Proteomics to Diagnose Human Tumors and Provide Prognostic Information

Completion of the human genome project and the development of high-throughput gene expression analysis has ushered in the era of "omics," bringing the promise of "molecular medicine" closer to reality. Genomics refers to the study of the human genome (ie, DNA sequences), and functional genomics is the study of gene expression (ie, messenger RNA [mRNA] levels). Because of the availability of robust technologies for DNA and mRNA analysis, most translational research studies of human cancer have focused on these two information reservoirs. However, although these studies can provide valuable information, adapting their use to clinical practice has not been easy due to biologic and technologic limitations. Proteomics is the characterization of biologic processes by quantitative and qualitative assessment of protein expression patterns. Unlike genomic studies, proteomics provides a dynamic picture of normal and abnormal cellular physiology. Although mRNA expression studies also provide dynamic information, they provide an indirect measure of protein expression (as mRNA merely directs protein expression). Proteins are responsible for all controlled biologic functions and are the true determinants of the malignant phenotype. Because glycosylation, phosphorylation, cellular trafficking, and degradation can affect protein function, important information may be missed by mRNA expression studies. Proteomic investigations can detect these posttranslation modifications and will provide additional information that is complementary to studies of gene expression. In light of the fact that virtually all US Food and Drug Administration- approved diagnostic/prognostic tests and cancer therapies are protein-based, findings from proteomic studies will be easier to develop into clinically useful tools than are findings from genetic- based studies. Early proteomic studies tended to detect changes in highly abundant proteins, but as technology has progressed, the ability to detect and quantify changes in less abundant proteins has become easier. Typically, the cancerous tissue is the most fertile source of relevant molecular information. For many human cancers, however, adequate tumor material is unobtainable without invasive procedures, and thus, is not available for molecular analysis. Molecular events within cancerous tissue (even in the case of a solid tumor) may be reflected by changes in the proteome of circulating body fluids, and analysis of these body fluids may provide important clinical information. The most appropriate tissue or body fluid source to study depends on the type of cancer and on the defined purpose of the investigation. Traditionally, the goal of most biomarker-based proteomic discovery efforts is to identify proteins that change as a cause or consequence of the disease process and then to develop enzyme-linked immunosorbent assay-based clinical tests to directly measure the analytes in question. Recently, however, efforts have focused on the development of proteomic technologies such as mass spectroscopy and protein arrays as the clinical test platform of choice.[1,2] Improvements in proteomic and bioinformatics technologies provide translational researchers with new opportunities to develop better methods of diagnosing and determining prognosis for human cancer. These new technologies should improve the health of cancer patients and people at risk for developing cancer. This article reviews these technologic advances and their potential clinical applications. Laser Capture Microdissection Established cultured cell lines are commonly used for molecular studies of human cancer. Although these studies can provide valuable information regarding cancer biology, some of the findings may not be applicable to patients because the behavior of cancer cells grown in culture may not be representative of the in vivo situation. In fact, studies comparing protein expression patterns of prostate and bladder cancer from tumor tissue and patient-matched cultured cells have demonstrated profound differences.[3,4] One of the major limitations of the direct study of molecular changes in clinical specimens has been the difficulty of separating malignant cells of interest from stroma, inflammatory cells, and benign epithelial cells. To overcome this investigative hurdle, various microdissection techniques have been developed for procuring pure populations of cells from human tissue sections. Laser capture microdissection is a relatively new technique that allows researchers to visualize a tissue section via light microscopy and procure the desired cells by activating a 7.5- to 30-μm diameter infrared laser beam to "weld" the tissue to a plastic cap. Intact DNA, RNA, and protein can then be extracted from the "welded" tissue and analyzed by conventional methods.[ 5,6] Binding properties of proteins are preserved, and laser capture microdissection can be used to study differences in protein-protein interaction within different tissue types. For example, prostate-specific antigen (PSA) recovered from cells procured via this technique retains the ability to bind inhibitors such as alpha- 1-antichymotrypsin. Studies utilizing laser capture microdissection have demonstrated that PSA exists as an unbound enzyme in both benign and malignant prostate epithelium and that this "free" form of PSA can bind to alpha-1-antichymotrypsin in either setting.[7] Discovery-Based Proteomics Two-Dimensional Gel Electrophoresis
Traditional proteomic studies have relied on multiplexed two-dimensional polyacrylamide gel electrophoresis (2D-PAGE) to compare protein expression patterns from different tissues or cell lines. The first dimension separates proteins by pH (isoelectric focusing), and the second dimension, by molecular weight (SDS-PAGE). Although 2D-PAGE has been available for several decades, advances in this technology have dramatically improved its sensitivity, spot resolution, and reproducibility. The use of fluorescent- based dyes (such as SYPRO Red) has improved the dynamic range and sensitivity of protein detection, while the development of immobilized pH gradients and image analysis software has improved the reproducibility of such tests. The primary use of 2D-PAGE is to facilitate identification of differentially expressed proteins or protein isoforms. Protein identification can be accomplished by direct sequencing or by comparing spot patterns to "standard" gels in which all spots have been microsequenced. Improvements in matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry and nanoelectrospray technology along with the availability of searchable protein sequence databases have greatly enhanced protein identification. With current technology, a spot on a 2D gel containing several hundred femtomoles of protein can be identified and fully characterized.[8-10] Multiple investigators have successfully used 2D-PAGE to identify protein expression changes associated with a wide variety of human cancers.[ 3,11-16] For example, 2D-PAGE analysis of procured patient-matched benign and cancerous cells showed that annexin I was downregulated in both prostate and esophageal cancers (Figure 1).[11,17] These findings were subsequently confirmed by immunohistochemical studies of large patient study sets.[18,19] Another study utilizing 2D-PAGE found that Rho Gprotein dissociation inhibitor and glyoxalase I are overexpressed in invasive ovarian cancers as compared to low-malignant-potential ovarian tumors.[20] Despite the tremendous utility of standard 2D-PAGE, this technology has significant limitations. The primary limitations are related to sensitivity of detection and spot separation. In particular, extremely basic or acidic proteins as well as low-molecularweight proteins are poorly separated with standard 2D-PAGE protocols. Therefore, the test can only survey a fraction of the cellular proteome and is most useful for analysis of abundant proteins larger than 10 kd. Because most secreted proteins fall into this category, 2D-PAGE is a powerful tool with which to search for clinically useful biomarkers. Differential In-Gel Electrophoresis
Differential in-gel electrophoresis (DIGE) is an emerging technology[21] that compares protein expression patterns by labeling protein samples with unique fluorescent dyes (ie, Cy2, Cy3, and Cy5) and then separating them on a single 2D-PAGE gel. This allows the simultaneous comparison of two to three protein samples and provides a relative quantitative assessment of protein expression levels. The primary advantage of DIGE is that it eliminates intrinsic gel-to-gel variability, which can compromise comparative studies. DIGE has been used successfully to identify the changes in protein expression associated with esophageal cancer.[22] In this study, protein lysates from normal and cancer esophageal cells were labeled with Cy3 and Cy5 fluorescent dyes, respectively, and separated by 2D-PAGE. Of more than 1,000 spots identified in both samples, 58 were found to be upregulated more than threefold, and 107 were downregulated more than threefold. One of the downregulated proteins was identified by capillary highperformance liquid chromatography/ tandem mass spectrometry to be annexin I, and one of the upregulated proteins was found to be tumor rejection antigen (gp96). DIGE is more sensitive than Coomassie Blue staining but detects 40% fewer spots than SYPRO Ruby dye. This reduced sensitivity compared to other fluorescent-based stains is due to the requirement that only 1% to 2% of lysine(Drug information on lysine) amino acid residues that form each protein be fluorescently labeled in order to maintain protein solubility. Currently, DIGE is only useful for the analysis of relatively abundant proteins; however, in the near future, improvements will likely lead to enhanced sensitivity and broader application of this technology. DIGE is a robust method of making quantitative comparisons of global protein expression levels among different tissue types, and thus a powerful tool with which to search for diagnostic and prognostic biomarkers. Isotope-Coded Affinity Tagging
Isotope-coded affinity tagging (ICAT) distinguishes two populations of proteins by labeling each with different isotope tags-a light reagent derived from eight hydrogen atoms or a heavy reagent derived from eight deuterium atoms.[23] These tags are linked to a chemical agent that specifically binds the thiol group of cysteine(Drug information on cysteine) residues in proteins and peptides. Following labeling, protein mixtures are subjected to proteolytic cleavage and fractionated by affinity chromatography. The relative amount and identity of each protein is revealed by mass spectroscopy. Qualitative information, based on the relative ratio of isotopic molecular mass peaks that differ by 8 Da (the mass difference between the light and heavy reagent), is ascertained by nanoscale liquid chromatography/ electrospray ionization mass spectroscopy. This technology is particularly useful in the analysis of membrane and hydrophobic proteins that can be difficult to dissolve and separate by 2DPAGE.[ 24] ICAT also facilitates the analysis of low-molecular-weight proteins and peptide fragments. Compared to 2D-PAGE (particularly with the application of DIGE), ICAT is less quantitative. Inaccuracies in relative quantitative assessment can result from differential fragmentation of the light and heavy tags, which can alter elution times and subsequent ionization. Moreover, standard ICAT technology does not analyze all proteins, because first-generation tags only label proteins with cysteine residues flanked by appropriately spaced protease cleavage sites. Despite these limitations, technologic improvements may well enhance the utility of ICAT as a tool for biomarker discovery. One potentially useful strategy is to label proteins with different ICAT reagents and then separate them with 2D-PAGE.[25] Because the most advantageous aspect of ICAT is its independence from gel-based separation, improvements in ICAT reagents and labeling protocols that facilitate uniform and efficient labeling of all proteins holds the greatest promise for improving the utility of this technology. Clinical Proteomics Tissue Microarrays
A major obstacle in translating findings from biomarker discovery studies into clinical practice is reliable validation of initial findings in large clinical data sets. Traditionally, validation studies have relied on immunohistochemical analysis of tissue slides from individual patients and protein quantification by visual scoring. Because substantial variation in staining occurs from tissue slide to tissue slide (and visual scoring can only provide semiquantitative information), it may not be valid to generate data by standard immunohistochemical studies. Tissue microarray has been developed to facilitate high-throughput immunohistochemistry and reduce experimental variability.[26] Tissue microarrays are constructed by incorporating multiple (0.6 mm wide * 3-4 mm high) tissue cores onto a single paraffin(Drug information on paraffin) block. From this block, 5-μm sections are cut onto a glass slide and analyzed by standard immunohistochemistry. This approach facilitates the simultaneous analysis of as many as 1,000 clinical samples including many different stages and grades of cancer. Tissue microarrays have been particularly useful in validating and determining the clinical significance of findings from cDNA microarray studies. For example, researchers used cDNA microarrays to discover that EZH2 was overexpressed in highgrade prostate cancers.[27] Immunohistochemical studies utilizing tissue microarrays confirmed that EZH2 was commonly overexpressed in highgrade prostate cancers and that this finding predicted increased risk for failure of local therapy.[28] Researchers have developed techniques to quantitatively assess protein expression levels, such as digital image analysis. These methods involve staining tissue sections with standard immunohistochemical protocols and then measuring the level of peroxidase or fluorescent staining with an optical scanner.[29] Digital image analysis was recently used to demonstrate that androgen-receptor protein expression was 81% higher in black American men with prostate cancer than in white American men.[30] Combining digital image analysis with tissue microarrays is a powerful strategy for validating the clinical utility of previously identified diagnostic and prognostic biomarkers. Protein Lysate Arrays
Another new technology that can not only facilitate clinical biomarker validation but, importantly, can be used to quantify changes in cellular signaling processes from extremely small cellular samples is reverse-phase protein arrays (ie, protein lysate arrays).[ 31,32] This technology involves arraying protein lysates from several hundred clinical samples in serial dilutions on a single nitrocellulose membrane (Figure 2). For many tissue samples, laser capture microdissection is required to procure a pure population of the cells of interest, which can be obtained even from a biopsy specimen. Protein expression levels are measured with standard antibody staining protocols and optical scanning.[32] This technology was used to demonstrate that progression from benign prostatic epithelium to invasive prostate cancer was associated with increased phosphorylation of AKT, suppression of apoptosis, and decreased phosphorylation of ERK. Compared to tissue microarrays, reverse-phase arrays are much more sensitive, can analyze dozens of end points from only a few thousand cells, are nonsubjective (thereby providing more accurate quantitative information), and are not affected by antigen retrieval issues. The main disadvantage of this technology is the frequent need for laser capture microdissection, which in some cases can be technically challenging and labor intensive, although newer automated laser capture technology is eliminating this roadblock. Work is under way to automate the entire process so that protein lysate arrays will likely become a highly useful tool not only for validation studies, but also for clinical decision-making. This technology is especially germane to patient-tailored therapy when analysis of cellular signaling pathways and posttranslational modifications is required. Protein array technology may be more adaptable to clinical practice than are tissue microarray methods, because it is less subjective and more reproducible. Proteomic Pattern Analysis
An emerging body of data suggests that for most cancers, the assessment of a pattern of multiple biomarkers provides more robust diagnostic and prognostic information than the measurement of a single biomarker. Advances in proteomic technologies have made it possible to rapidly assess complex protein expression patterns in a large number of clinical samples. MALDI-TOF and surfaceenhanced laser desorption ionization time-of-flight (SELDI-TOF) mass spectrometry are new technologies that can profile low-molecular-weight proteins.[33-36]

• SELDI-TOF-This proprietary technology utilizes a ProteinChip System and ProteinChip Reader (Ciphergen Biosystems, Fremont, Calif) to facilitate protein capture, purification, and analysis on a single platform (Figure 3). It produces crude but rapid protein purification and signal amplification and is a potentially valuable cancer biomarker screening tool because it rapidly generates a reproducible low-molecular-weight protein fingerprint from a minuscule sample (ie, 1 μL). SELDI-TOF mass spectrometry can accomplish high-throughput protein expression profiling from human tissue and body fluids. It has been shown to identify protein signatures from nipple aspirates that discriminate women with breast cancer from healthy women.[37] It has also been used to analyze protein expression patterns from pure populations of human cells procured by laser capture microdissection. With SEDLI-TOF, unique protein fingerprints characteristic of benign prostatic epithelium, high-grade prostatic intraepithelial neoplasia, and prostate cancer have been identified.[38]
• Pattern Recognition Algorithms- Because of its ability to rapidly analyze a large number of samples, SELDI-TOF is particularly well suited to generate informative proteomic patterns from serum. Visual analysis only detects gross changes in protein expression, but bioinformatics tools detect subtle differences in patterns of protein expression. Importantly, because of the huge dimensionality of the data, advanced pattern recognition algorithms are required to find the hidden, nonapparent signatures in a background of noise and chaos. Bioinformatics tools that utilize artificial intelligence-based pattern recognition algorithms can facilitate analysis of complex data sets. An analytic bioinformatics tool recently developed to analyze SELDITOF data streams-Proteome Quest beta version 1.0 (Correlogic Systems Inc, Bethesda, Md)-combines elements of genetic algorithms and selforganized cluster analysis. Proteomic data sets or spectra composed of 15,200 mass/charge ratio (m/z) values on the x-axis, with the corresponding amplitude on the y-axis are generated by this technique and imported into the genetic algorithm as an ASCII file. The genetic algorithm functions in a manner similar to natural selection, determining the subset of amplitudes at defined m/z values that best separates a "training" data set into predetermined groups. In other words, the genetic algorithm randomly analyzes multiple pattern combinations until one that discriminates the two groups of interest is found. This pattern is then recombined ("mated") with additional data. Nondiscriminatory patterns are discarded, and discriminatory ones further refined.
Once this fitness test has been successfully applied to all of the "training" data, the resultant set of y-axis- defined amplitudes that fully discriminate the training set is determined. Spectra are generated by SELDI-TOF from a set of "blinded" samples. These data are compared for their similarity to the previously defined patterns generated with the "training" set. A decision is then made that classifies the unknown samples either into one of the previously defined groups or into an "unclassified" group. As more data are input, existing clusters are refined and new clusters formed. Thereby, the genetic algorithm "learns" by experience, and in theory, will become more accurate over time.
• Clinical Utility-Artificial intelligence- based pattern recognition of serum proteomic profiles has been applied to the detection of ovarian and prostate cancer. Using this approach, a diagnostic algorithm was generated that yielded an overall positive predictive value of 94% for the diagnosis of ovarian cancer, and all 18 women with stage I ovarian cancer were correctly classified by the algorithm.[ 39] Although these preliminary studies have generated highly promising data and demonstrated the feasibility of a new diagnostic paradigm, the introduction of serum proteomic pattern diagnostics into clinical practice will be hindered by machine-tomachine, day-to-day, and platformto- platform variations, which may limit the ability to generate reproducible data streams. This problem is compounded by other factors: Human disease arises from a heterogeneous population, the disease process itself is multifactorial and heterogeneous, and clinics vary in their sample collection methodology.
A major limitation for clinical implementation may be the mass spectrometer platforms themselves. The use of high-end mass spectrometers is now being explored as a possible solution to the problems of reproducibility. The QSTAR Pulsar LC/MS/MS System (Applied Biosystems Inc, Foster City, Calif) is a high-performance hybrid quadrupole time-of-flight mass spectrometer that can analyze protein samples applied to Ciphergen's ProteinChip Arrays. The QSTAR has higher resolution and can generate far more data points than the ProteinChip Biology System II (PBS II) instrument, and most importantly, the increase in mass accuracy reduces machine-tomachine differences in mass drift. Moreover, because the source is uncoupled from the mass analyzed, this type of machine generates much truer time of flight and far less laserinduced fragmentation than a linear instrument (which means fewer confounding peaks that are not related to the disease but artificially induced by the process itself). Unlike the PBS II, the QSTAR can accomplish direct tandem mass spectometry protein identification. Because of these differences, it is likely that proteomic patterns generated from the QSTAR will be more robust than those generated by the PBS II, and indeed, this was found to be true. In a recent report, patterns were discovered that identified 100% of ovarian cancers, including all stage I cases and 63 of 66 cases of nonmalignant disease.[40] Based on these results, the investigators have extended this paradigm to more advanced highresolution instrumentation for upcoming National Cancer Institute/Center for Cancer Research-based clinical trials of ovarian cancer detection. This concept is not limited to just one type of cancer. Researcher found an algorithm for prostate cancer that yielded a positive predictive value of 41%. The genetic algorithm correctly identified 36 of 38 men with prostate cancer (ie, 95% sensitivity) and 177 of 228 men with benign biopsies (ie, 76% specificity). Among men with total PSA levels between 4.0 and 10.0 ng/mL, 97 of 137 (71%) were correctly classified as having benign prostates. Thus, if serum proteomic analysis had been used to determine the need for prostate biopsy, 70% of "unnecessary" biopsies could have been prevented, whereas only 5% of cancers would have been missed. Importantly, the genetic algorithm "correctly" classified all the men with prostate cancer. In addition, serum samples from seven men with moderate-grade, organ-confined prostate cancer were obtained prior to radical prostatectomy and 6 weeks postoperatively.[41] Another analytic strategy utilizes a decision-tree algorithm that relies on binomial decisions based on heights of a predefined set of specific protein peaks. Using this approach in a blinded test set of 60 men (30 with prostate cancer and 30 with benign prostates) yielded a sensitivity of 83% and a specificity of 97%.[42] Conclusions and Future Directions Proteomics is a multifaceted discipline that encompasses biomarker discovery to clinical diagnostics. Technologic advances have greatly improved 2D-PAGE technologies as well as non-gel-based protein-profiling strategies and have facilitated the discovery of protein changes associated with malignant transformation and progression. The development of tissue and protein arrays has provided high-throughput quantitative tools with which to validate the clinical utility of biomarkers identified through discovery-based studies. Protein microarray technology is envisioned as a means of profiling the changing state of cellular circuitry-before, during, and after therapy-to monitor multiple protein phosphorylation events at once. This type of technology could have immediate impact and utility in the new era of targeted molecular medicine. Advances in mass spectroscopy have made it possible to rapidly generate complex proteomic profiles from serum, and powerful bioinformatics tools have been developed to analyze these extremely complex data sets. The technology has advanced to the point that proteomic studies will likely have a major impact on how cancer patients are diagnosed and treated. In the shortterm, proteomics will enhance the discovery of highly predictive biomarkers to help clinicians diagnose cancer while it is still at a curable stage and determine the most appropriate therapy for any given patient. Ultimately, however, the proteomic analysis itself may become the diagnostic or prognostic test. This approach will likely provide the most accurate diagnostic and prognostic information for a given patient. Proteomic technologies have advanced to the point of making molecular diagnostics and tailored therapies possible. It is now incumbent upon clinicians and translational scientists to make the promise of "molecular medicine" a reality.