Experimental analysis of the rice mitochondrial proteome, its biogenesis and heterogeneity

Mitochondria in rice are vital in expanding our understanding of cellular response to reoxygenation of tissues after anaerobiosis, the cross-roads of carbon and nitrogen metabolism, and the role of respiratory energy generation in cytoplasmic male sterility. We have combined density gradient and surface change purification techniques with proteomics to provide an in-depth proteome of rice shoot mitochondria covering both soluble and integral membrane proteins. Quantitative comparisons of mitochondria purified by density gradients and after further surface charge purification has been used to ensure the proteins identified co-purify with mitochondria and to remove contaminants from the analysis. This rigorous approach to defining a sub-cellular proteome has yielded 322 non-redundant rice proteins and highlighted contaminants in previously reported rice mitochondrial proteomes. Comparative analysis to the Arabidopsis mitochondrial proteome reveals conservation of a broad range of known and unknown function proteins in plant mitochondria with only ~20% not having a clear homolog in the Arabidopsis mitochondrial proteome. Like in Arabidopsis, only ~60% of the rice mitochondrial proteome is predictable using current organelle targeting prediction tools. Use of the rice protein dataset to explore rice transcript data provided insights into rice mitochondrial biogenesis during seed germination, leaf development and heterogeneity in the expression of nuclear-encoded mitochondrial components in different rice tissues. Highlights include identification of components involved in thiamine synthesis, evidence for co-expressed and unregulated expression of specific components of protein complexes, a selective anther-enhanced subclass of the decarboxylating segment of the TCA cycle, the differential expression of DNA and RNA replication components, and enhanced expression of specific metabolic components in photosynthetic tissues.


INTRODUCTION
As rice is the one of the major food supplies for the expanding world population, especially in developing countries, exploiting a molecular understanding of rice biology has the potential to aid humanity in a profound way, as has been seen in the development and use of Vitamin A enhanced 'Golden Rice' (Paine et al., 2005). Mitochondria are essential for all plant species as the energy production factory for ATP production via respiratory oxidation of organic acids and the transfer of electrons to O 2 . But the role and nature of mitochondria in rice takes on special significance due to their early growth habitat in hypoxic or even anaerobic environments (Perata and Voesenek, 2007) and the need for mitochondrial biogenesis during the reoxygenation phase (Millar et al., 2004a;Howell et al., 2007). Rice seed embryos contain highly reduced protomitochondrial structures that are matured to fully functional mitochondria through a complex biogenesis process involving induction of the general import pathway (Howell et al., 2006) and oxygen signalling of transcription (Howell et al., 2007). Further, the farming practise of using hybrid rice production to boost crop yields relies on cytoplasmic male sterile lines which have dysfunctional mitochondria in their pollen and restorer lines that recover mitochondrial function and thus fertility to the hybrid (Eckardt, 2006;Wang et al., 2006). Mitochondria in dicots are known to play critical roles in the synthesis of vitamins and cofactors important for human nutrition including vitamin C (Bartoli et al., 2000;Millar et al., 2003), folate (Ravanel et al., 2001), biotin (Picciocchi et al., 2003) and lipoic acid (Yasuno & Wada 2002) but there is little research in rice to confirm these roles or investigate these processes at the molecular level. Photorespiration in C 3 plants like rice depends on the prevailing CO 2 concentrations and involves a critical role of mitochondria in carbon recycling. But despite attempts to engineer C 4 metabolism in rice and thus eliminate photorespiration (Ku et al., 1999), there has been little analysis of the photorespiratory machinery and related metabolism as integral components in rice mitochondrial function.
The co-ordination of biochemical processes to perform the functions of mitochondria requires many hundreds of different proteins working together in protein complexes, in two membrane systems and several aqueous spaces. The majority of mitochondrial proteins are encoded in the nucleus and transported into mitochondria as cytosolic precursor proteins by the mitochondrial protein import machinery. Prediction tools based on N-terminal portions of protein sequences are unable to predict localisation to a high fidelity (Heazlewood et al., 2005), so the best option is direct experimental analysis of the rice mitochondrial proteome. We have previously reported rice mitochondrial isolation and analysis, using Percoll gradient purification, 2-D IEF/ SDS-PAGE, blue native (BN)-PAGE and LC-MS/MS, and the identification of 122 non-redundant rice mitochondrial proteins (Heazlewood et al., 2003). Subsequently, a separate set of 112 nonredundant rice mitochondrial proteins was identified and listed in the rice proteome database (Komatsu, 2005) using mitochondria isolated by sucrose gradient centrifugation and gel-based spot analysis. However, there is less than 20% overlap between the protein lists reported in these two studies.
The removal of contaminants is essential for accurate curation of subcellular organelle proteomes. While dual-targeting of some proteins to multiple compartments occurs in plants (Peeters and Small., 2001), the question of contamination between compartments needs to be resolved in a quantitative fashion before such a claim can be considered. Isolation of mitochondria using the traditional differential and gradient centrifugation methods based on size and density have been applied to mitochondrial proteomic analysis in a variety of plant species (Kruft et al., 2001;Millar et al., 2001;Bardel et al., 2002;Heazlewood et al., 2003;Heazlewood et al., 2004).
However, a range of contaminants have been found when data obtained by these methods is compared to proteins identified in other cellular organelles by mass spectrometry and/or independent experiments (Heazlewood et al., 2005). Free-flow electrophoresis in zone electrophoresis mode (ZE-FFE) has been used to purify yeast mitochondria to an increased homogeneity based on the surface charge of the organelles (Zischka et al., 2006). Recent separation of plants organelles using ZE-FFE has allow a deeper and more comprehensive analysis of Arabidopsis organellar proteomes and highlighted that some proteins reported as dual-targeted can be explained as contaminants through quantitative analysis (Eubel et al., 2007).
In this study, traditional differential and gradient centrifugation was combined with FFE separation to isolate rice mitochondria. Through the direct analysis of trypsin-digested mitochondrial peptides by liquid chromatography-tandem mass spectrometry (LC-MS/MS) and gelbased analysis of rice mitochondrial proteins and the removal of contaminants by quantitative comparison of mitochondria prior to FFE separation, a refined rice mitochondrial dataset of 322 proteins is presented. The expanded rice mitochondrial dataset is comparable in size and complexity with the previous published Arabidopsis dataset (Heazlewood et al., 2004). Analysis revealed that rice and Arabidopsis mitochondria share conserved energy production and metabolism proteins. Interestingly, a significant proportion of the set of proteins with unknown function identified have clear homologs in Arabidopsis mitochondria. This indicates a range of conserved functions exist that are carried out by unknown function proteins in plant mitochondria that deserve future investigation. The use of this protein dataset to explore rice transcript data has given several insights into rice mitochondrial biogenesis during seed germination and heterogeneity between rice tissues. Highlights include evidence for co-expressed and unregulated expression of specific components of protein complexes, a selective anther-enhanced version of the decarboxylating segment of the TCA cycle, the differential expression of DNA and RNA replication components, and enhanced expression of specific mitochondrial metabolism in photosynthetic tissues.

Purification of isolated rice mitochondria using free flow electrophoresis (FFE)
The integrity of the mitochondrial proteome is largely dependent on the purification of the isolated organelles away from other cellular contaminants. A two-Percoll gradient density separation technique to isolate mitochondria from dark-grown Arabidopsis cells (Millar et al., 2001;Heazlewood et al., 2004) works efficiently in dark-grown rice shoots and yield extracts largely free of contamination by cytosol, peroxisomes, plastids, and other membranes (Heazlewood et al., 2003). To further purify these organelles, washed and Percoll-free organelles were then injected into the separation chamber of the FFE instrument. The visible turbidity pattern in the separation chamber was similar to that observed in separation of Arabidopsis organelles with two main streams and two minor streams (Eubel et al., 2007) and these streams were reflected in the pattern of the 280 nm absorbance of collected fractions in 96-well plates (Supplemental Figure 1). Based on comparison with the separation pattern in Arabidopsis (Eubel et al., 2007), the major peak 2 (fractions 25-36) would likely contain enriched mitochondria and the peaks 1 (fractions 16-22) and 3 (fractions 38-44) would contain plastids and peroxisomes, respectively.
To confirm the distribution of organelles in different FFE fractions, an aliquot of every third fraction was separated by one dimensional SDS-PAGE ( Figure 1A). Western blotting was applied for analysis using antibodies raised against protein markers for mitochondria (mtHSP70), plastids (small subunit of Rubisco, RbcS) and peroxisomes (3-ketoacyl-CoA thiolase, KAT2) ( Figure 1A). indicating that plastids were enriched in those fractions after FFE. The minor RbsS peak that copurified with mitochondrial fractions is most likely due to RbsS from ruptured plastids that is adhered to the mitochondrial membrane, but could be due to a very small proportion of chloroplast having the same surface charge as mitochondria. A similar bimodal distribution of RbsS is seen in Arabidopsis mitochondria purified by FFE (Eubel et al., 2007). The peroxisomal KAT2 marker was enriched in the fractions 33-45, deflecting to the right (cathodic) side of the mitochondrial fractions.
The FFE-purified mitochondrial samples were compared with Percoll gradient prepared mitochondria samples before FFE purification using differential two-dimensional iso-electric focusing (IEF)/SDS-PAGE by labelling proteins with fluorescent Cydyes (Figure 2). The overall spot patterns were consistent with the rice mitochondrial profiles previous published (Heazlewood et al., 2003). Spots with the decreased abundance after FFE (indicated by arrows in Figure 2) were selected for protein identification using LC-MS/MS. The proteins identified as putative contaminants are listed in Table 1. Most of them are plastidic, peroxisomal or cytosolic proteins but also include bovine serum albumin that was added in the mitochondrial extraction buffer.

Gel based analysis of FFE purified rice mitochondrial proteins
To identify the proteins in FFE purified rice mitochondria, preparative IEF/SDS-PAGE gels were run and analysed. A set of 291 abundant spots from these gels (Supplemental Figure 2) were excised and subjected to in-gel digestion followed by MS/MS-based analysis of the resultant peptides. Comparison of spot locations to the quantitative data in Figure 2 ensured these spots were not decreased during the FFE process indicating they were retained or enriched by the mitochondrial purification. This analysis led to identification of a set of 146 non-redundant proteins from rice. We also re-analyzed a set of 89 spots excised from BN/SDS-PAGE gels to separate rice mitochondrial protein complexes as described by Heazlewood et al. (2003), this allowed identification of these proteins against the latest set of predicted rice protein sequences (Osa5).
There were a set of 88 non-redundant protein sequences identified from BN-PAGE separated proteins, 57 were unique protein sequences not identified by the IEF/SDS-PAGE because most were membrane protein subunits of the respiratory chain complexes which do not enter IEF gels.
Taken together, a set of 203 non-redundant proteins were identified based on the gel separations.
Non-gel based analysis of FFE purified rice mitochondrial proteins using complex mixture

LC-MS/MS
Non-gel based LC-MS/MS of rice mitochondrial peptides allowed us to identify the highly hydrophobic, basic, and small or large molecular mass proteins excluded from polyacrylamide gelbased analysis. The whole mitochondrial samples before and after FFE purification were analysed with three independent biological samples using LC-MS/MS. This analysis allowed us to quantitatively compare the ratio of peptide numbers found for each protein before and after FFE purification, which provided additional criterion to remove contaminants. There were a total of 357 non-redundant proteins found by LC-MS/MS from the samples before and after FFE purification ( Figure 3). A set of 56 proteins were only found in the samples before FFE purification, and most of these proteins were contaminants from the plastids, cytosol and peroxisomes (Supplemental Table 2A). For example, 17, 11 and 10 peptides were found for peroxisomal hydroxyacid oxidase 1 (Os07g05820), cytosolic alanine aminotransferase 2 (Os07g01760) and plastidic ATP synthase CF1 beta chain (Osp1g00410), respectively (Supplemental Table 2A). There were 262 proteins for which peptides were found both in samples before and after FFE purification. Nearly 71% of proteins were enriched (ratio of peptide number identified  Table 1) but provide a much deeper assessment of low level contaminants and low abundance mitochondrial proteins. The possible contaminants were removed based on peptide ratios before and after FFE purification and DIGE data as described above. For the 52 proteins only found in FFE purified samples, MS spectral quality, peptide number and protein coverage were also used as additional criteria to distinguish contaminants from mitochondrial proteins (Supplemental Table 2B).

Broad analysis of mitochondrial functions identified in rice
Combining the gel and non-gel based approaches and after removal of contaminants, a nonredundant set of 322 proteins can be conservatively defined as rice mitochondrial proteins (Supplemental Table 3). In this set of rice mitochondrial proteins, 168 proteins were found using gel-based method and 307 proteins were found using LC-based methods (Supplemental Table 3).
Seventy eight of the 122 non-redundant rice mitochondrial proteins reported previously using Percoll gradient centrifugation purification methods (Heazlewood et al., 2003) were confirmed in this study (Supplemental Table 3). Half of the unconfirmed proteins (21 of 43) from Heazlewood et al (2003) were proteins predicted from retrotransposon sequences and unknown function proteins. Surprisingly, only six proteins were confirmed from a set of 112 non-redundant proteins listed as mitochondrial in the rice proteome database, (Komatsu 2005; http://gene64.dna.affrc.go.jp/RPD/), apparently due to the heavy contamination of the rice mitochondrial samples used to generate these identifications.
Each rice mitochondrial protein was assigned to one of 17 functional categories (Supplemental Table 3, Figure 4). In this dataset, known function proteins were highly represented by those involved in energy production (Complex I to V, 22%) and metabolism (TCA and general metabolism, 28%) (Figure 4), while the proteins with unknown function represented 17% of the mitochondrial protein set ( Figure 4). The number of proteins involved in energy production and metabolism are very similar in the rice and Arabidopsis mitochondrial datasets ( Figure 4). There are less proteins identified to be involved in ETC assembly and signalling, stress defence, carriers and transporters, protein import/fate and unknown proteins in the rice compared with the Arabidopsis mitochondrial datasets ( Figure 4).

Prediction of rice mitochondrial proteins
This experimentally determined rice mitochondrial protein set provides an opportunity to test the sensitivity of different organelle targeting prediction programs in rice. When our set of 313 nucleus-encoded rice mitochondrial proteins were analysed, the accuracy of mitochondrial prediction by four leading targeting prediction programs ranged from 61% to 66% (Table 2), which was higher than the 40-50% prediction rate observed for the Arabidopsis mitochondrial protein set (Heazlewood et al., 2004). The high accuracy of prediction by the four programs in this study may be due to the higher purity of the FFE purified mitochondrial sample and the removal of proteins during FFE that only bind to the surface of the mitochondrial outer membrane, but are not imported.
Proteins most often predicted by these programs belonged to ETC assembly, DNA and RNA replication, TCA cycle, and complex III (Figure 5, Supplemental Table 3). On the other hand, the proteins from the functional groups of carriers, protein import and fate, complex V and unknown proteins had a low proportion of proteins predicted to be mitochondrially localised (Figure 5, Supplemental Table 3). Some of these mitochondrial proteins have known internal targeting sequences rather than classical N-terminal sequences and these are not detected by these prediction programs. However, for many proteins the mechanism of sorting to mitochondria is still unknown.

Global analysis of expression pattern of the identified rice mitochondrial proteins
The plant mitochondrial proteome changes during development of plant organs as well as differing in various cell and tissue types. We have already shown in Arabidopsis that 40% of proteins change more than 2-fold in abundance when comparing mitochondrial proteomes from photosynthetic and non-photosynthetic tissues (Lee et al., 2008). The combination of proteome and gene expression data can provide a more global understanding of gene functions in particular organs, developmental stages and in response to stresses. To analyse the gene expression pattern, we extracted the available rice microarray data from the NCBI gene expression omnibus (http://www.ncbi.nlm.nih.gov/geo) of 6 independent studies with relevance for mitochondrial function (Walia et al., 2005;Jain et al., 2007;Lasanthi-Kudahettige et al., 2007;Li et al., 2007;Walia et al., 2007;Ribot et al., 2008) and our own rice microarray data during the first 24 h of germination. From 322 mitochondrial proteins identified, 306 had representative probeset IDs on the microarrays. Analysis of the combined microarray expression data was described in Materials and Methods. The genes were divided into 12 hierarchical clusters ( Figure 6A) based on differential expression patterns in different organs or treatments with 7 major clusters containing more than 16 genes in each cluster and 5 minor clusters containing less than 10 genes in each cluster. Cluster 1 was defined by high expression of photorespiratory components in leaf tissue, that were largely absent in most other samples except young seedlings. Cluster 4 was defined by expression in anoxic grown 4 day coleoptiles, and cluster 5 by a set of genes that peak in expression in mature anthers. Cluster 6 displayed high expression in suspension cells and 12 and 24 h imbibed seeds and to a lesser extent in developing seeds. Clusters 8 and 11 were defined by expression in most of arrays except 0 to 3 h imbibed seed, while cluster 10 members were evenly expressed in all rice tissues. The proportion of each cluster in different functional classes is summarized in Figure   6B. Detailed information is given in Supplemental Figure 3 and Table 5 and highlights are described in the following sections based on the functional classification of proteins and their expression patterns.

The respiratory apparatus and its expression
The mitochondrion is an energy factory for ATP production coupled to the respiratory oxidation of organic acids and the transfer of electrons to O 2 . Over 71 proteins of the five electron transport chain complexes and 28 protein subunits of TCA cycle enzymes have been identified (Figure 4), representing 31% of the rice mitochondrial protein set. Most of these proteins had orthologs in Arabidopsis and most have also been experimentally shown to be mitochondrial proteins in Arabidopsis (Supplemental Table 3). This highlights that rice and Arabidopsis have a very conserved composition of the respiration chain complexes and the TCA cycle. Six proteins involved in the alternative pathway respiration were also found, namely cytosolic facing NADH dehydrogenases that donate electrons to UQ and bypass complex I, and components of the electron-transfer flavoprotein pathway that reduces UQ and is linked to the matrix branched-chain amino acid degradation pathway (Supplemental Table 3) (Taylor et al., 2004). No alternative oxidases (AOX) that oxidise UQH 2 and consume O 2 were found in this rice study, which is consistent with the very low expression of AOX in rice shoots under normal conditions (Saika et al., 2002) and the very low cyanide-insensitive respiratory rate of isolated rice mitochondria (Heazlewood et al., 2003;Millar et al., 2004a). The genes encoding the subunits of the oxidative phosphorylation complexes (Complex I-V) were highly expressed across all tissues (mainly cluster 11 in Figure 6), confirming the essential function of respiration chain for energy production. Comparison of the expression profiles of the genes for the subunits in each respiratory complex separately revealed in each case a core of co-expressed subunits, and a series of apparently aberrantly expressed subunits (Supplementary Figure 3).
A series of genes encoding TCA cycle subunits were highly expressed in the anther, these were nearly all components of the pyruvate dehydrogenase complex (PDH) and the initial steps of the TCA cycle (citrate synthase, aconitase and isocitrate dehydrogenase(ICDH)). In most cases another isoform was also in our list of TCA cycle components and had a much more ubiquitous gene expression pattern (most notably, PDH E1α Os02g50620 vs Os12g08260, PDH E1β Os09g33500 vs Os08g42410, PDH E2 Os06g01630 vs Os02g01500, aconitase Os03g04410 vs Os08g09200, and ICDH Os04g40310 vs Os02g38200). In contrast to this, later steps of the TCA cycle such as malate dehydrogenase (Os01g46070, Os05g49880) were more evenly expressed across tissue types (Supplemental Figure 3). Mature anthers might thus possess a very high metabolic activity for energy production during pollen formation, potentially this is mediated by a specific highly expressed form of the TCA cycle in pollen.

General metabolism in rice mitochondria
Plant mitochondria are also involved in synthesis of vitamins, cofactors and lipids, metabolism of amino acids, photorespiratory glycine oxidation, and export of organic acid intermediates for other cellular biosynthesis. Rice mitochondrial proteins involved in these processes were also evident in our protein lists. In total, 64 proteins were identified as being involved in a range of metabolic pathways (Supplemental Table 3). Subdivision of this functional classification showed 16 proteins involved in amino acid metabolism, seven proteins involved in aldehyde/alcohol metabolism, six proteins involved in lipid synthesis and six proteins involved in metabolism of nucleotides had been identified. Photorespiratory genes (glycine decarboxylase complex subunits and serine hydroxymethyltransferase) were selectively expressed in leaf tissues (cluster 1, Figures 6A), while components linked to C1-metabolism -glyoxylate, formate and tetrahydrofolate metabolism -did not show leaf enhanced expression profiles (Supplemental Figure 3). Two isoforms of the H protein of glycine decarboxylase complex (Os06g45670, Os02g07410) did not show leaf enhanced expression patterns, but were broadly expressed with the C1 metabolism genes (Supplemental Figure 3). The role of glycine decarboxylase in plant mitochondrial C1 metabolism outside of its photorespiratory role is still largely unexplored in plants. A 4-methyl-5-thiazole monophosphate biosynthesis protein (Os01g11880) predicted to be involved in thiamine biosynthesis, was enriched after FFE purification (Supplemental Table 3), representing the first component of thiamine synthesis found in rice mitochondria.

DNA replication, transcription and translation
There were 19 proteins in the DNA replication and transcription category (Supplemental Table 3).
Among them, five were DAG-like protein and eight were pentatricopeptide repeat (PPR) proteins.
Genetic evidence shows DAG proteins influence DNA synthesis and altering chloroplast differentiation (Bisanz et al., 2003), but while DAG proteins have also been reported in the OsPPR_02g58300 is an ortholog of the P class PPR336 protein At1g61870. PPR336-like proteins are known to be extrinsically attached to the inner mitochondrial membrane and to be associated with polysomal RNA (Uyttewaal et al., 2008). OsPPR_02g02020 is an ortholog of At2g37230, which was found in the thylakoid membrane Arabidopsis (Peltier et al., 2004) even though these proteins were predicted to be mitochondrial in both rice and Arabidopsis. To our knowledge the possibility of orthologous PPRs swapping location between mitochondria and chloroplasts in different plants species or being dual-targeted has not been reported. OsPPR_04g09530 is, to our knowledge, the first DYW type PPR found by mass spectrometry. This class of PPR often have an RNA editing role and are typically expressed at very low levels. Further studies are required to investigate the function of these eight PPRs in rice mitochondria.
Fourteen proteins are listed in the group of proteins involved in translation. Six subunits of the putative mitochondrial ribosome were identified, the L1, L27 and L30 of the 50S complex and S12, S18 and S19 of the 30S complex. Even with the small number of predicted subunits of the mitochondrial ribosome (~60), clear differences between Arabidopsis and rice are apparent. The nuclear encoded S12 protein in rice (Os12g33930) is most similar to the mitochondrial encoded S12 in Arabidopsis (AtMg00980), while the mitochondrial-encoded S19 in rice (OsM1g00450) is orthologous to a nuclear-encoded ribosomal subunit in Arabidopsis (At5g47320). Extensive studies have been performed on ribosomal proteins shifting or swapping their location during recent evolution in plants (Adams et al., 2000;Adams et al., 2002). The S19 gene has been transferred to the nucleus in many cereals, with the notable exception of rice where an intact gene and transcript has been reported in mitochondria (Fallahi et al., 2005). Here we show clear evidence for this S19 protein accumulating in rice mitochondria. The S12 is mitochondrially-encoded and co-transcribed with nad3 in many dicots and monocots (Perrotta et al., 1996) but our match here is to a nuclear gene. So it seems that in rice it has been transferred to the nucleus, this has also be reported for the

Components of protein import and fate
There were 19 proteins identified involved in protein import and fate (Supplemental Table 3).
These included proteins involved in protein import and sorting, pre-sequence cleavage, and proteolysis and all of them had clear orthologs in the Arabidopsis mitochondrial proteome. The translocase of the outer membrane (TOM) was represented by TOM40, 20 and 22 subunits while the only TIM components found were the intermembrane space members of the carrier import pathway, Tim8, 9 and 13 homologs. Lon, ClpX and FtsH homologs were found, representing the three main classes of mitochondrial proteases in plants. The expression of genes encoding these proteins were mainly grouped into clusters 6 and 10, with notable expression in suspension cells, seeds and in embryos during the early stages of germination ( Figures 6, Supplemental Figure 3).
These results were consistent with our reports of the substantial mRNA pool in dry rice seeds for the genes encoding import components (Howell et al., 2006).

Heat shock and stress response proteins
There are 15 heat-shock proteins and putative or well known molecular chaperones listed in our current dataset (Supplemental Table 3). These included the classical 60/10, 70 and 80 kDa chaperone classes. While these chaperone and heat shock protein classes are sometimes components induced by stresses with roles in stress-tolerance (Schöffl et al., 1998), we found little evidence for this in the transcriptional data from stress responses (such as drought, salt and cold).
Instead, the expression of genes encoding these proteins was grouped into clusters 6 and 10, due mainly to high expression in suspension cells, seeds and during early germination (Figures 6, Supplemental Figure 3). This suggests primary transcriptional regulation of these proteins in response to cell division, expansion and growth rather than stress tolerance.
There were 9 proteins with putative roles in stress response or oxidative stress in the current dataset (Supplemental Table 3). Mitochondria are often exposed to self-generated ROS, primarily through  Table 3). Most of genes encoding stress-responsive proteins and anti-oxidants were highly expressed in the suspension cells (cluster 6, Figures 6).
Surprisingly, except for a few selective proteins that were induced (several of the ascorbate/glutathione cycle components), these components were not a group positively induced in response to the different environmental stresses tested, including drought, salt and cold (Figures 6, Supplemental Figure 3).

Proteins of unknown function in rice mitochondria
A total of 55 proteins were identified as proteins with unknown functions. Thirty-five of these rice unknown proteins were predicted as mitochondrial localised by at least one targeting prediction program, but only five of these unknown proteins have been identified in our previous study of the rice mitochondrial proteome (Heazlewood et al., 2004, Supplemental  These results indicate a substantial conservation of unknown function proteins between mitochondria from different plant species, even though there is great divergence between unknown function mitochondria proteins from different eukaryotic lineages (Heazlewood et al., 2004). The identification of these conserved unknown function mitochondrial proteins provides a focus for the use of reverse genetics to identify novel mitochondrial functions in plants.
The overall expression pattern of genes encoding proteins with unknown function was dispersed among other mitochondrial components as shown in Figure 6,

DISCUSSION
In this study, the combination of FFE base plant mitochondria separation (Eubel et al., 2007) and traditional differential and gradient centrifugations was applied to determine the rice mitochondrial proteome. This was achieved by organellar fraction selection using marker antibodies as well as differences in spot abundance in DIGE and peptide number ratios before and after FFE purification.
The final rice mitochondrial data set provides evidence to confirm most of the fundamental biological processes in mitochondria previously uncovered in Arabidopsis (Figure 4; Heazlewood et al., 2004). For example, the majority of mitochondrial proteins involved in energy production have highly conserved amino acid sequences between rice and Arabidopsis (Supplemental Table 3). A notable exception is the subunits of succinate dehydrogenase or complex II. Differences between complex II in higher plants and mammals have been known for sometime, most notably in the similarity of sequences for SDH1 and SDH2, but great divergence in sequences for the hydrophobic membrane anchor proteins SDH3 and SDH4 (Burger et al., 1996). In dicots, plant mitochondrial complex II appears to have 4 additional subunits in BN-PAGE purified preparations, termed SDH5, 6, 7 and 8 (Millar et al., 2004b). In rice we identified classical SDH1 and SDH2 subunits, but the SDH3 isoform we identified Os02g02940 is very poorly related to SDH3 At5g09600 in Arabidopsis. We did not identify the small SDH4 protein in rice, and the annotated SDH4 gene from rice (Os01g70980) bares little resemblance at all to the Arabidopsis SDH4. Furthermore, we found conserve homologs for SDH5, 6 and 7 (Supplemental Table 3 Glycolytic enzymes are traditionally regarded as cytosolic abundant proteins. Interestingly, 5 to 10% of the cytosolic isoforms of each glycolytic enzyme, at least in Arabidopsis, is associated with the outer membrane surface of mitochondrion (Giegé et al., 2003). It appears that glycolytic enzymes are associated dynamically with mitochondrial to support respiration and that substrate channelling restricts the use of intermediates by competing metabolic pathways (Graham et al., 2007). In plants, hexokinase is associated with the outer mitochondrial membrane (Dry et al., 1983;da-Silva et al., 2004;Damari-Weissler et al., 2006;Kim et al., 2006). In the present study, three hexokinases were identified as mitochondrial proteins (Os01g53930, Os01g71320, Os05g44760, Supplemental Table 3). In animal cells, mitochondrial-associated hexokinase play a pivotal role in the regulation of apoptosis (Downward, 2003;Birnbaum, 2004;Majewski et al., 2004). In plants such as Nicotiana benthamiana, a similar function of mitochondrial-associated hexokinases in the control of programmed cell death has been reported (Kim et al., 2006), suggesting a link between glucose metabolism and apoptosis in plant cells. Apart from hexokinases, however, other glycolytic enzymes, such as fructose-bisphosphate aldolase (Os01g02880), pyruvate kinase (Os03g20880), D-3-phosphoglycerate dehydrogenase (Os04g55720) and triosephosphate isomerase (Os09g36450) were only found in rice mitochondrial samples before, but not after FFE purification (Supplemental Table 2A). The separation of glycolytic enzymes from Arabidopsis mitochondria by FFE has also been observed (Eubel et al., 2007) suggesting that the FFE process allows a removal of peripherally-associated cytosolic proteins from the organelle surface.
The relatively low levels of transcript abundance observed for the majority of genes in mature leaf tissue (last four columns in Figure 6A A range of proteins identified that are involved in aldehyde/alcoholic metabolism might be related to alcoholic metabolism in rice mitochondria. Ethanolic fermentation is not only observed in anaerobic plant tissues, but also in aerobic tissues such as anthers (Tadege et al., 1999). The mechanism and regulation of aerobic ethanolic fermentation in the anther is still unclear, but it likely involves a new steady state in which pyruvate is distributed between PDH and an aerobic TCA cycle in mitochondria and a cytosolic fermentation pathway involving ADH and PDC. Our data suggests metabolism of pyruvate might be changed in anther mitochondria by differential expression of PDH complexes and early steps of the TCA cycle which might co-exist with aerobic alcoholic fermentation. The PDH complex is the key entry point of carbon into the TCA cycle and is considered to be a key point in regulation as phosphorylation/dephosphorylation controls its activity. It would be particularly interesting to investigate whether there are any differences in the kinetics of PDH between pollen and other tissues that allow low affinity PDC (K m ~mM) and higher affinity PDHs (K m ~µM) to simultaneously utilise a common pyruvate pool. Aldehyde dehydrogenases (ALDH) have been reported as major mitochondrial proteins in pea leaves and roots (Bardel et al., 2002) and two ALDHs were also observed in the Arabidopsis' mitochondrial proteome (Millar et al., 2001;Heazlewood et al., 2004). In the present study, three abundant rice mitochondrial ALDHs (Os06g15990, Os02g49720; Os05g45960) were found. Os06g15990 and Os02g49720 are orthologs of rf2a and rf2b encoding ALDHs in maize, respectively (Liu and Schnable, 2002). In maize, rf2a is involved in restoring male fertility to Texas cytoplasmic male sterility plant (Cui et al., 1996). The mechanism of restoration of male fertility in maize by rf2 encoding ALDH is still unclear. One possible role of rf2-encoded ALDH is the removal of the products of lipid peroxidation that would be expected to accumulated preferentially in T-cytoplasm cells if these cells produce more reactive oxygen species than normal cytoplasm cells (Liu et al., 2001;Moller, 2001b). The mitochondrial ALDH (Os02g49720) can also be induced in rice seedlings by submergence, having a very similar pattern of expression to classic anaerobic proteins such as ADH1 and PDC1 (Nakazono et al., 2000). These authors suggested induction of mitochondrial ALDH in rice under submergence might protect mitochondrial damage by diffused acetaldehyde produced by PDC during aerobic ethanolic fermentation (Nakazono et al., 2000).
Logically, the restoration of maize male fertility by mitochondrial ALDH might be partially due to protection of pollen mitochondria from acetaldehyde during aerobic ethanolic fermentation. There are also four sex determination TASSELSEED-2 like proteins (Os07g40250, Os07g46840, Os07g46920, Os07g46930) found in our proteome set, which encode short-chain alcohol dehydrogenases (SDRs) based on annotation. Their maize homologue, TASSELSEED-2, is required for stage-specific floral organ abortion (DeLong et al., 1993) and can reduce a broad range of substrates tested including steroids, dicarbonyl and quinone compounds (Wu et al., 2007). The substrate range of these rice mitochondrial SDR-like proteins and their potential involvement in alcoholic metabolism in specific rice organs deserves further investigation. Further exploring the expression and kinetics of the components of rice mitochondrial pyruvate, aldehyde and alcohol metabolism identified here may uncover a key mechanistic basis of cytoplasmic male sterility.
Thiamine (vitamin B1) synthesis in plants was thought at one time to be plastid specific (Belanger et al., 1995), while in yeast it is a mitochondrial process (Eijssen et al., 2008). It is synthesized through the convergence of two independent biosynthetic pathways, one that synthesises 2-methyl-4-amino-5-hydroxymethylpyrimidine pyrophosphate to form the pyrimidine moiety and another that synthesises 4-methyl-5-(-hydroxyethyl) thiazole phosphate to form the thiazole moiety, but the specific number and identity of the enzymes involved in plants is still unclear. In Arabidopsis, the product of the thiamine biosynthesis gene (THI1), that catalyses glycine, NAD + and a sulphur donor into hydroxyethylthiazole, is dual-targeted to chloroplasts and mitochondria (Chabregas et al., 2001;Chabregas et al., 2003). A 4-methyl-5-thiazole monophosphate biosynthesis protein (Os01g11880) that is unrelated to THI1 but is a probable THIJ (catalysing phosphorylation of hydroxymethylpyrimidine (HMP) to HMP monophosphate in thiamine biosynthesis) was enriched after FFE purification (ratio = 0.33) and is predicted as a mitochondrial protein by three different targeting programs (Supplemental Table 3). This finding provides further experimental evidence that plant mitochondria are involved in thiamine synthesis in plants.
The plant mitochondrial proteome is a changing entity over time, and in different cells and tissues. This is evident by looking at mitochondria from photosynthetic and non-photosynthetic plant tissues (Bardel et al., 2002;Lee et al., 2008) and at the mitochondrial components in the Arabidopsis proteome map generated from different organs, developmental stages and undifferentiated culture cells (Baerenfaller et al., 2008). The combination of proteome and gene expression on this scale in rice will greatly benefit our global understanding of the functions of genes for mitochondrial proteins in particular organ and developmental stages. Future combinations of data sets focusing on mitochondrial function will allow the common patterns of expression and thus putative regulators of mitochondrial biogenesis, stress-response and other aspects of nuclearmitochondrial interaction in rice to be uncovered.

Growth of rice seedlings
Batches of 200 g of (Oryza sativa cv Amaroo) seed were washed in 1% (v/v) bleach for 10 min, rinsed in distilled water, and grown in the dark in vermiculite trays (30 × 40 cm) at a constant 30 o C, watered daily, and the shoot tissues harvested at 10 d for mitochondrial isolation.

Free-flow electrophoresis
Free-flow electrophoresis was performed using the BD Free

One-dimensional SDS-PAGE and immunoblotting
Precast gels with 12% (w/v) acrylamide and 1 mM Tris-HCl (Biorad) were used for analytical purposes and western blotting. Protein assays (Bradford, 1976) were performed on pooled FFE fractions. Proteins were transferred onto nitrocellulose membranes and probed with 1:5000 dilution of the primary antibodies mtHSP70:PM003 from Dr. Tom Elthon (Lincoln, NE, USA); Kat2 (Germain et al., 2001); RbcS, raised against tobacco Rubisco in rabbits, supplied by Dr Spencer Whitney (Australian National University). Horseradish-peroxidase-conjugated secondary antibodies at a dilution of 1: 10000 were used for the chemiluminescent detection of the immune signal.

Two-dimensional gel electrophoresis
Mitochondrial protein samples (700 µg) were extracted by addition of cold acetone (-20 o C) to a final concentration of 80% (v/v). Samples were stored at -80 o C for 4 h and then centrifuged at 20 000g at 4 o C for 15 min. The pellets were resuspended in IEF sample buffer (7M urea, 2 M thiourea, 4% (w/v) CHAPS, 40 mM Tris-base, pH 8.5). Aliquots of 450 µL were used to re-swell immobilized pH gradient strips pH3-10 NL; 24 cm (GE healthcare) according to the manufacturer's instructions. The strips were run for 24 h in Ettan IPGphor3 (GE healthcare) according to the manufacturer's instruction. The strips were then transferred to an equilibration buffer (50mM Tris-HCl (pH 6.8), 4 M urea, 2% (w/v) SDS, 0.001% (w/v) Bromophenol blue and 100 mM β mecaptoethanol) and incubated for 20 min at room temperature with rocking. After a brief wash in 1 x gel buffer, the stripes were transferred on top of 12% acrylamide glycine gels and covered with 1.2% (v/w) agarose in gel buffer. Second dimensional gels were run at 50 mA per gel for 6 h.

DIGE two dimensional IEF/SDS-PAGE
Samples (50 µg x gel buffer, the stripes were transferred to the top of 12% (w/v) acrylamide glycine gels and were covered with 1.2% agarose in gel buffer. Second dimensional gels were run at 50 mA per gel for 6 h. Proteins were visualized on Typhoon TM laser scanner (GE Healthcare) and image comparison was performed using the DECYDER TM software package (version 6.5; GE Healthcare). Three independent experiments were performed and each of the resulting three gel sets were first analysed using differential in-gel analysis (DIA) mode DECYDER prior to a comprehensive biological variance analysis (BVA) including all three gel sets. Gel spots were filtered according to their presence, average ratio in abundance. Gel pictures were electronically overlaid using the IMAGE QUANT TL TM software (GE Healthcare).

Analysis of peptides from in-gel digested protein samples
Protein samples to be analysed were cut from the gels and were in-gel digested according to the method described by (Taylor et al., 2005). Samples were then dried in a vacuum centrifuge, number of hits' set to 'AUTO' and 'ion score cut-off' at 37. The significance threshold P≤0.05 and 'Require bold red' were also set. In order to estimate the false-positive rate (FPR) of this protein strategy, a single concatenated file was generated by MASCAT (Agilent Technologies) which comprised of all the MS/MS output data. The concatenated file was then used to search against Rice_osa5 (target), reversed (decoy) and randomized (decoy) database using the above search strategy. The false-positive rate in target-decoy searches was found to be 4.7% for peptides with ion scores > 37, which was calculated using the equation described previously (Elias et al., 2005).

Analysis of peptides from whole organelle digests
Whole organelle protein extracts were digested overnight at 37°C in the presence of trypsin and insoluble components were removed by centrifugation at 20,000 g for 5 mins. Samples were analysed on an Agilent 6510 Q-TOF mass spectrometer with an HPLC Chip Cube source (Agilent Technologies were also set. The false positive peptide identification rates under the matching criteria described above were estimated at 1.5%. The total number of times each protein was identified after FFE purification using non-gel based methods in four independent runs are listed in the Supplemental Table 4 with the removed contaminants listed in Supplemental Table 2B.

Rice germination microarray analysis and comparison with public rice microarray data
Dehulled, sterilized rice seeds (Oryza sativa cv. Amaroo) were grown under aerobic conditions in the dark at 30 ºC as previously described (Howell et al., 2006). RNA isolation, cDNA synthesis and quantitative RT-PCR were conducted according to methods as previously described (Howell et al., 2006). Transcriptomic analysis was performed using Affymetrix GeneChip™ Rice Genome Arrays (Affymetrix, Santa Clara, CA) and three biological replicates were analysed for each time point. Preparation of labelled cRNA from 2-3 μ g of total RNA, target hybridisation, as well as washing, staining and scanning of the arrays was carried out exactly as described in the Affymetrix  Table 5).   Figure 1) and selected protein spots identified from DIGE 2D gels that decreased in abundance after FFE purification of mitochondria ( Figure 2). Proteins were identified by tandem mass spectrometry (MS/MS), the predicted molecular mass in kD (MM) and isoelectric point (pI) of the match is shown along with the MOWSE score (p < 0.05 when score >37), number of peptides matched to tandem mass spectra and the percentage coverage of the matched sequence. The localisation of the identified proteins were listed based on their description. The average ratio in the section on DIGE represents the decrease of spot abundance after FFE purification (Figure 2).    Table 1. The numbers on the right represent molecular mass in kD.    predicted to be localised in mitochondria using the four major prediction programs listed in Table 2.
In each bar, white region was not predicted by any program, black region was predicted by all four programs, while increasingly grey bars indicate mitochondrial prediction by 1,2 or 3 programs. The functional classifications of total 322 proteins were taken from Supplemental Table 3. Euclidean distance of the 306 mitochondrial genes across all the Affymetrix rice genome arrays carried out on various tissue types and conditions. Ten distinct clusters were coloured and numbered (data shown in Supplemental Table 5). The proportion of components from each FUNCAT that are present in each of the 12 clusters define in (A) in shown in (B), the number of components in each FUNCAT is shown above each column.

Supplemental Table and Figure Index
Supplemental Table 1. Identification of proteins from FFE purified rice mitochondria using MS/MS, after separation of proteins by IEF-SDS/PAGE. Table 2. Proteins removed as contaminants in rice mitochondrial preparations.

Supplemental
Supplemental Table 3. Final list of proteins classified as rice mitochondrial proteins.
Supplemental Table 4. The total protein identifications found by 4 independent non-gel based LC runs after FFE purification of rice mitochondria Supplementary Table 5. Hierarchical clustering of all the arrays using average linkage clustering based on Euclidian distance, which was used for generating Figure 6.