INTRODUCTION

TOP ABSTRACT INTRODUCTION RESULTS AND DISCUSSION CONCLUSIONS MATERIALS AND METHODS LITERATURE CITED

Plastids exist in a wide range of differential forms, including proplastids, chloroplasts, etioplasts, amyloplasts, leucoplasts, and chromoplasts, depending on the developmental stage and function of the plant cells in which they reside. To a large extent, the types of plastids that are carried by cells determine the metabolic function and products of the particular plant tissue (Kirk and Tilney-Bassett, 1978).

One constant feature among the various types of plastids is the double membrane envelope structure that surrounds the organelle. The envelope, especially the inner envelope, effectively separates plastid metabolism from that of the cytosol. At the same time, almost all carbon and a major flux of other metabolites, various polypeptides, and signals must move through the envelope to coordinate and integrate metabolism with the entire cell. Plastid envelopes contain protein transport machinery (Schnell, 1995; Cline and Henry, 1996; Heins et al., 1998), and are a major site for membrane biogenesis. Metabolite transporters from chloroplast and/or nongreen plastids accommodate the requirements of the different photosynthetic or heterotrophic tissues (Kammerer et al., 1998; Neuhaus and Emes, 2000). Membrane constituents (glycerolipids, pigments, and prenylquinones) are synthesized on the envelope membrane as well as the porphyrin ring and phytyl chain used in chlorophyll synthesis, and the enzymes required for chlorophyll breakdown in senescing plastids (for review, see Joyard et al., 1998). It is also hypothesized that fatty acids that are synthesized within the stroma are exported through a channeled system on the envelope (Pollard and Ohlrogge, 1999; A.J. Koo, J.B. Ohlrogge, and M. Pollard, unpublished data). The acyl group modification of lipids such as desaturation takes place on the inner envelope and lipid-derived signals are produced on the envelope membranes (Miquel and Browse, 1992).

Despite the importance and complexity of the plastid envelope, only a small fraction of envelope proteins have been purified or characterized (Joyard et al., 1998). Recently, several groups have initiated proteomic studies to identify the constituents of plant subcellular organelles and membranes (Santoni et al., 1998; Sazuka et al., 1999; Seigneurin-Berny et al., 1999; Ferro et al., 2000; Peltier et al., 2000). Most of these studies have been based on one- or two-dimensional electrophoretic methods followed by mass spectrometric analysis of peptides derived from fractionated proteins. The continued development and improvement in these methods resulted in the identification of hundreds of proteins, particularly from organisms with sequenced genomes. Nevertheless, there are some limitations in current proteomic technologies. In particular, many membrane proteins are found in very low abundance and/or expressed only in certain tissues, making comprehensive samples difficult to obtain. In addition, many integral membrane proteins remain difficult to solubilize and often do not resolve well by gel electrophoresis or isoelectric focusing (Adessi et al., 1997; Santoni et al., 2000). Finally, trypsin, most frequently used to digest proteins before mass spectrometric analysis, has relatively few sites in the more hydrophobic integral membrane proteins. Together, these limitations make it likely that a significant fraction of the integral proteins present in membranes will remain difficult to detect by most proteomic technologies.

One of the first steps in understanding the function of a gene is to determine the subcellular localization of the gene product (Somerville and Dangl, 2000). To augment the proteomic approaches described above, an additional opportunity for the discovery of putative localization of proteins is now possible due to complete sequencing of the Arabidopsis genome (Arabidopsis Genome Initiative, 2000). To determine subcellular localization of plastidial proteins, one can make use of certain features of plastid-targeted proteins. The plastid genome itself encodes only about 120 single-copy genes and, therefore, must rely on proteins from the nucleus. The protein constituents that are encoded in the nuclear genome are synthesized in the cytoplasm as higher M_r precursors containing an amino-terminal transit peptide that is cleaved after entry into the plastid (Cline and Henry, 1996). These presequences of nuclear-encoded plastid proteins, even though not strictly conserved, share some common features that can be used to predict localization using computer algorithms (Emanuelsson et al., 2000). This, in combination with transmembrane -helical domain characteristics, namely charge bias and hydrophobicity (Sonnhammer et al., 1998; Krogh et al., 2001), make it possible to identify candidates for plastid membrane proteins. Considering the nature of predictors and the selection criteria used (see "Results and Discussion"), we expect our collection described in this paper to contain mostly inner envelope integral membrane proteins.

Our laboratory has a long-standing interest in enzymes related in lipid metabolism and proteins that may facilitate transport of lipids across membranes. To identify additional proteins involved in these processes, we reasoned that such proteins would be integral membrane proteins and might not be easily detected by standard proteomic technologies. We considered that a bioinformatics analysis of the Arabidopsis genome might provide a useful complement to other approaches toward characterization of integral proteins of the plastid envelope. This study and its associated database of predicted candidates, together with ongoing global functional analysis, will provide insight into proteins associated with the plastid envelope as well as an initial step to further elucidate their function.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION RESULTS AND DISCUSSION CONCLUSIONS MATERIALS AND METHODS LITERATURE CITED

Plastid Envelope Membrane Protein Candidates

The strategy to select candidates of plastid envelope integral membrane proteins from Arabidopsis nuclear genome sequences by computational methods and the results are summarized in Figure 1.

View larger version (24K): [in this window] [in a new window]   Figure 1.   Summary scheme of plastid envelope protein prediction from the Arabidopsis nuclear genome. The graph displays the number of proteins after each selection step. Subcellular localization was predicted using the TargetP version 1.01 Web server. Transmembrane region prediction was done using the TMHMM version 2.0 Web server. cTP, Chloroplast transit peptide. RC, TargetP reliability class. TM, Transmembrane domain. See "Materials and Methods" for Web site addresses for the predictors.

The first step was to predict plastid-targeted proteins from the entire Arabidopsis genomic sequences (see "Materials and Methods" for Arabidopsis genomic sequence retrieval). This was done using TargetP (http://www.cbs.dtu.dk/services/TargetP/). TargetP is considered the best subcellular location predictor that is publicly available (Emanuelsson and von Heijne, 2001; Bannai et al., 2002; Peltier et al., 2002) and was also chosen by the Arabidopsis Genome Initiative to analyze the recently finished genome sequence of Arabidopsis (Arabidopsis Genome Initiative, 2000). Protein targeting to the plastid usually relies on its amino-terminal presequence (Cline and Henry, 1996) and TargetP, a neural network-based tool, is trained to recognize those signals. Overall success rate on test sets analyzed by the developers was 85% (Emanuelsson et al., 2000). The sensitivity for chloroplast-targeted proteins was 85%, whereas the specificity was lower (69% and 84% on two different test sets). This means that more false positives are expected than false negatives. To increase the specificity, cutoff restraints can be applied. However, to include a maximum number of possible candidates we instead chose default decision rule (winner takes all) and designated in our database the "reliability classes" from 1 to 5 (Emanuelsson et al., 2000). In total, 3,665 of 25,552 (14.3%) proteins encoded by the Arabidopsis nuclear genome were predicted to be plastidial, which is similar to the reported value of 3,574 from the Arabidopsis Genome Initiative (2000).

These 3,665 predicted plastid-targeted proteins were next analyzed for transmembrane -helical domains (Fig. 1). There are several Web-based predictors available (Moller et al., 2001). A recent study compared the performance of 14 different methods against 883 membrane-spanning regions of biochemically characterized proteins, and concluded that Transmembrane Hidden Markov Model (TMHMM) is the most accurate method for transmembrane segment detection (Moller et al., 2001). Subsequently, Hidden Markov Model for TOpology Prediction (HMMTOP) also has been shown to have similar performances (Moller et al., 2002). Most recent methods for prediction of transmembrane helices rely on hydrophobicity and charge bias of the transmembrane regions. There were variations in the accuracy of overall correct topology prediction (number and location and orientation of transmembrane regions); however, correct prediction of the presence of transmembrane segments per se was overall 95% in a comparison of three best predictors (Sonnhammer et al., 1998). TMHMM is reported to discriminate between soluble and membrane proteins with both specificity and sensitivity better than 99% (Krogh et al., 2001). TMHMM, like TargetP, is also a neural network-based program, but built with an architecture that corresponds more closely to the biological system (Hidden Markov model). The probability of a predicted integral protein to be a true membrane protein is positively correlated with the expected number of amino acid residues within the transmembrane segments and also with the expected number of transmembrane segments in the protein. In other words, the more amino acid residues per predicted transmembrane segments and the more predicted number of transmembrane segments per gene protein, the more likely that it is a truly integral protein. To include as many integral proteins as possible, we selected all proteins that have at least one predicted transmembrane domain, regardless of amino acid residue numbers. One of the advantages of TMHMM is that it provides plots of probabilities for prediction throughout the analyzed proteins. In our study, proteins that were overall not predicted to be transmembrane proteins, yet contained weak -helical domains with probabilities over 50%, were also included in the database as potential membrane proteins with their probabilities. Because the accuracy of TMHMM prediction drops when signal peptides are present (Krogh et al., 2001), transmembrane predictions within the predicted targeting peptides (according to the TargetP cleavage site predictions) were removed and those domains that were predicted close to the targeting peptides were designated in the database as "N-term." After processing as described above, of 3,665 predicted plastidial proteins, 562 (about 15%) contained potential transmembrane -helical domains (Fig. 1) and, thus, are considered as plastid membrane protein candidates.

The third step was to eliminate the thylakoid integral membrane proteins. Although there are suborganelle location predictors such as PSORT, the overall accuracy of such programs is not high enough (about 50%) to use for discrimination between the thylakoid and the envelope-localized proteins (Nakai and Horton, 1999). It is estimated that there are at least 200 proteins in the lumenal space and in the periphery of the thylakoid membrane (Peltier et al., 2000). Four multisubunit protein complexes with 75 to 100 peptides involved in photosynthesis comprise the majority of the thylakoid membrane proteome (Peltier et al., 2000). These proteins are relatively well studied and many of them are known to be encoded by the plastid genome, which contains about 90 protein-encoding genes (Sugita and Sugiura, 1996). If the known nuclear-encoded thylakoid membrane proteins are removed from our candidate list of 562, the remaining candidates are expected to be highly enriched in envelope proteins. After manual removal of 21 known thylakoid-associated proteins and their homologs, 541 proteins remained as potential candidates for Arabidopsis plastid envelope membrane proteins (AtPEM).

It is important to note that these are predicted to be integral membrane protein candidates and the candidate list will contain omissions and inclusions of various types. First, as mentioned above, the specificity of TargetP prediction is lower than the sensitivity. Because we did not employ any specificity cutoff for TargetP prediction and used low stringency for transmembrane prediction (transmembrane domain probability > 50%), these 541 candidates may contain many false positives. If instead the TargetP specificity is set to >0.95 and proteins with low probabilities of transmembrane domain are excluded, about 250 candidates remain and these represent a higher reliability core set. These higher reliability candidates are color coded on our Web site. Second, there will be many membrane proteins that do not have transmembrane domains but are associated with the envelope through different types of interactions such as -sheet conformation, polyisoprenylation, acylation, glycolipid anchors, protein-protein interaction-mediated associations, etc. Third, most outer membrane-localized proteins appear to insert directly into the outer envelope membrane without going through the general import apparatus (Cline and Henry, 1996; Fulgosi and Soll, 2001; Schleiff and Klosgen, 2001). Therefore, the 541 AtPEM candidates will underrepresent the outer membrane proteins. However, the protein composition of outer membranes is much simpler than that of inner membrane. The lipid to protein ratio of outer membrane is about 3 times higher than that of inner membrane and the inner envelope membrane, which contains various metabolite translocators, is the main permeability barrier for solutes, although there is growing evidence suggesting that the outer envelope may contain regulated ion channels (Cline et al., 1981; Block et al., 1983; Douce and Joyard, 1990; Pohlmeyer et al., 1998; Flugge, 2000; Bolter and Soll, 2001). Thus, we consider it likely that AtPEM candidates represent the majority of integral plastid envelope proteins.

What Are They?

Figure 2A indicates the functional identification status for the selected 541 AtPEM candidates based on the short descriptions in the annotation databases (MIPS and GenBank). The proteins with identified function (identified) or that have homology with known proteins (X like, putative X) comprised only about 29% of the 541, whereas 71% of the candidate proteins could not be assigned to any known function (putative, unknown, and hypothetical). This is a very low-characterized status compared with the overall 69% functional assignment of the whole genome (Arabidopsis Genome Initiative, 2000) and reflects the past difficulties in studying membrane proteins (Wilkins et al., 1998; Seigneurin-Berny et al., 1999).

View larger version (36K): [in this window] [in a new window]   Figure 2.   Functional classification of the Arabidopsis plastid envelope protein (AtPEM) candidates. A, Overall functional annotation status based on the short descriptions of each candidate from the Munich Information Center for Protein Sequences (MIPS; chromosomes 1, 3, and 5) and from GenBank (chromosomes 2 and 4). X, Any protein/gene/enzyme name. B, Detailed functional classification using automatically derived functional categories from the Protein Extraction Description and Analysis Tool (PEDANT) Web server (see "Materials and Methods"). C, Subclassification of the proteins classified as "transport facilitation" and "metabolism" by PEDANT.

More detailed functional classification of the AtPEM candidates was done according to the automatically derived functional categories from the MIPS Arabidopsis Database using the PEDANT Web server (http://pedant.gsf.de/; Frishman et al., 2001). Among the 541 candidates, 183 (34%) were found in the PEDANT database and were classified into 17 classes and 89 subclasses. Figure 2B shows the number of AtPEM candidates found in each category. The largest number of AtPEM candidates (39% of 183 classified predicted proteins) fell into the class of "transport facilitation." Thus, membrane transporters are highly enriched in the selected AtPEM candidates. Figure 2C shows the subclassification of "transport facilitation" and "metabolism" classes. The three major subclasses for "transport facilitation" were "ion transporters," "C compound and carbohydrate transporters," and "ATP binding cassette (ABC) transporters."

There were also several candidates involved in protein translocation into the plastids. None of the outer membrane components of the protein translocon were found among AtPEM candidates as was expected (many outer membrane proteins lack obvious plastid-targeting sequences as discussed above), whereas five Arabidopsis homologs to the known pea (Pisum sativum) translocon at the inner membrane of chloroplast (Tic110, Tic20, Tic40, and Tic55; Jackson-Constan and Keegstra, 2001) were present among the candidates (Table I). Two Arabidopsis homologs of Tic22 were predicted to be targeted to the plastids by the TargetP, but did not have obvious transmembrane-spanning domain (by TMHMM prediction). Tic22 is thought to be peripherally associated with the inner envelope membrane facing the intermembrane space (Jackson-Constan and Keegstra, 2001).

In agreement with the known role of plastid envelopes in lipid metabolism, "lipid-, fatty acid-, and isoprenoid metabolism-" related proteins represented the largest subset among the "metabolism" class. Table I presents the results of a more detailed survey of proteins predicted or known to be involved in plant lipid metabolism (the search was carried out using our database of predicted lipid-related genes in Arabidopsis (Mekhedov et al., 2000; F. Beisson and J.B. Ohlrogge, unpublished data). Seven proteins potentially responsible for known enzymatic reactions to produce glycerol lipids in the plastid envelope were found among AtPEM candidates (Table I; for review, see Ohlrogge and Browse, 1995). Arabidopsis homologs of plastidial 2-lysophosphatidic acid acyltransferase (EC 2.3.1.51), which mediates acyl group transfer from acyl-acyl carrier protein to the sn-2 position of lyso-phosphatidic acid (1-acyl-sn-glycerol 3-phosphate) toproduce phosphatidic acid (1,2-diacyl-sn-glycerol 3-phosphate), was missing. All three Arabidopsis homologs of this enzyme were predicted to a different subcellular location ("mitochondria" and "others") by TargetP. Digalactosyldiacylglycerol synthase (EC 2.4.1.184), which was known to localize on the outer envelope (Joyard et al., 1998), was not among the AtPEM candidates. Although it had plastid transit peptide recognized by TargetP, no membrane-spanning region was detected by TMHMM.

In plants, fatty acid synthesis takes place in the plastid. Although a portion of the newly synthesized acyl chains are then used for lipid synthesis within the plastid, a major portion is exported into the cytosol for glycerolipid assembly at the endoplasmic reticulum or other sites. In addition, some of the extraplastidial glycerolipids return to the plastid. Thus, there is considerable flux of lipid exchange and intermixing between the plastid and the endoplasmic reticulum (Ohlrogge and Jaworski, 1997). This involves lipid transport across the plastid envelope membranes. Evidence is emerging that some of the ABC transporters are involved in the lipid transport (Ruetz and Gros, 1994; Zhou et al., 1998; Hettema and Tabak, 2000). There were 15 potential ABC transporters among the AtPEM candidates, which, thus, are candidates for lipid transporters of the plastid envelope.

Surprisingly, there were several candidates that are homologous to proteins involved in cell wall biogenesis and modifications, such as arabinogalactan-protein homolog, cellulose synthase catalytic subunit-like protein, pectinesterase-like protein, pectin methylesterase-like protein, putative xyloglucan fucosyltransferase, etc. These unexpected relationships between the plastid envelope and predicted proteins annotated with functions that are generally considered non-plastidial may reflect inaccurate prediction by TargetP software. Alternatively, because these putative functions are assigned based solely upon sequence similarities, an equally likely possibility is that the annotations point toward related, but previously undescribed, functions in the plastid. Therefore, this incongruence represents an example where bioinformatics analysis can direct attention toward potential novel functions in the plastid envelope.

Characteristic Features of Plastid Envelope Candidates

The distribution of the peptide length of AtPEM candidates is shown in Figure 3A. Eighty percent of the candidates were smaller than 600 amino acids in length. The average length of the peptides was 411 amino acids, which is slightly smaller than the average peptide length of 434 amino acids for the nuclear genome (Arabidopsis Genome Initiative, 2000). As indicated in Figure 3B, 63% of the AtPEM candidates contained two or less membrane-spanning domains, whereas there were 40 candidates that had 10 or more membrane-spanning domains. The spanning domains of proteins with only a few such domains might serve as an anchor to the envelope or to tether the peripheral proteins. Interestingly, 70% of candidates with 10 or more membrane-spanning domains were classified as "transport facilitation." Due to their high hydrophobicity, these potential transporters would be particularly difficult to display on two-dimensional electrophoretic gels of purified plastid envelopes.

View larger version (15K): [in this window] [in a new window]   Figure 3.   Characteristics of AtPEM candidates. A, Peptide length distribution. L, Number of amino acid residues. Plastid-targeting sequences were removed based on TargetP cleavage site predictions. B, Distribution of the number of membrane-spanning domains. The TMHMM version 2.0 Web server was used for the membrane-spanning domain prediction. prob<1, Proteins with spanning domain probability less than 1.

AtPEM candidates provide an opportunity to compare predicted plastid envelope proteome with the proteome of other organisms, i.e. Cyanobacteria. The Synechocystis PCC6803 genome contains about 3,167 protein encoding genes, which is roughly the number predicted to be plastidial in Arabidopsis. Highly homologous proteins for about 32% (175) of AtPEM candidates were in the Synechocystis PCC6803 genome (data not shown), compared with 44% and 47% of thylakoid and stromal proteins, respectively. Thus, the boundary envelop of the plastid may have a more "mixed" evolutionary origin than the internal components.

We also compared our list of candidates with approximately 180 proteins identified by mass spectrometric analysis of Arabidopsis chloroplast envelope preparations (B.S. Phinney and J.E. Froehlich, unpublished data). Among these 180 proteins, about 70 had at least one transmembrane domain predicted by TMHMM (TM probability > 50%) and 38 proteins overlapped with AtPEM candidates. Thus, the proteomics approach identified many nonintegral membrane proteins and outer membrane proteins not selected by our approach, whereas the bioinformatic identifications used here may provide novel information on a set of inner envelope integral membrane proteins not easily characterized by proteomics.

Digital mRNA Expression Profile

In addition to the subcellular localization, indirect information on cellular or developmental function can be obtained from spatial and temporal expression patterns of genes. Gene chips, microarrays, expressed sequence tags (ESTs), and serial analysis of gene expression all provide useful means to study mRNA expression profiles (Bouchez and Hofte, 1998).

A large proportion (68%) of AtPEM candidates have at least one EST in the GenBank dbEST (113,330 ESTs as of January 4, 2002; Table II) and 87% (470 of 541) were represented by The Institute for Genomic Research (TIGR) Tentative Consensus (TC) sequences (163,752 TCs as of May 23, 2001; http://www.tigr.org/tdb/agi/; Quackenbush et al., 2000; see "Materials and Methods"). Thus, the majority of the AtPEM candidates are transcriptionally active.

The abundance of ESTs sequenced from different cDNA libraries can provide an estimate of relative transcript abundance provided a number of conditions are met (Audic and Claverie, 1997). To obtain tissue-specific expression profiles of AtPEM candidates, 110,000 ESTs deposited in GenBank and sequenced from 55 different cDNA libraries were grouped into eight "library pools" according to the source tissues from which the cDNA libraries were derived (Table II, F. Beisson, J.B. Ohlrogge, unpublished data). We then surveyed the abundance of ESTs within each library pool for AtPEM candidates after normalization by dividing the number of AtPEM candidate ESTs in the given "library pools" by the total number of ESTs of that "library pool." The relative abundances of ESTs for AtPEM candidates in each tissue were similar, ranging from 1.4% to 2.5% of the total ESTs, except for the "flowers" pool, which showed the highest abundance at 4.2% of ESTs.

Although these aggregate EST frequencies varied little, a number of individual proteins were found to have more distinct tissue-specific expression patterns. To study this in detail, tissue-specific expression patterns of individual candidates were analyzed using statistical equations developed by Audic and Claverie (http://igs-server.cnrs-mrs.fr) that differentiate between random EST sampling fluctuations versus significant change in EST frequencies. The number of ESTs corresponding to a given gene (AtPEM candidate) found in each of three tissue-specific library pools ("flowers," "roots," or "seeds") was compared with that in the reference "mixed" pool; the results are presented in Table III. A total of 21 AtPEM candidates (6% of 365 candidates that had at least one EST) displayed tissue-specific expressions when P < 0.005 (P, a probability of the compared EST abundances being different by chance) was applied. Figure 4 visualizes the digital expression profile of 21 AtPEM candidates that showed tissue-specific expression pattern at P < 0.005. Among the differentially expressed genes were Glc-6-phosphate/phosphate translocator (GPT) and phosphoenolpyruvate/phosphate translocator (PPT). GPT was shown previously to be highly expressed in developing maize (Zea mays) kernel and potato (Solanum tuberosum) tuber (Kammerer et al., 1998) and the ESTs of GPT were abundant in the Arabidopsis developing seed EST database (White et al., 2000). In agreement with these observations, GPT had higher EST abundance in the "seeds" library pool than the reference at P < 0.004. Similarly, for PPT, biochemical analysis and mRNA-blotting results indicated a high expression of PPT in nongreen tissue (Kammerer et al., 1998), especially roots (Fischer et al., 1997), and in our EST analysis, PPT expression was highest in the "roots" (P < 0.003). Thus, the comparisons in Table III agree with previous biochemical characterizations and support the validity of the digital expression profile approach. Many of the proteins shown in Table III are of unknown function, including several with high representation in specific tissues.

View larger version (43K): [in this window] [in a new window]   Figure 4.   Digital differential display analysis. AtPEM candidates with statistically different levels (P < 0.005, where P is probability of difference being by chance) of EST frequencies are displayed. Number of ESTs in the "flowers," "roots," or "seeds" pools was compared with that in the reference "mixed" library pool. The bars indicate the abundances of the ESTs (EST frequency axis) corresponding to the proteins (proteins axis) in each library pool (tissue axis).

The bioinformatics predictions of suborganellar localization, together with tissue-specific expression, provide initial clues that may help in discovering the functions of these proteins. However, it should be noted in such analysis that the normalization procedures used in the construction of the cDNA libraries for EST projects can undermine the statistical analysis, and even with the high probability such as P < 0.005 used in Table III, there could still be false positives (Audic and Claverie, 1997). Therefore, the digital expression profile analysis should be considered as an initial step to find possible candidates for differentially expressed genes from a large collection of data and needs to be verified by experiments such as northern blots, reverse transcription-PCR, or western blots, etc.

Plastids undergo massive changes during differentiation into their various forms (chloroplast, chromoplast, amyloplast, leucoplast, etioplast, etc.) and need to import different spectra of proteins according to their changed biochemical properties. However, the structure of the concentric pair of envelope membranes remains constant (Douce and Joyard, 1990; Joyard et al., 1998). It is yet unknown the extent to which the proteome of the envelope will also remain unchanged. The low number (21) of candidates differentially expressed between the different tissues may indicate that the envelope proteome candidates are relatively constitutive in terms of transcript abundance across the different tissues. Furthermore, when the tissue-specific EST distributions of AtPEM candidates and that of 39 thylakoid-localized proteins and 1,802 stromal proteins were compared, AtPEM candidates showed the least changes (Table IV). Although only 8% of the AtPEM candidates were tissue specific at P < 0.05, 69% of the analyzed thylakoid-localized proteins showed specific EST tissue distributions. The stromal proteins also displayed a higher level (19% at P < 0.05) of differential expression than the AtPEM candidates. Thus, based on EST frequencies from different cDNA libraries, the transcript abundance for the envelope membrane proteins appears relatively constant compared with that of the stromal or thylakoid proteins.

Microarray Analysis

To further investigate the tissue-specific gene expression pattern of AtPEM candidates, we analyzed publicly available cDNA microarray data from the Stanford Microarray Database (SMD; http://genome-www.stanford.edu/microarray; Sherlock et al., 2001) and from the Arabidopsis developing seed array Web site (http://www.bpp.msu.edu/Seed/SeedArray.htm; Girke et al., 2000; Ruuska et al., 2002). Expression profiles of 230 AtPEM candidates for about 110 different microarray experiments were found in the SMD public domain (as of March 12, 2002). We specifically examined tissue comparison data that were publicly available for comparisons between flower, leaf, or root versus the whole plant for 147 AtPEM candidates and for seed versus leaf or seed versus plantlets for 59 AtPEM candidates. Figure 5 presents a summary of genes which displayed more than 2-fold changes in at least two different tissue comparisons and the pattern of which agreed in all combinations (i.e. At1g29390 expression is higher in leaf compared with that in flower and higher in leaf also when compared with that in root). Twenty-two candidates showed higher levels of transcript abundance in leaf when compared with that of root, whereas only three showed higher levels in root than in leaf. Twenty-one of these 22 candidates (expressed higher in leaf than in root) also had higher transcript abundance in leaf than that in flower, whereas only three candidates showed higher expression in flower than in leaf. Comparison between the seed and the leaf also showed higher levels of transcript abundance in leaf (four higher in seed and 23 higher in leaf). Although it is well known that thylakoid and stromal chloroplast proteins are very abundant in leaves, very little comparative information is available on the envelope proteins. The data of Figure 5 suggest that transcripts for many plastid envelope proteins are also more abundant in leaves than from heterotrophic tissues.

View larger version (61K): [in this window] [in a new window]   Figure 5.   Tissue-specific microarray analysis of AtPEM candidates. AtPEM candidates with >2-fold tissue-specific expression in at least two different tissue comparisons are shown clustered according to the tissue specificity (indicated with the brackets and labeled). The intensity of green and red colors represents the relative level of transcripts (refer to the color keys at the bottom for the actual ratios) in tissues under comparison (the tissues under comparison are indicated above the image and colored with representative colors). The data for flower, leaf, and root comparisons were from SMD (http://genome-www.stanford.edu/microarray) and data for seed versus leaf and seed versus plantlet were from the Arabidopsis developing seed array (Girke et al., 2000; Ruuska et al., 2002). The graphic was generated by Tree View software (Eisen et al., 1998).

Six genes selected by the digital expression profile analysis in Table III as tissue specific were also present in the microarray data shown in Figure 5. Although the digital expression profiles used mixed tissues as a reference and microarray data compared individual tissues or plantlets, data for all six genes agree at least partially between the two types of analysis. However, the cross validity between the statistical EST analysis and the microarray analysis should be further characterized by larger sets of genes that have data analyzed by both methods, combined with additional experiments (northern blots, RT-PCR, etc.) to verify results.

Changes in transcript abundance in seeds during 5 to 13 d after flowering were recently reported by Ruuska et al. (2002). Twenty-six AtPEM candidates changed more than 2-fold during this period (data not shown). The expression of many genes involved in seed storage protein and starch and lipid biosynthesis also change during these stages (Ruuska et al., 2002); therefore, these 26 AtPEM candidates may be involved in the role of plastids in seed filling and development.

Toward the Function of Unknown Candidates

Gene knockouts often can provide key information to link genes of unknown function to a phenotype. Large collections of T-DNA insertion mutants are publicly available and when the DNA flanking the T-DNA insertion sites are sequenced and aligned with the Arabidopsis genome sequence, provide mutants useful for studying gene function. We searched the Sequence-Indexed Library of Insertion Mutations generated by the Salk Institute Genome Analysis Laboratory (http://signal.salk.edu/tabout.html; containing 32,758 T-DNA sequences as of March 8, 2002). In total, 388 insertion lines corresponding to 217 AtPEM candidates (approximately 40% of AtPEM candidates) were found. The "insertion Ids" can be used to search and to order the mutant line seed stocks from the Arabidopsis Biological Resource Center (http://www.biosci.ohio-state.edu/~plantbio/Facilities/abrc/abrchome.htm) at Ohio State University (Columbus). These "Insertion Ids," as well as the accession numbers to search the publicly available microarray data in SMD (http://genome-www.stanford.edu/microarray), were deposited on our Web site (http://www.plantbiology.msu.edu/PlastidEnvelope/).

	CONCLUSIONS

TOP ABSTRACT INTRODUCTION RESULTS AND DISCUSSION CONCLUSIONS MATERIALS AND METHODS LITERATURE CITED

Plastids draw the attention of plant biologists in large part because of their defining roles in establishing the character of the plant cell. Envelope proteins may hold one key to the understanding of coordinated control between the plastid and the rest of the cell. So far, there is no established inventory of plastid envelope proteins. In this study, we attempted to predict integral plastid envelope proteins from the Arabidopsis nuclear genome using computational methods.

As the word "candidates" implies, the results of our study are not definitive due to the nature of prediction software and due to the possibility of errors in protein annotations in the genome databases (van Wijk, 2001). Although it is likely that at least 10% of the AtPEM candidates represent incorrect predictions, it is reasonable to assume that the selected candidates represent a large portion of the real envelope proteome. These candidates can be used as a starting point for designing further biological experiments. For example, one alternative strategy to complement "proteomics" approaches is to experimentally characterize envelope localization of selected candidates. The AtPEM candidates could be analyzed for their targeting to the chloroplasts and partitioning to the envelopes by in vitro reconstitution of import into the chloroplast followed by fractionation into the suborganellar compartments. This method can be especially powerful in identifying highly hydrophobic, low-expressed transporters and also can circumvent the difficulties in isolation of pure plastid envelope from many non-photosynthetic tissues. This approach is currently being developed and evaluated in our laboratory. Another application to further characterize selected candidates will be to define subsets of AtPEM proteins that show concerted spatial, developmental, or conditional expression profiles. Finally, based on our database, and such selections, a more focused analysis of the phenotypes and biochemical compositions altered in T-DNA insertion lines can be carried out.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION RESULTS AND DISCUSSION CONCLUSIONS MATERIALS AND METHODS LITERATURE CITED

Establishment of a Database for Plastid Envelope Protein Candidates

The results of TargetP prediction for Arabidopsis chromosomes 2 and 4 were from http://www.cbs.dtu.dk/services/TargetP/predictions/pred.html. The sequences for these two chromosomes were from TIGR and the European Union Arabidopsis Sequencing Consortium (as of January 7, 2000). Additional sequence retrieval was from the National Center for Biotechnology Information Batch Entrez Web server (http://www.ncbi.nlm.nih.gov:80/Entrez/batch.html). Chromosomes 1, 3, and 5 sequences were downloaded from the MIPS Arabidopsis Database ftp site (ftp://ftpmips.gsf.de/cress/, as of June 2001). Subcellular localization was predicted using TargetP version 1.01 from the Center for Biological Sequence Analysis (http://www.cbs.dtu.dk/services/TargetP/). No cutoff was applied but instead "Reliability Class" values from 1 to 5 were designated for each predicted proteins. Proteins with plastid transit peptides were then evaluated for membrane-spanning domains using the TMHMM version 2.0 Web server at http://www.cbs.dtu.dk/services/TMHMM-2.0/.

Proteins that were not strongly predicted to be transmembrane proteins by the program, yet contained weak -helical domains with probabilities over 50%, were also included in the database as possible membrane proteins. Proteins with transmembrane -helical domains within the range of possible plastid-targeting sequences (plastid-targeting sequence prediction was according to the TargetP cleavage site prediction) were removed and in cases where the domain located close to but not within the predicted cleavage site were marked "N-term" to reduce false prediction of hydrophobic targeting sequences as membrane-spanning domains. Proteins that are known to locate in thylakoid were removed manually.

Classification by Function

Functional classification was based on the MIPS Arabidopsis Database automatically derived functional categories. The catalogue was downloaded from the PEDANT Web server (http://pedant.gsf.de/). The catalogue of genes for plant glycerolipid biosynthesis was from http://www.canr.msu.edu/lgc/(Mekhedov et al., 2000). An updated inventory of lipid metabolism-related genes was available at our regional database (F. Beisson and J.B. Ohlrogge, unpublished data). The full inventory of Arabidopsis ABC proteins was downloaded from http://www.arabidopsisabc.net/(Sanchez-Fernandez et al., 2001) and was queried for AtPEM candidates.

Digital mRNA Expression Profiling

A set of all public Arabidopsis ESTs was obtained through a structured query language query of our in-house "SeqStore" database that contained 103,109 EST sequences from the GenBank EST database (dbEST, http://www.ncbi.nlm.nih.gov/dbEST/index.html). The EST sequences were used as queries (BLASTN version 2.2.1) against the target database of all predicted transcripts from the Arabidopsis genome (ftp://ftp.tigr.org/pub/data/a_thaliana/ath1/, January 10, 2002). Then the plastid envelope protein candidates were queried against this resulting database using common chromosomal locus identifiers (i.e. At2g39040). TC sequences assembled from ESTs and expressed transcript sequences were retrieved from TIGR Arabidopsis Gene Index (http://www.tigr.org/tdb/agi/). To match TCs with the chromosomal locus identifiers, TCs retrieved were mapped to all the predicted transcripts of Arabidopsis genome by Blast alignments (BLASTN version 2.2.1). TCs corresponding to the plastid envelope protein candidates were selected by chromosomal locus identifiers.

To identify possible differential expression of the AtPEM candidates, the relative frequencies of ESTs between tissue-specific library pools were compared (55 different EST libraries were grouped into eight tissue-specific "library pools" according to the source tissues from which the library was derived; F. Beisson and J.B. Ohlrogge, unpublished data). The influence of random fluctuations and sampling size was considered statistically to discern the reliability of the digital expression profiling (Audic and Claverie, 1997). For this analysis, software was downloaded from http://igs-server.cnrs-mrs.fr. EST analysis for the thylakoid and stromal proteins was done using 39 selected known thylakoid-localized proteins, most of which are related to photosynthesis, and 1,802 stromal proteins that were chosen randomly from plastidial proteins predicted by TargetP and subtracting those with transmembrane domains and obvious thylakoid proteins. ESTs for the thylakoid- and stroma-localized proteins were from TIGR Arabidopsis Gene Index TCs.

Microarray Data

Public microarray data for AtPEM candidates were searched from a local database that contained most of the microarray data downloaded from SMD (as of March 1, 2002; Sherlock et al., 2001) and was courtesy of Rodrigo Gutierrez (Michigan State University-Department of Energy [MSU-DOE] Plant Research Laboratory, East Lansing). The duplicates in the tissue comparison data sets (SMD experiment identifiers: 7,197, 7,199, 7,200, 7,201, 7,203, and 7,205) were averaged and divided each other in a way to give pair-wise comparisons between flower, leaf, and root tissues. Arabidopsis developing seed array data were from a local database and are available at http://www.bpp.msu.edu/Seed/SeedArray.htm (Girke et al., 2000; Ruuska et al., 2002). The cluster image (Fig. 5) was generated by Tree View software downloaded from http://rana.lbl.gov/(Eisen et al., 1998).

T-DNA Insertion Mutants

T-DNA insertion mutant lines were searched from the Sequence-Indexed Library of Insertion Mutations generated by the Salk Institute Genome Analysis Laboratory using the Arabidopsis Gene Mapping Tool (http://signal.salk.edu/cgi-bin/tdnaexpress).

Information on AtPEM candidates and their various attributes, including tissue-specific EST frequencies, accession numbers to search SMD, and T-DNA insertion mutant stock numbers from Salk Institute Genome Analysis Laboratory can be downloaded from our Web site (http://www.plantbiology.msu.edu/PlastidEnvelope/).