Bioscience Reports, Vol. 25, Nos. 1/2, February/April 2005 (Ó 2005) DOI: 10.1007/s10540-005-2851-3
Clinical Proteomics: From Biomarker Discovery and Cell Signaling Proﬁles to Individualized Personal Therapy Katherine R. Calvo,1,3 Lance A. Liotta,1 and Emanuel F. Petricoin2
The discovery of new highly sensitive and speciﬁc biomarkers for early disease detection and risk stratiﬁcation coupled with the development of personalized ‘‘designer’’ therapies holds the key to future treatment of complex diseases such as cancer. Mounting evidence conﬁrms that the low molecular weight (LMW) range of the circulatory proteome contains a rich source of information that may be able to detect early stage disease and stratify risk. Current mass spectrometry (MS) platforms can generate a rapid and high resolution portrait of the LMW proteome. Emerging novel nanotechnology strategies to amplify and harvest these LMW biomarkers in vivo or ex vivo will greatly enhance our ability to discover and characterize molecules for early disease detection, subclassiﬁcation and prognostic capability of current proteomics modalities. Ultimately genetic mutations giving rise to disease are played out and manifested on a protein level, involving derangements in protein function and information ﬂow within diseased cells and the interconnected tissue microenvironment. Newly developed highly sensitive, speciﬁc and linearly dynamic reverse phase protein microarray systems are now able to generate circuit maps of information ﬂow through phosphoprotein networks of pure populations of microdissected tumor cells obtained from patient biopsies. We postulate that this type of enabling technology will provide the foundation for the development of individualized combinatorial therapies of molecular inhibitors to target tumor-speciﬁc deranged pathways regulating key biologic processes including proliferation, differentiation, apoptosis, immunity and metastasis. Hence future therapies will be tailored to the speciﬁc deranged molecular circuitry of an individual patient’s disease. The successful transition of these groundbreaking proteomic technologies from research tools to integrated clinical diagnostic platforms will require ongoing continued development, and optimization with rigorous standardization development and quality control procedures. KEY WORDS: Clinical Proteomics; mass spectroscopy; protein microarrays; combinatory theraphy; oncology; pathology; microdissection.
Human disease is thought to be largely genetic in etiology. Underlying genetic mutations are either inherited through the germline or acquired somatically over time. Such mutated genes encode altered proteins that perturb normal cellular 1 Laboratory of Pathology, National Cancer Institute, National Institutes of Health, Bethesda, MD 20892, USA. 2 Oﬃce Cell Therapies and Gene Therapies, Center for Biologic Evaluation and Research, U.S. Food and Drug Administration, Bethesda, MD 20852, USA. 3 To whom correspondence should be addressed. E-mail: [email protected]
0144-8463/05/0400-0107/0 Ó 2005 Springer Science+Business Media, Inc.
Calvo, Liotta, and Petricoin
physiology resulting in disease. The current ongoing revolution in molecular medicine has sought to understand the molecular basis of human disease with an ultimate goal of developing rationally designed therapies. It consists of multiple evolving phases. The gene discovery phase has been largely driven by key technological advances including PCR, high-throughput sequencing and bioinformatics. This phase recently culminated in the completion of the Human Genome Project in 2003 [1,2], 50 years following the discovery of the DNA double helix. Now that the genome is sequenced, there are ongoing eﬀorts to identify genetic polymorphisms (e.g. single nucleotide polymorphisms [SNPs]) that may point to disease predisposition, or unique response to therapy such as untoward drug side eﬀects . Development of microhybridization arrays has powered the functional genomics phase in which gene expression proﬁling is being used to correlate gene expression patterns with disease classiﬁcation and predict response to therapy. Gene expression proﬁles have been demonstrated to be able to further subclassify and predict outcomes for complex entities such as lymphoma [4–6], prostate cancer , breast cancer , and ovarian cancer . Although the ‘‘blueprints’’ of human disease may be genetically encoded, the execution of the disease process occurs through altered protein function. Hence, identifying the genetic or epigenetic events leading to disease requires subsequent understanding of the proteomic consequences of these events. While gene microarray studies elucidate gene expression patterns associated with disease, they give no indication of the complexity of protein–protein interactions, their localization, or whether the encoded proteins are stably expressed, phosphorylated, cleaved, acetylated, glycosylated or functionally ‘‘active’’. For many diseases, such as cancer, protein function is altered in the context of key signaling pathways that regulate critical cellular functions including proliferation, apoptosis, diﬀerentiation, survival, immunity, metabolism, invasion and metastasis. Understanding which combinations of protein regulatory networks are dysfunctional, and at which speciﬁc nodes in the cell circuitry, may be fundamental for the development of eﬀective combinations of pharmacologic inhibitors . Proteomic expression proﬁling provides an opportunity for a synergistic systems biology approach to the understanding of disease that, when combined with gene transcript proﬁling, can amplify our knowledge repertoire. The next phase of the molecular medicine revolution involves the use of genomic technologies combined with newly evolving proteomic technologies to diagnose, subclassify, and drive the development of individualized molecularly targeted therapies, ushering in a new era of clinical medicine. The predominant technologies driving the proteomic phase of molecular medicine involve distinct modalities that approach diagnostics from fundamentally different but complementary starting points. A fundamental and underpinning hypothesis of mass spectrometry (MS) based proﬁling is that there exists in the LMW information archive molecules that, when measured in combination, can detect disease with a greater accuracy than any single marker alone. Proteomic proﬁling using mass spectroscopy (MS) technologies (e.g., SELDI-TOF and MALDI-TOF) generate complex ﬁngerprints of ion peaks from small amounts of human serum or tissue [11,12]. Multiple studies have revealed the existence of this information archive and the feasibility of developing this technology for biomarker discovery  for the early detection of diseases including ovarian cancer [14,15], breast cancer  and
prostate cancer [17,18]. SELDI-TOF mass spectrometry has been shown to be able to discriminate patients having prostate cancer with a speciﬁcity and sensitivity greater than the currently used prostate speciﬁc antigen test [18,19]. Potential serum biomarkers for early stroke have been identiﬁed via SELDI-TOF analysis . Clarke et al.  reported the successful analysis of urine by SELDI-TOF MS to distinguish rejection vs. no rejection in a renal transplant population. This is particularly promising for immunocompromised transplant patients. Urine proteomic pattern diagnostics represents a diagnostic method with no morbidity and mortality compared with standard invasive kidney biopsies. Analysis of cerebrospinal ﬂuid by MS has led to the identiﬁcation of potential biomarkers for Alzheimer’s disease . Some commercial approaches utilize the concept of pattern analysis where the identity of the individual molecules generating the pattern peaks may not be known, but the speciﬁc uncharacterized pattern itself can still be utilized as a diagnostic . While previous approaches to identify biomarkers have sought to ﬁnd single biomarkers indicative of disease, evidence provided from serum proﬁling eﬀorts indicates that an endpoint comprised of multiple simultaneously measured analytes may be more powerful at diagnosis than any of the individual proteins making up the portrait. In essence, the whole is greater than the mere sum of its parts, much akin to gene transcript proﬁling applications. Interestingly, the concept of patternbased diagnostics is not unfamiliar to the anatomic pathologist physician with a ‘‘well-trained eye’’, who currently diagnoses human disease based largely on the morphologic patterns gleaned from histologic sections of diseased tissue under the microscope. Proteomic pattern analysis approach oﬀers several advantages over previous technologies , but also has several roadblocks ahead, which must be overcome for routine clinical use. The most obvious current impediment is the lack of reported day-to-day and machine-to-machine reproducibility for the generation of identical looking spectra. Until now, mass spectrometry has primarily functioned as a research and biodiscovery tool. The transition of this technology from a research tool to a reliable clinical diagnostic platform will require rigorous standardization, spectral quality control and assurance, standard operating procedures for robotic and automatic sample application, and standardized controls to insure the generation of highly reproducible spectra . Since many investigators are independently developing their own methods, optimization procedures and in-process controls, there appears to be a notable and somewhat understandable lack of coordinated eﬀorts to standardize methodology within the community at this infant stage of research and development. The development of standardized technology MS platforms, which are constantly evolving, reference standards for controls and calibrators will certainly help the ﬁeld accelerate to rigorous evaluation for clinical applications. This transition is anticipated to require widespread collaborative eﬀorts between government, industry, university and community based health care delivery systems. Certainly an advantage of a proteomic proﬁle of uncharacterized and unidentiﬁed molecules has over standard complexed immunoassay measurements is that the ﬁngerprint can be rapidly obtained from as little as 1 ll of raw, unfractionated serum from patients. The small serum sample can be analyzed by MS type approaches generating a unique proteomic signature of the serum quite rapidly (Fig. 1).
Calvo, Liotta, and Petricoin
Fig. 1. Proﬁling and Characterization of the Low Molecular Weight Biomarker Archive. Low molecular weight components are harvested from the circulation ex vivo using specialized nanoparticles designed to concentrate and harvest low molecular weight biomarkers prior to analysis. The information content contained on the harvested particles can be directly sequenced and/or input into MS based proﬁling analysis.
BIOMARKERS ARISING FROM THE MICROECOLOGY AT THE TUMOR–HOST INTERFACE Based on the discovery that these LMW molecules exist in the circulation and comprise much of the mass spectral information output generated by MS proﬁling, the opportunity now exists to extend simple pattern analysis as an observation of unknown peak collections that appear to correlate with disease to a state where these molecules can be puriﬁed, sequenced, and identiﬁed. Once each candidate biomarker is identiﬁed, the next objective is to develop capture reagents (e.g., antibodies) that can be used to measure multiplexed panels of analytes consisting of subsets of the candidate biomarkers. However, in contrast to direct MS proﬁling of blood or tissue, sequencing and characterization of the underlying constituents is a very laborious process. In fact, the cycle time for protein sequencing, characterization, antibody (or analyte speciﬁc ligand) development, validation in clinical research study sets and immunoassay development is the biggest impediment for the direct characterization approaches. The obvious advantage of this path is that once characterized, reproducibility of measurements of the analytes using well-tested and validated immunoassay platforms is not an issue. Additionally once the molecules are identiﬁed, bias and over-ﬁtting can be assessed directly.
Where do these molecules come from and how do they achieve an aggregate concentration that can be measured successfully? Complex diseases like cancer are products of their proteomic tissue microenvironment, involving communication networks between cells, stroma and extracellular matrix . The tumor–host interface system is thought to involve unique enzymatic events, ﬂow of information, and sharing of growth factors and metabolic substrates. The blood proteome is presumably altered in the diseased population as a consequence of constant perfusion of the diseased organ. Unique disease-related diﬀerences in protein levels could theoretically be due to several factors including (1) the overexpression and/or abnormally shedding of speciﬁc proteins into the serum proteome, (2) the shedding of proteins that are uniquely cleaved or modiﬁed as a consequence of the disease process, or (3) the subtraction of speciﬁc proteins from the serum proteome owing to abnormal activation of proteolytic degradation pathways in the diseased state. Quaternary protein structure relationships due to disease-related protein–protein interactions and protein-complex formation may also contribute to changes in the serum proteome. Many tumors, such as breast carcinoma, induce prominent desmoplastic reactions in the adjacent stromal tissue. These tumor-induced stromal reactions can result in hyalinization or ﬁbrosis, which contributes to the ﬁrm hard quality of a tumor ‘‘lump’’. It is reasonable to hypothesize that protein fragments associated with the unique and active biological processes occurring at the tumor stromal interface would be shed into the extracellular interstitium. A unique combination of protein fragments derived from the microenvironment of the tumor–host interaction would be drained from the tumor site via the lymphatics and ultimately shed into the serum. Hence, the circulation is a protein-rich information reservoir that contains the traces of what has been encountered by the blood during its constant perfusion of tissues throughout the body. If a sensitive and speciﬁc reliable method of detection and characterizing LMW tumor markers is developed it is theoretically possible to detect the presence of a tumor while it is still microscopic in size. Although proteomic ﬁngerprinting based approaches where patterns of unknown entities are used as a classiﬁcation tool , we and others are currently orienting our eﬀorts for the identiﬁcation and sequencing the LMW range of the circulatory proteome that underpin the MS proﬁles. The identiﬁcation of the speciﬁc analytes will be very important to enhance our understanding of disease mechanism, potentially provide new novel eﬀective drug targets, and perhaps lead to the discovery of analytes that can be measured as a multiplex immunoassay using conventional antibody-base approaches. Since it is likely that such biomarkers would be smaller cleaved proteins arising from larger molecules, it will be necessary to develop methods to eﬀectively distinguish the diagnostic fragments from the larger parental molecules, as epitopes contained in the low molecular weight forms might cross react with the wild type normal proteins. Regardless of whether or not these biomarker candidates are measured as unknown entities by MS proﬁling, or via a multiplexed immunoassay, we would predict that these ‘‘combinatorial diagnostics’’ approaches would be inherently superior to single marker antibody based tests for early disease detection . Mathematically, the measurement of a constellation of rigorously validated multiple biomarkers should contain a higher level of discriminatory power than a single biomarker alone. This may be
Calvo, Liotta, and Petricoin
particularly relevant in the context of heterogenous patient populations and heterogenous disease states. NANOTECHNOLOGY AND THE CIRCULATORY PROTEOME Future developments in biomarker-based proteomics technologies will be dramatically impacted by the recent realization that a high percentage of the diagnostically useful lower molecular weight serum protein entities are bound to higher molecular weight carrier proteins such as albumin . In fact, these carrier proteins likely serve to amplify and protect lower molecular weight biomarkers from clearance by the renal system . Conventional protocols for biomarker discovery have begun by discarding the abundant high molecular weight carrier species without realizing the valuable cargo they harbor. In the future we anticipate the development of novel nanotechnology platforms that will allow the ampliﬁcation and abundant harvesting of diagnostic low molecular weight biomarkers in vivo or ex vivo . Such tools might consist of derivatized gold nanoparticles that actively bind biomarkers providing enriched signature proﬁles elucidated via mass spectrometry platforms . Initial studies with magnetic nanoparticle probes coated with bait antibodies and unique ‘‘bar code’’ DNA fragments are able to amplify signals of low abundant biomolecules at concentrations as low at 3 attomolar (approximately 18–20 copies per 10 ll of ﬂuid) . This ampliﬁcation is comparable to PCR ampliﬁcation of nucleotide sequences, and can theoretically be used to detect hundreds of protein targets at a time in patient samples. CLASSIFICATION OF DISEASE: TISSUE FINGERPRINTING METHODS In the last century, tissue based diagnosis of human disease has largely occurred in under the purview of the medical specialty of anatomic pathology. Despite the many advances in medical technology, nothing to date has consistently outperformed the well-trained human eye of a pathologist at tissue diagnosis and subclassiﬁcation. Biopsies and surgical specimens are formalin ﬁxed, parafﬁn embedded, sectioned onto microscopic slides and stained. Diagnosis is largely made on the basis of morphology and pattern recognition involving multiple variables including tissue architecture, cellular conﬁgurations, pleomorphism, nuclear shape and contour, and staining patterns. For example, cancer cells typically have higher nuclear to cytoplasmic ratios, prominent nucleoli, distinctive chromatin patterns and a high mitotic index. In general, aggregates of tumor cells are disorganized and distort the normal tissue architecture. Accurate diagnosis requires years of experience as many benign reactive conditions can also exhibit similar characteristics. Often the diagnosis of benign vs. malignant may be made solely on the basis of the behavior of cells, as in capsular invasion of a thyroid follicular neoplasm. Subclassiﬁcation of tumors can be challenging, with signiﬁcant clinical ramiﬁcations. For example a benign breast lesion such as sclerosing adenosis may require little treatment and have an excellent prognosis. However, to the inexperienced eye, this lesion could be mistaken for a malignant aggressive invasive carcinoma, which might erroneously result in a mastectomy.
The advent of immunohistochemistry in the last century and use of antibody stains for subclassiﬁcation of tumors has added a signiﬁcant dimension to clinical diagnostics. On the other hand, within tumors bearing the same histologic and immunophenotypic diagnosis there is often a wide range of patient response to treatments. This suggests that there is a diverse biology of tumors on a molecular level that is not necessarily apparent by outward microscopic morphology. A particularly good example of this concept is in the diagnostic entity of diffuse large B-cell lymphoma, which has a very heterogeneous outcome pattern. Gene microarray studies have been able to subclassify several distinct gene expression patterns that correlate with distinct patient outcome patterns [4,5,31]. These distinct groups identiﬁed via gene expression analysis are not apparent by traditional histopathologic analysis. It is likely that the diﬀerential gene expression patterns reﬂect unique combinations of protein products that cooperate along multiple deranged signaling pathways to orchestrate the malignant behavior of an individual patient’s tumor. The complex pattern of protein expression and functional state is presumed to contain important information about the pathologic process taking place in the cells within their tissue microenvironment. This proteomic information should have relevance for diagnostic subclassiﬁcations of tumors and more importantly, valuable information for therapeutic targeting. The other spectrum of clinical proteomics involves a nascent ﬁeld that can be though of as functional ‘‘microproteomics’’ . Once disease has been diagnosed and characterized, the identiﬁcation of speciﬁc derangements within functional regulatory protein networks serves as the basis for the formulation of personalized molecularly targeted therapies . The overriding goal of this ﬁeld is to characterize information ﬂow through known protein–protein signaling networks that interconnect the extracellular tissue microenvironment with the control of gene transcription within normal and diseased cells. Using cancer as a model, the malignant phenotype is the culmination of multiple genetic or epigenetic ‘‘hits’’ [34,35] which cooperate to dysregulate protein function along multiple protein signaling pathways regulating cellular physiologic processes including proliferation, diﬀerentiation, apoptosis, metabolism, immune recognition, invasion and metastasis. Many approaches to elucidating altered protein function in human disease have relied on the use of in vitro cultured cell lines originally derived from fresh tissue. However, cultured cells may not accurately represent the molecular events taking place in the actual tissue they were derived from. Protein expression levels and post-translational modiﬁcations affecting protein activity of the cultured cells are inﬂuenced by the culture environment, and may be quite different from the proteins expressed in the native tissue state. Cultured cells are separated from the tissue elements that regulate gene expression, such as soluble factors, extracellular matrix molecules and cell–cell communication. Human disease occurs in the context of complex tissue microenvironments  involving host stroma, immune cells, cytokines and growth factors that may not be adequately reﬂected either in in vitro studies or in non-human animal studies. In the context of clinical medicine and patient treatment, individual biologic heterogeneity is an additional layer of complexity that must be factored. Each individual patient may harbor unique attributes that are critical, for example, to understanding a distinct tumor–host behavior that can be exploited therapeutically.
Calvo, Liotta, and Petricoin
TISSUE MICRODISSECTION TECHNOLOGY The analysis of pure populations of cells in their native tissue environment is a critical component of proteomic analysis. Achieving meaningful analysis requires more than mere grinding up pieces of patient tissue with subsequent analysis of cell lysate. This is particularly important for diseases such as cancer, where the tumor cells of interest may comprise only a small portion of the biopsy material. For example, a breast biopsy containing invasive carcinoma may contain numerous cell populations in addition to the tumor cells: (1) adipose cells in the tissue surrounding the mammary ducts, (2) normal epithelial and myoepithelial cells (3) ﬁbroblasts and endothelial cells in the stroma and blood vessels, (4) premalignant carcinoma cells in the ductal carcinoma in situ or lobular carcinoma in situ lesions, and (5) regions of invasive carcinoma. Assuming we want to map the aberrant protein signaling circuitry in the tumor cells, these subpopulations may be located in microscopic regions occupying less than 15% of the total biopsy. If the entire biopsy (containing multiple populations of benign cells admixed with tumor cells), were analyzed using an MS or protein microarray technology, the subsequent output data may be severely compromised or confounded if the goal was to characterize the tumor cells. Therefore, cellular heterogeneity in patient specimens can be a barrier to the proteomic analysis of normal and diseased tissue. The problem of cellular heterogeneity in tissue specimens can be resolved using Laser Capture Microdissection (LCM). LCM allows the microdissection and extraction of a microscopic homogeneous cellular sub-population from its complex tissue milieu under direct microscopic visualization [36,37]. With this technology, a pure sub-population can be analyzed and compared to adjacent stromal cells, epithelial cells, or any interacting populations of cells within the same tissue. The integrity of cellular proteins is preserved during microdissection facilitating subsequent quantitative analysis in gene or protein microarrays or MS analysis. The molecular analysis of pathologic processes in clinical specimens can be signiﬁcantly enhanced by procurement of pure populations of cells from complex tissue biopsies using LCM, [38,39]. New generation automated LCM platforms would allow a pathologist to microscopically inspect patient biopsies, identify and circle tumor populations on a computer monitor, with subsequent automated microdissection of selected diseased cells for molecular analysis.
MASS SPECTROMETRY IN TISSUE PROTEOMICS Mass spectrometry (MS) analysis of patient specimens is an area of intense interest within tissue proteomics. This interest is fueled by the dual applicability of MS for biomarker discovery and for the development of clinically applicable tissue proteomic pattern diagnostics . As previously noted, MS platforms require no speciﬁc preliminary knowledge of the identity of proteins that are surveyed, rather the system can generate protein signature ‘‘bar codes’’ that can be used to discriminate of ‘‘normal’’ vs. ‘‘diseased’’. The clinical utility of this technology for tissue biopsy analysis would be the discrimination of (1) reactive processes vs. benign neoplasia vs. malignant neoplasia, and (2) complex subclassiﬁcations within disease types that can predict prognosis or response to current treatment modalities. One of
the ﬁrst studies to utilize MS in tissue analysis demonstrated that MS spectral patterns from tissue biopsies could be used to distinguish benign or premalignant cells from invasive cancer . Solubilized cellular proteins from lysates of pure populations of microdissected premalignant and tumor cells were applied to treated metal chips for SELDI-TOF MS analysis with subsequent analysis by bioinformatics data mining programs as previously described. Unique patterns correlated with the progression of atypical premalignant cells to malignant counterparts. More recently others have performed MS analyses of whole heterogeneous tissue pieces directly without prior microdissection into pure populations of constituent cells. In this context, a frozen section of patient tissue is dried onto a matrix-assisted laser/desorption ionization (MALDI) plate, then directly subjected to a laser in a vacuum chamber. Spectral signature patterns are derived directly from proteins on the surface of the tissue. Using this method Yanagisawa et al. [42,43] reported the classiﬁcation of lung tumors according to disease stage with 85% accuracy. Schwartz et al.  used 20 snap frozen sections of normal brain and brain tumor specimens to generate MS spectral patterns that resulted in the accurate discrimination of glial neoplasms vs. normal brain tissue. One of the beneﬁts of MS based proteomic pattern analysis is that small amounts of tissue are adequate. Hence it may be suitable for analysis of clinical tissue biopsies. Interestingly, in these small initial studies, MS proteomic pattern analysis was able to partition patients into diagnostic groups consistent with pathologic diagnoses obtained via traditional microscopic morphology based analysis. Once a signature has been identiﬁed, MS–MS sequencing technologies can be employed to sequence and identify the underlying molecular component of the peak itself. MICROPROTEOMICS: DIAGNOSING ABERRANT PROTEIN SIGNALING CIRCUITS IN CLINICAL SPECIMENS Within cells, proteins are assembled into complex networks through a variety of protein–protein interactions. The underlying three-dimensional shape of a protein is determined by its amino acid sequence. The structural conformation of a protein and presentation of nested interaction domains (e.g., SH2 and SH3 domains), enables the highly selective recognition between protein partners in a communication circuit. Proteins can undergo conformational changes that functionally permit or prevent protein activity within networks. Conformational changes are largely dictated by post-translational modiﬁcations that include phosphorylation, cleavage, acetylation, glycosylation and ubiquitinylation. Such modiﬁcations functionally deﬁne regulated protein–protein interactions and in essence turn proteins ‘‘on’’ and ‘‘oﬀ’’ at nodes within circuits of information ﬂow. These protein signaling networks regulate key biologic processes deﬁning cell function within larger tissue and organ speciﬁc contexts. In cancer, speciﬁc protein signaling networks are typically deranged resulting in unregulated proliferation, aberrant diﬀerentiation and immortality. Aberrant activity through speciﬁc signaling pathways can be monitored by evaluating the activity of proteins within key nodes. This can be achieved by using antibodies that recognize the active form (e.g., phosphorylated) of a protein vs. the inactive form (e.g. unphosphorylated). Disruption of key regulated protein–protein interactions in diseased cells may serve as important targets of drug therapy .
Calvo, Liotta, and Petricoin
Coupled with LCM, protein microarrays offer the advantage of allowing the evaluation of native proteins in normal and diseased cells and the post-translational modiﬁcations associated with protein–protein interactions [45,46]. Assume that information ﬂow through a speciﬁc node in the proteomic network requires the phosphorylation of a known protein at a speciﬁc amino acid sequence. By measuring the proportion of those protein molecules that are phosphorylated (see Raggiaschi et al., this issue) we can infer the level of activity of that signal node. If we compare this measurement over time, or at stages of disease progression, or before and after treatment, a correlation can be made between the activity of the node and the biologic or disease state. The development of highly sensitive protein microarrays (see Maercker, this issue) now make it possible to proﬁle the states of protein signal pathways in tissue biopsies, aspirates or body ﬂuid samples The application of this technology to clinical molecular diagnostics is greatly enhanced by increasing numbers of high quality antibodies that are speciﬁc for the modiﬁcation or activation state of target proteins within key pathways. Antibody speciﬁcity is particularly critical given the complex array of biological proteins at vastly different concentrations contained in cell lysates. Given that there are no standard PCR-like direct ampliﬁcation methods for proteins, the sensitivity of antibodies must be achieved in near femtomolar range. Moreover, the labeling and ampliﬁcation method must be linear and reproducible. A cubic centimeter of biopsy tissue may contain approximately 109 cells, while a needle biopsy or cell aspirate may contain less than 100,000 cells. If the cell population of the specimen is heterogeneous the ﬁnal number of actual tumor cells microdissected or procured for analysis may be as low as a few thousand. Assuming that the proteins of interest, and their phosphorylated counterparts, exist in low abundance, the total concentration of analyte proteins in the sample will be very low. Newer generations of protein microarrays combined with highly sensitive and speciﬁc validated antibodies are now able to achieve adequate levels of sensitivity for analysis of clinical specimens. For analysis of clinical patient specimens, we have found reverse phase arrays (RPAs, also known as tissue lysate arrays) to have many advantages over forward phase arrays. In the RPA system (Fig. 2a), individual patient cellular lysates are immobilized on the array. Each array can contain many patient samples, which are incubated with one antibody. The antibody levels are measured and directly compared across many samples. For this reason, RPAs do not require direct labeling of the patient proteins and do not utilize a two-site antibody sandwich. Hence, there is no experimental variability introduced due to labeling yield, eﬃciency, or epitope masking. As each array is comprised of dozens of patient samples, subtle diﬀerences in a target protein can be measured because each sample is exposed for the same amount of time to the same concentration of primary and secondary antibody and ampliﬁcation reagents. Additionally, each patient sample can be applied in a miniature dilution curve on the RPA array. This provides an excellent means of matching the antibody concentration with the target protein concentration so that the linear range of detection is insured to exist on at least one or more diluted spots. The high sensitivity of RPAs is in part because the antibody can be tagged and the signal ampliﬁed independent from the immobilized patient sample. For example, coupling the detection antibody with highly sensitive tyramide based avidin/biotin signal ampliﬁcation systems can yield detection sensitivities down to fewer than 1000–5000
Fig. 2. Protein microarray and bioinformatic analysis of patient tissue samples. (a) Reverse phase protein microarrays. Reverse phase microarrays immobilize patient tissue lysates to a substratum such as a nitrocellulose coated glass slide. An antibody or analyte speciﬁc ligand is applied in solution phase. Bound antibodies are detected by secondary tagging and signal ampliﬁcation using standard methods. For high throughput analysis of clinical samples, reverse phase protein microarrays have multiple advantages over forward phase arrays that immobilize antibodies to a substratum sandwiching the test sample between the antibody and the secondary labeled antibody. (b) Scanning and bioinformatic analysis of protein microarray data. Pure populations of diseased or normal cells from patient biopsy specimens are obtained via microdissection, FACS or other methodology. Lysates are arrayed onto nitrocellulose covered glass slides in a miniature dilution curve, which insures subsequent detection within the linear range of the detection system. Multiple patient lysates are contained on one array. Each array is assayed with a speciﬁc antibody. Multiple arrays may be assayed using standard high throughput standard machines currently found in most immunohistochemistry labs. Assayed arrays are scanned using a ﬂexible open source image quantiﬁcation program. Spot intensities are calculated and normalized. The data output is suitable for analysis by traditional supervised and unsupervised computer software learning systems using powerful Bayesian clustering analysis with generation of traditional ‘‘heat maps’’.
molecules/spot. RPAs have been successfully applied to analyze the state of mediators of apoptosis [47,48] and mitogenesis pathways within microdissected premalignant lesions, compared to adjacent normal epithelium, invasive carcinoma, and host stroma [49,50]. RPAs are particularly well suited to the mapping of signal transduction pathways in cancer cell lines  and patient specimens [52,53]. A variety of methods have been used to analyze data obtained from protein microarrays [54–58]. These methods have been primarily adopted from those used in gene microarray analysis. The analysis RPAs presents a new set of challenges, compared with conventional spotted arrays. Multiple RPAs, each analyzing a different phosphorylated protein, are scanned and the images are analyzed by software
Calvo, Liotta, and Petricoin
programs speciﬁcally designed for the RPA format. Spot intensities are calculated and normalized, and the dilution curve is collapsed to a single intensity value (Fig. 2b). This value is then assigned a relative normalized intensity value referenced to the other patient samples on the array. The data output is in a spreadsheet format suitable for analysis by traditional unsupervised and supervised bioinformatics computer systems. Ultimately, protein array data is displayed as traditional ‘‘heat maps’’ which can be analyzed by Bayesian clustering methods for signal pathway proﬁling. INDIVIDUALIZED THERAPY: INTEGRATION OF GENOMICS AND PROTEOMICS WITH TRADITIONAL MEDICINE Cancer progression is characterized by the accumulation of multiple genetic mutations or epigenetic events that cooperate to drive malignancy. This orchestration occurs via protein alterations along multiple pathways that together drive proliferation, block differentiation, or inhibit apoptosis conferring ‘‘immortality’’ (Fig. 3a). Each of these cellular processes is regulated by complex protein networks with multiple interconnected nodes of activity. Theoretically aberrant protein function at any key node can impact the ﬂow of downstream information. In cancer, there could be numerous combinations of ‘‘mutated’’ or aberrant regulatory proteins in diﬀerent cooperating pathways that are suﬃcient to drive malignancy. Likewise each individual patient’s tumor might have a unique complement of pathogenic molecular derangements. It is also possible that each metastasis, originating from the same primary, may have a unique protein signaling circuitry that is in part dictated by the speciﬁc tissue microenvironment of the new organ it has seeded. Protein kinases, and phosphatases are key regulatory proteins that play a role in controlling information ﬂow between nodes in the cellular signaling circuitry ultimately regulating gene transcription. Their aberrant function is frequently central to the pathogenesis of cancer and other diseases . For example, c-Src is implicated as a cooperative partner with multiple other oncoproteins in aberrantly remodeling signaling pathways in cancer [60,61]. Constitutive activation of Ras has been associated with uncontrolled tumor growth in many cancers including pancreatic, colon and lung . Ras has also been shown to drive expression of IL-8 thereby eliciting a stromal response that fosters angiogenesis and tumor progression . Until recently, the concept of using molecularly targeted therapies has focused on the development and use of single agent inhibitors [64–67]. Imatinib (Gleevec, STI-571) is a prime example of the promise of molecularly targeted therapies . Treatment with Imatinib inhibits the abberant protein kinase Bcr-Abl in chronic myelogenous leukemia (CML) by binding to and blocking its ATP-binding domain. Imatinib has a striking ability to induce remission in CML patients even when their leukemia is resistant to traditional chemotherapy. Despite the initial success of Imatinib, after a period of remission many patients relapse with resistant tumor cells. This underscores the need to combine multiple agents that would theoretically be more eﬀective than single agents alone. Likewise there is a need for diagnostic modalities to identify multiple targets within patient tumors that would be ideal for combinatorial pharmacologic targeting.
Fig. 3. Functional proteomic proﬁling of human disease serves as the foundation for individualized combinatorial therapy of molecularly targeted inhibitors. (a) Protein microarrays map functional state of key nodes in signal transduction pathways orchestrating human disease. Protein signaling pathways consist of complex networks of regulatory proteins that become activated via post-translational modiﬁcations (e.g. phosphorylation) through protein–protein interactions. Signaling networks control basic cellular physiologic processes such as proliferation, apoptosis, and diﬀerentiation. Signaling pathways are commonly dysregulated in human disease and serve as targets of molecular inhibitors. Protein microarrays are able to map the state of signaling pathways by measuring the functional state or activity of key nodes within the molecular circuitry. Microarrays containing pure populations of diseased cells (e.g., tumor cells) are assayed for the phosphorylated form of key phosphoproteins using phosphospeciﬁc antibodies. Activity of up to 50–100 signaling nodes can be assayed from a single patient biopsy. These data are used to create a signaling circuitry map of the diseased cells identifying pathologic pathways suitable for targeting with pharmacologic inhibitors. (b) Combinatorial therapy as a model for treatment of disease using molecularly targeted inhibitors. Traditional pharmacologic inhibitors have been used to inhibit one key activated regulator within a signal transduction cascade in eﬀorts to eﬀectively shut down up to 90% of the pathway using a high dose of the agent. In a combinatorial therapy model, the same level of inhibition may be achieved using inhibitors to multiple nodes along the pathway using lower doses of pharmacologic inhibitors. Multiple nodes within multiple key pathologic pathways may be targeted with decreased side eﬀects and toxicities. Combinatorial therapy cocktails would be designed for individual patients using circuitry maps obtained through protein microarray analysis integrated with information from traditional histopathologic diagnosis and gene-based assays. Proteomic response of disease can be monitored with adjustment of combinatorial inhibitor therapy as appropriate.
At the FDA-NCI Clinical Proteomics Program we have piloted studies that prove the feasibility of using highly sensitive, speciﬁc and linearly dynamic protein microarrays to create circuitry maps of activated phosphoprotein networks within patient tumor biopsies . Until recently it was not feasible to perform detailed
Calvo, Liotta, and Petricoin
proteomic analysis of individual patient biopsy specimens due to the relatively large amount of material that is required by traditional protein-based technologies such as two-dimensional gel analysis, or Western blotting. However, highly sensitive, speciﬁc and linearly dynamic RPA microarray systems [46,69] can capture snapshots of information ﬂow between dynamic phosphoprotein signaling networks in small patient biopsies with as little as 10,000 cells. Pure populations of diseased cells and surrounding host tissue are segregated via Laser Capture Microdissection (LCM)  technologies. Mapping of cell signaling networks is achieved by using multiple validated panels of antibodies (we, with help from the laboratory of Dr. Gordon Mills at MD Anderson, have validated over 400 antibodies to date, and posted at http://home.ccr.cancer.gov/ncifdaproteomics/) against key regulatory proteins and their post-translationally modiﬁed forms (i.e., phosphorylated, cleaved, acetylated, etc.). The functional state of over 100 phosphoproteins can be assayed in a single 1 cm biopsy. These maps reveal the functional state of multiple protein ‘‘nodes’’ along multiple interconnected pathways  thereby determining the activity of information ﬂow driving aberrant gene transcription and cell function in the context of disease. Armed with this knowledge, it its theoretically possible to devise cocktails of speciﬁc inhibitors to pharmacologically target multiple nodes along pathogenic pathways in efforts to shut down aberrant signaling within an individual patient’s speciﬁc tumor or disease . Response to therapy could be monitored over time with appropriate adjustments. Microproteomic proﬁling of protein signaling circuits in diseased or malignant cells assays the functional state of known regulatory proteins and serves as a foundation for development of personalized molecularly targeted therapies [10,52]. Initial studies have shown that individual patients sharing the same type of cancer, with identical histopathologic diagnoses, have tumors that display unique in vivo proteomic signaling proﬁles with measurable signaling responses to perfused chemotherapy during surgery . Consequently, a given class of therapy might be eﬀective for only a subset of patients who harbor tumors with susceptible molecular derangements. This idea is further supported by recent studies by Irish et al.  in acute myeloid leukemia. They noted striking patterns of diﬀerences in the remodeling of phosphoprotein signaling networks in stimulated leukemic cells that produced patient classiﬁcations predictive of outcome. Overall, prognostic and therapeutic decisions can be informed by protein signaling network interrogation. Synergistic therapeutic eﬀects that target diﬀerent points in the signaling networks, with lower doses of individual agents, may lower the toxicities associated with treatment. The concept of using combinatorial inhibitor therapy has been explored, for example, with the EGFR tyrosine kinase inhibitor ZD1839 (Iressa) and the anti-Erb-B2 monoclonal antibody trastuzumab (Herceptin) . With this concept in mind, a redeﬁned goal of molecular proﬁling is to map the cellular circuitry so as to deﬁne the optimal set of interconnected drug targets. Drug discovery efforts are intensely focused on the development additional kinase inhibitors . Such inhibitors are necessary components in the development of combinatorial individualized therapies. In addition to kinase inhibitors, other classes of molecules may also prove to be useful targets such as (1) caspases which are modiﬁed by cleavage and are involved in apoptosis or (2) histone acetylases and
deacetylases which regulate gene transcription, (3) farnesyl transferase inhibitors of Ras proteins . At present, multiple molecules that block kinase activity are being investigated in Phase III trials, and as many as 30 kinase inhibitors are being evaluated in Phase I/II trials [75,76]. SPECTRUM OF CLINICAL CARE: THE FUTURE OF CLINICAL PROTEOMICS We can foresee a future where clinical medicine will integrate proteomics and genomics with traditional pathology-based diagnostics (Fig. 4). Treatment will transition from radiation and chemotherapy type medications that have globally undesirable side eﬀects to individualized molecularly targeted therapies. Diseases like cancer will be detected early from screening serum or urine tests using future
Fig. 4. Clinical proteomics: Model for Spectrum of Care. We envision a day when patients will be screened and diagnosed for disease using serum-based assays. Low molecular biomarkers can be harvested ex vivo or in vivo using nanotechnology-based strategies. Discovery and rigorous validation of low molecular weight biomarkers identiﬁed in patient serum, plasma or tissue could be employed to predict health or disease. Once disease is detected, follow-up with diagnostic imaging and tissue biopsy allows staging of disease with feedback to bioinformatics systems for expansion of data base sets and validation. Proteomic signal transduction circuitry mapping of pure populations of diseased cells using reverse phase protein microarrays identiﬁes aberrant protein networks driving disease. Based on this information, ‘‘designer cocktails’’ of molecular inhibitors can be prescribed to target multiple nodes along key pathways identiﬁed via microarray analysis. Molecular response to therapy can be monitored with adjustments to combinatorial therapy for maximal patient beneﬁt.
Calvo, Liotta, and Petricoin
generation mass spectrometry-based platforms and multiplexed immunoassays. Patients may ingest nanoparticles prior to blood tests that will harvest LMW biomarkers for increased sensitivity and speciﬁcity of diagnostic testing. When cancers are detected, tissue biopsies will likely continue to be analysed by traditional morpho-histologic methods. However, the pathologist of the future will also utilize genomic and proteomic technologies such as gene expression and protein microarrays to further subclassify human disease and predict outcomes. In vivo cell signaling and protein network pathway proﬁles will characterize the speciﬁc aberrant molecular circuitry of an individual patient’s disease. With this knowledge, an individualized molecular cocktail of inhibitors may be prescribed that best targets the entire disease speciﬁc protein network of the tumor. The pathologist and the diagnostic imaging physician will assist the clinical team to perform real-time in vivo assessment of therapeutic efﬁcacy and toxicity. Proteomic and genomic analysis of recurrent tumor lesions could be the basis for rational redirection of therapy because it may reveal remodeling of the diseased protein network that is associated with drug resistance. The paradigm shift will directly affect clinical practice as it impacts all of the crucial elements of patient care and management.
REFERENCES 1. Lander, E. S. et al. (2001) Initial sequencing and analysis of the human genome. Nature 409(6822):860–921. 2. Venter, J. C. et al. 3. Badano, J. L. and Katsanis, N. (2002) Beyond Mendel: an evolving view of human genetic disease transmission. Nat. Rev. Genet. 3(10):779–789. 4. Staudt, L. M. (2002) Gene expression proﬁling of lymphoid malignancies. Annu Rev Med. 53:303–318. 5. Shipp, M. A. et al. (2002) Diﬀuse large B-cell lymphoma outcome prediction by gene-expression proﬁling and supervised machine learning. Nat. Med. 8(1):68–74. 6. Dave, S. S. et al. (2004) Prediction of survival in follicular lymphoma based on molecular features of tumor-inﬁltrating immune cells. N. Engl. J. Med. 351(21):2159–2169. 7. Singh, D. et al. (2002) Gene expression correlates of clinical prostate cancer behavior. Cancer Cell. 1(2):203–209. 8. Sorlie, T. et al. (2003) Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc. Natl. Acad. Sci. USA. 100(14):8418–8423. 9. Schwartz, D. R. et al. (2002) Gene expression in ovarian cancer reﬂects both morphology and biological behavior, distinguishing clear cell from other poor-prognosis ovarian carcinomas. Cancer Res. 62(16):4722–4729. 10. Petricoin, E. F. et al. (2002) Clinical proteomics: translating benchside promise into bedside reality. Nat. Rev. Drug Discov. 1(9):683–695. 11. Petricoin, E. F. and Liotta, L. A. (2004) SELDI-TOF-based serum proteomic pattern diagnostics for early detection of cancer. Curr. Opin. Biotechnol. 15(1):24–30. 12. Rosenblatt, K. P. et al. (2004) Serum proteomics in cancer diagnosis and management. Annu. Rev. Med. 55:97–112. 13. Rapkiewicz, A. V. et al. (2004) Biomarkers of ovarian tumours. Eur. J. Cancer. 40(17):2604–2612. 14. Petricoin, E. F. et al. (2002) Use of proteomic patterns in serum to identify ovarian cancer. Lancet. 359(9306):572–577. 15. Conrads, T. P., et al. (2004) High-resolution serum proteomic features for ovarian cancer detection. Endocrine Related Cancer (in press). 16. Li, J. et al. (2002) Proteomics and bioinformatics approaches for identiﬁcation of serum biomarkers to detect breast cancer. Clin. Chem. 48(8):1296–1304.
17. Ornstein, D. K. et al. (2004) Serum proteomic proﬁling can discriminate prostate cancer from benign prostates in men with total prostate speciﬁc antigen levels between 2.5 and 15.0 Ng/Ml. J. Urol. 172(4, Part 1 of 2):1302–1305. 18. Petricoin, E. F. 3rd et al. (2002) Serum proteomic patterns for detection of prostate cancer. J. Natl. Cancer Inst. 94(20):1576–1578. 19. Petricoin, E. F., Ornstein, D. K., and Liotta, L. A. (2004) Clinical proteomics: applications for prostate cancer biomarker discovery and detection. Urol. Oncol. 22(4):322–328. 20. Prados, J. et al. (2004) Mining mass spectra for diagnosis and biomarker discovery of cerebral accidents. Proteomics. 4(8):2320–2332. 21. Clarke, W. et al. (2003) Characterization of renal allograft rejection by urinary proteomic analysis. Ann. Surg. 237(5):660–664discussion 664–665. 22. Carrette, O. et al. (2003) A panel of cerebrospinal ﬂuid potential biomarkers for the diagnosis of Alzheimer’s disease. Proteomics. 3(8):1486–1494. 23. Petricoin, E. F. et al. (2004) Lessons from Kitty Hawk: from feasibility to routine clinical use for the ﬁeld of proteomic pattern diagnostics. Proteomics. 4(8):2357–2360. 24. Wulfkuhle, J. D., Liotta, L. A., and Petricoin, E. F. (2003) Proteomic applications for the early detection of cancer. Nat. Rev. Cancer 3(4):267–275. 25. Petricoin, E. F. and Liotta, L. A. (2004) Proteomic approaches in cancer risk and response assessment. Trends Mol. Med. 10(2):59–64. 26. Liotta, L. A. and Kohn, E. C. (2001) The microenvironment of the tumour–host interface. Nature 411(6835):375–379. 27. Liotta, L. A., Ferrari, M., and Petricoin, E. (2003) Clinical proteomics: written in blood. Nature. 425(6961):905. 28. Mehta, A. I. et al. (2003) Biomarker ampliﬁcation by serum carrier protein binding. Disease Markers. 19(1):1–10. 29. Geho, D. H. et al. (2004) Opportunities for nanotechnology-based innovation in tissue proteomics. Biomed. Microdevices 6(3):231–239. 30. Nam, J. M., Thaxton, C. S., and Mirkin, C. A. (2003) Nanoparticle-based bio-bar codes for the ultrasensitive detection of proteins. Science 301:1884–1886. 31. Calvo, K. R. et al. (2004) Molecular proﬁling provides evidence of primary mediastinal large B-cell lymphoma as a distinct entity related to classic Hodgkin lymphoma: implications for mediastinal gray zone lymphomas as an intermediate form of B-cell lymphoma. Adv. Anat. Pathol. 11(5):227– 238. 32. Petricoin, E. F. and Liotta, L. A. (2003) Clinical applications of proteomics. J. Nutr. 133(7 Suppl):2476S–2484S. 33. Liotta, L. A., Kohn, E. C., and Petricoin, E. F. (2001) Clinical proteomics: personalized molecular medicine. JAMA 286(18):2211–2214. 34. Moolgavkar, S. H. and Knudson, A. G. Jr. (1981) Mutation and cancer: a model for human carcinogenesis. J. Natl. Cancer Inst. 66(6):1037–1052. 35. Hanahan, D. and Weinberg, R. A. (2000) The hallmarks of cancer. Cell. 100(1):57–70. 36. Emmert-Buck, M. R. et al. (1996) Laser capture microdissection. Science 274(5289):998–1001. 37. Fuller, A. P. et al. (2003) Laser capture microdissection and advanced molecular analysis of human breast cancer. J. Mammary Gland Biol. Neoplasia 8(3):335–345. 38. Ma, X. J. et al. (2003) Gene expression proﬁles of human breast cancer progression. Proc. Natl. Acad. Sci. USA 100(10):5974–5979. 39. Paweletz, C. P., Liotta, L. A., and Petricoin, E. F. 3rd (2001) New technologies for biomarker analysis of prostate cancer progression: laser capture microdissection and tissue proteomics. Urology 57(4 Suppl 1):160–163. 40. Geho, D. H., Petricoin, E. F., and Liotta, L. A. (2004) Blasting into the microworld of tissue proteomics: a new window on cancer. Clin. Cancer Res. 10(3):825–827. 41. Paweletz, C. P. et al. (2000) Rapid protein display proﬁling of cancer progression directly from human tissue using a protein biochip. Drug Develop. Res. 49:34–42. 42. Yanagisawa, K. et al. (2003) Proteomic patterns of tumour subsets in non-small-cell lung cancer. Lancet 362(9382):433–439. 43. Yanagisawa, K. et al. (2003) Molecular ﬁngerprinting in human lung cancer. Clin. Lung Cancer 5(2):113–118.
Calvo, Liotta, and Petricoin
44. Schwartz, S. A. et al. (2004) Protein proﬁling in brain tumors using mass spectrometry: feasibility of a new technique for the analysis of protein expression. Clin. Cancer Res. 10(3):981–987. 45. Espina, V. et al. (2003) Protein microarrays: molecular proﬁling technologies for clinical specimens. Proteomics 3(11):2091–2100. 46. Liotta, L. A. et al. (2003) Protein microarrays: meeting analytical challenges for clinical applications. Cancer Cell. 3(4):317–325. 47. Paweletz, C. P. et al. (2001) Reverse phase protein microarrays which capture disease progression show activation of pro-survival pathways at the cancer invasion front. Oncogene 20(16):1981–1989. 48. Herrmann, P. C. et al. (2003) Mitochondrial proteome: altered cytochrome c oxidase subunit levels in prostate cancer. Proteomics 3(9):1801–1810. 49. Grubb, R. L. et al. (2003) Signal pathway proﬁling of prostate cancer using reverse phase protein arrays. Proteomics 3(11):2142–2146. 50. Wulfkuhle, J. D. et al. (2003) Signal pathway proﬁling of ovarian cancer from human tissue specimens using reverse-phase protein microarrays. Proteomics 3(11):2085–2090. 51. Nishizuka, S. et al. (2003) Proteomic proﬁling of the NCI-60 cancer cell lines using new high-density reverse-phase lysate microarrays. Proc. Natl. Acad. Sci. USA 100(24):14229–14234. 52. Petricoin, E. F., et al. (2004) Mapping molecular networks using proteomics: a vision for patienttailored combination therapy (in submission). 53. Calvo, K. R., et al. (2004) Clinical proteomics: in vivo molecular signaling proﬁles of human tumors, pre and post tumor perfusion with experimental chemotherapy (in preparation). 54. Brazma, A. et al. (2001) Minimum information about a microarray experiment (MIAME)-toward standards for microarray data. Nat. Genet. 29(4):365–371. 55. Carlisle, A. J. et al. (2000) Development of a prostate cDNA microarray and statistical gene expression analysis package. Mol. Carcinogen. 28(1):12–22. 56. Cutler, P. (2003) Protein arrays: the current state-of-the-art. Proteomics 3(1):3–18. 57. Sreekumar, A. et al. (2001) Proﬁling of cancer cells using protein microarrays: discovery of novel radiation-regulated proteins. Cancer Res. 61(20):7585–7593. 58. Miller, J. C. et al. (2003) Antibody microarray proﬁling of human prostate cancer sera: antibody screening and identiﬁcation of potential biomarkers. Proteomics 3(1):56–63. 59. Blume-Jensen, P., and Hunter, T. (2001) Oncogenic kinase signalling. Nature 411(6835):355–365. 60. Ishizawar, R. and Parsons, S. J. (2004) c-Src and cooperating partners in human cancer. Cancer Cell. 6(3):209–214. 61. Kloth, M. T. et al. (2003) STAT5b, a mediator of synergism between c-Src and the epidermal growth factor receptor. J. Biol. Chem. 278(3):1671–1679. 62. Bos, J. L. (1989) Ras oncogenes in human cancer: a review. Cancer Res. 49(17):4682–4689. 63. Sparmann, A. and Bar-Sagi, D. (2004) Ras-induced interleukin-8 expression plays a critical role in tumor growth and angiogenesis. Cancer Cell. 6(5):447–458. 64. Jones, A. L. and Leyland-Jones, B. (2004) Optimizing treatment of HER2-positive metastatic breast cancer. Semin. Oncol. 31(5 Suppl 10):29–34. 65. Leyland-Jones, B. (2002) Trastuzumab: hopes and realities. Lancet Oncol. 3(3):137–144. 66. Sebolt-Leopold, J. S. (2004) MEK inhibitors: a therapeutic approach to targeting the Ras-MAP kinase pathway in tumors. Curr. Pharm. Des. 10(16):1907–1914. 67. Traxler, P. (2003) Tyrosine kinases as targets in cancer therapy—successes and failures. Expert Opin. Ther. Targets 7(2):215–234. 68. Druker, B. J. (2004) Imatinib as a paradigm of targeted therapies. Adv. Cancer Res. 91:1–30. 69. Espina, V. et al. (2004) Use of proteomic analysis to monitor responses to biological therapies. Expert Opin. Biol. Ther. 4(1):83–93. 70. Petricoin, E. et al. (2004) Clinical proteomics: revolutionizing disease detection and patient tailoring therapy. J. Proteome Res. 3(2):209–217. 71. Irish, J. M. et al. (2004) Single cell proﬁling of potentiated phospho-protein networks in cancer cells. Cell. 118(2):217–228. 72. Normanno, N. et al. (2002) Cooperative inhibitory eﬀect of ZD1839 (Iressa) in combination with trastuzumab (Herceptin) on human breast cancer cell growth. Ann. Oncol. 13(1):65–72. 73. Tibes, R., Trent J., and Kurzrock R. (2004) Tyrosine kinase inhibitors and the dawn of molecular cancer therapeutics. Annu. Rev. Pharmacol. Toxicol.
74. Crul, M. et al. (2001) Ras biochemistry and farnesyl transferase inhibitors: a literature survey. Anticancer Drugs. 12(3):163–184. 75. Traxler, P. et al. (2001) Tyrosine kinase inhibitors: from rational design to clinical trials. Med. Res. Rev. 21(6):499–512. 76. Zwick, E., Bange, J., and Ullrich, A. (2002) Receptor tyrosine kinases as targets for anticancer drugs. Trends Mol. Med. 8(1):17–23.