- Email: [email protected]
Application of proteomic technologies in the drug development process
Toxicology Letters 149 (2004) 377–385
Application of proteomic technologies in the drug development process Jennie L. Walgren, David C. Thompson∗ Worldwide Safety Sciences, Pfizer, Inc., 700 Chesterfield Parkway West, Chesterfield, MO 63017, USA
Abstract Proteins are the principal targets of drug discovery. Most large pharmaceutical companies now have a proteomics-oriented biotech or academic partner or have started their own proteomics division. Common applications of proteomics in the drug industry include target identification and validation, identification of efficacy and toxicity biomarkers from readily accessible biological fluids, and investigations into mechanisms of drug action or toxicity. Target identification and validation involves identifying proteins whose expression levels or activities change in disease states. These proteins may serve as potential therapeutic targets or may be used to classify patients for clinical trials. Proteomics technologies may also help identify protein–protein interactions that influence either the disease state or the proposed therapy. Efficacy biomarkers are used to assess whether target modulation has occurred. They are used for the characterization of disease models and to assess the effects and mechanism of action of lead candidates in animal models. Toxicity (safety) biomarkers are used to screen compounds in pre-clinical studies for target organ toxicities as well as later on in development during clinical trials. Complementary approaches such as metabolomics and genomics can be used in conjunction with proteomics throughout the drug development process to create more of a unified, systems biology approach. © 2004 Elsevier Ireland Ltd. All rights reserved. Keywords: Proteomics; Drug development; Biomarkers
1. Introduction As proteins are the principal targets of drug discovery, the evolution of proteomics techniques is of major importance to the drug development process. While the complexity of structure and function in the proteome presents a significant challenge, the pharmaceutical industry and its biotech and academic ∗ Corresponding author. Tel.: +1-314-274-7226; fax: +1-314-274-4226. E-mail address: [email protected] (D.C. Thompson).
partners are expending a tremendous amount of resources to decipher and utilize proteomic data to make drug development more efficient and successful. The practice of proteomics ranges from the identification of thousands of proteins in a particular model system, to the detailed analysis of the 3D structure, possible modifications/isoforms, and function of a single protein. All of these explorations can be equally valuable depending upon the issue addressed and the phase of drug development. This review discusses the role of proteomics in the stages of drug development, including target identification, target validation, drug design, lead optimization, and pre-clinical and clinical
0378-4274/$ – see front matter © 2004 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.toxlet.2003.12.047
378
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
development. We focus particularly on relevant examples of proteomics applications as well as recent developments in proteomics techniques and the systems biology approach.
2. Target identification High-throughput proteomics, identifying potentially hundreds to thousands of protein expression changes in model systems following perturbation by drug treatment or disease, lends itself particularly well to target identification in drug discovery. Yet, the identification of proteins is only the beginning of the process: the data analysis and validation of potential protein targets that follows is a time-consuming and labor-intensive process as well. That effort can assume incredible value, however, if higher drug candidate success rates are achieved. The cost of invalid targets can be enormous, considering that the current estimate for capitalized new drug development is over US$ 800 million (DiMasi et al., 2003), and in recent history the overall success rate for investigational drugs, after IND filing, is estimated at approximately 20% (DiMasi, 2001). In addition to the hope that proteomics technologies can help achieve higher drug development success rates, the recent emphasis on developing disease-modifying compounds makes proteomic analyses of disease etiology and progression of critical importance. Heart disease, as the number one cause of morbidity and mortality in the US, is a common focus for drug development in the pharmaceutical industry. Researchers at Pfizer, Inc., using protein identification by two-dimensional gel electrophoresis (2-DE) and mass spectrometry, profiled protein expression changes in a rat model of endothelin-induced cardiac hypertrophy and myocardial infarction (Macri and Rapundalo, 2001). Several academic laboratories have also been involved in proteomic profiling of other cardiac abnormalities such as ischemia, hypertension, hypertrophy, and cardiomyopathy, as well as identifying proteins in fibroblasts that are phosphorylated in signaling pathways of pertinent growth factors such as platelet-derived growth factor (PDGF). Collaboration on such studies has helped drug companies build large databases of possible protein targets in heart disease. The goal is to create additional pub-
lic databases such as HP-2DPAGE, a 2-DE database of human myocardial proteins (http://www.mdcberlin.de/∼emu/heart/), which will include proteins found in various species and in various disease states. There are numerous other examples of research aimed at mapping pathways that are involved in disease processes. A group of researchers at GlaxoSmithKline and their collaborators at Toyama Medical and Pharmaceutical University investigated apoptotic pathways involved in the selection of T-lymphocytes by examining the differential display of soluble proteins from incubations of nuclei with and without exposure to extract from T-lymphocytes stimulated with anti-CD3 antibody (Kawakami et al., 2003). Using 2-DE display followed by in-gel digestion and nanoLC/nanoelectrospray/ion trap MS analysis of the resulting peptides, the researchers found an upregulation and possible modification of HMG2, a protein involved in DNA binding and “bending” that may also enhance nuclease activity. Understanding pathways at the subcellular level has been the focus of several other recent papers including a study of human nucleolar proteins by Scherl et al. (2002). In studies such as this a major step in the proteomic process is the purification of the organelle, followed by protein separation and identification. By combining their efforts with existing data this team has established a database of 350 nucleolar proteins that is helping to elucidate the role of nucleoli in mRNA and ribosome transport and in the control of the cell cycle and aging. While two-dimensional gel electrophoresis can be very effective at identifying protein expression changes in a given system, efforts have been made to develop techniques that enable higher throughput, greater coverage of the proteome, or comprehensive protein maps for protein profiling. In a technique using protein sequence tags (PST), each protein is terminally tagged and digested, and then only the terminal peptides are isolated and sequenced, allowing for rapid identification of an entire proteome (Jain, 2002). Two other methods that have become more common for protein profiling are MudPIT and ICAT. In the multidimensional protein identification technology (MudPIT) method, proteins or peptides are identified via LC/MS using strong cation exchange and reverse-phase adsorbent separation columns. By separating the proteins by both charge and
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
hydrophobicity, theoretically a greater number of proteins can be resolved. In isotope-coded affinity tagging (ICAT), an alkylating reagent consisting of a reactive group which binds to a particular amino acid (often cysteine), a “light” or “heavy” isotopic linker, and an affinity tag such as biotin, are incubated with each sample. Samples from different treatment groups or disease states can be incubated with different isotopes. The samples are then combined 1:1 and digested, and the labeled peptides are identified by LC/MS. This technique has the advantage of being able to not only identify the proteins in a sample, but also determine the abundance of each protein under the different conditions (Burbaum and Tobal, 2002). Obtaining more detailed protein maps from diseased and control tissues has been the pursuit of Caprion Pharmaceuticals (Montreal, Canada), who developed CellCarta, a technology which allows the construction of intracellular maps of a protein’s location, orientation, movement, and post-translational modifications. Caprion has applied their subcellular approach in protein expression profiling for the development of cancer treatments in collaboration with IDEC pharmaceuticals (Jain, 2002). 2.1. Biomarker identification In addition to using proteomic tools to identify an array of proteins that are modified in the diseased state, a significant focus of proteomic activities in drug discovery has turned to the identification of biomarkers in easily accessible biological fluids. The importance of the development of such markers is evident when one considers the influence of such a tool on all stages of drug development. Not only can a biomarker aid in the understanding of the disease process and progression and what molecular pathways are involved, but also this biomarker can then serve as a monitoring tool in later stages of development. For instance, a change in the status of this marker may be useful in determining the efficacy of various drug candidates in the process of lead optimization, and then can also be used in the selection of appropriate animal models for pre-clinical studies as well as in patient profiling for clinical trials. Good examples of such markers are the serum and urine biomarkers used to identify arthritis. Numerous biomarkers from synovial fluid, blood, and urine have been used to identify and study the stages
379
of osteoarthritis (OA) (e.g. urinary collagen fragments; Downs et al., 2001). Researchers from Pfizer, Inc., in collaboration with investigators from McGill University (Montreal, Canada), compared 14 different serum and urine biomarkers in OA patients to determine how they could be grouped and which markers best distinguished patients from controls (Otterness et al., 2000). Following principal components analysis of marker values collected at baseline and at various later stages of disease progression, the markers were grouped into clusters which represented inflammation markers, bone markers, and markers of cartilage anabolism and catabolism. A group of three markers best distinguished the OA patients: one inflammatory marker and markers of both cartilage anabolism and catabolism. So, in this example, biomarkers that are directly indicative of the disease process (skeletal metabolism) are readily measured in body fluids and can be used throughout the development of drug candidates. This study also illustrates the point that often more than one biomarker is necessary or preferable for distinguishing disease states. A single marker may not readily track with disease progression in each patient, especially if the marker is not related to the molecular basis of the disease. As an additional example in the case of arthritis, researchers at Lund University in Sweden monitored serum levels of some of the markers also measured in the OA study, including cartilage oligomeric matrix protein, 846 epitope, C-propeptide of type II collagen, and bone sialoprotein in patients. By relating the levels of the various markers, the group was able to predict prognosis, in particular the rapidity of disease progression, in these patients (Mansson et al., 1995). 2.2. Identifying protein modifications Beyond the identification of proteins involved in disease progression, it is often critical to explore protein modifications to elucidate the changes in molecular pathways involved in disease. In a study of lymphocytes with normal or Scott syndrome (inherited bleeding disorder) phenotype, changes in the tyrosine phosphorylation status of Ig-chain precursor, fascin, and actin-associated proteins were identified by immunoprecipitation of tyrosine-phosphorylated proteins followed by separation of the proteins by 2-DE (Imam-Sghiouar et al., 2002). Using an alternative
380
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
method for immunoprecipitation of phosphorylated proteins, a group of researchers at the University of Dundee (Allan et al., 2003) investigated the phosphorylation of caspase-9 (which inhibits apoptosis) via the ERK-MAPK pathway, which may help to resolve the mechanism of tumor induction during constitutive activation of this pathway. In this study extracts containing glutathione-S-transferase (GST)-labeled caspase-9 were incubated with 32 P-ATP under conditions of maximal kinase stimulation, followed by affinity purification of caspase-9-GST. After reduction, alkylation of cysteine residues, and separation on a gradient gel, caspase-9 was excised and digested with trypsin. The peptide fragments were analyzed by MALDI-TOF MS and the identity of each peptide was confirmed by Edman degradation. The advantages of this technique of phosphorylation exploration are that it is more sensitive at detecting phosphorylated peptides, it is quantitative, and that the site(s) of phosphorylation are mapped by Edman degradation sequencing. Another protein modification that is clearly modified in many disease states is glycosylation; in fact, many proteins known to be involved in disease processes or that are already being used as biomarkers or possible therapeutic candidates are glycoproteins. A group of scientists at Oxford University have reported a powerful procedure for examining glycoproteins. Peracaula et al. (2003a) found that secreted RNAse 1 from pancreatic adenocarcinoma tumor cells displayed vastly different glycosylation patterns than RNAse 1 from healthy pancreas. The general procedure used to determine the glycosylation patterns was to isolate RNAse 1 by affinity chromatography, digest the RNAse 1 with N-glycosidase F to release the glycans, and to either label the glycans with fluorescent probes for HPLC analysis, or examine the glycans by MALDI-TOF. The distinct patterns of glycosylation that they identified could actually be selected by monoclonal antibodies specific for the carbohydrate side chains, providing an easy detection method for a potential biomarker of pancreatic adenocarcinoma. Similarly, the same group found that the glycoprotein prostate cancer marker prostate-specific antigen (PSA) from a cancer source (LNCaP cell line) contained neutral glycans with a high fucose and GalNAc content in contrast to PSA from normal origins (Peracaula et al., 2003b). This data may potentially provide a way to distinguish benign from malignant prostate
tumors if the differences are consistent in patients and can be identified from serum PSA. The results of these glycosylation studies highlight the importance of identifying potential modifications (i.e. shift in gel mobility) in proteins of interest, as these modifications may be indicative of the disease state or may prove to be valuable ways to develop biomarker assays. 2.3. Exploring protein–protein interactions The study of molecular pathways in diseased and normal states necessitates the development of tools to study protein–protein interactions. Such techniques have greatly expanded in recent years. The ability to affinity-purify a protein facilitates the identification of protein associations via isotopic labeling in cell culture using a method known as stable isotopic amino acids in culture (SILAC). Using this strategy, Blagoev et al. (2003) incubated HeLa cells cultured in the presence or absence of epidermal growth factor (EGF) with either [12 C] or [13 C] arginine. The lysates were mixed and the EGF receptor was affinity-purified over a column displaying the SH2 domain of the adapter Grb2. Proteins which bound to the activated receptor were identified after digestion and LC–MS/MS analysis. Other methods used to study protein interaction include two-hybrid systems, phage display, Biacore® surface plasmon resonance (SPR) technology, and the more recent techniques of reverse transfection and tandem affinity purification (TAP) (Burbaum and Tobal, 2002; Jain, 2002). Reverse transfection involves creating an array of the cDNA of interest on a glass slide, followed by transfection of cells that attach to the slide. The protein is then generally visualized with fluorescent antibodies to assess its localization. This technique can be combined with a two-hybrid approach to assess protein interactions (Burbaum and Tobal, 2002). The TAP method involves the fusion of a TAP tag to the N- or C-terminus of a protein of interest so that it can be expressed and then affinity purified from a host cell. The TAP tag is made up of two IgG binding domains of protein A and a calmodulin binding peptide, separated by a TEV (tobaccoetch virus) protease cleavage site. The tag is introduced in-frame in the coding region of the target protein in an expression vector which is expressed in the appropriate host. Generally expression is maintained at close to normal levels so that erroneous protein
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
associations are kept at a minimum. The protein can then be purified in its native state, concentrated, and complexes can be identified by mass spectrometry, Edman degradation, or Western blot (Puig et al., 2001). Advances in mass spectrometry have also facilitated the study of protein–protein complexes. Fourier transform ion cyclotron resonance mass spectrometry (FTICR-MS) provides high mass resolution and accuracy of whole proteins and protein–protein complexes via the circulation of ions in a superconducting magnet; isotopes can be clearly resolved in high charge states. FTICR-MS has been used to study the contributions of electrostatic and hydrophobic interactions to the binding of calmodulin to synthetic mutated peptides derived from smooth muscle myosin light chain kinase in the presence or absence of calcium (Nousiainen et al., 2003).
3. Target validation Thus far this discussion has focused on examples of the methods used to identify changes in protein expression, form, or interactions in disease states. Once these proteins are identified, the questions of relevance to disease etiology and generality of the disease biomarker still remain. While it is important to establish that a potential protein target is present in the disease-relevant cell or human tissue, the next phase in development is to begin to validate that target by modulating the protein’s activity in a model system to determine the outcome on disease phenotype. Transgenic animals have been classically used to study protein function in a model system. In addition to transgenic mice, transgenic zebrafish are now being used, as they are easier to care for and have shorter life cycles than mice, and actually serve as excellent models for some human diseases and processes such as angiogenesis, inflammation, and insulin regulation (Smith, 2003). A novel type of mouse “knockin” model, called the analog-sensitive kinase alleles (ASKA) mouse, goes a step beyond transgenics and knockout mice in that the native protein kinase expressed in this model is a mutated version that can be inhibited by an ATP analog “P-inhibitor.” The advantage in this model is that the mouse can develop normally, and the enzyme can be “knocked out” at a given time later in its lifespan and the effect of this inhibition on its health and aging
381
can be assessed (Szymkowski, 2003). Researchers at Sangamo BioSciences (Richmond, CA) have also developed a system for regulating the expression of particular proteins in cell and whole animal models. Their system uses zinc-finger transcription factors where the DNA recognition domain is attached to a functional domain and allows the target gene expression to be regulated. This group has been very active in studying the importance of the regulation of different protein isoforms (Smith, 2003). One of the most powerful tools in target validation, though, has been the use of models where the protein target of interest has been completely “knocked out.” While knockout mice have been very useful in this area, they are costly and time-consuming models. Alternative systems to the knockout animal are becoming available to facilitate target validation; one of the most exciting of these is RNA interference (RNAi), a technique where protein expression is “silenced” at the post-transcriptional level. Briefly, cells are presented with short segments of double-stranded RNA specific to the protein target that are then processed into short single-stranded fragments which bind to the complementary antisense strands and form the RNA-induced silencing complex (RISC). It is thought that such RISC complexes inhibit the protein expression by promoting RNA degradation and translational inhibition. While RNAi has been in use for the last several years in systems such as plants and C. elegans, it is only in the last 2 years that RNAi has been successfully demonstrated in mammalian cells. Improvements are currently being made to this technique, including direct treatment of mammalian cells with the single-stranded RNA fragments (siRNAs), which is effective in many cell types but are transient “knockouts.” To overcome the transient nature of this technique in mammalian cells, researchers have used treatment of the cells with short hairpin RNAs (shRNAs) that are composed of an inverted repeat sequence in a hairpin structure, which support stable expression of the inhibitory RNA (Hannon, 2002). RNAi should prove to be a very powerful method to validate target protein function. However, concern that the reduction in RNA levels will not always lead to a reduction in protein expression has led some researchers to develop intracellular antibody capture technology (IACT) to select antibodies that will directly target the protein of interest in a cell (Szymkowski, 2003).
382
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
Two additional techniques used for turning off protein function in cellular systems are selected interacting domains (SIDs) and chromophore-assisted laser inactivation (CALI). SIDs are dominant-negative acting fragments that can modulate protein function in molecular pathways. CALI is a procedure where proteins are photochemically modified and therefore acutely “knocked out” of a system. CALI has been used in functional screens as well as in individual protein studies, rendering it useful in target discovery as well (Jain, 2002). Protein and antibody microarrays are also being used for screening of protein interaction and function. Protein “biochips” or microarrays are ideal for studying receptor–ligand interactions, enzyme activity, and antibody–antigen interaction with rapid throughput. These tools can be useful for both understanding molecular pathways in which the target protein participates, as well as screening model systems for changes that occur in disease states or following drug treatment (i.e. enzyme activity testing) (Huels et al., 2002). Tools for screening interactions with entire “proteome sets” of both known and unknown proteins are also available in microarray format. Xencor’s ProCode and Sense Proteomics (Cambridge, UK) COVET technology allow for the creation of a set of proteins from a given protein expression library which are all tagged with plasmids. These proteins can be arrayed and screened and the cDNA tags allow for rapid protein identification (Smith, 2003). The creation of very specific and sensitive assays for proteins of interest has also been achieved through the use of aptamers, or synthetic nucleic acid ligands. In addition to being used in the place of antibodies or protein ligands in microarrays, stabilized aptamer forms can be used to validate the function of various protein targets, and aptamers designed against target proteins can be applied in competition assays to screen lead compounds. The advantages of using aptamers for microarray assays are their high specificity achieved by rigorous selection, and potential greater sensitivity when the aptamers are modified to cause covalent linkage to the microarray surface. The detection of aptamers has been aided by the development of aptazymes, or aptamers conjugated with ribozyme domains, and the construction of pairs of aptamers recognizing adjacent epitopes on the target protein. The binding of an aptazyme to its target causes a
conformational change which alters the activity of the ribozyme. The binding of adjacent aptamers to their respective target sequences creates an oliogonucleotide bridge and produces ligation of the connector nucleotide, which is subsequently amplified for detection. The latter technique has the potential to detect a protein with sensitivity 1000-fold greater than a sandwich ELISA (Burgstaller et al., 2002).
4. Drug design and lead optimization Once a target protein has been validated, the task of identifying chemical compounds that can appropriately modulate the target can also be aided by proteomic techniques: namely structural proteomics. If a specific protein target has been identified, in many cases now the expression and purification of the protein for crystallization can be achieved. While crystallization is still a significant hurdle to overcome, such procedures are constantly being updated and optimized (Syrrx, San Diego, CA). Once crystallization is achieved, X-ray diffraction and solving of the protein structure is being aided by the development of multiple-wavelength anomalous diffraction (MAD) phasing, which entails a high-energy tunable X-ray source with enhanced computational abilities. The knowledge of the crystal structure of the protein facilitates the production of suitable drug candidates, as it can be directly input into virtual screening software that will select fragments or compounds that will bind the protein surface of interest. Such virtual screening is invaluable not only to the selection of a group of chemical entities which will “hit” the protein target, but as a screen for structure-activity relationships among those agents to optimize the lead compound (e.g. SkelGen, de novo Pharmaceuticals, Cambridge, UK and software products of Accelrys, San Diego, CA). Such structural proteomic screening has been very useful in the development of HIV protease inhibitors, influenza virus neuraminidase inhibitors, and antimicrobial agents, where the technique may also help in developing new broad-spectrum antibiotics (Schmid, 2002). Even if the crystal structure of the protein is not available, software which will produce a virtual protein structure is available (e.g. Quasi2, de novo Pharmaceuticals, Cambridge, UK) (Smith, 2003).
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
Beyond virtual screening surveys, the interaction of potential drug compounds with particular proteins can be examined with the use of activity-based probes (ABPs). These chemically reactive probes provide the ability to measure the protein activity based on the active component of the particular class of proteins, and can be used for inhibition studies with potential drug agents (Burbaum and Tobal, 2002).
5. Safety and diagnostic biomarkers for pre-clinical and clinical development Once a class of compounds is in pre-clinical development, lead optimization and the viability of the project can depend on assays developed to assess the efficacy and toxicity of the lead agents. Many pharmaceutical companies are beginning to use proteomic techniques to accomplish these tasks. When searching for toxicity (safety) biomarkers, pre-clinical development groups are often hoping to find not only biomarkers specific to the toxicity of the current compound in development, but also general tissue toxicity biomarkers to assess compounds in future studies. Scientists at GlaxoSmithKline have used conventional 2-DE proteomics techniques to identify protein alterations involved in gentamicin-induced nephrotoxicity (Charlwood et al., 2002). The proteins identified were consistent with the mechanism of toxicity, and these researchers suggested that these markers may be useful in screening antibiotics in development for nephrotoxicity potential. In another study at GlaxoSmithKline, researchers used Ciphergen’s ProteinChip® /SELDI (surface-enhanced laser desorption/ionization) technology to discover a urinary biomarker which could be useful in detecting skeletal muscle toxicity (Dare et al., 2002). SELDI allows for the enrichment of proteins with particular binding affinities (e.g. hydrophobic, hydrophilic, metal binding, receptor affinity) through the chemically modified surfaces of the proprietary ProteinChips. Using this technology, one can screen a complex mixture of proteins such as those found in serum or urine through fractionation of the sample directly on the chip. By profiling samples from both treated/disease and control, a pattern of protein differences can distinguish the treatment groups. Using SELDI profiling, protein purification, and on-chip trypsin digestion for peptide mapping, the researchers
383
found that urinary parvalbumin-␣ was increased in rats treated with 2,3,5,6-tetramethyl-p-phenylenediamine, a compound which induces skeletal muscle toxicity. SELDI technology has also been shown to be a successful profiling tool in populations of cancer patients. Li et al. (2002) report that a set of three proteins selected from Ni2+ -affinity ProteinChip profiles of patient serum can serve as biomarkers to distinguish normal individuals from patients with stage 0–I breast cancer, and between patients with stage II or stage III breast cancer. Markers that can help determine not only the existence of the disease but also the stage of the disease provide benefits first to the patients in more accurate diagnoses and treatment, and second to those in pharmaceutical R&D who can then profile patients for clinical trials. It is reasonable to assume that the better the model animal or patient in pre-clinical and clinical trials, the more beneficial the research in advancing health care.
6. Systems biology approach One fact which proteomics research has made quite clear is that there is rarely a single target of a disease or chemical; often several protein targets or biomarkers are discovered and validated. Part of the reality of trying to develop agents to treat diseases such as heart disease and diabetes is that these diseases are enormously complex and involve simultaneous pathologies of multiple organ systems. A relatively new wave of technology that is attempting to integrate a greater amount of the potential information available on disease processes, and that incorporates proteomics as one of its major tools, is systems biology. This approach involves the simultaneous measurement of genomic, proteomic, and metabolomic parameters in a given system under defined conditions (Davidov et al., 2003). Such measurements are then subjected to rigorous computational biology analyses to define the link which ties the data to the particular disease or perturbation in question. In this way, instead of using just one “omics” technique to produce a laundry list of genes, proteins, or metabolites that are altered in the disease state, the combined analysis of all data sets may be able to define the common thread, such as the regulation of a gene, that results in the given phenotype. Further, once a systems biology analysis
384
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
for a particular disease model is completed, that data, fed into an in silico biology platform, can be used to simulate the effects of a particular drug candidate in the system. Also, the model could provide information about pathways that may contribute to side effects of a given agent, and can therefore be utilized in lead compound optimization. In addition to several biotech companies such as Beyond Genomics and Bioseek, Eli Lilly has recently developed a center for the pursuit of systems biology (Davidov et al., 2003).
7. Conclusions The results of recent proteomics pursuits have demonstrated the potential value that proteomics has to offer in drug development. Proteomics techniques are providing precise and fairly rapid methods to screen both target proteins and potential therapeutic compounds. In-house, drug companies have mainly focused on protein profiling for target identification, to develop efficacy and toxicity biomarkers, and to create valuable protein databases for access in future projects. Through collaboration with biotech firms and universities, the pharmaceutical industry is gaining access to exciting innovations in the proteomics field such as RNA interference and systems biology, and is testing the results of proteomic techniques in patient populations. The key to drug development in the future is the integration and application of the vast amount of information that techniques such as proteomics make available to researchers.
References Allan, L.A., Morrice, N., Brady, S., Magee, G., Pathak, S., Clarke, P.R., 2003. Inhibition of caspase-9 through phosphorylation at Thr 125 by ERK MAPK. Nat. Cell. Biol. 5 (7), 647–655. Blagoev, B., Kratchmarova, I., Ong, S-E., Nielsen, M., Foster, L.J., Mann, M., 2003. A proteomics strategy to elucidate functional protein–protein interactions applied to EGF signaling. Nat. Biotech. 21, 315–318. Burbaum, J., Tobal, G.M., 2002. Proteomics in drug discovery. Curr. Opin. Chem. Biol. 6, 427–433. Burgstaller, P., Jenne, A., Blind, M., 2002. Aptamers and aptazymes: accelerating small molecule drug discovery. Curr. Opin. Drug Discov. Develop. 5 (5), 690–700. Charlwood, J., Skehel, J.M., King, N., Camilleri, P., Lord, P., Bugelski, P., Atif, U., 2002. Proteomic analysis of rat kidney
cortex following treatment with gentamicin. J. Prot. Res. 1, 73–82. Dare, T.O., Davies, H.A., Turton, J.A., Lomas, L., Williams, T.C., York, M.J., 2002. Application of surface-enhanced laser desorption/ionization technology to the detection and identification of urinary parvalbumin-␣: a biomarker of compoundinduced skeletal muscle toxicity in the rat. Electrophoresis 23, 3241–3251. Davidov, E.J., Holland, J.M., Marple, E.W., Naylor, S., 2003. Advancing drug discovery through systems biology. Drug Discov. Today 8 (4), 175–183. DiMasi, J.A., 2001. Risks in new drug development: approval success rates for investigational drugs. Clin. Pharmacol. Ther. 69 (5), 297–307. DiMasi, J.A., Hansen, R.W., Grabowski, H.G., 2003. The price of innovation: new estimates of drug development costs. J. Health Econ. 22 (2), 151–185. Downs, J.T., Lane, C.L., Nestor, N.B., McLellan, T.J., Kelly, M.A., Karam, G.A., Mezes, P.S., Pelletier, J-P., Otterness, I.G., 2001. Analysis of collagenase-cleavage of type II collagen using a neoepitope ELISA. J. Immun. Meth. 247, 25–34. Hannon, G.J., 2002. RNA interference. Nature 418, 244–251. Huels, C., Muellner, S., Meyer, H.E., Cahill, D.J., 2002. The impact of protein biochips and microarrays on the drug development process. Drug Discov. Today 7 (Suppl 18), S119–S124. Imam-Sghiouar, N., Laude-Lemaire, I., Labas, V., Pflieger, D., Caron, M., Nabias, D.K., Joubert-Caron, R., 2002. Subproteomics analysis of phosphorylated proteins: application to the study of B-lymphoblasts from a patient with Scott syndrome. Proteomics 2, 828–838. Jain, K.K., 2002. Using proteomics to identify targets and validate leads. Trends Biotech. 20 (8), 319–320. Kawakami, T., Nagata, T., Muraguchi, A., Nishimura, T., 2003. Proteomic approach to apoptotic thymus maturation. J. Chromatogr. B 787, 223–229. Li, J., Zhang, Z., Rosenzweig, J., Wang, Y.Y., Chan, D.W., 2002. Proteomics and bioinformatics approaches for identification of serum biomarkers to detect breast cancer. Clin. Chem. 48 (8), 1296–1304. Macri, J., Rapundalo, S.T., 2001. Application of proteomics to the study of cardiovascular biology. TCM 11 (2), 66–75. Mansson, B., Carey, D., Alini, M., Ionescu, M., Rosenburg, L.C., Poole, A.R., Heinegard, D., Saxne, T., 1995. Cartilage and bone metabolism in rheumatoid arthritis. Differences between rapid and slow progression of disease identified by serum markers of cartilage metabolism. J. Clin. Invest. 95 (3), 1071–1077. Nousiainen, M., Derrick, P.J., Lafitte, D., Vainiotalo, P., 2003. Relative affinity constants by electrospray ionization and Fourier transform ion cyclotron resonance mass spectrometry: calmodulin binding to peptide analogs of myosin light chain kinase. Biophys. J. 85 (1), 491–500. Otterness, I.G., Swindell, A.C., Zimmerer, R.O., Poole, A.R., Ionescu, M., Weiner, E., 2000. An analysis of 14 molecular markers for monitoring osteoarthritis: segregation of the markers into clusters and distinguishing osteoarthritis at baseline. Osteoarthr. Cart. 8, 180–185.
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385 Peracaula, R., Royle, L., Tabarés, G., Mallorqúi-Fernández, G., Barrabés, S., Harvey, D.J., Dwek, R.A., Rudd, P.M., de Llorens, R., 2003a. Glycosylation of human pancreatic ribonuclease: differences between normal and tumor states. Glycobiology 13 (4), 227–244. Peracaula, R., Tabarés, G., Royle, L., Harvey, D.J., Dwek, R.A., Rudd, P.M., de Llorens, R., 2003b. Altered glycosylation pattern allows the distinction between prostate-specific antigen (PSA) from normal and tumor origins. Glycobiology 13 (6), 457–470. Puig, O., Caspary, F., Rigaut, G., Rutz, B., Bouveret, E., BragadoBilsson, E., Wilm, M., Séraphin, B., 2001. The tandem affinity
385
purification (TAP) method: a general procedure of protein complex purification. Methods 24, 218–229. Scherl, A., Couté, Y., Déon, C., Callé, A., Kindbeiter, K., Sanchez, J-C., Greco, A., Hochstrasser, D., Diaz, J-J., 2002. Functional proteomic analysis of human nucleolus. Mol. Biol. Cell 13, 4100–4109. Schmid, M.B., 2002. Structural proteomics: the potential of high-throughput structure determination. Trends Microbiol. 10 (Suppl 10), S27–S31. Smith, C., 2003. Hitting the target. Nature 422, 341–347. Szymkowski, D.E., 2003. Target validation joins the pharma fold. Targets 2 (1), 8–9.
Application of proteomic technologies in the drug development process Jennie L. Walgren, David C. Thompson∗ Worldwide Safety Sciences, Pfizer, Inc., 700 Chesterfield Parkway West, Chesterfield, MO 63017, USA
Abstract Proteins are the principal targets of drug discovery. Most large pharmaceutical companies now have a proteomics-oriented biotech or academic partner or have started their own proteomics division. Common applications of proteomics in the drug industry include target identification and validation, identification of efficacy and toxicity biomarkers from readily accessible biological fluids, and investigations into mechanisms of drug action or toxicity. Target identification and validation involves identifying proteins whose expression levels or activities change in disease states. These proteins may serve as potential therapeutic targets or may be used to classify patients for clinical trials. Proteomics technologies may also help identify protein–protein interactions that influence either the disease state or the proposed therapy. Efficacy biomarkers are used to assess whether target modulation has occurred. They are used for the characterization of disease models and to assess the effects and mechanism of action of lead candidates in animal models. Toxicity (safety) biomarkers are used to screen compounds in pre-clinical studies for target organ toxicities as well as later on in development during clinical trials. Complementary approaches such as metabolomics and genomics can be used in conjunction with proteomics throughout the drug development process to create more of a unified, systems biology approach. © 2004 Elsevier Ireland Ltd. All rights reserved. Keywords: Proteomics; Drug development; Biomarkers
1. Introduction As proteins are the principal targets of drug discovery, the evolution of proteomics techniques is of major importance to the drug development process. While the complexity of structure and function in the proteome presents a significant challenge, the pharmaceutical industry and its biotech and academic ∗ Corresponding author. Tel.: +1-314-274-7226; fax: +1-314-274-4226. E-mail address: [email protected] (D.C. Thompson).
partners are expending a tremendous amount of resources to decipher and utilize proteomic data to make drug development more efficient and successful. The practice of proteomics ranges from the identification of thousands of proteins in a particular model system, to the detailed analysis of the 3D structure, possible modifications/isoforms, and function of a single protein. All of these explorations can be equally valuable depending upon the issue addressed and the phase of drug development. This review discusses the role of proteomics in the stages of drug development, including target identification, target validation, drug design, lead optimization, and pre-clinical and clinical
0378-4274/$ – see front matter © 2004 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.toxlet.2003.12.047
378
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
development. We focus particularly on relevant examples of proteomics applications as well as recent developments in proteomics techniques and the systems biology approach.
2. Target identification High-throughput proteomics, identifying potentially hundreds to thousands of protein expression changes in model systems following perturbation by drug treatment or disease, lends itself particularly well to target identification in drug discovery. Yet, the identification of proteins is only the beginning of the process: the data analysis and validation of potential protein targets that follows is a time-consuming and labor-intensive process as well. That effort can assume incredible value, however, if higher drug candidate success rates are achieved. The cost of invalid targets can be enormous, considering that the current estimate for capitalized new drug development is over US$ 800 million (DiMasi et al., 2003), and in recent history the overall success rate for investigational drugs, after IND filing, is estimated at approximately 20% (DiMasi, 2001). In addition to the hope that proteomics technologies can help achieve higher drug development success rates, the recent emphasis on developing disease-modifying compounds makes proteomic analyses of disease etiology and progression of critical importance. Heart disease, as the number one cause of morbidity and mortality in the US, is a common focus for drug development in the pharmaceutical industry. Researchers at Pfizer, Inc., using protein identification by two-dimensional gel electrophoresis (2-DE) and mass spectrometry, profiled protein expression changes in a rat model of endothelin-induced cardiac hypertrophy and myocardial infarction (Macri and Rapundalo, 2001). Several academic laboratories have also been involved in proteomic profiling of other cardiac abnormalities such as ischemia, hypertension, hypertrophy, and cardiomyopathy, as well as identifying proteins in fibroblasts that are phosphorylated in signaling pathways of pertinent growth factors such as platelet-derived growth factor (PDGF). Collaboration on such studies has helped drug companies build large databases of possible protein targets in heart disease. The goal is to create additional pub-
lic databases such as HP-2DPAGE, a 2-DE database of human myocardial proteins (http://www.mdcberlin.de/∼emu/heart/), which will include proteins found in various species and in various disease states. There are numerous other examples of research aimed at mapping pathways that are involved in disease processes. A group of researchers at GlaxoSmithKline and their collaborators at Toyama Medical and Pharmaceutical University investigated apoptotic pathways involved in the selection of T-lymphocytes by examining the differential display of soluble proteins from incubations of nuclei with and without exposure to extract from T-lymphocytes stimulated with anti-CD3 antibody (Kawakami et al., 2003). Using 2-DE display followed by in-gel digestion and nanoLC/nanoelectrospray/ion trap MS analysis of the resulting peptides, the researchers found an upregulation and possible modification of HMG2, a protein involved in DNA binding and “bending” that may also enhance nuclease activity. Understanding pathways at the subcellular level has been the focus of several other recent papers including a study of human nucleolar proteins by Scherl et al. (2002). In studies such as this a major step in the proteomic process is the purification of the organelle, followed by protein separation and identification. By combining their efforts with existing data this team has established a database of 350 nucleolar proteins that is helping to elucidate the role of nucleoli in mRNA and ribosome transport and in the control of the cell cycle and aging. While two-dimensional gel electrophoresis can be very effective at identifying protein expression changes in a given system, efforts have been made to develop techniques that enable higher throughput, greater coverage of the proteome, or comprehensive protein maps for protein profiling. In a technique using protein sequence tags (PST), each protein is terminally tagged and digested, and then only the terminal peptides are isolated and sequenced, allowing for rapid identification of an entire proteome (Jain, 2002). Two other methods that have become more common for protein profiling are MudPIT and ICAT. In the multidimensional protein identification technology (MudPIT) method, proteins or peptides are identified via LC/MS using strong cation exchange and reverse-phase adsorbent separation columns. By separating the proteins by both charge and
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
hydrophobicity, theoretically a greater number of proteins can be resolved. In isotope-coded affinity tagging (ICAT), an alkylating reagent consisting of a reactive group which binds to a particular amino acid (often cysteine), a “light” or “heavy” isotopic linker, and an affinity tag such as biotin, are incubated with each sample. Samples from different treatment groups or disease states can be incubated with different isotopes. The samples are then combined 1:1 and digested, and the labeled peptides are identified by LC/MS. This technique has the advantage of being able to not only identify the proteins in a sample, but also determine the abundance of each protein under the different conditions (Burbaum and Tobal, 2002). Obtaining more detailed protein maps from diseased and control tissues has been the pursuit of Caprion Pharmaceuticals (Montreal, Canada), who developed CellCarta, a technology which allows the construction of intracellular maps of a protein’s location, orientation, movement, and post-translational modifications. Caprion has applied their subcellular approach in protein expression profiling for the development of cancer treatments in collaboration with IDEC pharmaceuticals (Jain, 2002). 2.1. Biomarker identification In addition to using proteomic tools to identify an array of proteins that are modified in the diseased state, a significant focus of proteomic activities in drug discovery has turned to the identification of biomarkers in easily accessible biological fluids. The importance of the development of such markers is evident when one considers the influence of such a tool on all stages of drug development. Not only can a biomarker aid in the understanding of the disease process and progression and what molecular pathways are involved, but also this biomarker can then serve as a monitoring tool in later stages of development. For instance, a change in the status of this marker may be useful in determining the efficacy of various drug candidates in the process of lead optimization, and then can also be used in the selection of appropriate animal models for pre-clinical studies as well as in patient profiling for clinical trials. Good examples of such markers are the serum and urine biomarkers used to identify arthritis. Numerous biomarkers from synovial fluid, blood, and urine have been used to identify and study the stages
379
of osteoarthritis (OA) (e.g. urinary collagen fragments; Downs et al., 2001). Researchers from Pfizer, Inc., in collaboration with investigators from McGill University (Montreal, Canada), compared 14 different serum and urine biomarkers in OA patients to determine how they could be grouped and which markers best distinguished patients from controls (Otterness et al., 2000). Following principal components analysis of marker values collected at baseline and at various later stages of disease progression, the markers were grouped into clusters which represented inflammation markers, bone markers, and markers of cartilage anabolism and catabolism. A group of three markers best distinguished the OA patients: one inflammatory marker and markers of both cartilage anabolism and catabolism. So, in this example, biomarkers that are directly indicative of the disease process (skeletal metabolism) are readily measured in body fluids and can be used throughout the development of drug candidates. This study also illustrates the point that often more than one biomarker is necessary or preferable for distinguishing disease states. A single marker may not readily track with disease progression in each patient, especially if the marker is not related to the molecular basis of the disease. As an additional example in the case of arthritis, researchers at Lund University in Sweden monitored serum levels of some of the markers also measured in the OA study, including cartilage oligomeric matrix protein, 846 epitope, C-propeptide of type II collagen, and bone sialoprotein in patients. By relating the levels of the various markers, the group was able to predict prognosis, in particular the rapidity of disease progression, in these patients (Mansson et al., 1995). 2.2. Identifying protein modifications Beyond the identification of proteins involved in disease progression, it is often critical to explore protein modifications to elucidate the changes in molecular pathways involved in disease. In a study of lymphocytes with normal or Scott syndrome (inherited bleeding disorder) phenotype, changes in the tyrosine phosphorylation status of Ig-chain precursor, fascin, and actin-associated proteins were identified by immunoprecipitation of tyrosine-phosphorylated proteins followed by separation of the proteins by 2-DE (Imam-Sghiouar et al., 2002). Using an alternative
380
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
method for immunoprecipitation of phosphorylated proteins, a group of researchers at the University of Dundee (Allan et al., 2003) investigated the phosphorylation of caspase-9 (which inhibits apoptosis) via the ERK-MAPK pathway, which may help to resolve the mechanism of tumor induction during constitutive activation of this pathway. In this study extracts containing glutathione-S-transferase (GST)-labeled caspase-9 were incubated with 32 P-ATP under conditions of maximal kinase stimulation, followed by affinity purification of caspase-9-GST. After reduction, alkylation of cysteine residues, and separation on a gradient gel, caspase-9 was excised and digested with trypsin. The peptide fragments were analyzed by MALDI-TOF MS and the identity of each peptide was confirmed by Edman degradation. The advantages of this technique of phosphorylation exploration are that it is more sensitive at detecting phosphorylated peptides, it is quantitative, and that the site(s) of phosphorylation are mapped by Edman degradation sequencing. Another protein modification that is clearly modified in many disease states is glycosylation; in fact, many proteins known to be involved in disease processes or that are already being used as biomarkers or possible therapeutic candidates are glycoproteins. A group of scientists at Oxford University have reported a powerful procedure for examining glycoproteins. Peracaula et al. (2003a) found that secreted RNAse 1 from pancreatic adenocarcinoma tumor cells displayed vastly different glycosylation patterns than RNAse 1 from healthy pancreas. The general procedure used to determine the glycosylation patterns was to isolate RNAse 1 by affinity chromatography, digest the RNAse 1 with N-glycosidase F to release the glycans, and to either label the glycans with fluorescent probes for HPLC analysis, or examine the glycans by MALDI-TOF. The distinct patterns of glycosylation that they identified could actually be selected by monoclonal antibodies specific for the carbohydrate side chains, providing an easy detection method for a potential biomarker of pancreatic adenocarcinoma. Similarly, the same group found that the glycoprotein prostate cancer marker prostate-specific antigen (PSA) from a cancer source (LNCaP cell line) contained neutral glycans with a high fucose and GalNAc content in contrast to PSA from normal origins (Peracaula et al., 2003b). This data may potentially provide a way to distinguish benign from malignant prostate
tumors if the differences are consistent in patients and can be identified from serum PSA. The results of these glycosylation studies highlight the importance of identifying potential modifications (i.e. shift in gel mobility) in proteins of interest, as these modifications may be indicative of the disease state or may prove to be valuable ways to develop biomarker assays. 2.3. Exploring protein–protein interactions The study of molecular pathways in diseased and normal states necessitates the development of tools to study protein–protein interactions. Such techniques have greatly expanded in recent years. The ability to affinity-purify a protein facilitates the identification of protein associations via isotopic labeling in cell culture using a method known as stable isotopic amino acids in culture (SILAC). Using this strategy, Blagoev et al. (2003) incubated HeLa cells cultured in the presence or absence of epidermal growth factor (EGF) with either [12 C] or [13 C] arginine. The lysates were mixed and the EGF receptor was affinity-purified over a column displaying the SH2 domain of the adapter Grb2. Proteins which bound to the activated receptor were identified after digestion and LC–MS/MS analysis. Other methods used to study protein interaction include two-hybrid systems, phage display, Biacore® surface plasmon resonance (SPR) technology, and the more recent techniques of reverse transfection and tandem affinity purification (TAP) (Burbaum and Tobal, 2002; Jain, 2002). Reverse transfection involves creating an array of the cDNA of interest on a glass slide, followed by transfection of cells that attach to the slide. The protein is then generally visualized with fluorescent antibodies to assess its localization. This technique can be combined with a two-hybrid approach to assess protein interactions (Burbaum and Tobal, 2002). The TAP method involves the fusion of a TAP tag to the N- or C-terminus of a protein of interest so that it can be expressed and then affinity purified from a host cell. The TAP tag is made up of two IgG binding domains of protein A and a calmodulin binding peptide, separated by a TEV (tobaccoetch virus) protease cleavage site. The tag is introduced in-frame in the coding region of the target protein in an expression vector which is expressed in the appropriate host. Generally expression is maintained at close to normal levels so that erroneous protein
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
associations are kept at a minimum. The protein can then be purified in its native state, concentrated, and complexes can be identified by mass spectrometry, Edman degradation, or Western blot (Puig et al., 2001). Advances in mass spectrometry have also facilitated the study of protein–protein complexes. Fourier transform ion cyclotron resonance mass spectrometry (FTICR-MS) provides high mass resolution and accuracy of whole proteins and protein–protein complexes via the circulation of ions in a superconducting magnet; isotopes can be clearly resolved in high charge states. FTICR-MS has been used to study the contributions of electrostatic and hydrophobic interactions to the binding of calmodulin to synthetic mutated peptides derived from smooth muscle myosin light chain kinase in the presence or absence of calcium (Nousiainen et al., 2003).
3. Target validation Thus far this discussion has focused on examples of the methods used to identify changes in protein expression, form, or interactions in disease states. Once these proteins are identified, the questions of relevance to disease etiology and generality of the disease biomarker still remain. While it is important to establish that a potential protein target is present in the disease-relevant cell or human tissue, the next phase in development is to begin to validate that target by modulating the protein’s activity in a model system to determine the outcome on disease phenotype. Transgenic animals have been classically used to study protein function in a model system. In addition to transgenic mice, transgenic zebrafish are now being used, as they are easier to care for and have shorter life cycles than mice, and actually serve as excellent models for some human diseases and processes such as angiogenesis, inflammation, and insulin regulation (Smith, 2003). A novel type of mouse “knockin” model, called the analog-sensitive kinase alleles (ASKA) mouse, goes a step beyond transgenics and knockout mice in that the native protein kinase expressed in this model is a mutated version that can be inhibited by an ATP analog “P-inhibitor.” The advantage in this model is that the mouse can develop normally, and the enzyme can be “knocked out” at a given time later in its lifespan and the effect of this inhibition on its health and aging
381
can be assessed (Szymkowski, 2003). Researchers at Sangamo BioSciences (Richmond, CA) have also developed a system for regulating the expression of particular proteins in cell and whole animal models. Their system uses zinc-finger transcription factors where the DNA recognition domain is attached to a functional domain and allows the target gene expression to be regulated. This group has been very active in studying the importance of the regulation of different protein isoforms (Smith, 2003). One of the most powerful tools in target validation, though, has been the use of models where the protein target of interest has been completely “knocked out.” While knockout mice have been very useful in this area, they are costly and time-consuming models. Alternative systems to the knockout animal are becoming available to facilitate target validation; one of the most exciting of these is RNA interference (RNAi), a technique where protein expression is “silenced” at the post-transcriptional level. Briefly, cells are presented with short segments of double-stranded RNA specific to the protein target that are then processed into short single-stranded fragments which bind to the complementary antisense strands and form the RNA-induced silencing complex (RISC). It is thought that such RISC complexes inhibit the protein expression by promoting RNA degradation and translational inhibition. While RNAi has been in use for the last several years in systems such as plants and C. elegans, it is only in the last 2 years that RNAi has been successfully demonstrated in mammalian cells. Improvements are currently being made to this technique, including direct treatment of mammalian cells with the single-stranded RNA fragments (siRNAs), which is effective in many cell types but are transient “knockouts.” To overcome the transient nature of this technique in mammalian cells, researchers have used treatment of the cells with short hairpin RNAs (shRNAs) that are composed of an inverted repeat sequence in a hairpin structure, which support stable expression of the inhibitory RNA (Hannon, 2002). RNAi should prove to be a very powerful method to validate target protein function. However, concern that the reduction in RNA levels will not always lead to a reduction in protein expression has led some researchers to develop intracellular antibody capture technology (IACT) to select antibodies that will directly target the protein of interest in a cell (Szymkowski, 2003).
382
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
Two additional techniques used for turning off protein function in cellular systems are selected interacting domains (SIDs) and chromophore-assisted laser inactivation (CALI). SIDs are dominant-negative acting fragments that can modulate protein function in molecular pathways. CALI is a procedure where proteins are photochemically modified and therefore acutely “knocked out” of a system. CALI has been used in functional screens as well as in individual protein studies, rendering it useful in target discovery as well (Jain, 2002). Protein and antibody microarrays are also being used for screening of protein interaction and function. Protein “biochips” or microarrays are ideal for studying receptor–ligand interactions, enzyme activity, and antibody–antigen interaction with rapid throughput. These tools can be useful for both understanding molecular pathways in which the target protein participates, as well as screening model systems for changes that occur in disease states or following drug treatment (i.e. enzyme activity testing) (Huels et al., 2002). Tools for screening interactions with entire “proteome sets” of both known and unknown proteins are also available in microarray format. Xencor’s ProCode and Sense Proteomics (Cambridge, UK) COVET technology allow for the creation of a set of proteins from a given protein expression library which are all tagged with plasmids. These proteins can be arrayed and screened and the cDNA tags allow for rapid protein identification (Smith, 2003). The creation of very specific and sensitive assays for proteins of interest has also been achieved through the use of aptamers, or synthetic nucleic acid ligands. In addition to being used in the place of antibodies or protein ligands in microarrays, stabilized aptamer forms can be used to validate the function of various protein targets, and aptamers designed against target proteins can be applied in competition assays to screen lead compounds. The advantages of using aptamers for microarray assays are their high specificity achieved by rigorous selection, and potential greater sensitivity when the aptamers are modified to cause covalent linkage to the microarray surface. The detection of aptamers has been aided by the development of aptazymes, or aptamers conjugated with ribozyme domains, and the construction of pairs of aptamers recognizing adjacent epitopes on the target protein. The binding of an aptazyme to its target causes a
conformational change which alters the activity of the ribozyme. The binding of adjacent aptamers to their respective target sequences creates an oliogonucleotide bridge and produces ligation of the connector nucleotide, which is subsequently amplified for detection. The latter technique has the potential to detect a protein with sensitivity 1000-fold greater than a sandwich ELISA (Burgstaller et al., 2002).
4. Drug design and lead optimization Once a target protein has been validated, the task of identifying chemical compounds that can appropriately modulate the target can also be aided by proteomic techniques: namely structural proteomics. If a specific protein target has been identified, in many cases now the expression and purification of the protein for crystallization can be achieved. While crystallization is still a significant hurdle to overcome, such procedures are constantly being updated and optimized (Syrrx, San Diego, CA). Once crystallization is achieved, X-ray diffraction and solving of the protein structure is being aided by the development of multiple-wavelength anomalous diffraction (MAD) phasing, which entails a high-energy tunable X-ray source with enhanced computational abilities. The knowledge of the crystal structure of the protein facilitates the production of suitable drug candidates, as it can be directly input into virtual screening software that will select fragments or compounds that will bind the protein surface of interest. Such virtual screening is invaluable not only to the selection of a group of chemical entities which will “hit” the protein target, but as a screen for structure-activity relationships among those agents to optimize the lead compound (e.g. SkelGen, de novo Pharmaceuticals, Cambridge, UK and software products of Accelrys, San Diego, CA). Such structural proteomic screening has been very useful in the development of HIV protease inhibitors, influenza virus neuraminidase inhibitors, and antimicrobial agents, where the technique may also help in developing new broad-spectrum antibiotics (Schmid, 2002). Even if the crystal structure of the protein is not available, software which will produce a virtual protein structure is available (e.g. Quasi2, de novo Pharmaceuticals, Cambridge, UK) (Smith, 2003).
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
Beyond virtual screening surveys, the interaction of potential drug compounds with particular proteins can be examined with the use of activity-based probes (ABPs). These chemically reactive probes provide the ability to measure the protein activity based on the active component of the particular class of proteins, and can be used for inhibition studies with potential drug agents (Burbaum and Tobal, 2002).
5. Safety and diagnostic biomarkers for pre-clinical and clinical development Once a class of compounds is in pre-clinical development, lead optimization and the viability of the project can depend on assays developed to assess the efficacy and toxicity of the lead agents. Many pharmaceutical companies are beginning to use proteomic techniques to accomplish these tasks. When searching for toxicity (safety) biomarkers, pre-clinical development groups are often hoping to find not only biomarkers specific to the toxicity of the current compound in development, but also general tissue toxicity biomarkers to assess compounds in future studies. Scientists at GlaxoSmithKline have used conventional 2-DE proteomics techniques to identify protein alterations involved in gentamicin-induced nephrotoxicity (Charlwood et al., 2002). The proteins identified were consistent with the mechanism of toxicity, and these researchers suggested that these markers may be useful in screening antibiotics in development for nephrotoxicity potential. In another study at GlaxoSmithKline, researchers used Ciphergen’s ProteinChip® /SELDI (surface-enhanced laser desorption/ionization) technology to discover a urinary biomarker which could be useful in detecting skeletal muscle toxicity (Dare et al., 2002). SELDI allows for the enrichment of proteins with particular binding affinities (e.g. hydrophobic, hydrophilic, metal binding, receptor affinity) through the chemically modified surfaces of the proprietary ProteinChips. Using this technology, one can screen a complex mixture of proteins such as those found in serum or urine through fractionation of the sample directly on the chip. By profiling samples from both treated/disease and control, a pattern of protein differences can distinguish the treatment groups. Using SELDI profiling, protein purification, and on-chip trypsin digestion for peptide mapping, the researchers
383
found that urinary parvalbumin-␣ was increased in rats treated with 2,3,5,6-tetramethyl-p-phenylenediamine, a compound which induces skeletal muscle toxicity. SELDI technology has also been shown to be a successful profiling tool in populations of cancer patients. Li et al. (2002) report that a set of three proteins selected from Ni2+ -affinity ProteinChip profiles of patient serum can serve as biomarkers to distinguish normal individuals from patients with stage 0–I breast cancer, and between patients with stage II or stage III breast cancer. Markers that can help determine not only the existence of the disease but also the stage of the disease provide benefits first to the patients in more accurate diagnoses and treatment, and second to those in pharmaceutical R&D who can then profile patients for clinical trials. It is reasonable to assume that the better the model animal or patient in pre-clinical and clinical trials, the more beneficial the research in advancing health care.
6. Systems biology approach One fact which proteomics research has made quite clear is that there is rarely a single target of a disease or chemical; often several protein targets or biomarkers are discovered and validated. Part of the reality of trying to develop agents to treat diseases such as heart disease and diabetes is that these diseases are enormously complex and involve simultaneous pathologies of multiple organ systems. A relatively new wave of technology that is attempting to integrate a greater amount of the potential information available on disease processes, and that incorporates proteomics as one of its major tools, is systems biology. This approach involves the simultaneous measurement of genomic, proteomic, and metabolomic parameters in a given system under defined conditions (Davidov et al., 2003). Such measurements are then subjected to rigorous computational biology analyses to define the link which ties the data to the particular disease or perturbation in question. In this way, instead of using just one “omics” technique to produce a laundry list of genes, proteins, or metabolites that are altered in the disease state, the combined analysis of all data sets may be able to define the common thread, such as the regulation of a gene, that results in the given phenotype. Further, once a systems biology analysis
384
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385
for a particular disease model is completed, that data, fed into an in silico biology platform, can be used to simulate the effects of a particular drug candidate in the system. Also, the model could provide information about pathways that may contribute to side effects of a given agent, and can therefore be utilized in lead compound optimization. In addition to several biotech companies such as Beyond Genomics and Bioseek, Eli Lilly has recently developed a center for the pursuit of systems biology (Davidov et al., 2003).
7. Conclusions The results of recent proteomics pursuits have demonstrated the potential value that proteomics has to offer in drug development. Proteomics techniques are providing precise and fairly rapid methods to screen both target proteins and potential therapeutic compounds. In-house, drug companies have mainly focused on protein profiling for target identification, to develop efficacy and toxicity biomarkers, and to create valuable protein databases for access in future projects. Through collaboration with biotech firms and universities, the pharmaceutical industry is gaining access to exciting innovations in the proteomics field such as RNA interference and systems biology, and is testing the results of proteomic techniques in patient populations. The key to drug development in the future is the integration and application of the vast amount of information that techniques such as proteomics make available to researchers.
References Allan, L.A., Morrice, N., Brady, S., Magee, G., Pathak, S., Clarke, P.R., 2003. Inhibition of caspase-9 through phosphorylation at Thr 125 by ERK MAPK. Nat. Cell. Biol. 5 (7), 647–655. Blagoev, B., Kratchmarova, I., Ong, S-E., Nielsen, M., Foster, L.J., Mann, M., 2003. A proteomics strategy to elucidate functional protein–protein interactions applied to EGF signaling. Nat. Biotech. 21, 315–318. Burbaum, J., Tobal, G.M., 2002. Proteomics in drug discovery. Curr. Opin. Chem. Biol. 6, 427–433. Burgstaller, P., Jenne, A., Blind, M., 2002. Aptamers and aptazymes: accelerating small molecule drug discovery. Curr. Opin. Drug Discov. Develop. 5 (5), 690–700. Charlwood, J., Skehel, J.M., King, N., Camilleri, P., Lord, P., Bugelski, P., Atif, U., 2002. Proteomic analysis of rat kidney
cortex following treatment with gentamicin. J. Prot. Res. 1, 73–82. Dare, T.O., Davies, H.A., Turton, J.A., Lomas, L., Williams, T.C., York, M.J., 2002. Application of surface-enhanced laser desorption/ionization technology to the detection and identification of urinary parvalbumin-␣: a biomarker of compoundinduced skeletal muscle toxicity in the rat. Electrophoresis 23, 3241–3251. Davidov, E.J., Holland, J.M., Marple, E.W., Naylor, S., 2003. Advancing drug discovery through systems biology. Drug Discov. Today 8 (4), 175–183. DiMasi, J.A., 2001. Risks in new drug development: approval success rates for investigational drugs. Clin. Pharmacol. Ther. 69 (5), 297–307. DiMasi, J.A., Hansen, R.W., Grabowski, H.G., 2003. The price of innovation: new estimates of drug development costs. J. Health Econ. 22 (2), 151–185. Downs, J.T., Lane, C.L., Nestor, N.B., McLellan, T.J., Kelly, M.A., Karam, G.A., Mezes, P.S., Pelletier, J-P., Otterness, I.G., 2001. Analysis of collagenase-cleavage of type II collagen using a neoepitope ELISA. J. Immun. Meth. 247, 25–34. Hannon, G.J., 2002. RNA interference. Nature 418, 244–251. Huels, C., Muellner, S., Meyer, H.E., Cahill, D.J., 2002. The impact of protein biochips and microarrays on the drug development process. Drug Discov. Today 7 (Suppl 18), S119–S124. Imam-Sghiouar, N., Laude-Lemaire, I., Labas, V., Pflieger, D., Caron, M., Nabias, D.K., Joubert-Caron, R., 2002. Subproteomics analysis of phosphorylated proteins: application to the study of B-lymphoblasts from a patient with Scott syndrome. Proteomics 2, 828–838. Jain, K.K., 2002. Using proteomics to identify targets and validate leads. Trends Biotech. 20 (8), 319–320. Kawakami, T., Nagata, T., Muraguchi, A., Nishimura, T., 2003. Proteomic approach to apoptotic thymus maturation. J. Chromatogr. B 787, 223–229. Li, J., Zhang, Z., Rosenzweig, J., Wang, Y.Y., Chan, D.W., 2002. Proteomics and bioinformatics approaches for identification of serum biomarkers to detect breast cancer. Clin. Chem. 48 (8), 1296–1304. Macri, J., Rapundalo, S.T., 2001. Application of proteomics to the study of cardiovascular biology. TCM 11 (2), 66–75. Mansson, B., Carey, D., Alini, M., Ionescu, M., Rosenburg, L.C., Poole, A.R., Heinegard, D., Saxne, T., 1995. Cartilage and bone metabolism in rheumatoid arthritis. Differences between rapid and slow progression of disease identified by serum markers of cartilage metabolism. J. Clin. Invest. 95 (3), 1071–1077. Nousiainen, M., Derrick, P.J., Lafitte, D., Vainiotalo, P., 2003. Relative affinity constants by electrospray ionization and Fourier transform ion cyclotron resonance mass spectrometry: calmodulin binding to peptide analogs of myosin light chain kinase. Biophys. J. 85 (1), 491–500. Otterness, I.G., Swindell, A.C., Zimmerer, R.O., Poole, A.R., Ionescu, M., Weiner, E., 2000. An analysis of 14 molecular markers for monitoring osteoarthritis: segregation of the markers into clusters and distinguishing osteoarthritis at baseline. Osteoarthr. Cart. 8, 180–185.
J.L. Walgren, D.C. Thompson / Toxicology Letters 149 (2004) 377–385 Peracaula, R., Royle, L., Tabarés, G., Mallorqúi-Fernández, G., Barrabés, S., Harvey, D.J., Dwek, R.A., Rudd, P.M., de Llorens, R., 2003a. Glycosylation of human pancreatic ribonuclease: differences between normal and tumor states. Glycobiology 13 (4), 227–244. Peracaula, R., Tabarés, G., Royle, L., Harvey, D.J., Dwek, R.A., Rudd, P.M., de Llorens, R., 2003b. Altered glycosylation pattern allows the distinction between prostate-specific antigen (PSA) from normal and tumor origins. Glycobiology 13 (6), 457–470. Puig, O., Caspary, F., Rigaut, G., Rutz, B., Bouveret, E., BragadoBilsson, E., Wilm, M., Séraphin, B., 2001. The tandem affinity
385
purification (TAP) method: a general procedure of protein complex purification. Methods 24, 218–229. Scherl, A., Couté, Y., Déon, C., Callé, A., Kindbeiter, K., Sanchez, J-C., Greco, A., Hochstrasser, D., Diaz, J-J., 2002. Functional proteomic analysis of human nucleolus. Mol. Biol. Cell 13, 4100–4109. Schmid, M.B., 2002. Structural proteomics: the potential of high-throughput structure determination. Trends Microbiol. 10 (Suppl 10), S27–S31. Smith, C., 2003. Hitting the target. Nature 422, 341–347. Szymkowski, D.E., 2003. Target validation joins the pharma fold. Targets 2 (1), 8–9.