measured as described by Oliver (42). The activity of NAD-malic enzyme was measured according to Jenner et al. (43). Fumarase was assayed as described by Hatch (44). The amination and deamination activity of glutamate dehydrogenase (GDH) at A 340 was measured according to Turano et al. (45). Aconitase was assayed using a modification of the method of MacDougall and ap Rees (46). The rate of change in A 340 was monitored in the following mixture: 80 mM HEPES-NaOH (pH 7.5), 0.5 mM NADP, 0.5 mM MnCl 2 , 2 units of NADP-isocitrate dehydrogenase, 0.05% (v/v) Triton X-100. The reaction was initiated by the addition of 8 mM aconitate. The activity of malate dehydrogenase was measured by following NADH oxidation to NAD ϩ at 340 nm. The assay was carried out using the following reaction mixture: 90 mM KH 2 PO 4 -KOH (pH 7.4), 0.05% (v/v) Triton X-100, 5 mM MgCl 2 , 2 M NADH. The reaction was initiated by the addition of 750 M oxaloacetate (OAA). The activity of citrate synthase was measured by following APAD reduction to APADH at 365 nm by citrate synthase-dependent malate dehydrogenase (MDH) reaction. The assay was carried out using the following reaction mixture: 90 mM triethanolamine-KOH (pH 8.5), 0.05% (v/v) Triton X-100, 10 mM L-malate, 20 M APAD, 2 units of malate dehydrogenase. The reaction equilibrated at room temperature for 15 min, and citrate synthase activity was measured after the addition of 17 M acetyl-CoA.
Microarray Analysis-Microarray analysis of the changes in transcript abundance in cell culture and shoot was performed using Affymetrix GeneChipா Arabidopsis ATH1 Genome Arrays (catalog number 900386). Total RNA was isolated using the RNeasy plant miniprotocol and RNase-free DNase kit (Qiagen, Clifton Hill, Victoria, Australia). The quality of isolated RNA was assessed by spectrophotometric analysis of A 260 /A 280 ratio using an ND-1000 UV-visible spectrophotometer (NanoDrop Technologies, Wilmington, DE) and qualitative analysis using an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Synthesis of double-stranded cDNA, biotin labeling and fragmentation of cDNA, target hybridization, washing, staining, and scanning of arrays were performed as specified in the Affymetrix GeneChip Expression Analysis Technical Manual. Oligo B2 and Biotinylated Hybridization Controls (Affymetrix) were included in the hybridization as controls. Prior to hybridization to an ATH1 Ge-neChip, the quality of the target cRNA was assessed by hybridization of the prepared cRNA to a Test3 array (catalog number 900341, Affymetrix). Hybridization was performed in an Affymetrix GeneChip Hybridization Oven 640. Washing and staining were performed using an Affymetrix Fluidics Station 450. Scanning was performed with an Affymetrix GeneChip Scanner 3000 7G. Washing, staining, scanning, and data extraction were performed using Affymetrix GeneChip Operating Software (version 1.4). Affymetrix CEL files generated were further analyzed using Avadis analysis software (version 4.3; Strand Life Sciences, Carlsbad, CA). Data were normalized using the MAS5 algorithm and subjected to log 2 transformation. "Absent" probe sets were filtered out before averaging three biological replicates to get the expression value. The locus number for each probe identity was retrieved from TAIR database. The differential analysis was performed taking cell culture as reference to identify genes that are expressed more than 2-fold in shoot with false discovery rate-adjusted p values less than 0.05. The Arabidopsis subcellular localization database SUBA (47) was used to determine all the genes that are likely to be found in the mitochondria. All microarray data are available from ArrayExpress under accession number E-ATMX-31, experiment name Millar_Athal_cells.
Supplemental Material-Images of the replicated DIGE IEF/SDS-PAGE and BN/SDS-PAGE gels are in supplemental Data 1. Scores and matched peptides for identifications in Table II are in supplemental Data 2. .mgf files for each protein spot, spectra for identifications based on single peptides available for reanalysis (using the spectra drawing tool developed by Carroll et al. (48), and a graph of the false-positive rate versus ion score for the concatenated MS/MS data set are provided as supplemental Data 3. Primary data from Forner et al. (49) and Kislinger et al. (50) were reanalyzed to performed pairwise comparisons using Pearson correlation analysis on their mitochondrial components from different organs to compare with our own data in Fig.  6; this reanalysis is presented in supplemental Data 4.

Respiratory Characteristics of Mitochondrial Cell Culture and Shoot Mitochondria
Mitochondria were purified from Arabidopsis cell culture and a hydroponic shoot culture based on density purification techniques using Percoll. The respiratory characteristics of these organelles were compared using oxygen electrode measurements with different respiratory substrates (Table I). This analysis showed that the outer membrane integrity of these isolated organelles exceeded 93-97% and that both oxidized externally supplied NADH via the cytosol-facing rotenone-insensitive NADH dehydrogenases and used tricarboxylic acid cycle substrates succinate, glutamate plus malate, and malate plus pyruvate at similar rates on a protein basis. Glutamate can be utilized by mitochondria due to a matrix-located glutamate dehydrogenase, generating NADH and making 2-oxoglutarate that can enter the tricarboxylic acid cycle. In both cell culture-and shoot-derived mitochondria electron transport could be supported by glutamate as a sole substrate. In contrast, shoot mitochondria could use glycine or formate as a respiratory substrate, whereas for cell culture mitochondria these were very poor substrates with respiratory rates 5-10 times lower than those in shoot mitochondria. Comparison of the respiratory rates compares favorably with other published data on the rate of Percollpurified mitochondria from cell culture (34,51) and Arabidopsis leaf (52).

Quantitative Proteomics Comparison of Shoot and Cell Culture Mitochondria
A detailed proteomics analysis was then planned to quantitatively compare the composition of these organelles. Early analysis on gels revealed that despite the high quality of these preparations higher purity isolates were required to minimize minor contaminants in both preparations of peroxisome and chloroplast proteins from being revealed as differences between the organelles (data not shown). To alleviate this problem we adopted a free flow electrophoresis separation of mitochondria after Percoll-based purification to minimize this contamination. We have separately shown that this reduces peroxisomal and plastid contamination of cell culture and shoot mitochondria by ϳ5-10 times (36). Using Western blots to immunodetect cytosolic (fructose bisphosphatase), plastid (small subunit of ribulose-3,5-bisphosphate carboxylase/decarboxylase), and peroxisome (Kat2) marker proteins we can show the reduction in contamination in the samples used in this study (Fig. 1). In comparison the mitochondrial markers (porin, AOX, and mtHSP70) are increased in abundance.
With mitochondria prepared in this manner we performed a quantitative comparison of shoot and cell culture mitochondria using DIGE CyDye technology (GE Healthcare). In this approach differential 2D (DIGE) IEF/SDS-PAGE is performed using fluorescent dyes Cy3 and Cy5 separately bound to the two samples to be compared and a mixture of all samples in the experiment bound to Cy2 as an internal standard common to all gels (Fig. 2). A substantial set of changes are apparent by the colored overlay of the fluorescence images. Using four independent cell culture and shoot mitochondrial isolates, a set of four gels were run and analyzed using DeCyder quantitation software (GE Healthcare; all gels are available in   88 supplemental Data 1). Protein spots that reproducibly changed in abundance between the two samples with p Ͻ 0.05 (n ϭ 4) were selected for further analysis. Individual protein spots on paired preparative Coomassie-stained gels were selected and analyzed using LC-MS/MS (Fig. 2, indicated by arrows). In total 88 differentially abundant protein spots were analyzed, and the identities of most of the proteins found are summarized in Table II (summaries of peptides matching for each protein are in supplemental Data 2; primary data as .mgf files are available in supplemental Data 3). It was apparent that for some protein spots the staining intensity was different for fluorophore labeling and Coomassie staining. Therefore, apparent changes in abundance of several protein spots in Fig. 2 could not been analyzed because they could not be identified with confidence from the Coomassie-stained gel; also several analyzed proteins could not be identified by mass spectrometry with confidence (Spots 62, 83, 86, and 90). In Table II, identified proteins have been organized ac- Samples of shoot (labeled with Cy3, shown in green) and cell culture (labeled with Cy5, shown in red) were compared. Top, gel pictures, as derived from the scanner, analyzed with the DeCyder software package (GE Healthcare). The abundance of protein spots encircled in red or green were significantly increased or decreased compared with the profile in the other tissue. Bottom, gel pictures were electronically overlaid using Image Quant TL software (GE Healthcare). Yellow protein spots represent proteins of equal abundance between the tissues. Protein spots that are more abundant in shoot sample are green, and those that are more abundant in the cell culture sample are red. Arrows indicate proteins unambiguously identified by MS; the numbers correlate with Table II.   TABLE II Proteins from mitochondrial samples found to vary in abundance (ratio Ͼ 2, p Ͻ 0.05) between shoot and cell culture samples The identity of proteins was determined by tandem MS. The predicted molecular mass (MM) and pI of the matched protein and the gel sample are shown along with the molecular weight search (MOWSE) score (Score); number of independent, non-redundant peptides matched to tandem mass spectra; and the percentage of coverage of the matched sequence. Protein spots with at least 2-fold change (p Ͻ 0.05, n ϭ 4 from IEF and n ϭ 3 from BN gels) are presented as increased in shoot mitochondria (ϩ) or increased in cell culture mitochondria (Ϫ). Spot numbers shown correspond to protein spots in Figs. 2

New Mitochondrial Proteins Identified in This Study
The 88 protein spots were matched to the sequences of 65 non-redundant proteins that changed in abundance between cell culture and shoot mitochondria. Although most of these were known mitochondrial proteins with defined functions, there were nine proteins in this set that have not been reported previously in mass spectrometry analysis of plant mitochondria. Not all of these proteins were more abundant in shoots. Five showed greater expression in shoots and most likely were not identified before because of their low abundance in cell culture preparations, whereas four were more abundant in cell culture so were most likely identified here because the DIGE analysis focused efforts on the identification of these low abundance protein spots. Two subunits of glycine decarboxylase complex (GDC) were identified by MS for the first time, T protein (At1g11860) and H protein (At2g35370); both were more than 20-fold greater in abundance in shoot mitochondrial samples. New subunits of the branched chain amino acid catabolism pathway were also identified by mass spectrometry, the electron transfer flavoprotein-ubiquinone oxidoreductase (ETFQO, At2g43400) that has previously been shown by GFP tagging experiments to be a mitochondrial protein (53) and the ␣ subunit of the MCCase (At1g03090) that is predicted to be mitochondrial (54) but not previously identified experimentally in mitochondria. We also identified a peptide deformylase that has been shown by GFP tagging experiments to be localized in mitochondria in Arabidopsis (55). This enzyme is proposed to remove the N-formyl group on modified initiator Met residues found at the N terminus of proteins as the first step toward post-translational Met removal by a methionine aminopeptidase. This protein was nearly 7-fold more abundant in cell culture than in shoot mitochondria. A protein annotated as an FAD-linked oxidase (At4g36400) was also identified. This protein is closely related to At5g06580 that has been shown to be a glycolate dehydrogenase present in mitochondria according to GFP tagging experiments (56) and mass spectrometry (57). Both proteins share sequence similarity and the same Pfam domains for FAD/FMN binding that are characteristic of glycolate oxidases (58). Knock-out of At5g06580 in Arabidopsis shows that mitochondrial glycolate oxidation contributes to photorespiration (59). The increased abundance of At4g36400 in shoot mitochondria (Table II) is consistent with these published data. An arginase (At4g08870) was also found that is more than 10-fold more abundant in shoot mitochondria. Biochemical evidence for mitochondrial arginases in plants has been reported (60), and a pea arginase has been localized to plant mitochondria (14). An additional member of the mitochondrial glycoprotein family of proteins was also found (At4g32605) that appears to be 4-fold more abundant in cell culture mitochondria.

Insights into Shoot-enhanced Mitochondrial Metabolism
Considering the breadth of function of the proteins identified, there are a range of mitochondrial metabolic pathways that appear to be altered in shoots compared with cell culture in a manner consistent with the changing role of mitochondria in these tissues.
Increased Photorespiratory Apparatus-The most obvious difference between the mitochondrial samples is the 6 -30fold difference in abundance of most of the components of the GDC (subunits H, T, and P) and the serine hydroxymethyltransferase (SHMT) in shoot mitochondria. These proteins can also be easily observed in Fig. 2 as the major changes in total abundance between the samples. The L protein of GDC is shared with a number of other enzyme complexes in plant mitochondria, namely the pyruvate, 2-oxoglutarate, and branched chain amino acid dehydrogenase complexes. Two genes encode isoforms of this protein in Arabidopsis (At1g48030 and At3g17240); the former has been proposed as the isoform linked to GDC based on enhanced transcript abundance in photosynthetic tissues (61), but in our analysis, both isoforms are probably more abundant proteins in the shoot compared with cell culture mitochondria (Spots 19 and 21). Looking at the isoforms of the other GDC components it is apparent that both P protein isoforms (At2g26080 and At4g33010) are more abundant in shoot mitochondria, whereas of the three H protein isoforms, redundant matches to two are more than 30-fold higher in shoot mitochondria (At1g32470 and At2g35370), and specific matches to the third (At2g35120) are more abundant in cell culture mitochondria. These changes in combination underlie the 10fold greater glycine-dependent oxygen consumption in shoot mitochondria (Table I) but provide a much more detailed insight into the protein isoforms responsible for this specialized function.
Decreased Tricarboxylic Acid Cycle Functions-With mitochondrial respiration in the light significantly shifted to a role in photorespiration, a number of studies have suggested that tricarboxylic acid cycle function is reduced in the light (5,62,63). But as the same tissues must respire at night in a glycineindependent fashion, the nature of the tricarboxylic acid cycle in shoot mitochondria is worthy of further investigation. In Table II a range of differences in abundance of tricarboxylic acid cycle components are apparent. Two isoforms of aconitase are found to differ in abundance: ACON-2 (At4g26970) is more abundant in shoot (Spot 3), whereas ACON-1 (At2g05710) is the dominant isoform in cell culture and is reduced in abundance in shoot (Spot 3). Two very low abundance breakdown products of ACON-1 were identified (Fig. 2, Spots 14 and 20) and varied in abundance between shoot and cell culture, but these protein spots represent only a very small percentage of the total ACON-1 on the gels. Analysis of co-expression data using Gene Expression Angler (64) shows that ACON-1 (At2g05710) is closely co-expressed not with other tricarboxylic acid cycle enzymes but with components of the fatty acid ␤ oxidation pathway, Kat2 (At2g33150; r value ϭ 0.648) and enoyl-CoA hydratase (At4g16210; r value ϭ 0.652), suggesting that it has a role in mitochondrial utilization of citrate synthesized by the peroxisome rather than citrate synthesized inside the mitochondrion. To determine the impact of this switch of isoforms on aconitase function, spectrophotometric assays of aconitase activity were performed (Table III). These assays showed a 5-fold lower aconitase activity in shoot than in cell culture mitochondria. Two enzymes around aconitase in the cycle, namely citrate synthase and 2-oxoglutarate dehydrogenase, were also lower in abundance (Table II) and in activity in shoot mitochondria (Table III). On the non-decarboxylating side of the tricarboxylic acid cycle, subunits of succinyl-CoA synthetase, fumarase, and malate dehydrogenase were all higher in shoot than in cell culture. However, measurements of fumarase and malate dehydrogenase activities could not find changes in maximal activities of these enzymes (Table III).
PDC is a key entry point of carbon into the tricarboxylic acid cycle and is considered to be a key point in regulation as phosphorylation/dephosphorylation controls its activity. This complex is made up of a variety of components, and changes in the abundance of several subunits were observed. The dihydrolipoamide acetyltransferases (E2) form the core of the enzyme, and both single lipoyl and double lipoyl isoforms are present in Arabidopsis (41), but it is apparent that the balance between these isoforms is altered with the double lipoyl form (At3g52200) being more abundant in cell culture and the single lipoyl form (At1g54220) being more abundant in shoot mitochondria. Two isoforms of the catalytic E1␣ subunit are also present in both shoot and cell culture samples, but in both cases the shoot protein spots show a more acidic pI. This is consistent with the known changes in these protein subunits during phosphorylation (65,66), suggesting that more of the E1␣ is phosphorylated and thus inactive in the shoot mitochondrial samples. Assaying PDC in the presence of pyruvate and absence of ATP, to fully dephosphorylate and activate the enzyme, revealed a similar activity in both shoot and cell culture samples (Table III).
Bypasses of Tricarboxylic Acid Cycle for 2-Oxoglutarate Formation-A number of enzymes linking amino acid metabolism and the tricarboxylic acid cycle were highlighted in the DIGE comparisons. Alanine aminotransferase and aspartate aminotransferase were both more abundant in shoot than in cell culture mitochondria, and glutamate dehydrogenase subunits moved on gels to more acidic pI values in shoot mitochondria (Table II). This is interesting because the combination of these enzymes could be used to bypass the early steps of the tricarboxylic acid cycle, feeding amino acid pools and generating 2-oxoglutarate for export from the mitochondrion (Fig. 3).
Although glutamate dehydrogenase catalyzes a reversible reaction, mitochondrial GDH is generally understood to work in the deaminating direction to form 2-oxoglutarate and contribute to NADH production that can then be used by the respiration chain to couple to ATP production (67)(68)(69). Assay of GDH from our mitochondrial samples in both directions showed that although there was no increase in the maximal activity in the deaminating direction in shoots there was a 40% decrease in the maximum aminating activity in shoots (Table III). This change in enzyme properties correlates with the pI shift in this protein (Table II) and would be consistent with an increased potential for net 2-oxoglutarate production of GDH in shoot mitochondria. Aspartate aminotransferase and alanine aminotransferase can also catalyze 2-oxoglutarate formation from glutamate through transamination of pyruvate and OAA. This scenario provides pathways for 2-oxoglutarate formation that are independent of the early down-regulated steps of the tricarboxylic acid cycle that can be used in further assimilation of nitrogen outside mitochondria or could fuel the later steps of the tricarboxylic acid cycle at a higher rate (Fig. 3). This proposal supports previous evidence showing that the tricarboxylic acid cycle partially operates in the light (5,70) and that flux through mitochondrial isocitrate dehydrogenase is not essential for 2-oxoglutarate formation in Arabidopsis (71). Transport properties of isolated plant mitochondria and the principles of the malate-aspartate shuttle are consistent with this pathway (72).
The OAA entering the mitochondria in shoots can be converted into malate by MDH, condensed with acetyl-CoA by citrate synthase, or used to feed the amino acid pool via aspartate aminotransferase (Fig. 3). OAA also facilitates the transport of malate and glutamate thus playing an important role in the balance of photosynthesis and mitochondrial respiration (73,74). Malate in mitochondria can also enter the tricarboxylic acid cycle via either pyruvate or OAA, catalyzed by malic enzyme or MDH, respectively. As alanine aminotransferase catalyzes the reversible transfer of an amino group from glutamate to pyruvate to form 2-oxoglutarate and alanine, it does not only play a role in amino acid metabolism but also in limiting entry into the down-regulated portion of the tricarboxylic acid cycle by depleting the pyruvate pool (Fig. 3).
Differences in Branched Chain Amino Acid Degradation-Four proteins of the branched chain amino acid degradation pathway were altered in abundance but in different directions: ETFQO (Spot 15) and isovaleryl-CoA dehydrogenase (Spot 65) are higher in cell culture, whereas both MCCase and ␣ and ␤ subunits (Spots 13 and 25) are higher in shoot mitochondria. This apparent discrepancy can be understood when the pathway is considered: isovaleryl-CoA dehydrogenase and ET-FQO are required for a step in metabolism common to Leu/ Ile/Val metabolism, whereas MCCase is specifically for Leu metabolism. We have shown previously that respiration by cell culture mitochondria via ETFQO best utilizes Val and the Val-derived organic acid (␣-ketoisovaleric acid) over Leu or Ile derivatives (41), whereas plant leaf mitochondria are best known for catabolism of Leu (75), and knock-out of ETFQO in Arabidopsis leads to accumulation primarily of Leu-derived organic acids in leaves (53,76).

Coupling of Glycolate, Cysteine, Formate, and One-carbon
Metabolism with Photorespiratory Metabolism-A number of other mitochondrial metabolic networks that might be linked to both the photorespiratory pathways and the tricarboxylic acid cycle can also be proposed based on the proteins upregulated in shoot mitochondria (Fig. 4). Glycolate dehydrogenase/oxidase (At4g36400), an FAD-oxidase domain-containing protein that catalyzes the oxidation of glycolate to glyoxylate, is increased in abundance in shoot mitochondria by 2-fold. Although peroxisomal glycolate oxidation is essential for the photorespiratory pathway, mitochondrial glycolate degradation is not essential as mitochondrial glycolate dehydrogenase mutants grow under ambient CO 2 concentration (59). However, it is noteworthy that the abundance of alanine: glyoxylate aminotransferase 2 (AGT2; At4g39660) was lower in shoot than in cell culture mitochondria. AGT2 could couple with glycolate dehydrogenase in converting glycolate into glycine using alanine as an amino donor, providing substrate for photorespiratory GDC. This is consistent with evidence that the release of radioactive CO 2 from 14 C-labeled glycolate is significantly reduced when the concentration of glycine added to isolated mitochondria increases (59). Thus, it is possible that, in the light where the total mitochondrial glycine pool is high in shoot mitochondria, the amount of glycine input into the photorespiratory pool by AGT2 may be restricted.
Glyoxylate can also be readily oxidized non-enzymatically to produce formate (77). This pathway could be the preferred fate of glyoxylate in shoot mitochondria because the abundance of FDH (At5g14780) in shoot mitochondria is significantly higher. In addition, a 5-fold increase in FDH activity in shoot mitochondria compared with cell culture (Table III) suggests a higher influx to the mitochondrial formate pool in shoots. FDH catalyzes the oxidation of formate to produce CO 2 . NADH produced from formate oxidation can then be used by the respiratory chain (78). Oxidation of formate, therefore, may offer an alternative pathway to energy production in mitochondria. The dysfunction of GDC and SHMT can lead to the production of formaldehyde (14,79) that can be readily detoxified to produce less toxic formate by aldehyde dehydrogenase (At3g48000) (Fig. 4). It has also been shown that a substantial amount of serine can be derived from supplied formate (80). Mitochondrial formate can enter the C 1 metabolic pathway via its reaction with tetrahydrofolate, and because 5,10-methylene tetrahydrofolate is incorporated into serine in the reaction catalyzed by SHMT, formate can be used as an alternative to glycine for the formation of serine and thus enter the photorespiratory cycle by bypassing GDC (81). Indeed a GDC mutant in barley has also shown a lightdependent increase in the rate of serine formation via a GDCindependent pathway (82). Recent studies of Arabidopsis overexpressing formate dehydrogenase demonstrated that formate oxidation is the preferred metabolic fate of formate (83,84). Arabidopsis double knock-out mutants of GDC P proteins do not develop beyond cotyledon stage under non-photorespiratory conditions, and even the addition of formate could not improve their growth (85). These results indicated that the GDC route of photorespiration cannot be bypassed, and GDC-independent serine formation can only partially compensate for a deficiency in the glycine cleavage system (80,84).
The cysteine biosynthetic pathway in the mitochondria involves the assimilation of inorganic sulfur and its integration into the amino acid pool. In this study, we identified cysteine synthase (CysS; A3g61440) at 10-fold higher levels in shoot mitochondria. CysS forms part of a cysteine synthase complex with serine acetyltransferase and incorporates serine and acetyl-CoA with free sulfide in a two-step cysteine biosynthetic pathway (86 -88). This pathway thus not only couples with photorespiration through serine utilization but may also act as a control at the entry to the tricarboxylic acid cycle. This enzyme has both O-acetylserine (thiol)-lyase activity and ␤-cyanoalanine synthase (CAS) activity in Arabidopsis (88,89). The CAS activity in CysS allows the detoxification of cyanide, an inhibitor of complex IV, to form a less toxic sulfide, providing an extra line of defense to avoid cyanide inhibition of the cytochrome c oxidase (89). In potato the activity of both CAS and CysS was higher in leaves than in tuber (90). Together with the results presented here it appears that the abundance of CAS and CysS is significantly higher in photosynthetic tissues; this might be influenced by the level of operation of both the tricarboxylic acid cycle and photorespiration.

Alterations in the Respiratory Chain and ATP Synthase
Although 2D (DIGE) IEF/SDS-PAGE allowed the identification of several respiratory chain proteins that were differentially expressed, only a limited number of hydrophobic proteins were resolved due to their lack of solubility under IEF conditions. To assess any possible changes in the structure and composition of the respiratory complexes, digitonintreated mitochondrial fractions isolated from cell culture and shoot were fractionated by differential 2D (DIGE) blue native/ SDS-PAGE (Fig. 5). The composition and relative abundance of the major respiratory complexes were very similar in shoot and cell culture samples (Fig. 5). Five protein spots that differed in abundance between the samples were identified using LC-MS/MS (Table II). Subunit 6 of succinate dehydrogenase (SDH6, At1g08480) appeared to be post-translationally modified in the shoot mitochondrial samples as seen by an apparent increase in the molecular weight of the shoot protein band in BN/SDS-PAGE (Spots 93 and 94; Fig. 5 and Table II) and as increased acidity of the shoot protein spot in IEF/SDS-PAGE (Spots 48 and 49; Fig. 1 and Table II). This protein was identified as one of the plant-specific subunits in complex II (91), but its exact role in complex II function is not certain. Analysis of the peptide spectra obtained from both IEF and BN gel samples from both tissues for this protein failed to identify spectra for altered mass peptides that might reveal the nature of this modification. The other proteins identified were membrane-bound SHMT1 (Spot 92; Fig. 5), prohibitin 2 (Spot 89; Fig. 5), and a minor protein spot of ATP9 that appeared to be more abundant in the shoot sample (Spot 91). Thus there appeared to be relatively little difference in the respiratory chain. The 16-and 24-kDa subunits of complex I (At2g27730 and At4g02580, respectively) were identified by 2D DIGE IEF/SDS-PAGE separation to be differentially expressed, surprisingly in opposite directions in shoot and cell culture. The BN/SDS-PAGE showed no significant difference in the abundance of the assembled complex I or its individual components (Fig. 5). The differences in the amount of these particular subunits may represent variations between shoot and cell culture mitochondria in the state of the protein assembly into complex I and the abundance of these subunits in the matrix. The plant-specific 16-kDa subunit is one of the few complex I subunits that can be readily identified in matrix samples from Arabidopsis mitochondria (92). Most of the other changes seen in the 2D DIGE IEF/SDS-PAGE separation were matched to ␣ or ␤ subunits of the ATP synthase (Table  II). These were all minor protein spot variants for these proteins or low level degradation products: there was no apparent change in the major protein spots containing the bulk of ATP synthase subunits on either IEF/SDS-PAGE or BN/SDS-PAGE.

Do Transcriptional or Post-transcriptional Processes Maintain Differences in the Mitochondrial Proteome?
To assess to what extent changes in mitochondrial protein abundance correlate with elevated levels of transcripts in Arabidopsis shoot and cell culture, we performed Affymetrix GeneChip experiments using triplicate RNA samples prepared from the same plant material we used for mitochondrial isolations and compared our proteomics data with corresponding transcript data. This analysis showed that 3727 genes were significantly up-regulated in shoots (-fold change Ն2) and 5799 genes were up-regulated in cell culture (-fold change Ն2) of the 18,721 gene products that were called positive in the array replicates. Of the non-redundant set of 65 proteins found to be different in abundance between cell culture and shoot mitochondria (Table II), 58 of the transcripts were detected in the array experiments, and these data were matched for further analysis. The change in abundance of transcript was assigned to all the corresponding protein spots detected in the DIGE experiment, excluding protein spots that were previously found to be the minor protein spots for these proteins in 2D gels (data not shown). In addition a further series of protein spots, present in the DIGE analysis but found not to change significantly, were matched to genes by comparison with reference gels for Arabidopsis mitochondria (data not shown); 107 of these could be matched to transcripts detected in the replicated array experiments.
This provided a set of 165 different data points that are shown in a scatter plot of mRNA to protein expression ratio in cell culture and shoot mitochondria (Fig. 6). The diagonal line (y ϭ x) would be expected if the protein and mRNA abundance ratio is perfectly correlated. As expected, the majority of the mitochondrial components are clustered within quadrants B and C, indicative of a positive correlation between protein and mRNA abundance. A parametric correlation analysis of the plotted data using the Pearson correlation method gives a correlation coefficient of 0.49 (p Ͻ 0.0001), indicating that the mRNA and protein abundance ratio in shoot and cell culture mitochondria is positively and moderately correlated. Notably only 22% of the data points fall into quadrants A and D, indicating discordant changes in transcript and protein abundance. This subset includes components involved in the tricarboxylic acid cycle, stress defense, and also branched chain amino acid catabolism. The correlation between transcript and protein abundance in either shoot or cell culture separately was measured using Spearman rank correlation, giving a correlation coefficient of 0.45 and 0.46 (p Ͻ 0.0001), respectively (data not shown). Overall this suggests that protein abundance within these systems is likely to be controlled at a range of different levels but does contain a significant element of control by transcript abundance. This is consistent with a number of other studies broadly comparing mRNA level and protein abundance in a variety of species (93)(94)(95)(96). DISCUSSION We found that nearly 40% of the 165 major proteins observed on 2D gels changed in abundance more than 2-fold between mitochondrial samples from Arabidopsis shoot and cell culture ( Fig. 1 and Table II), suggesting that overall mitochondrial composition in plants is likely to be very dynamic. The complexity of this heterogeneity across a range of tissues in Arabidopsis is also evident from the abundance of transcripts for mitochondrial components in previous studies (17,(97)(98)(99)(100). The heterogeneity of mitochondria has been investigated extensively in the last 2 years in mammals using qualitative and quantitative proteomics analysis of mitochon-dria from a variety of tissues and organs of rat, mouse, and human (49,(101)(102)(103)(104). These analyses sit on the backdrop of many reports of differences in mitochondrial function, structure, and abundance between mammalian tissues (105)(106)(107). The proteomics studies show modification of the abundances of mitochondrial central machinery, a varying percentage of tissue-specific mitochondrial proteins (some of unknown function), and apparent biochemical links between the observed alteration of mitochondrial composition and the known specialization of the cell type. The claimed proportion of the mammalian mitochondrial proteome that is ubiquitously or constitutively expressed and the proportion that is more tissue-specific vary depending on the type and extent of each analysis. The study by Mootha et al. (103) suggests that only ϳ50% of the mouse mitochondrial proteome is ubiquitously present and that ϳ50% is tissue-specific. A more detailed and quantitative study of the mitochondrial matrix proteome from rat tissues, by the same authors, found fewer qualitative (tissue-specific) differences but many quantitative (tissue-selective) differences (49). Thus our plant mitochondrial data are consistent with what is found in mammals not only in terms of the number of proteins changing in abundance but the clear linkage of these changes with altering requirements of the cellular metabolism in the tissues studied (Figs.  3 and 4).
We found in Arabidopsis that there were very few differences in protein abundance of oxidative phosphorylation (Ox-Phos) complexes ( Fig. 5 and Table II). In agreement, far fewer differences in the abundance of components in the OxPhos apparatus between mammalian tissues have been observed perhaps because of the constraint of defined supercomplex stoichiometries between OxPhos complexes. However, large tissue-specific changes in ubiquinone and cytochrome c contents and alterations in maximal activities and catalytic con-stants and also in total cellular content/capacity of OxPhos have been reported in mammals (104,105).
A key functional difference between mitochondria from photosynthetic and non-photosynthetic tissues appears to be substrate choice or availability. Amino acid and carbon metabolism is altered in photosynthetic tissues by the elevated role of GDC and SHMT that act not simply as a pathway of serine biosynthesis in plant mitochondria but as a major carbon recycling route in photorespiration (108). Further the role of plant mitochondria as an important source of 2-oxoglutarate for nitrogen assimilation can vary between plant tissues (5), and because OAA/malate rather than pyruvate is the primary carbon entry point to mitochondrial carbon metabolism in plants (109), the operation of both malic enzyme and malate dehydrogenase increases the flexibility in plants to up-and down-regulate only portions of the tricarboxylic acid cycle in different tissues ( Fig. 3 and Table III). A related pattern of changes in substrate choice is apparent in mitochondria from different mammalian tissues. This is observed as differences in the abundance of components in the tricarboxylic acid cycle and ␥-aminobutyric acid shunt, fatty acid ␤ oxidation, the urea cycle, and amino acid metabolic pathways. These are seen in the proteomics studies (49,(101)(102)(103)(104) but also in functional studies and modeling of mammalian mitochondrial metabolism (101). The metabolic pathways involved differ in plants because mitochondria have a different set of substrate choices in vivo; for example ␤ oxidation of fatty acids does not occur in mitochondria but has been relocated to peroxisomes in plants (110).
Transcriptional Versus Post-transcriptional Control of Heterogeneity of the Mitochondrial Proteome-Comparison of changes in transcript and protein abundance in a relatively short term response to a stimulus generally shows a robustly  6. Scatter plot representation of the correlation between mRNA and protein expression levels. The protein abundance ratio of mitochondrial components between shoot and cell culture samples (y axis) was plotted against the transcript abundance ratio for the same components (x axis) in log 10 scale (n ϭ 165). Each component was assigned to a functional category as indicated. The Pearson correlation coefficient of the logarithm of mRNA abundance ratio versus the logarithm protein abundance ratio is 0.49 (p Ͻ 0.0001). The plotted line (y ϭ x) represents a hypothetical perfect correlation between the two data sets. Four quadrants (A-D) were assigned as indicated. Genes within quadrants A and D are components that are discordant, i.e. having opposite transcript and protein abundance ratios. TCA, tricarboxylic acid. significant positive correlation (111). However, there has been significant debate about the level of correlation between protein abundance and transcripts in comparisons of different tissues or organ types (112). In these cases nearly the whole life history of the materials being compared differ, and thus a much wider variety of factors can influence the degrees of mRNA and protein correlation (113). A variety of specific factors could also negatively influence the significance of correlation in the specific case of plant mitochondria: for example, the presence of multitargeted proteins in mitochondria that may be significantly or even predominantly located elsewhere in the cell (114,115), variation in total mitochondrial content compared with total cell protein in different cell types, and coordination of nuclear and organelle genomes for assembly of respiratory protein complexes and ribosomes that are largely understood to occur post-translationally (116).
However, despite these predicted problems, the correlations experimentally observed are often quite significant using a variety of statistical approaches. The positive correlation of our protein and mRNA abundance data and the only 22% discordance found in our data (Fig. 6) suggest a significant control of protein abundance by transcript abundance. A number of recent reports have compared the expression level of mRNA and protein of mitochondria from a variety of organs in mouse and rat using different statistical models (49,50). Using their primary data, we performed pairwise comparisons using Pearson correlation analysis on the mitochondrial components from different organs (supplemental Data 4) to compare with our own data (Fig. 6). Pairwise correlation analysis between muscle, heart, and liver in rat mitochondria shows a strong positive concordance between protein and mRNA abundance ratio; only ϳ11-30% of the genes are discordant (data from Ref. 49). However, the mitochondrial components in mouse brain, heart, kidney, and liver show a positive and moderate correlation, and again ϳ10 -30% of the genes fell outside the linear relationship (data from Ref. 50). Earlier data comparing mitochondrial components in mouse brain, heart, kidney, and liver found that 75% of the mRNA and protein abundance were significantly concordant (103).
The Importance of Specialization of the Mitochondrial Proteome-Wider heterogeneity of the plant mitochondrial proteome is very likely to occur beyond the differences observed between photosynthetic and non-photosynthetic tissues. This can be seen in the expression of genes for mitochondrial components across a wide range of plant organs and organs at different developmental stages (data not shown; available through Genevestigator), the differential response of mitochondrial genes to biotic and abiotic stresses (117), and the significant positive correlation observed between nuclear transcript abundance and protein abundance in mitochondria from both mammals and plants. Further understanding this heterogeneity of plant mitochondria may give new insights into a wide variety of mitochondria-linked functions and phenotypes in plants.
Although the expression and abundance of the OxPhos components seem to be constitutive in different parts of a plant (Fig. 5), it appears that differences in the photosynthetic capacity can lead to the selective induction or repression of metabolic enzymes to compensate for the variations in energy demands in different tissues (Figs. 3 and 4). Deficiency in OxPhos components has been shown to alter the expression of nuclearly encoded HSPs (118) and AOX (119), suggesting that retrograde signaling pathways can be activated upon changes in energy demand. Many of the nuclearly encoded genes for mitochondrial components share the site II regulatory element that has been suggested to direct tissue-specific expression (17,120,121). However, it is likely that different regulatory properties of the site II element (e.g. positive or negative), binding of transcription factors to this type of promoter, and the presence of other promoter motifs may also have a combinatorial effect on the expression of a gene in a particular tissue. This can be seen by the tissue-specific response to changes in the metabolic role of mitochondria by various developmental and/or environmental challenges as shown in the differences in the AOX expression in leaves and cotyledons upon antimycin A and monofluoroacetate treatment (122) and the tissue-specific response of pyruvate dehydrogenase kinase to gibberellic acid (123).
Cytoplasmic male sterility phenotypes have often been linked to mutations in the mitochondrial genome (124 -126). However, it has been hard to explain why mutations that are shared by mitochondrial genomes in all tissues only induce a fatal phenotype in male floral organs. The main explanation offered has been that different energy demands in particular tissues exert a strain that mutationally damaged mitochondria are selectively unable to meet (125,127). However, heterogeneity of mitochondria protein composition in a particular tissue could also be an explanation if it is incompatible with a mutated genome due to the expression of particular isoforms of proteins, the assembly of particular complexes, and/or the stoichiometry of different components in pathways.
Mutation of genes for nuclearly encoded mitochondrial components have also been reported to yield tissue-specific plant phenotypes, including delayed or modified flowering by loss of citrate synthase or pentatricopeptide repeat proteins (128,129); increased photosynthetic efficiency by loss of malate dehydrogenase (25); altered leaf morphology and/or chloroplast development by loss of complex I, complex IV, or mitochondrial ribosome function (130 -132); inhibition of stomatal function by loss of fumarase (133); and decreased shoot growth at low temperature by loss of AOX (134). These examples indicate that heterogeneity in the expression of the mitochondrial proteome has functional consequences but also that there are possibly boundaries in variability that are detrimental to cross. Understanding these boundaries will be important in defining both formal protein complexes that have not been found by other experiments but also in identifying looser metabolons in mitochondrial metabolism. Boundaries to heterogeneity will also be important in selection of tissues for metabolic engineering that rely on modifying features of mitochondria function: for example, the prospects of increasing vitamin and cofactor synthesis (135,136), making terpenes via the ubiquinone synthesis pathway as protectants against herbivory (137), or altering amino acid or organic acid profiles in plant products (138,139). * This work was supported by Grant CE0561495 from the Australian Research Council (ARC) through the Centres of Excellence Program and by Western Australia State Government support to the Centre for Computational Systems Biology. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
All microarray data are available from ArrayExpress under accession number E-ATMX-31.