The use of two-dimensional gels

During the early years of proteomics and until relatively recently, profiling of protein expression in disease relied primarily on the use of two-dimensional polyacrylamide gel electrophoresis (2D PAGE), which was later combined with mass spectrometry². Most studies of this nature followed an approach in which a cocktail was used to solubilize the protein contents of an entire cell population, tissue or biological fluid, followed by separation of the protein contents of the lysate using 2D gels and visualization of the separated proteins using silver staining. It became clear that such an approach allows only a limited display of protein content that consisted of relatively abundant proteins. Nevertheless, profiling of disease tissues using this approach has had some utility. For example, it was demonstrated long before the use of DNA microarrays that leukaemias could be classified into their different subtypes using 2D PAGE³.

Numerous other studies have also identified disease-related changes in protein expression, primarily using 2D PAGE and mass spectrometry. One such example is provided by studies of heart disease⁴, the spectrum of which encompasses a broad set of pathological conditions, some with acute onset of severe disease and others with slow, chronic progression. To assist in data gathering and mining, online 2D gel-derived databases of protein expression in the myocardium for human and other species were constructed^{5, 6}. These databases have allowed investigators to compare data and establish reference standards. Relevant findings have emerged from studies of changes in myocardial proteins associated with human heart failure as well as from studies of animal models of heart failure and of isolated rat myocytes⁴. Although only a small proportion of the proteome has been analysed, pronounced changes in the composition of the cardiac proteome have been found, affecting proteins with diverse functions. Altered overall levels of specific proteins or altered post-translational modifications of proteins such as myosin light chain 2 have been reported in the failing heart⁷. And protein expression studies have uncovered proteins that exhibited new disease-related post-translational modifications with predicted functional relevance^{8, 9, 10, 11}. This level of progress typifies that made in profiling disease tissue in a wide variety of diseases.

An important development in 2D PAGE is the use of immobilized pH gradients (IPGs) in which the pH gradient is fixed within the acrylamide matrix. IPGs also allow production of gels that cover a defined pH range, from wide to narrow^{12, 13}. A variation on this theme is the use of so-called 'zoom gels' in which the protein contents of an individual sample are first fractionated into narrow pH ranges under low resolution, and then each fraction undergoes high-resolution separation by 2D PAGE¹⁴. For example, the pre-fractionation of serum using this approach has enhanced the ability to detect low-abundance and potentially new circulating disease-marker proteins¹⁴. Yet another innovation in 2D gels is the use of differential in-gel electrophoresis (DIGE), in which two pools of proteins are labelled with different fluorescent dyes¹⁵. The labelled proteins are mixed and separated in the same 2D gel. In one study¹⁶, 2D DIGE was applied to quantify the differences in protein expression between oesophageal carcinoma cells and normal epithelial cells. A large number of proteins were found to be either upregulated or downregulated in cancer cells.

Sample fractionation prior to analysis

Some of the main challenges facing expression proteomics, be it using 2D PAGE or any other approach, include the great dynamic range of protein abundance and a wide range of protein properties, including mass, isoelectric point, extent of hydrophobicity and post-translational modifications. Reducing sample complexity prior to analysis — for example, by analysing protein subsets and subcellular organelles separately — improves the reach of 2D gels or other separation techniques for the quantitative analysis of low-abundance proteins. An elegant demonstration of the power of sub-proteome analysis is illustrated in studies of phagosomes¹⁷; these have led to the identification of over 250 proteins from this organelle and the demonstration that phagosomes are formed by direct association and fusion of endoplasmic reticulum to the plasma membrane during early phagocytosis.

The isolation of sub-proteomes may be combined with protein tagging to further enhance sensitivity, as in the case of surface-membrane proteins, a compartment rich in diagnostic and therapeutic targets. Protein tagging technologies are currently being implemented for the comprehensive analysis of the cell-surface proteome (Fig. 1). A surface-protein biotinylation strategy, coupled with the use of mass spectrometry, was applied to the gastric pathogen Helicobacter pylori, leading to the identification of new surface-membrane proteins¹⁸. This strategy has also led to the detection and identification of many new proteins on the surface of cancer cells (Fig. 1)¹⁹.

Figure 1: Affinity capture of surface-membrane proteins.

Biotinylation reagents provide a 'tag' that transforms poorly detectable surface-membrane proteins into probes that can be recognized by a labelled detection reagent. a, Intact cells are tagged using lipid-insoluble biotin reagents. Tagged proteins are captured after cell lysis using avidin columns and subsequently eluted. Following a separation step using 2D gels or other means, tagged proteins are detected with a labelled avidin conjugate. Individual proteins are identified by mass spectrometry. b, Close-up section of a 2D pattern in which biotinylated proteins are selectively visualized (top) in contrast with the pattern of the same whole-cell lysate visualized by silver staining (bottom). Several selectively visualized proteins were found to have chaperone functions¹⁹.

High resolution image and legend (50K)

Beyond 2D gels for disease expression profiling

Even with all the improvements that could be introduced, 2D gels will probably remain a rather low-throughput approach that requires a relatively large amount of sample. The latter is particularly problematic for clinical samples, as such samples are generally procured in limited amounts. Furthermore, tissue heterogeneity complicates the analysis of clinical samples. Various tissue microdissection approaches are beneficial to reduce heterogeneity, but they further reduce the amount of sample available. In particular, the use of laser-capture microdissection, which allows defined cell types to be isolated from tissues, yields amounts of proteins that are difficult to reconcile with the need for greater amounts for 2D gels¹.

Undoubtedly, various non-gel-based schemas that rely on liquid-based separations of proteins or peptides, with or without tagging, will have utility for disease proteomics, particularly given their potential for automation. Additionally, advances in microfluidic technology will likely allow automated separation of proteins in complex lysates using much reduced sample amounts. Microfluidic systems already have been integrated with mass spectrometry for protein digestion and identification²⁰.

Non-separation-based strategies, including direct profiling using mass spectrometry or the use of protein microarrays, are important developments. Mass spectrometry has been applied to the in situ proteomic analysis of tissues, an approach that allows imaging of protein expression in normal and disease tissues²¹. By this method, frozen tissue is sliced and sections are applied on a matrix-assisted laser desorption/ionization (MALDI) plate and analysed at regular spatial intervals. The mass spectra obtained at different intervals are compared, yielding a spatial distribution of individual masses across the tissue section (Fig. 2). Mass profiles of tissue sections obtained from normal and disease tissues may be compared to detect altered protein expression. Tumour analyses using this approach have uncovered differences in protein expression between normal and tumour tissues that may have specificity for different tumour types²¹.

Figure 2: Imaging mass spectrometry.

Transverse sections of rat brain were cut, thaw-mounted on the target plate and coated with matrix²¹. A survey scan was performed first with data acquisition taken randomly across the section to generate an average protein profile. Over 200 individual mass peaks were detected in a mass-to-charge (m/z) range of up to 40,000. The figure presents an optical image of the brain section prior to matrix deposition. The section was scanned by acquiring 74 75 points with a resolution of 180 m by averaging spectra produced by 15 laser shots using an automated imaging computer algorithm. In this scan, the intensity of all of the different mass signals was monitored. Fifteen ion-density maps are shown, each obtained for different protein signals; some of these, in particular m/z 6,844, have low intensities. As expected, some proteins were found to be highly specific for a given brain region. This is particularly striking for the density maps of the proteins detected at m/z 5,631 and m/z 18,388, which are almost 'negatives' of each other.

High resolution image and legend (128K)

DNA and protein microarrays in disease investigation

The quest for disease biomarkers using proteomics

There is substantial interest in applying proteomics to the identification of disease markers. Approaches include comparative analysis of protein expression in normal and disease tissues to identify aberrantly expressed proteins that may represent new markers, analysis of secreted proteins in cell lines and primary cultures, and direct serum protein profiling. The potential of mass spectrometry to yield comprehensive profiles of peptides and proteins in biological fluids without the need to first carry out protein separations has attracted interest. In principle, such an approach is highly suited for marker identification because of reduced sample requirements and high throughput.

This approach is currently popularized, particularly for serum analysis, by the technology referred to as surface-enhanced laser desorption/ionization¹. Microlitre quantities of serum from many samples are applied to the surface of a protein-binding plate, with properties to bind a class of proteins. The bound proteins are treated and analysed by MALDI. The mass spectra patterns obtained for different samples reflect the protein and peptide contents of these samples. Patterns that distinguish between cancer patients and normal subjects with remarkable accuracy have been reported for several types of cancer¹. The main drawbacks of direct analysis of tissues or biological fluids by MALDI are the preferential detection of proteins with a lower molecular mass and the difficulty in determining the identity of proteins owing to post-translational modifications obscuring the correspondence of measured and predicted masses. Occasionally the masses observed match precisely the predicted masses of specific proteins. This was the case in a study of proteins secreted by stimulated CD8 T cells, which led to the identification of the small proteins -defensin 1, 2 and 3 as contributing to the anti-HIV-1 activity of CD8 antiviral factor³⁴.

A productive approach for the identification of cancer markers has been the analysis of serum for autoantibodies against tumour proteins. There is increasing evidence for an immune response to cancer in humans, demonstrated in part by the identification of autoantibodies against a number of intracellular and surface antigens detectable in sera from patients with different cancer types³⁵. The identification of panels of tumour antigens that elicit an antibody response may have utility in cancer screening, diagnosis or in establishing prognosis, and in immunotherapy against the disease.

There are several approaches for the detection of tumour antigens that induce an immune response³⁵. A number of antigens have been detected by screening expression libraries with patient sera^{36, 37, 38, 39, 40, 41} or, more recently, by using a random peptide-library approach⁴². Multiple proteins that induce autoantibodies that are specific for different types of cancer have been identified using 2D gels to separate tumour proteins, followed by western blotting and incubation with patient sera⁴³. For most antigenic proteins identified using this approach, post-translational modifications contributed to the immune response. In a study of lung cancer, sera from 60% of patients with lung adenocarcinoma and from 33% of patients with squamous-cell lung carcinoma, but from none of the non-cancer controls, exhibited immunoglobulin--based reactivity against proteins identified as glycosylated annexins I and II⁴⁴. Microarrays that contain proteins derived from tumour cells have the potential of substantially accelerating the pace of discovery of tumour antigens and yielding a molecular signature for immune responses directed against protein targets in different types of cancer³³.

The increased emphasis on proteomics for disease investigations is stimulating a reassessment of strategies for sample procurement and preservation to render them compatible with proteomics, because of the inherent instability of proteins. For example, the manner in which biological fluids such as serum or plasma are collected is not ideally suited for proteomics. There is a need to reduce protein degradation and other forms of modifications that may substantially alter protein content and interfere with global profiling.

Disease-related functional proteomics

Although data obtained by various expression proteomics strategies have functional relevance by uncovering altered levels or post-translational modification states of proteins in disease, additional technologies are needed for more direct functional analysis. This is exemplified by the need to analyse protein complexes and their disruption in disease, to assay in a high-throughput fashion the activity of various classes of proteins, and to manipulate the levels and activities of individual proteins, in a cellular context, to determine their role in different biological processes and disease states.

Various strategies are currently in use for studies of protein complexes and protein–protein interactions (see review in this issue by Fields and co-workers, page 208). So far, such strategies have been applied to disease investigations on a limited basis, with a relatively narrow focus on particular complexes. For example, using affinity pull-down assays, 2D gels and mass spectrometry, myocardial protein kinase C (PKC) was found to be associated physically with at least 36 other proteins with a multitude of functions. Cardioprotection induced by activation of PKC was found to be coupled with dynamic modulation and recruitment of PKC-associated proteins⁴⁵. Previously unrecognized functions of PKC have relevance to heart function and heart disease. Clearly, at the present time, most systematic studies of protein–protein interactions have dealt with normal or physiological states, as the field is still in its early stages and the merits of various technologies, particularly for disease investigations, have yet to be fully appreciated.

Likewise, there has been limited application of activity-based proteomics to disease investigations, although the field itself is still in its early stages^{46, 47, 48}. Activity-based assays of normal and disease tissue have substantial relevance to the study of disease, but probably will be limited to the analysis of one class of proteins at a time. The potential contribution of this technique to the measurement of global dynamics in protein function is illustrated by the application of a chemical proteomics strategy to quantitatively compare enzyme activities across normal and disease tissue⁴⁷ (Fig. 3). A global analysis of the activity, subcellular distribution and glycosylation state for the serine hydrolase superfamily in a panel of human breast and melanoma cell lines resulted in the identification of a cluster of proteases lipases and esterases that distinguished cancer lines based on tissue of origin. Remarkably, the majority of these enzyme activities were downregulated in the most invasive cancer lines examined, which instead upregulated a distinct set of secreted and membrane-associated enzyme activities.

Figure 3: Activity-based protein profiling.

This chemical strategy is used to monitor changes in the functional state of enzyme superfamilies directly in complex proteomes⁴⁷. a, The method uses chemical probes that comprise three general elements: an active-site-directed reactive group (shown here as an ethoxy fluorophosphonate group that targets the serine hydrolase class of enzymes), a linker (shown here as a polyethylene glycol group) and a tag for the visualization of active enzymes (shown here as a rhodamine group). Active enzymes are denoted by shapes with open active sites; their inactive counterparts have their active sites shaded in black. b, Labelled proteome activity profile. Representative in-gel fluorescence analysis of the secreted proteome (labelled with fluorescent pigment) derived from invasive MUM-2B and non-invasive MUM-2C human melanoma cancer cells. Enzyme activities selectively associated with either invasive cells (for example, urokinase) or non-invasive cells (for example, sialic acid 9-O-acetylesterase or SAE) are noted.

High resolution image and legend (42K)

Approaches are being developed or implemented to allow directed protein manipulations other than by decreasing or increasing overall amounts of protein, as can be done by gene manipulations. One advantage of these approaches is the ability to inactivate a specific site in a protein in a time-dependent and localization-restricted manner, as with chromophore-assisted laser inactivation (CALI). Such inactivation allows assessment of the disease relevance of the site and provides a means to revert a cellular phenotype to a more normal state⁴⁹ (Fig. 4). Inactivation of protein targets involved in different signal-transduction pathways in cancer using CALI has been demonstrated⁴⁹.

Figure 4: Chromophore-assisted laser inactivation.

This technology involves the generation of short-lived radicals that induce covalent modifications at spatially restricted sites on a protein⁴⁹. A transient functional inactivation occurs if the radicals modify amino acids of the target protein that have a functional role. A target protein (a) is complexed with a ligand (yellow; for example, antibody, antibody fragments such as scFv and Fab, peptide, nucleic acid aptamers or small molecules) labelled with dye molecules (green; for example, malachite green or fluorescein) (b). The complex is irradiated with laser or incoherent light (red lightning), leading to the generation of reactive species (concentric circles) that travel a short distance (c). The reactive species in turn lead to modifications (black) of nearby amino acids (d). If the modified amino acids are responsible for a function, that particular functional protein domain will be inactivated, leaving the other functional domains intact.

High resolution image and legend (57K)

Contributions of proteomics to studies of pathogens

Despite earlier predictions to the contrary, infectious diseases remain as a leading cause of death worldwide. A complicating factor in therapy for infectious disease is the development of resistance to commonly used drugs (for example, as has occurred in tuberculosis), which heightens the need for developing effective new therapies. Interest in the application of proteomics to microbiology goes back at least two decades, with the pioneering work of Fred Neidhardt to characterize protein expression patterns in Escherichia coli under different growth conditions⁵⁰. The complete sequencing of a number of microbial genomes has provided a framework for identifying proteins encoded in these genomes using mass spectrometry. A case in point is the sequencing of the genome of the malaria parasite Plasmodium falciparum, which has provided a basis for conducting comparative proteomics studies of this pathogen, leading to the identification of new potential drug and vaccine targets^{51, 52}. Aside from comprehensive identification of microbial proteins, proteomics is relevant to numerous aspects of microbial disease pathogenesis and treatment^{53, 54, 55, 56, 57} (Table 1).

Table 1: Targeted proteomic approaches to microbial pathogens

Full table

Contribution of proteomics to drug development

There is currently a burgeoning interest in proteomics on the part of the pharmaceutical industry, evidenced by implementation of proteomics programmes by most major pharmaceutical companies. The notion has been advanced that, as the vast majority of drugs target proteins, proteomics should have substantial utility for drug development. But the industry has so far adopted a cautious attitude, and it is too early to make a critical assessment of the contributions of proteomics to drug development, relative to other approaches. The caution stems from the prior heavy investment in genomics and other approaches and some uncertainty surrounding the adequacy and scalability of proteomics to meet the needs of the pharmaceutical industry. Provided suitable technology platforms become available, the use of proteomics may permeate numerous aspects of drug development, by identifying new targets and facilitating assessment of drug action and toxicity both in the preclinical and clinical phases.

Several published studies illustrate the application of functional proteomics for identification of regulated targets in specific pathways. Lewis et al.⁵⁸ have combined functional proteomics with selective activation and inhibition of mitogen-activated protein kinase (MAPK) kinase (MKK), in order to identify cellular targets regulated by the MKK/ERK cascade. Twenty-five targets of this signalling pathway were identified, of which only five were previously characterized as MKK/ERK effectors. The remaining targets suggest new roles for this signalling cascade in cellular processes of nuclear transport, nucleotide excision repair, nucleosome assembly, membrane transport and cytoskeletal regulation.

In another study to identify proteases most suitable for drug targeting, an automated microtitre-plate assay was modified to allow detection of the four main classes of proteases in tissue samples (matrix metalloproteases, cathepsins, and the cell serine proteases, tryptase and chymase)⁵⁹. Fifteen sets of colorectal carcinoma biopsies representing primary tumour, adjacent normal colon and liver metastases were screened for protease activity. Matrix metalloproteases were expressed at higher levels in the primary tumour than in adjacent normal tissue. The mast cell proteases, in contrast, were found at very high levels in adjacent normal tissue, but were not detectable in the metastases. Cathepsin B activity was significantly higher in the primary tumour, and highest in the metastases. The proteases detected by activity assays were then localized in biopsy sections by immunohistochemistry. Mast cell proteases were abundant in adjacent normal tissue, because of infiltration of the lamina propria by mast cells. Matrix metalloproteases were localized to the tumour cells themselves, whereas cathepsin B was expressed predominantly by macrophages at the leading edge of invading tumours.

Such activity-based screening provides a basis for selecting targets in the development of inhibitors to specific proteases. Protein biochips potentially could provide a high-throughput platform for target identification. Biochips developed for interaction studies could be important in lead-compound optimization and could accelerate drug development by allowing efficient evaluation of lead compounds for specificity and selectivity in binding to drug targets. Proteomics also may provide increased efficiency of clinical trials through the availability of biologically relevant markers for drug efficacy and safety.

Organizing proteomics initiatives

It is clear that while some progress has been made in disease proteomics, the field is still in its infancy. Knowledge of the sequence of the human genome has provided a framework for genomic approaches to unravel disease processes. A similar knowledge of the human proteome is currently lacking. Developing a comprehensive knowledge framework of the proteome is considerably more complex than sequencing the human genome. Ideally, such a proteome framework would encompass knowledge of all human proteins, from their sequence to their post-translational modifications, to their interactions among each other, their cellular and subcellular distribution, and their temporal pattern of expression. Although such an exhaustive framework will not materialize in the foreseeable future, more modest goals may well be within reach.

To that effect, there is a need to begin an organized effort, the goals of which include developing an infrastructure in proteomics that would substantially facilitate unravelling the complexity of the proteome in health and in disease. The Human Proteome Organisation (HUPO, http://www.hupo.org) was founded to regroup scientists in the public and private sectors engaged throughout the world in various aspects of proteomics. HUPO's mission is threefold: to consolidate national and regional proteome organizations into a worldwide organization; to engage in scientific and educational activities to encourage the spread of proteomics technologies and disseminate knowledge pertaining to the human proteome and that of model organisms; and to assist in the coordination of public proteome initiatives aimed at characterizing specific tissue and cell proteomes. Initiatives currently in the pilot phase include an international effort to identify proteins detectable in normal serum and plasma and their range of variation with age, ethnicity and physiological state, and a liver proteome study to identify proteins expressed in the liver. These initiatives have attracted substantial interest and will be integrated with efforts in protein informatics to achieve data standardization on the one hand, and data curation on the other.

Concluding remarks

Proteome alterations in disease may occur in many different ways that are not predictable from genomic analysis, and it is clear that a better understanding of these alterations will have a substantial impact in medicine. A useful repertoire of proteomics technologies is currently available for disease-related applications, although further technological innovations would be beneficial to increase sensitivity, reduce sample requirement, increase throughput and more effectively uncover various types of protein alterations such as post-translational modifications. The use of these technologies will likely expand substantially, particularly to meet the need for better diagnostics and to shorten the path for developing effective therapy.