Eukaryotic cells segregate and organize specific molecules that carry out defined functions in organelles. Organelles — for example, the nucleus, endoplasmic reticulum (ER) and Golgi complex — are dynamic membrane-bounded compartments that have distinct structures and functions. Vesicles and tubules transport proteins and lipids between organelles, which results in organelles having a constantly changing molecular composition. Lipid bilayers form a barrier between organelle lumens and the cytoplasm, and these bilayers contain many transmembrane proteins that can interact with components of the cytoskeleton and signalling pathways. Organelles are composed of molecules that are present almost all of the time ('resident' molecules), molecules that are in transit and molecules that transiently interact with the organelle to carry out distinct functions. Approaches to identify and delineate all of these molecules will provide the data necessary to understand organelle function.
In the post-genomic era, large-scale proteomics technologies and tools have made organelle-scale studies feasible, and have become powerful methods for studying organelles, their components and dynamics1, 2, 3, 4, 5. Historically, the presence of lipid bilayers limited the protein chemistry technologies that could be applied, but new strategies are emerging to identify and characterize integral membrane proteins6, 7, 8. These methods and strategies have created opportunities for the complete characterization of the different types of transmembrane protein, including obtaining information on protein domains and post-translational modifications7. Combining large-scale proteomics studies with traditional cell-biology techniques is providing a strategy for the functional characterization of organelles and the molecules they interact with in the cytoplasm, for example, those in signalling pathways9.
Organelle proteomics studies begin with the isolation, by classic cell-biology techniques (Box 1), of an enriched fraction that contains the organelle of interest. This allows the proteins that are isolated with the organelle to be identified using proteomics techniques (Boxes 2,3). There are five important caveats to this analysis. First, the extent of the enrichment will vary and molecules that are not bona fide components of the organelle will be present in the fraction and identified as being a part of the proteome (Box 4). Second, as mentioned above, the composition of organelles is always changing, with molecules being delivered to and transported from organelles by vesicles and tubules. This trafficking is mediated by components of the cytoskeleton — actin, tubulin and intermediate filaments — as well as by the motors and accessory molecules that function with each of the cytoskeletal components. These components might or might not be isolated with the organelle, which results in varying, but seldom logical, data that are used to classify these molecules when they are identified in proteomics studies. Third, many molecules in the cytoplasm interact with organelles to carry out defined functions, especially signalling molecules that are specifically activated by hormones, growth factors and other signals. In this case, whether these molecules will be associated with a specific organelle depends on the previous history of the cells that are being analysed. Fourth, the composition of an organelle from different cells and different tissues will vary. Last, many proteins in organelles, as well as the transmembrane proteins and the cytoplasmic proteins that are associated with organelles, are post-translationally modified. Understanding these post-translational modifications is paramount to dissecting function. Although there are now many methods to study post-translational modifications, comprehensive methods that uncover all post-translational modifications are time consuming and probably never carried out in a comprehensive way.
Insights from organelle proteomics analyses
The analysis of organelles using proteomics methods is an active field of research, and significant progress is being made in defining the proteomes of organelles. So far, a sufficient number of proteomics studies have been published so that we can ask if the methods and technologies are suitable for organelle studies and, after the initial 'shock and awe' that surrounded proteomics technologies, if new functional or biological insights are being derived from these experiments.
Organelles and other large cellular structures have been analysed using proteomics techniques (Box 5). However, no systematic or organized effort is underway to characterize the subcellular proteome of a particular cell. Rather than facilitating a comprehensive analysis of a single cell type, cell types and organisms are chosen more on the basis of the interest of the investigator, the convenience of sample preparation, and their relevance to disease. In the following sections, we review recent proteomics analyses and discuss the biological insights that have been derived.
Exosomes. Many of the vesicles that transport lipids and proteins in cells have been studied using different proteomics approaches. EXOSOMES (Box 5) are vesicles that are discharged into the extracellular milieu when multivesicular endosomes fuse with the plasma membrane. Recent studies have investigated the role of exosomes in the immune response10, 11, 12, 13, which has led to the hypothesis that these secreted vesicles have a role in intercellular signalling14. To analyse this, exosomes were purified from the supernatants obtained from melanoma cell lines and analysed using two-dimensional (2D) gel electrophoresis (2DGE) followed by tandem mass spectrometry (MS/MS)15. The analysis of 49 protein spots that were not obvious in cell-lysate controls resulted in the identification of 41 different proteins, some of which had been previously identified in exosomes, but some of which were novel to exosomes — for example, p120 catenin, radixin and immunoglobulin-superfamily member-8. In another study, urinary epithelial-cell-derived exosomes from six normal male volunteers were analysed by one-dimensional (1D) SDS-PAGE followed by nanospray liquid chromatography (LC)–MS/MS, which identified 295 unique proteins16 (Box 5). Of these, 73 were known membrane-trafficking proteins, mainly endosomal, which further confirmed the proposed endosomal origin of exosomes. In addition, a number of proteins that are known to be involved in specific kidney diseases were identified, which indicates that the analysis of exosomes that are excreted in the urine might provide a method for the early detection of renal disease.
Phagosomes. Phagocytosis is used by cells to internalize large macromolecular complexes, especially pathogens, for degradation. The complex or pathogen binds to cell-surface receptors, which results in the invagination of the plasma membrane and the internalization of the complex or pathogen. As mentioned in Box 1, latex-bead internalization, which is a model system for phagocytosis, has allowed the phagosomal compartment to be significantly enriched for proteomics analysis. Using this strategy, a total of 520 PHAGOSOME proteins have been identified9. As phagosomes fuse with organelles of the endocytic and biosynthetic pathways, separating the stages of maturation is difficult, and the proteins that are identified probably represent several stages of the phagocytic process. The interesting and novel finding of this proteomics analysis was the presence of many ER proteins in the PHAGOLYSOSOME proteome, because no obvious phagosome–ER link was known. However, studies by Desjardins and colleagues proposed that a portion of the phagosome membrane originates from the ER17, 18 and, by using a proteomics strategy, they showed that phagosomes have a full cross-presentation machinery for exogenous antigens19 (Box 5). Further proteomics studies investigated the role of protein kinase C
(PKC) in the regulation of phagolysosome biogenesis. Phagosomes isolated from macrophages overexpressing a dominant-negative kinase-dead mutant of PKC showed inhibition of the recruitment of the phagosome components Rab7, cathepsin-D and cathepsin-S when compared to phagosomes isolated from control mouse macrophages. These data indicate that PKC has a role in phagolysosome biogenesis20 and show how functional insights can be obtained from proteomics studies.
Clathrin-coated vesicles. Clathrin-coated vesicles (CCVs) (Box 5) that form using adaptor protein-1 (AP1) are involved in transport from the trans-Golgi network (TGN) to the endosomal system. Adaptor protein-2 (AP2)-positive CCVs function in a range of endocytic processes that start at the plasma membrane. These include the uptake of signalling receptors, plasma-membrane pumps and nutrients.
A proteomics analysis of a CCV fraction from rat brain, which used 1D SDS-PAGE and LC–QUADRUPOLE TIME-OF-FLIGHT–MS/MS, identified 209 proteins including 8 novel proteins21, 22. These proteins were ranked according to their abundance by counting the number of fragmented peptides that were assigned to each protein. Using this method, the expected 1:1 stoichiometry was observed for the clathrin heavy chain and clathrin light chain. Interestingly, a 3:1 stoichiometry was found for AP2 to AP1, which indicates that most CCVs in the brain function in endocytosis rather than in budding from the TGN22. These data are consistent with the identification of known protein components of synaptic vesicles in the CCV proteome. As an important function of brain CCVs is to recycle synaptic vesicles, the proteomics data support a full-fusion model for synaptic vesicle exocytosis22. Furthermore, enthoprotin/epsinR and adaptin ear-binding coat-associated protein-2 (NECAP2) were identified and shown to be new components of CCVs that form at the TGN and plasma membrane, respectively21, 23. Other studies showed that these proteins interact with AP1 and AP2, respectively, through previously unrecognized novel peptide motifs that are based around sequences with a 
XX core ( represents a bulky hydrophobic residue and X represents any amino acid)24. This increases the known complexity of the molecular sorting that is carried out by CCVs. Finally, as mentioned above, this proteomics analysis discovered the protein enthoprotin, which is probably responsible for susceptibility to a subtype of schizophrenia25.
Mitochondria. Mitochondria function in oxidative phosphorylation and ATP production in all cells, but their protein composition varies depending on cell type, condition and disease state. Proteomics analyses can provide clues regarding how mitochondria physiologically adapt.
Four proteomics analyses of mitochondria that were isolated from three different species have identified 416–750 mitochondrion-associated proteins26, 27, 28, 29, 30. A study of the proteome of human-heart mitochondria identified 615 proteins using a 1D SDS-PAGE approach for protein fractionation26, and the authors of this study subsequently used multidimensional-liquid-chromatography–MS/MS to obtain an expanded coverage of this proteome27. Sickmann et al.28 used a combination of 2DGE and two-dimensional liquid chromatography (LC/LC)–MS/MS to identify 750 proteins in the mitochondria of Saccharomyces cerevisiae, whereas a study by Prokisch et al.30 used a 'SHOT GUN' PROTEOMICS STRATEGY to identify 546 S. cerevisiae mitochondrial proteins. Of the 546 proteins identified by Prokisch et al.30, 337 overlapped with the Sickmann et al.28 study, and most of the remaining proteins could be classified as low abundance proteins28, 30. Mootha et al.29 carried out a proteomics study of mouse mitochondria, and produced a list of 591 mitochondrial proteins.
Both Mootha et al.29 and Prokisch et al.30 integrated data from expression analyses, deletion phenotype screening, protein-interaction analyses, computational predictions and subcellular localization studies to achieve a comprehensive view of mitochondria. Mootha et al.29 isolated mitochondria from mouse brain, heart, kidney and liver, analysed the digested proteins using LC–MS/MS, and compared the proteomics data to RNA-expression profiles for the same tissues. These data provided insights into both the tissue-specific differences in mitochondrial composition and the correlation between protein and RNA levels29. They identified 591 mitochondrial proteins, including 163 proteins that had not been previously linked to this organelle29. An important aspect of this study was the integration of proteomics and RNA-expression data with genotype data from patients with French-Canadian-type LEIGH SYNDROME, as this allowed the identification of a single candidate gene for this disease, LRPPRC31 (Box 5).
Such proteomics studies have provided important insights into mitochondrial function — for example, they have provided a greater understanding of the proteins that are imported into this organelle and have allowed the identity of DNA-repair enzymes for the small mitochondrial genome to be predicted. The broader value of these datasets is that they enable researchers to compare protein orthologues between species to gain a better understanding of the processes that occur in mitochondria.
Peroxisomes. These ubiquitous organelles are specialized for oxidative reactions that produce hydrogen peroxide and are linked to numerous functions including fatty acid metabolism and PLASMALOGEN synthesis. A proteomics strategy has been used to discriminate contaminants from bona fide peroxisomal proteins in S. cerevisiae, and 70 proteins were thought to be peroxisomal32. An important insight was provided by the discovery of Rho1 on peroxisome membranes. This provided the first evidence of a link between peroxisomes and the actin cytoskeleton, which indicates how peroxisome location and movement is regulated in the cell (Box 5).
Endoplasmic reticulum. The mRNAs of transmembrane and secretory proteins are translated at the ER and the newly synthesized proteins subsequently enter the protein-folding 'factory' of the ER lumen. A proteomics analysis of the lumenal content of the ER that was fractionated from the livers of Balb/C mice identified 141 proteins33 (Box 5). Lumenal proteins were separated by 2DGE, and matrix-assisted laser desorption/ionization (MALDI) MS and MALDI MS/MS were used to analyse the excised spots. Of the 141 proteins, six novel proteins were identified, including ERp19 and ERp46. These proteins contain thioredoxin motifs, which functionally substituted for protein disulphide isomerase in S. cerevisiae complementation studies. This indicates that ERp19 and ERp46 are protein disulphide isomerases. This family of protein-folding enzymes catalyses productive protein folding by reducing and oxidizing disulphide bonds in substrate proteins until their correct tertiary structure is attained. The discovery of two new enzymes is indicative of the substrate specificity that exists in the molecular-chaperone function of these protein-folding enzymes.
ERGIC. Transport in the exocytic pathway from the ER to the Golgi is mediated by an intermediate compartment that is composed of tubular–vesicular structures and is known as the endoplasmic-reticululm–Golgi intermediate compartment (ERGIC). This compartment was isolated from the human hepatoma cell line HepG2 for proteomics analysis using an affinity technique34. The cells were treated with BREFELDIN A to induce the Golgi complex to fuse with the ER, leaving behind the brefeldin-A-resistant ERGIC compartment. ERGIC was isolated by subcellular fractionation using NYCODENZ gradients and immunopurification using magnetic beads that were coupled to a cytoplasmically exposed epitope of KDEL RECEPTOR. Following the tryptic digestion of its component proteins, 1D SDS-PAGE gels were used for peptide separation and MALDI–time-of-flight MS and nanospray MS/MS were used for peptide analysis.
Twenty-four proteins were identified including two that were previously uncharacterized — SURF4 (Surfeit locus protein-4) and ERGIC-32 (Box 5). ERGIC-32 has sequence homology to the S. cerevisiae proteins ER vesicle-41 (Erv41) and Erv46, and was found to be a novel membrane protein that cycles between the ER and Golgi. Erv proteins are constituents of vesicles that were derived from the ER of S. cerevisiae using a cell-free assay that selects vesicles carrying newly synthesized cargo to the Golgi. These proteins are candidate receptors for newly synthesized cargo and they were also characterized by proteomics techniques35. In the study of Breuza et al.34, ERGIC-32 was found to be localized to ERGIC by immunofluorescence microscopy. RNA interference (RNAi) of ERGIC-32 increased the turnover of human ERV46, which indicated that these proteins might form a complex, the stability of which requires the presence of both proteins. Remarkably, the novel protein SURF4 has sequence homology to Erv29, which has also been implicated as a cargo receptor36. The total complement of molecules that are required to move cargo between the ER and Golgi has not been defined, and it is not clear where in the dynamic ERGIC–Golgi compartments retrograde transport is initiated. Future studies, such as the one on Erv29 (Ref. 36), in 'well-timed' systems will allow these questions to be addressed.
Golgi complex The Golgi complex is the central organelle of the exocytic pathway. It is responsible for many of the post-translational modifications of newly synthesized proteins and lipids, as well as for the sorting of these molecules to their site of function. Four proteomics studies of the Golgi complex from rat-liver or mammary-epithelial cells have been reported, two using 2DGE37, 38, one using 1D SDS-PAGE39, and another using multidimensional protein-identification technology (MudPIT)40. The studies by Taylor, Wu and colleagues37, 38, 40 are the result of a collaboration between two groups, and represent an increased refinement of the proteomics technology. The starting material — a Golgi fraction — for the studies by Bell et al.39 and by Taylor, Wu and colleagues37, 38, 40 was isolated by different methods, but represented Golgi fractions that were enriched in stacked flattened cisternae in both cases. Bell et al.39 first extracted the Golgi fraction in Triton-X 114 to enrich for transmembrane proteins and then separated the detergent-phase proteins using 1D SDS-PAGE. Both traditional EDMAN SEQUENCING and MALDI MS identified 81 proteins, 45 of which were considered to be Golgi proteins and 24 of which were contaminants, mostly from the ER and mitochondria. Many (32) of the 81 proteins were not transmembrane proteins but were probably proteins that interact with transmembrane proteins and that did not dissociate and partition into the detergent phase. Importantly, several novel proteins were identified, and the group focused on one called Golgi peripheral protein of 34 kDa (GPP34). The same protein was also identified by Taylor, Wu and colleagues. It was named Golgi matrix protein of 33 kDa (GMx33), was studied extensively37, 41 and was proposed to be a member of a novel family of peripheral trans-Golgi proteins.
Wu et al.40 used LC/LC–MS/MS to obtain a more comprehensive Golgi proteome. They identified 421 proteins, 41 of which had no known function. Almost as many known ER proteins (96) were identified as known Golgi proteins (110). The remaining 215 proteins were sorted into categories on the basis of their reported localization and functions. Many of these proteins might functionally interact with the Golgi, but verification of their functional significance requires further localization and functional studies.
These proteomics studies identified novel proteins that can be targeted by future functional studies, and perhaps the most enlightening finding was that a large number of the proteins identified were ER proteins. Morphological studies have shown that the ER is adherent to numerous trans-Golgi cisternae42, 43, 44, and the proteomics data indicate that the ER is truly adherent and isolates with stacked Golgi fractions. The literature indicates that this interaction could be mediated by CERT (ceramide ER transfer protein), which transfers ceramide from the ER to sphingomyelin synthetase in the Golgi45.
Other new biological information came from mining the MudPIT data for post-translational modifications40. A novel cytoplasmic post-translational modification — arginine dimethylation — was identified on ten Golgi and five ER proteins, and one of the novel Golgi-localized proteins was predicted to be a methyltransferase that is itself dimethylated on an arginine residue (Box 5). As this post-translational modification is known to be important in the nucleus, its presence in the cytoplasm indicates functional significance. In the future, the MudPIT data will be further analysed and will hopefully reveal more information on post-translational modifications. The Golgi proteome will vary in different cells and tissues, and in different physiological states. The data from these studies will therefore function as a baseline for comparison with the results of future studies.
Chloroplast envelope. In plants and algae, the chloroplast is the organelle in which photosynthesis occurs. Two membranes surround chloroplasts and form the chloroplast envelope. A specialized procedure was devised to obtain highly purified chloroplast envelopes from Arabidopsis thaliana, and these membranes were subjected to three different extraction procedures to maximize the number of membrane proteins that could be identified by LC–MS/MS46. From these three extractions, 112 proteins were identified, of which 89 were predicted to be genuine envelope proteins. These predictions were made using the literature or rigorous bioinformatics prediction methods to determine protein hydrophobicity, the presence of TRANSIT PEPTIDES, and homology to previously identified chloroplast proteins (Box 5). Of the 89 proteins, 32% were involved — or were predicted to be involved — in transport functions, and even bioinformatics prediction methods could not propose a function for 30% of these proteins. Several of the transport proteins are involved in phosphate transport, which is particularly relevant to the regulation of stromal phosphate levels and the initiation of the CALVIN CYCLE. The final 38% were classified as metabolic proteins or as having other functions that remain to be elucidated46. These studies confirmed an important role for chloroplast envelope membranes in lipid metabolism, including fatty acid desaturation and in the synthesis of lipid growth regulators and defence molecules.
Lysosomes. Lysosomes are the main degradative compartment of the cell and represent the end point of the endocytic pathway. They contain many different hydrolytic enzymes that degrade cellular macromolecules. A proteomics analysis of lysosomes — specifically of Triton-WR1339-filled lysosomes, which are known as 'tritosomes' — identified 215 proteins47 (Box 5). The enrichment of membrane proteins using alkaline sodium carbonate treatment followed by further fractionation enabled the identification of presumed cargo proteins (for example, Golgi glycosyltransferases as well as trafficking proteins, fusion proteins and transporters). An abundance of lipid-raft proteins in the lysosomal membrane confirmed the presence of these domains, which were previously identified in phagosomes48. An abundance of ER proteins was also found. This most recent study extends earlier lysosomal proteomics efforts (reviewed in Ref. 9), which had identified signalling proteins such as CREG (cellular repressor of E1A-stimulated genes), the GTPase LRG-47 and the mitogen-activated-protein-kinase kinase (MEK)-binding protein MP1. Continuing efforts to extend the characterization of lysosomal membrane proteins, the lumenal proteins and the proteins that are bound to the cytoplasmic surface of this organelle might help to identify the proteins that regulate lysosomal abundance and location.
Nucleus. Proteins of the nuclear pore complex and the nuclear envelope in S. cerevisiae and mammalian cells have been identified by proteomics studies49, 50, 51 (Box 5). The study of the nuclear pore complex in S. cerevisiae elucidated a proteomics-based Brownian-affinity-gating model for nuclear transport49. The analysis of the nuclear envelope is complicated by the contiguous nature of the outer nuclear membrane and the ER. Schirmer et al.51 derived a SUBTRACTION STRATEGY for analysing the nuclear envelope (Fig. 1). A comprehensive proteomics analysis of a MICROSOMAL MEMBRANE FRACTION was carried out and the ER proteins that were identified were subtracted from the nuclear-envelope proteome51. All of the previously identified nuclear-envelope proteins (13 in total) were identified in this study, as were 67 novel proteins. Eight of these were tagged and their nuclear-membrane localization was confirmed by immunolocalization. On the basis of the chromosomal location of the human homologues of these proteins, Schirmer et al.51 suggested that some of these new nuclear-envelope proteins might be involved in the DYSTROPHIES that have not yet been linked to a disease gene (a number of dystrophies are thought to be linked to mutations in nuclear-envelope proteins). Twenty-three of the human homologues mapped to regions of the genome that have been linked to various dystrophies. These data therefore provide an excellent starting point for understanding what is a highly complicated inner-nuclear-membrane structure and its relationship to disease. The evaluation of the localization of the remaining 59 novel proteins is eagerly awaited in order to delineate the total complement and function of the nuclear-envelope-specific proteins.
Figure 1 | Subtractive proteomics methods to enrich proteins in organelles.
The analysis of proteins in the nuclear envelope was helped by identifying proteins that are present in microsomal membranes51. Microsomal membranes and nuclear envelopes were separately enriched by centrifugation techniques. Proteins in both fractions were digested using a protease such as trypsin and analysed using multidimensional liquid-chromatography–tandem mass spectrometry. Protein sequence databases were searched, using the data from the tandem mass spectra of the peptides that were identified in each fraction, to identify the proteins present. The proteins that were identified in the microsomal membranes were then subtracted from the list of proteins that were identified in the nuclear envelope fraction to produce an enriched set of protein identifications for the nuclear envelope.
A study of the nucleolus was recently reported by Andersen and colleagues52. The nucleolus is a ribosome-generating region of the nucleus, where transcription is coordinated with processing of ribosomal RNA and ribosome biogenesis. Nucleoli were isolated from HeLa cells using a standard protocol, and protein mixtures from individual 1D gel slices were in-gel digested. The resulting peptide mixtures were analysed using LC–MS/MS, as well as ion-trap–Fourier-transform mass spectrometry. Over 11,000 unique peptides were identified, which led to the identification of 692 proteins that were considered to be nucleolar. Remarkably, 90% of the previously characterized nucleolar proteins of S. cerevisiae53 were matched with those found in this study (an increase compared to the percentage that matched a previous study54), which highlights the high degree of conservation of the fundamental nature of this compartment. In a remarkable demonstration of quantitative temporal proteomics, 489 proteins were identified that were recruited to or lost from the nucleolus following transcription inhibition or proteasome inhibition (Box 5). Significant changes were found in the number of ribosomal proteins, which decreased following transcription inhibition and increased following proteasome inhibition. The roles of all of these proteins in nucleolar morphology and function can now be addressed.
Post-translational modifications are an important way to regulate cellular processes, and proteomics methods are beginning to focus on the identification of the different forms of post-translational modification. Most studies have focused on phosphorylation, and the large-scale characterization of phosphoproteins can provide information about the regulation of processes that are specific to an organelle. Beausoleil et al.55 developed a strategy for the large-scale identification of phosphopeptides and applied it to the analysis of the nucleus of HeLa cells. By using strong cation-exchange chromatography to enrich phosphopeptides, 2,002 phosphorylation sites were identified in 967 proteins55. This enabled the specificities of the kinases that are responsible for these phosphorylation events to be determined, and showed that proline-directed kinases (for example, extracellular signal-regulated kinase-1) and acidophilic kinases (for example, casein kinase-I and -II) are responsible for 77% of the observed phosphorylation events.
Plasma membrane. An important challenge for proteomics, and for subcellular fractionation methods, is the characterization of cell-surface proteins, in particular, in the context of disease. A important advance was made by Oh et al.56, who characterized the proteins in the plasma membranes of rat lung endothelial cells, which were isolated using an established silica-coated-bead methodology56. They identified 2,000 proteins, and showed that 2 were specific to lung-endothelial cells and were exposed on the cell surface. Comparing these data with information on the plasma-membrane proteins of lung endothelial cells from rats with breast adenocarcinomas highlighted 12 proteins that are enriched in tumour endothelium plasma membranes. One of these 12 proteins was annexin A1, and injecting antibodies against annexin A1 induced tumour remission. This application of plasma-membrane proteomics to cancer therapy adds to the work of Shin et al.57 on the cell-surface proteome of cancer cells, and also adds to the studies by Tam et al.58, Zhang et al.59 and Nuhse et al.60 on various plasma membranes. Recently, Nielsen et al.61 used a variation of the DIGITONIN SHIFT protocol62 to characterize 862 plasma-membrane proteins of the mouse brain cortex and 1,685 plasma-membrane proteins of the mouse hippocampus (Box 5). Collectively, these studies give insights into the proteins that are important for communication between the cell surface, the extracellular matrix and the intracellular milieu in the context of signalling59, 60 and disease56, 57, 58.
Membrane domains and lipid rafts. Specific functional domains are found in the membranes of all organelles. LIPID RAFTS are the focus of much recent work and surrounding controversy63. These domains are extremely dynamic and it has been proposed that this allows them to function as sorting and signalling platforms63. They are commonly extracted from cells with TritonX-100 at 4°C, and the extracted domains only partially reflect the in vivo structures, which means that interpreting the results of proteomics studies on these structures is not straightforward.
Using the SILAC (stable-isotope labelling by amino acids in cell culture) method, a large-scale proteomics analysis of lipid rafts was carried out64. The SILAC method represents a quantitative strategy in which, in this case, leucine residues that each contained three deuterium atoms were incorporated into proteins. Peptides that were derived from these labelled proteins were easily identified, because they were offset by three mass units for each leucine residue. All the proteins in one of two cell populations were metabolically labelled with deuterium-substituted leucine, and one population was treated with a cholesterol-disrupting drug to destabilize lipid rafts. MS/MS was then used to identify the proteins that were depleted from lipid rafts. This approach identified 241 lipid-raft-specific proteins, which were mostly signalling proteins (Box 5).
The problem of identifying lipid-raft proteins, which are extremely hydrophobic, was addressed by the use of an SDS-aided high-performance-LC–MALDI–MS/MS proteomics approach65, which bypasses 1D gels and the in-gel digestion of proteins. This technique enabled the identification of 71 hydrophobic raft proteins, 45 of which had not been detected in previous experiments using in-gel digestion65. In another study, the dynamic nature of the lipid-raft proteome was assessed following T-cell receptor (TCR) triggering using 2DGE and MALDI–time-of-flight MS66. Although limited by the sampling problems of 2D gels, this study was remarkable and showed that TCR triggering promotes the temporally regulated recruitment of proteins that participate in TCR signalling to lipid rafts.
Large cellular structures
In addition to the study of organelles and transport intermediates, proteomics technologies have been successfully used to study large cellular structures such as the cytoskeleton and CENTROSOME. The cytoskeleton interacts with all organelles, and many cytoskeletal proteins have been identified in organelle proteomes. Studies that specifically focus on the cytoskeleton and on cytoskeletal organizing centres are important to identify regulatory molecules and molecules that function at the interface between the cytoskeleton and organelles.
Cytoskeleton. The cytoskeleton is a complex network of filaments, associated motor proteins and regulatory molecules. Novel proteomics approaches have been applied to the study of the cytoskeleton. Tubulin-affinity chromatography was used to isolate tubulin-binding proteins from A. thaliana67. The bound proteins were resolved using preparative 2DGE and identified using LC–MS/MS. In total, 122 proteins were identified from 86 spots on the original 2D gel and were divided into six functional categories: microtubule-associated proteins, translation factors, RNA-binding proteins, signalling proteins, metabolic enzymes and proteins with other functions. Half of the proteins identified had been previously shown to interact with microtubules. An interesting finding was the identification of cell-wall modification enzymes, such as endo-1,4-glucanase and endoxyloglucan glycosyl transferase, which modify cell-wall architecture and are needed for cell elongation and fruit ripening (Box 5).
Another focus has been the identification of cytoskeleton-associated functional complexes — for example, human 14-3-3 COMPLEXES68. FLAG-TAGGED human 14-3-3
was stably expressed at a level equivalent to endogenous 14-3-3 isoforms in human embryonic kidney (HEK)293 cells, and its associated proteins were isolated using immobilized anti-Flag antibodies. LC–MS/MS analysis revealed 170 proteins that associated with 14-3-3, including the seven endogenous 14-3-3 isoforms and other proteins with a broad range of functions, including control of the cytoskeleton69. Blocking the ability of 14-3-3 complexes to bind to phosphorylated proteins in vivo modified membrane dynamics and cell shape69. These data indicate the importance of phosphodependant 14-3-3 interactions for membrane dynamics and cytoskeletal function (Box 5).
A study of detergent-resistant membrane fragments from bovine neutrophils highlighted a potential interaction of a cytoskeletal submembrane complex with transmembrane signalling molecules. MALDI–time-of-flight MS analysis identified 19 important proteins in this complex70. These proteins included the cytoskeletal proteins fodrin, myosin-IIA, myosin-IG, -actinin, vimentin and the filamentous (F)-actin-binding-protein supervillin. The lipid-raft proteins stomatin, flotillin-1 and flotillin-2 were also identified. Together with data from immunofluorescence microscopy, these results indicate the existence of a lipid-raft-associated membrane skeleton in neutrophil plasma membranes70.
Centrosome. Human centrosomes function as microtubule-organizing centres in INTERPHASE cells and as SPINDLE POLES during mitosis. Centrosomes from interphase cells were quantitatively analysed using PROTEIN-CORRELATION PROFILING71. Each fraction from a density separation was proteolytically digested and the resulting peptides were analysed using LC–MS/MS and database searching. Centrosome-specific proteins were distinguished from nonspecific proteins by tracking their abundance in a fractionation profile of peptides that were identified in five sucrose fractions of the centrosome preparation. A consensus profile for peptides that could be assigned to known centrosomal proteins was established, and peptides that deviated from this profile were considered nonspecific. When the abundance of the peptides correlated with the abundance of the peptides from proteins that are known to localize to the centrosome, the corresponding proteins were presumed to be localized to the centrosome. In this analysis 50 known centrosome proteins, 41 candidate centrosome proteins and 23 novel centrosome-localized components were identified (Box 5). Localizing the last group of proteins to the centrosome provided new insights into possible centrosome functions.
Mitotic spindle. The mitotic spindle encompasses microtubule spindle poles, centrosomes and the KINETOCHORE. Mitotic spindle poles that were enriched from HeLa cells were analysed using MS/MS after protein separation on 1D SDS-PAGE gels72. A total of 151 proteins that were previously known to be associated with the spindle apparatus, centrosomes or kinetochores were identified. However, most of the other proteins that were identified (644) had not been previously shown to be centrosome associated, and 154 of these proteins were uncharacterized. 17 of the uncharacterized proteins were tagged and transfected into mitotic cells, which led to the identification of 6 new spindle components72 (Box 5). This emphasizes the point that not all proteins that are identified in cellular fractions are bona fide components of the compartment under investigation.
Midbody During CYTOKINESIS, when membrane cleavage is almost complete, the plasma membrane of the cleavage furrow tapers to form the MIDBODY. The midbody is derived from the central spindle and persists as an attachment between the two daughter cells before final separation. A clear function has never been ascribed to this structure, but it is assumed that it will contain proteins that are involved in cytokinesis. The mammalian midbody was isolated from Chinese hamster ovary cells and analysed using MudPIT73. A total of 160 candidate mammalian midbody proteins were identified, and these were functionally characterized in Caenorhabditis elegans using RNAi. Of the 160 candidate proteins, 147 had obvious homologues in C. elegans, and the RNAi-mediated suppression of protein expression resulted in scorable defects for 141 of these proteins. RNAi against 85 of the 147 homologues caused defects in cytokinesis, and 10 novel proteins were localized to the midbody in HeLa cells using immunofluorescence. Unexpectedly, a role for the abundant ER molecular chaperone GRP94 was uncovered in chromosome segregation (Box 5). These identifications mean that many new molecules can now be tested for their involvement in the process of cytokinesis. The confirmation of proteomics results using techniques such as RNAi is exactly what is needed to exploit the power of proteomics.
Conclusions and future directions
The use of proteomics technologies to characterize organelles and large cellular structures has provided functional insights, although the data become more complete and convincing when the proteomics analyses are combined with protein-localization and protein-knockdown techniques. Identifying the phenotype that is produced by disrupting a protein that is localized to a specific organelle can clarify how the organelle functions and indicate the role of the individual protein. It is clear that MS-based proteomics methods are being used to study organelle function and will be used increasingly to understand diseases and to identify disease specific biomarkers. Recent advances in characterizing membrane proteins and post-translational modifications and in bioinformatics methods are essential for the future of organelle studies, and we anticipate continued advances in proteomics technologies. In our opinion, the time has come for an organized effort to characterize the subcellular proteomes of several different types of human cell comprehensively. Such an effort would allow us to improve technologies to enrich subcellular structures and to carry out high-throughput, follow-up studies such as immunofluorescence-localization and RNAi studies. The real power of proteomics will be realized when comprehensive and comparative analyses of organelles are carried out in different cell types, under different physiological states, and in normal and diseased cells.