The single most common application of proteomics is protein identification. Most investigators use proteomics approaches to isolate and display proteins based on their own specific criteria and then identify the proteins. Protein identification provides immediate information that will direct subsequent experimentation. For example, the identity of a protein can reveal an expected result, validate a proteomics approach, provide completely unexpected information, or reveal that your biochemical method is not working at all. We feel that the most critical stage of any proteomics approach is the strategic design for the isolation of protein targets. In recent years, as the technology of MS has improved, there has been a de-emphasis on the “front-end” of proteomics experiments compared to data analysis. This can result in the isolation of hundreds of irrelevant proteins for identification, consuming both time and effort. Our general strategy is to devise techniques that enrich for low-abundance proteins and then analyze only the proteins that appear on differential display or are isolated by affinity chromatography. To accomplish this, we use affinity columns and other strategies to select for protein targets. In each case, protein samples are subjected to a series of precolumns and high-stringency washes to remove nonspecific proteins. This reduces the number of irrelevant proteins for analysis.
Characterization of Protein Complexes
Many laboratories are now engaged in an effort to characterize protein complexes by MS. Examples include Link et al. utilizing multidimensional LC and MS/MS to identify proteins (95
) or Mann and colleagues identifying proteins present after immunoprecipitation of protein complexes (124
). Recently, Macara, Haystead, and coworkers used MS to identify interacting proteins with the Cdc42 effector, Borg3 (80
). In this case, the “bait” protein, Borg3, was produced as a glutathione S
-transferase (GST) fusion in E. coli
and then mixed with NIH 3T3 cell lysate. Four interacting proteins were identified by mixed-peptide sequencing: heat shock protein Hsp70 and three septins including Septin6, Cdc10, and Nedd5 (Fig. ). None of these proteins were present in the GST-only control sample. Although the interaction with Hsp70 was not pursued, it was shown from coimmunoprecipitation studies that endogenous Borg3 interacts with endogenous Cdc10 and Nedd5 (80
). Additional proof from expression and structure-function studies confirmed a role for the Borg proteins as regulators of septin organization. It should be noted that although several proteins were quickly identified as Borg3 interactors by the pull-down experiment, it took several more months of work to confirm this interaction.
FIG. 11. Identification of novel protein interactions by protein coprecipitation. (A) Pull-down experiment with a control (GST) or target (GST-Borg3) protein using 35S-labeled NIH 3T3 cell lysate. (B) Large-scale affinity purification of GST-Borg3 from the NIH (more ...)
Protein Expression Profiling
The largest application of proteomics continues to be protein expression profiling. Through the use of two-dimensional gels or novel techniques such as ICAT, the expression levels of proteins or changes in their level of modification between two different samples can be compared and the proteins can be identified. This approach can facilitate the dissection of signaling mechanisms or identify disease-specific proteins.
Expression profiling by two-dimensional electrophoresis.
Currently, the majority of protein expression profiling studies are performed by 2-DE. Several diseases have been studied, including heart disease (44
) and cancer (30
). Cancer cells are good candidates for proteomics studies because they can be compared to their nontransformed counterparts. Analysis of differentially expressed proteins in normal versus cancer cells can (i) identify novel tumor cell biomarkers that can be used for diagnosis, (ii) provide clues to mechanisms of cancer development, and (iii) identify novel targets for therapeutic intervention. Protein expression profiling has been used in the study of breast (121
), esophageal (121
), bladder (30
) and prostate (114
) cancer. From these studies, tumor-specific proteins were identified and 2-D protein expression databases were generated. Many of these 2-D protein databases are now available on the World Wide Web (15
Isotope-coded affinity tags.
Recently, a novel method for protein expression profiling was introduced that does not depend on the separation of proteins by 2-DE. This method is known as isotope-coded affinity tags (ICAT) and relies on the labeling of protein samples from two different sources with two chemically identical reagents that differ only in mass as a result of isotope composition (66
). Differential labeling of samples by mass allows the relative amount of protein between two samples to be quantitated in the mass spectrometer. An example of the methodology of ICAT is shown in Fig. . Cell extract from two different samples is reacted with one of two forms of the ICAT reagent, an isotopically light form in which the linker contains eight hydrogens or a heavy form in which the linker contains eight deuterium atoms. The ICAT reagent reacts with cysteine residues in proteins via a thiol-reactive group and contains a biotin moiety to facilitate purification (Fig. ). Peptides are recovered on the basis of the biotin tag by avidin affinity chromatography and are then analyzed by MS. The difference in peak heights between heavy and light peptide ions directly correlates with the difference in protein abundance in the cells. Thus, if a protein is present at a threefold higher level in one sample, this will be reflected in a threefold difference in peak heights. Following quantitation of the peptides, they can be fragmented by MS/MS and the amino acid sequence can be obtained. Thus, using this approach, proteins can be identified and their expression levels can be compared in the same analysis.
FIG. 12. The ICAT method for measuring differential protein expression. (A) Structure of the ICAT reagent. ICAT consists of a biotin affinity group, a linker region that can incorporate heavy (deuterium) or light (hydrogen) atoms, and a thiol-reactive end group (more ...)
The single biggest advantage of this method is the elimination of the 2-D gel for protein quantitation. As a result, an increased amount of sample can be used to enrich for low-abundance proteins. Alternatively, the cell lysate can be fractionated prior to reaction with the ICAT reagent. This can allow the enrichment of low-abundance proteins before the analysis begins. The main disadvantages are that currently this method works only for proteins containing cysteine, even though this includes the majority of proteins (68
). In addition, peptides must contain appropriately spaced protease cleavage sites flanking the cysteine residues. Finally, the ICAT label is large (~500 kDa) and remains with each peptide throughout the analysis. This can make database searching more difficult, especially for small peptides with limited sequence (4
). Sensitivity may also be of concern since tagged peptides derived from low-copy proteins are likely to be poorly recovered during the affinity step as a result of nonspecific interactions with avidin-Sepharose. Studies have been performed to optimize the labeling of proteins with the ICAT reagent (151
Protein arrays are undergoing rapid development for the detection of protein-protein interactions and protein expression profiling (17
). Recently, protein microarrays were created using ordinary laboratory equipment (98
). Proteins were immobilized by being covalently attached to glass microscope slides, and the protein microarrays were shown to be capable of interacting with other proteins, small molecules, and enzyme substrates (98
). In another report, 5,800 yeast proteins were expressed and printed onto microscope slides. These protein microarrays were used to identify novel calmodulin- and phospholipid-interacting proteins (180
). These reports indicate that protein arrays hold great promise for the global analysis of protein-protein and protein-ligand interactions. Undoubtedly, these arrays will improve as the technology for their creation is developed and refined.
Proteomics Approach to Protein Phosphorylation
Posttranslational modification of proteins is a fundamental regulatory mechanism, and characterization of protein modifications is paramount for understanding protein function. MS is one of the most powerful tools for the analysis of protein modifications because virtually any type of protein modification can be identified. Although we focus here on protein phosphorylation, the analysis of other types of protein modification by MS has been described (25
). Protein phosphorylation is one of the most common of all protein modifications and has been found in nearly all cellular processes (74
). MS can be used to identify novel phosphoproteins, measure changes in the phosphorylation state of proteins in response to an effector, and determine phosphorylation sites in proteins. Identification of phosphorylation sites can provide information about the mechanism of enzyme regulation and the protein kinases and phosphatases involved. A proteomics approach to protein phosphorylation has the advantage that instead of studying changes in the phosphorylation of a single protein in response to some perturbation, one can study all the phosphoproteins in a cell (the phosphoproteome) at the same time. A common approach to studying protein phosphorylation events is the use of in vivo labeling of phosphoproteins with inorganic 32
P. The phosphoproteomes of cells that differ in some way (e.g., normal versus diseased) can be analyzed by growing cells in inorganic 32
P and creating cell lysates. Changes in the phosphorylation state of proteins can then be examined by 2-DE and autoradiography. Proteins of interest are excised from the gel and microsequenced by MS. A major limitation of this approach is that while many phosphorylated proteins can be visualized by autoradiography, they cannot be identified because of their low abundance. One solution to this problem is enrichment of the phosphoproteome.
Enrichment of the phosphoproteome of a cell can allow the identification of low-copy phosphoproteins that would otherwise go undetected. In one approach, phosphoproteins were enriched by conversion of phosphoserine residues to biotinylated residues (118
). This method is an extension of techniques originally developed by Hielmeyer and colleagues (108
) and more recently by our laboratory (51
) for the identification of phosphorylation sites using Edman sequencing. Following derivatization, proteins that were formerly phosphorylated can be isolated by avidin affinity chromatography (118
). Proteins immobilized on avidin beads can then be eluted with biotin, theoretically resulting in the isolation of the entire phosphoserine proteome (Fig. ). By increasing the amount of cell lysate used for avidin affinity chromatography, low-abundance phosphoproteins can be enriched. However, this technique does not work for phosphotyrosine and the reactivity of phosphothreonine by this method is very poor (118
). Tyrosine-phosphorylated proteins can be isolated by the use of antiphosphotyrosine antibodies (124
). As an alternative, another method for phosphopeptide enrichment was devised to allow the recovery of proteins phosphorylated on serine, threonine, and tyrosine (179
). In this method, a protein or mixture of proteins is digested to peptides with a protease and then subjected to a multistep procedure for the conversion of phosphoamino acids into free sulfhydryl groups. To capture the derivatized peptides, the free sulfhydryl groups in the peptides are then reacted with iodoacetyl groups immobilized on glass beads. Using this method, several phosphopeptides were recovered from β-casein and from a yeast cell extract, although it was unclear whether all the proteins isolated from the yeast extract were phosphoproteins (179
FIG. 13. Phosphopeptide and phosphoprotein enrichment. (A) Enrichment of phosphopeptides. Phosphoproteins are digested with a protease, and the phosphate groups are converted to biotin tags (119). Once biotinylated, the peptides can be selectively recovered with (more ...)
Enrichment of the phosphoproteome can also be combined with protein profiling by 1- or 2-DE. In this way, changes in protein amount observed on electrophoresis will reflect the level of protein phosphorylation (Fig. ). Recently, the principle of protein quantitation by ICAT has been combined with phosphoprotein enrichment (60
). This was accomplished by the introduction of isotopic label into ethanedithiol, the reagent used to convert the alkene created by β-elimination of phosphoserine into a free sulfhydryl group. In this way, the differences in the amount of phosphoproteins in extracts can be analyzed quantitatively in the mass spectrometer (60
). It should be noted that because of the chemistry used in both of these methods, these techniques are relatively insensitive and require tens of picomoles of phosphoprotein. As a result, we have found that these methods as currently designed are impractical for the isolation and enrichment of low-abundance phosphoproteins.
Phosphorylation site determination by Edman degradation.
Edman sequencing is still a widely used method for determining phosphorylation sites in proteins labeled with 32
P, either in vitro or in vivo (5
). This is because sites can be determined at the sub-femtomolar level if enough radioactivity can be incorporated into the phosphoprotein of interest. In our hands, this can be as little as 1,000 cpm (not ideal). Briefly, a 32
P-labeled protein is digested with a protease and the resulting phosphopeptides are separated and purified by reverse-phase HPLC or thin-layer chromatography (TLC) (Fig. ). The isolated peptides are then cross-linked via their C termini to an inert membrane (e.g. Immobilon P; PerSeptive Biosystems). The radioactive membrane is subjected to several rounds of Edman cycles, and radioactivity is collected after the cleavage step. The released 32
P is counted in a scintillation counter. This method positionally places the phosphoamino acid within the sequenced phosphopeptide. Of course, this is meaningful only if the sequence of the phosphopeptide is already known. In addition, the analysis ceases to become quantitative beyond 30 Edman cycles (even with efficient, modern Edman machines) due to well-understood issues with repetitive yield associated with Edman chemistry.
FIG. 14. Strategies for determination of phosphorylation sites in proteins. Proteins phosphorylated in vitro or in vivo can be isolated by protein electrophoresis and analyzed by MS. (A) Identification of phosphopeptides by peptide mass fingerprinting. In this (more ...)
Recently, our laboratory has extended the usefulness of phosphorylation site characterization by Edman chemistry through the development of the cleaved radioactive peptide (CRP) program (J. A. MacDonald, A. J. Mackay, W. R. Pearson, and T. A. J. Haystead, submitted for publication). In CRP analysis, one requires only that the sequence of the protein be known. Purification and sequencing of individual peptides is not required. Radiolabeled proteins (isolated following immunoprecipitation from 32
P-labeled cells, for example) are cleaved at predetermined residues by the action of a protease. The phosphopeptides are then separated by HPLC or TLC (if only one site is present, no peptide separation is required), cross-linked to the inert membrane, and carried through 25 to 30 Edman cycles. The sequence of the target protein is entered into the CRP program. This program predicts how many Edman cycles are required to cover 100% of all the serines, threonines, and tyrosines from the site of cleavage. Generally, one round of CRP analysis narrows the number of possible sites to 5 to 10 for most proteins. Phosphoamino acid analysis can be used to reduce the number of possibilities still further. The CRP analysis is then repeated following cleavage with a second protease (usually one cutting at R, but M and F are alternatives). The second round of CRP usually unambiguously localizes the phosphoamino acid to one possible site. The technique does not work if sites are more than 30 amino acids away from all possible cleavage sites. The finding that CRP analysis is not applicable may in itself confine a phosphorylation site to a segment of the protein that is likely to produce very large proteolytic fragments. The Cleavage of Radioactive Proteins (CRP) program is accessible at http://fasta.bioch.virginia.edu/crp/
and was written in collaboration with Aaron Mackey and Bill Pearson of the University of Virginia (MacDonald et al., submitted).
Phosphorylation site determination by mass spectrometry.
Because of its sensitivity, MS can allow the direct sequencing of phosphopeptides, resulting in unambiguous phosphorylation site identification. Below, a brief overview of some common methods for phosphorylation site determination by MS are given. A more complete discussion of this topic is provided by Mitchelhill and Kemp (110
). Identification of phosphorylation sites in proteins provides several unique challenges for the mass spectrometrist. For example, unlike in protein identification, where analysis of any peptide within the protein can be informative, phosphorylation site analysis requires that the phosphorylated peptide be analyzed. This means that considerably more protein is required for analysis. In addition, phosphorylation can alter the cleavage pattern of a protein and the resulting phosphopeptides may require different purification methods. To isolate and purify the phosphopeptides of interest, it may be necessary to alter the way in which the phosphoprotein is digested and to alter the pH or the chromatographic material used for peptide purification (27
(i) Phosphopeptide sequencing by MS/MS.
In our laboratory, we have found that a combination of HPLC, Edman degradation, and phosphopeptide sequencing by MS/MS provides the best results for phosphorylation site determination (Fig. ). Following excision and digestion of a 32P-labeled protein, the peptides are resolved by HPLC. By monitoring HPLC fractions for radioactivity, the phosphopeptides can be selected for analysis. This reduces the complexity of the peptide mixture before MS is performed and facilitates phosphopeptide identification (Fig. ).
Phosphopeptides can be identified from a mixture of peptides by a method known as precursor ion scanning (116
). In this method, the second mass analyzer in the mass spectrometer is set at the mass of the reporter ion for the phospho group (PO3−
) of m/z
= 79. Peptides are sprayed under neutral or basic conditions, and phosphopeptides are identified in the precursor ion scan only if their fragmentation yields an ion of m/z
= 79. Once a phosphopeptide is identified, the peptide mixture is sprayed under acidic conditions and the phosphopeptide is sequenced by conventional tandem MS/MS. On fragmentation of the phosphopeptide, phosphoserine can be identified by the formation of dehydroalanine (69 Da), the β-elimination product of phosphoserine. Similarly, phosphothreonine can be identified by the formation of its β-elimination product, dehydroamino-2-butyric acid at 83 Da (116
(ii) Analysis of phosphopeptides by MALDI-TOF.
MALDI-TOF mass spectrometry can also be used to identify phosphopeptides (81
). When phosphorylated peptides are subjected to ionization by MALDI, phosphate groups are frequently liberated from the peptides. This is the case for phosphoserine- and phosphothreonine-containing peptides, which can liberate HPO3
, resulting in a neutral loss of 80 and 98 Da, respectively. Careful examination of the TOF spectrum for differences in peptide masses of 80 Da that are not found in the unphosphorylated peptide control can identify phosphopeptides. Phosphopeptides can also be identified by treating one of two identical samples with protein phosphatase to liberate phosphate groups (Fig. ). Once a phosphopeptide is identified, it can be sequenced by MS/MS for identification of the phosphorylation site (178
Yeast Genomics and Proteomics
One of the most exciting applications of proteomics involves combining this technology with the power of yeast genetics to delineate signaling events in vivo. Our laboratory has published two papers using this strategy to identify in vivo targets for protein phosphatases (9
). In one study (9
), we identified physiological substrates for the Glc7p-Reg1p complex by examining the effects of deletion of the REG1
gene on the yeast phosphoproteome. In S. cerevisiae
, PP-1 (Glc7p) and its binding protein, Reg1p, are essential for the regulation of glucose repression pathways. The target for this phosphatase complex was not known. Analysis by 2-D phosphoprotein mapping identified two distinct proteins that were greatly increased in phosphate content in reg1
Δ mutants. Mixed-peptide sequencing identified these proteins as hexokinase II (Hxk2p) and the E1α subunit of pyruvate dehydrogenase. We then went on to validate these findings in a comprehensive biochemical study. Consistent with increased phosphorylation of Hxk2p in response to REG1
deletion, fractionation of yeast extracts by anion-exchange chromatography identified a Hxk2p phosphatase activity in wild-type strains that was selectively lost in the reg1
Δ mutant. Having carried out these studies, we attempted to rescue the reg1
Δ phosphoprotein phenotype by overexpressing both wild-type and mutant Reg1p in the deletion strains. Here, both the phosphorylation state of Hxk2p and Hxk2p phosphatase activity were restored to wild-type levels in the reg1
Δ mutant by expression of a LexA-Reg1p fusion protein. In contrast, expression of a LexA-Reg1p protein containing mutations at phenylalanine in a putative PP-1C (the catalytic subunit) binding site motif (K/R)(X)(I/V)XF was unable to rescue Hxk2p dephosphorylation in intact yeast or restore Hxk2p phosphatase activity. These results demonstrate that Reg1p targets PP-1C to dephosphorylate Hxk2p in vivo and that the peptide motif (K/R)(X)(I/V)XF is necessary for its PP-1 targeting function. These studies therefore demonstrate how a proteomics approach can be used to first identify enzyme targets in cells and then direct all further analysis to verify the findings. It should be pointed out that often 6 to 12 months of work ensues following the initial sequencing of the targeted proteins. Nevertheless, clearly a combined proteomics and genetics approach greatly enhances one's ability to directly answer key biological questions. We believe that a similar strategy could be adopted with transgenic or knockout mouse work, particularly in cases where there is no obvious phenotype.
Proteome mining is a functional proteomics approach used to extract protein information from the analysis of specific subproteomes. The strategy of proteome mining is shown in Fig. . The principles of proteome mining are based on the assumption that all drug-like molecules selectively compete with a natural cellular ligand for a binding site on a protein target. In a proteome mine, natural ligands are immobilized on beads at high density and in an orientation that sterically favors interaction with their protein targets. The immobilized ligand is then exposed to whole-animal or tissue extract, and bound proteins are evaluated for specificity by protein sequencing. In the prototypic example from our laboratory, ATP is immobilized in the “protein kinase orientation” (via its gamma phosphate). Microsequencing of the proteins that were eluted with free ATP demonstrated that the nucleotide selectively recovered purine binding proteins including protein kinases, dehydrogenases, various purine-dependent metabolic enzymes, DNA ligases, heat shock proteins, and a variety of miscellaneous ATP-utilizing enzymes (P. R. Graves, J. Kwiek, P. Fadden, R. Ray, K. Hardeman, and T. A. J. Haystead, submitted for publication). This immobilized proteome represents ~4% of the expressed eukaryotic genome.
FIG. 15. Proteome-mining strategy. Proteins are isolated on affinity column arrays from a cell line, organ, or animal source and purified to remove nonspecific adherents. Then, compound libraries are passed over the array and the proteins eluted are analyzed by (more ...)
We have utilized this captured proteome (the purine binding cassette proteome) to test the selectivity of purine analogs that inhibit protein kinases and stress-induced ATPases in vitro. Using a proteome-mining ATP affinity array apparatus constructed in our laboratory, sufficient biomass was applied to ensure the recovery, per column, of 1 fmol of any protein expressed at 100 copies/cell (107 cells). After washing, each column in the array is eluted in parallel with molecules from a purine-based iterative library and fractions are collected. Eluates are screened for protein, and positive fractions generally contain a single protein, a small number of structurally related proteins, or a complex mixture. Only the first two categories are sequenced, since the third resulted from elution with a nonselective inhibitor. Once one has identified an eluted protein, one has all the necessary information on how to proceed. The first decision is biological relevance. Does the eluted protein(s) in any given fraction have relevance to any human disease? If the protein has no obvious use as a drug target, it is ignored. If the protein is deemed relevant, one immediately has a lead molecule and a defined target. In cases where a single protein is eluted, the lead is likely to be selective because it had an equal opportunity to interact with the rest of the captured proteome (~4% of the genome). Selectivity can be tested by increasing the concentration of the lead compound during elution from nanomolar to micromolar. Information concerning potential toxicity can be gained by sequencing other proteins that are simultaneously eluted or eluted at higher concentrations. If some of these are undesirable targets, iterative substitutions can be made around the lead scaffold to improve selectivity. Proof of principle of this technology was obtained by using an iterative library derived from the heat shock protein 90 inhibitor geldanamycin, and a new physiological target, ADE2, was identified (P. Fadden, V. J. Davisson, L. Neckers, and T. A. J. Haystead, unpublished data). Screening Combichem libraries through a proteome-mining approach exploits the serendipitous nature of drug discovery to its maximum, merely because it accelerates the hit rate over a conventional screen by a factorial of the proteome that is bound. In the case of purine binding proteins, this may be several hundredfold. Protein microsequencing, the data contained within the various genome projects, and the ability to instantly search the literature for relevance enable one to interpret the outcomes in a rationale way.
We are currently using proteome mining to discover new antimalarial drugs that target purine binding proteins in the blood stage of infection. Because of the essential roles of purine-utilizing enzymes in cellular function, it is our hypothesis that these proteins are attractive candidates for a new generation of antimalarial drugs. In our malaria project, the P. falciparum (blood stage) and human red blood cell purine binding proteome are captured on ATP affinity arrays and simultaneously screened against purine-based combinatorial libraries. Combining both proteomes enables the selectivity and potential toxicity of a lead molecule to be measured early in the discovery process. Microsequencing enables human proteins to be readily discriminated from malarial ones. An additional benefit of mining the entire malarial purine binding cassette proteome is that multiple leads and their targets will be identified. Combined therapies that target multiple genes simultaneously are likely to exert such tremendous selective pressure on the targeted pathogen that it cannot develop resistance. We are currently expanding our immobilized natural-ligand library in order to apply proteome mining to other areas of biology.
Challenges for Proteomics
The study of proteins, in contrast to that of DNA, presents a number of unique challenges. For example, there is no equivalent of PCR for proteins, so the analysis of low-abundance proteins remains a major challenge. In addition, in protein interaction studies, native conformations of proteins must be maintained to obtain meaningful results. Can proteins be studied on a large scale with speed, sensitivity, and reliability? In the last several years, recognition of the limitations of proteomics are beginning to point the field in new directions.
Although the technology for the analysis of proteins is rapidly progressing, it is still not feasible to study proteins on a scale equivalent to that of the nucleic acids. Most of proteomics relies on methods, such as protein purification or PAGE, that are not high-throughput methods. Even performing MS can require considerable time in either data acquisition or analysis. Although hundreds of proteins can be analyzed quickly and in an automated fashion by a MALDI-TOF mass spectrometer, the quality of data is sacrificed and many proteins cannot be identified. Much higher quality data can be obtained for protein identification by MS/MS, but this method requires considerable time in data interpretation. In our opinion, new computer algorithms are needed to allow more accurate interpretation of mass spectra without operator intervention. In addition, to access unannotated DNA databases across species, these algorithms should be error tolerant to allow for sequencing errors, polymorphisms, and conservative substitutions. New technologies will have to emerge before protein analysis on a large-scale (such as mapping the human proteome) becomes a reality.
Another major challenge for proteomics is the study of low-abundance proteins. In some eukaryotic cells, the amounts of the most abundant proteins can be 106-fold greater than those of the low-abundance proteins. Many important classes of proteins (that may be important drug targets) such as transcription factors, protein kinases, and regulatory proteins are low-copy proteins. These low-copy proteins will not be observed in the analysis of crude cell lysates without some purification. Therefore, new methods must be devised for subproteome isolation. Despite these limitations, proteomics, when combined with other complementary technologies such as molecular biology, has enormous potential to provide new insight into biology. The ability to study complex biological systems in their entirety will ultimately provide answers that cannot be obtained from the study of individual proteins or groups of proteins.