Clinical Appli

Clinica Chimica Acta 357 (2005) 151 – 158 www.elsevier.com/locate/clinchim

The clinical application of proteomics David A. Colantonio, Daniel W. Chan* Department of Pathology, Clinical Chemistry Division, Johns Hopkins Medical Institutions, 600 North Wolfe Street/Meyer B-121, Baltimore, MD 21287-7065, United States Received 4 March 2005; accepted 9 March 2005 Available online 17 May 2005

Abstract Background: Proteomics is defined as a scientific approach used to elucidate all protein species within a cell or tissue, and many researchers are taking advantage of proteomic technology to elucidate protein changes between healthy and diseased states. Methods: The application of proteomic techniques and strategies to the field of medicine is slowly transforming the way biomarker discovery is conducted. However, the complexity of serum is the source of both its promise to clinical applications and its challenge to proteomic analysis. Like any new technology when it is first introduced, proteomics has been touted with much hope and promise. Results and conclusions: We provide a review of the clinical application of proteomics with the emphasis on current practical issues and challenges facing proteomic research. D 2005 Elsevier B.V. All rights reserved. Keywords: Proteomics; Clinical proteomics; Biomarkers; Mass spectrometry; SELDI

1. General introduction Proteomics is defined as a scientific approach used to elucidate all protein species within a cell or tissue. It has been just over a decade since the term bproteomicsQ was coined [1], and even though this new term was used to introduce the concept of exploring changes in all proteins expressed by a genome, the

* Corresponding author. Tel.: +1 410 955 2674; fax: +1 410 955 0767. E-mail address: [email protected] (D.W. Chan). 0009-8981/$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.cccn.2005.03.020

tools used in proteomic analysis are borrowed from traditional biochemistry. Although this approach is commonly used to characterize all protein species within a given cell, many researchers are also taking advantage of proteomic technology to elucidate protein changes between healthy and diseased states. Proteomics, like any new technology when it is first introduced, has been touted with much hope and promise. As with all new technologies, time is required to work through all the glitches. This article provides a review of the clinical application of proteomics with emphasis on current practical issues in proteomics research.

152

D.A. Colantonio, D.W. Chan / Clinica Chimica Acta 357 (2005) 151–158

2. Proteomics technologies Like any area of research, careful attention must be given to proper experimental planning and design so that meaningful results can be obtained. Furthermore, an understanding of the strengths and limitations of the common tools used in proteomic research is important. Consideration should be given to the following steps in the proteomic approach: (i) sample preparation, (ii) protein separation, and (iii) identification of proteins and/or their post-translation modifications. These steps, as they are commonly employed in proteomic studies, are briefly discussed. Sample preparation involves efficient and effective solubilization of proteins from biological samples, whether tissue or blood. Since proteins are not a homogeneous entity, having different biophysical and chemical properties, it is important that the solubilization process be optimized for each sample type [2]. Protein/protein, protein/lipid, and protein/nucleotide interactions must be disrupted without inducing protein modifications resulting from sample manipulation. The appropriate solubilization conditions for specific classes of proteins, sharing similar biochemical characteristics, must be determined on a case-bycase basis. Detergents, chaotropic, and reducing agents are often used in sample preparations. In order to minimize introduction of non-experimental protein modifications, protease, kinase, and phosphatase inhibitors could be included in the buffer used for sample preparation. bSub-proteomicQ analysis refers to the proteomic analysis of a subset of proteins, sharing specific characteristics, which have been isolated from a complex mixture of proteins [2]. The sub-proteomic approach relies on enrichment techniques for isolation of proteins with similar characteristics or biophysical and chemical properties. Proteins may be classified according to various schemes such as (i) biochemical characteristics (e.g., isoelectric point, molecular weight, and hydrophobicity); (ii) cellular compartment (e.g., mitochondria, sarcoplasmic reticulum); or (iii) function. Sub-proteomic analysis allows for a detailed investigation of a specific cellular biochemical system relevant to a particular research question. Consolidation of data from separate proteomic analysis of distinct protein subsets should provide a more comprehensive view of the proteome as compared to

investigation of a heterogeneous protein mixture generated from a generic solubilization condition. The next step following sample preparation is separation of proteins. The most common method for separating complex mixtures of proteins is twodimensional sodium dodecyl sulphate polyacrylamide gel electrophoresis (2D-PAGE) [3,4]. Protein mixtures are resolved in the first dimension by isoelectric focusing (IEF), which separates proteins according to isolectric point (pI). These proteins are subsequently resolved in the second dimension, according to molecular weight (MW), by SDS-PAGE. Proteins can then be visualized using various staining methods and protein binding dyes (e.g., staining with either brilliant Coomassie blue or silver staining). If protein identification is the ultimate goal, it is important to use staining methods that are compatible with other techniques, such as mass spectrometry [5]. Although 2D-PAGE provides reasonable protein separation and resolution at a low cost, this method is insufficient in visualizing all protein species (e.g., membrane proteins or basic proteins, and proteins spanning a broad range in concentrations, especially those in low abundance) [6]. Furthermore, inadequate gel-to-gel reproducibility is another limiting factor as is the unlikely potential for high-throughput analysis due to its complex multi-step process. Other methods employed to separate proteins are high performance liquid chromatography (HPLC), or two-dimensional liquid chromatography (2D-LC) [7]. Equally ubiquitous, and one of the cornerstone technologies used in proteomics, is mass spectrometry (MS). Mass spectrometry is the current method of choice for the identification of proteins, since this method offers high analytical sensitivity and the capacity for high-throughput protein identification [8]. Matrix-assisted laser desorption ionization time-offlight (MALDI-TOF) MS is commonly employed because of its robustness and user-friendly qualities. Briefly, proteins, obtained from protein spots of interest excised from a silver-stained one-dimensional SDS-PAGE or 2D-PAGE gel, or by other various separation/fractionation methods such as salt precipitation or HPLC, are first digested with proteases or chemical agents to produce a mixture of peptide fragments. The mass-to-charge (m/z) ratio of these peptide fragments is then determined by MALDI-TOF MS, which generates a peptide mass fingerprint (PMF)

D.A. Colantonio, D.W. Chan / Clinica Chimica Acta 357 (2005) 151–158

from the sample of interest. The PMFs from the proteins of interest are then compared to theoretical PMFs of known and DNA sequence-derived proteins contained within protein databases. A high degree of similarity between the signature sequence PMF and a protein from the database is denoted by a high probability score, resulting in the possible identification of the protein. Surface-enhanced laser desorption/ionization timeof-flight (SELDI-TOF) is a modified MALDI-TOF. Compared to mass spectrometers, which are capable of analyzing small peptides (~ 30–300 Da) [9], MALDI is capable of analyzing high molecular weight proteins (N 100 kDa), a capability SELDITOF exploits. The advantage SELDI offers is the potential to separate complex protein mixtures easily, without much sample manipulation, prior to analysis by MS. This is achieved by utilizing chromatographic chip surface technology (ProteinChipR) to selectively capture proteins, which are not digested but analyzed intact. One advantage of the ProteinChipR technology is the number of chemical- and biochemical-treated chip surfaces that can be used to capture proteins from a complex mixture [10]. For example, peptides can be captured on a chromatographic chip surface through hydrophobic, cationic, anionic, metal binding matrixes; antibodies; DNA; or receptor interactions [11]. Once captured, the masses of these whole proteins are then determined by SELDI-TOF, which generates a protein profile spectra. Changes in the protein peaks, or m/z ratios within the spectra, can be used to identify protein changes that may underlie pathophysiological processes. Alternatively, SELDI can also be used to generate PMFs from a complex protein sample, which is then compared to theoretical PMFs of known and DNA sequence-derived whole proteins contained within databases. One limitation of using SELDI-TOF MS in proteomic analysis is related to the high abundance of a small number of proteins in serum. Since the matrixes used by SELDI to treat the sample are not specific to any protein, but rather based on the biophysical chemistry of the peptide, it is unlikely that lower abundant proteins would bind to the chip. They would be out-competed by the high abundant proteins [12]. This is a valid concern not only for SELDI but for most tools in the proteomic armamentarium. As will be discussed later, many investigators

153

are examining ways to remove high abundant proteins from serum. Additionally, MS can be employed to obtain amino acid sequence data to support PMF identification in situations where it is sometimes difficult to assign an unequivocal identity to a protein based on PMF alone or for novel protein identification. Tandem mass spectrometry (MS/MS), which is capable of providing amino acid sequence information on peptide fragments of the parent protein, is the connection of two MS in series. In the application of MS/MS, the first MS generates the PMF spectra; the peptide fragment of interest is selected and isolated by an ion gate voltage that excludes all other fragments but allows the passage of the selected peptide fragment of interest into the second MS where it is fragmented. Fragmentation may occur through metastable decay from the ions’ own internal energy, or be induced by collisions with a gas [9]. As amino acids are sequentially removed from the selected peptide fragment, the decreasing mass of the fragment is used to determine the amino acid sequence. Together, peptide mass information derived from MALDI-TOF analysis, in conjunction with peptide sequence information obtained from MS/MS analysis, provides a strong basis for protein identification. Unlike 2D-PAGE, which can only separate proteins, MS can provide protein ID and can be coupled with other proteomic applications, such as HPLC. Mass spectrometry is also amenable with highthroughput methods. However, MS is expensive and requires highly skilled technical capabilities.

3. Clinical proteomics Clinical proteomics is the application of proteomic techniques and strategies to the field of medicine. For example, changes observed in the proteome of an animal model of disease or a clinical subject can be utilized as a biomarker to detect disease, or used as the basis for the development of pharmacological targets for therapeutic intervention. A search of PubMed reveals that for the year 2004 alone, 1619 articles were published on the topic of proteomics. Of that number, 192 articles were on the topic of clinical proteomics, 71 of which discussed biomarker

154

D.A. Colantonio, D.W. Chan / Clinica Chimica Acta 357 (2005) 151–158

research. This was a significant increase in the number of published articles on the subject of clinical proteomics as compared to 89 in the year 2002. These numbers suggest that the application of clinical proteomic research is growing rapidly in the field of biomarker discovery. Since biomarker discovery is primarily centered on the proteomic exploration of blood, as it is an information-rich sample source that is easily accessible, some issues must first be addressed before proceeding with the proteomic analysis of this type of sample. Unlike other tissues, blood (plasma or serum) is the most complex human-derived proteome since, in addition to containing proteins specific to blood, it also contains proteins released, either through leakage, injury, or other factors, from other tissues in the body [13]. An added difference between blood and other body tissues is the dynamic range in concentration of the various protein species present in the blood. For example, albumin is present in the milligrams per milliliter range while the cytokine, interleukin-6, is present in the picograms per milliliter range [14]. These two serum proteins differ in their plasma abundance by a factor of 10 orders of magnitude [13]. The fact that both these proteins can be measured using current immunologically based analyzers is a testament to the power of the technology; finding a molecule of an analyte that is present at 10 pg/ml among albumin, which is present at 55 mg/ml, is analogous to finding a needle in a haystack. This dynamic range in concentration poses problems for proteomic analysis. Serum, which is derived by centrifugation of clotted plasma, contains 60–80 mg/ml protein, in addition to various small molecules such as salts, lipids, amino acids, and sugars [14]. It is estimated that there are approximately 10,000 proteins commonly found in serum, not including proteins secreted from various tissues, most of which are present in very low concentrations. Roughly 22 high abundant proteins–albumin, immunoglobulin, haptoglobin, and transferrin, to name a few–comprise approximately 97% of the protein content of serum. The remaining 3% of proteins are present in low concentrations and are referred to as low abundant proteins [13,15]. Some may debate as to which pool of serum proteins, high or low abundant, is best to explore for novel biomarkers of disease. In spite of

this debate, whether searching for novel serum biomarkers of disease or trying to elucidate the serum proteome, high abundant proteins could interfere with proteomic analysis and should be separated or removed. Most attempts to separate/remove high abundant proteins from serum have focused on albumin, the most abundant serum protein which comprises 45– 55% of all serum proteins, and immunoglobulin, which comprises approximately 15% of all serum proteins [14]. Thus, eliminating/reducing these two high abundant proteins will remove approximately 60% of the total serum proteins and make it easier to analyze lower abundant proteins. Attempts to remove albumin has relied on affinity-based methods that utilize dyes [16], antibodies, or ultra-centrifugation [17,18]. Although affinity-based methods are effective in removing albumin, the lack of specificity results in the binding and removal of other serum proteins, which may be of interest as a biomarker candidate [16,19]. Methods employing size exclusion via centrifugal filtration have also proven unsuccessful in removing high abundant proteins from plasma [17]. Although separating complex protein mixtures into fractions using salt precipitation results in the inefficient removal of high abundant proteins, rendering the fractions unusable–because significant amounts of high abundant proteins still present in the fractions interfere with proteomic analysis (one method), which employs the use of sodium chloride and ethanol–has demonstrated promising results in the removal of albumin from serum [20]. The removal of high abundant proteins from serum is requisite to conducting a proper proteomic analysis of low abundant proteins and is the intense focus of many research groups. Regardless of the method employed to remove high abundant proteins, it is important to keep in mind that the more complex a method or the more steps a fractionation protocol requires, the greater the likelihood of losing proteins due to sample handling. Thus, a simple fractionation protocol requiring the least amount of sample handling is preferred. When attempting to remove albumin from a sample, it is important to consider that small peptides, which may be potential biomarkers of interest, may co-bind to albumin and thus be removed with the high abundant proteins.

D.A. Colantonio, D.W. Chan / Clinica Chimica Acta 357 (2005) 151–158

After choosing an appropriate sample fractionation protocol, the next step is the analysis of the protein sample. Unlike 2D-PAGE, MS has the potential for rapid and high-throughput analysis for biomarker discovery. Two common approaches to the application of MS to biomarker discovery are the use of protein profiling gained from MS to identify potential disease biomarkers, or use of the protein profile itself as a bprotein fingerprintQ to diagnose/identify a diseased state. In general, it is important to know the identity of biomarkers resulting from proteomics profiling. Efforts towards biomarker discovery have focused on looking for a single over- or under-expressed protein or a disease-induced protein change that is released and detected in the circulation. One tenuous argument against this approach is the difficulty in finding a single biomarker amongst the hundreds to thousands of intact, modified, or cleaved protein isoforms in the serum proteome. The poor success in discovering new biomarkers reflects the inability of this approach [21]. A more substantial argument against this tactic is the idea that, rather than using a single biomarker to identify disease, a multi-marker approach may be better at identifying a disease state. Rai et al. [22] identified three potential biomarkers that were capable of differentiating ovarian cancer from healthy individuals and compared their performance against that of the tumor marker, cancer antigen 125 (CA 125) [22]. Individually, the three potential biomarkers did not perform better than CA 125, but the use of the biomarkers in combination with CA 125 significantly improved their performance. Combining two of the biomarkers with CA 125 achieved a sensitivity of 94% (95% CI, 85–100) in contrast to a sensitivity of 81% (95% CI, 68–95) for CA 125 alone. Similarly, Zhang et al. [23] identified three potential markers that were able to differentiate ovarian cancer from non-cancer controls. The three biomarkers, together with CA 125, showed better sensitivity (74%; 95% CI, 52–90%) at detecting ovarian cancer compared to CA 125 alone (65%; 95% CI, 43–84%) [23]. These results raise the important issue that potential disease biomarkers should be validated using various patient cohorts from multiple sites. The clinical performance of the newly discovered biomarkers could be improved by combining with existing biomarkers.

155

4. Current issues in clinical proteomics Although proteomics in general, and clinical proteomics in particular, offer much promise for biomarker discovery, issues of pre-analytical variables, analytical variability, and biological variation must be addressed if further progress is to be gained. A good understanding of how pre-analytical variables affect test results is important in any clinical laboratory and is equally important in biomarker research. Specimen manipulation, whether sample collection, pipetting, or diluting, contributes to preanalytical error. Whether looking for protein changes that act as potential biomarkers of disease, or using spectral patterns as bprotein fingerprints,Q understanding how the use of different blood collection tubes, coagulation times, storage conditions, or sample type, such as whole blood, plasma, or serum, affects MS protein spectra must be explored and considered. The Human Proteome Organization (HUPO) established a Specimen Committee in 2002 to study and address many of these issues. A report on the preliminary data will be published [24]. A study by Marshall et al. [25] highlights the importance of understanding the effects of pre-analytical variation in interpreting data produced through proteomic analysis [25]. Using MALDI-TOF MS, the authors analyzed blood samples obtained from patients with myocardial infarction and observed changes in protein profiles that were generated not by pathophysiological processes but rather by the amount of time between sample draw and analysis [25]. They found that the pattern of protein fragments in both serum and plasma changed within as little as 2 h at room temperature [25]. Using different concentrations of protease inhibitors, they determined that proteases in serum and plasma remain active after blood collection and their activity has an impact on the spectra produced. This example highlights that failure to understand how pre-analytical variables affect proteomic results can lead to misinterpretation of data, but a better understanding of these variables should lead to improved experimental design and interpretation of data. Another source of variation that must also be considered in biomarker research is biological variation [26]. In the context of proteomic research and biomarker discovery, biological variation, referred to as

156

D.A. Colantonio, D.W. Chan / Clinica Chimica Acta 357 (2005) 151–158

between-subject variation, is the sum of differences in protein expression between people, resulting, but not limited to, differences in age, gender, or race [26]. Another source of variance is within-subject variation. For any individual, many analytes fluctuate based on time of day, fasting state, or age. Although these fluctuations may not be clinically relevant, they do add an additional level of complexity in elucidating disease-induced protein changes from changes due to bio-rhythmic fluctuations. Comparison of complex protein samples, such as serum, between healthy and disease states adds yet an additional layer of complexity in elucidating disease-induced protein changes. Teasing out protein changes that are due to withinsubject variation, between-subject variation, and pathophysiological processes requires good experimental design, a solid understanding of the strengths and limitations of proteomic technologies used, and the proper utilization of statistical techniques.

5. A move in the right direction The enormous amount of data produced during proteomic analysis, coupled with the inherent variation in the instrumentation used as well as the biological variation of the subjects, requires the utilization of sound experimental design, proper calibration of instruments, and appropriate bioinformatics methods in order to generate good quality data by which valid conclusions can be drawn. A number of strategies, with the aim of improving the quality of data produced from the application of proteomic research to biomarker discovery, must be employed. All good research begin with experimental design. In proteomics, good experimental design should help to avoid or reduce the analytical error associated with MS instrumentation, as well as help reduce the effect of both pre-analytical and biological variation. In biomarker research, samples are generally collected from multiple sites and randomly divided into a discovery (training) data set and a validation (testing) data set. Differences in collection practices, sample handling, or storage conditions will differ between institutions and, as such, may influence the proteins present in a given sample [6,27]. For results to be meaningful, a sufficiently large number of collection sites must be employed and the sample population

must be diverse so as to best represent the target population of interest. Since both data sets are derived from the same pooled samples, it will naturally follow that the discovery set will represent the validation set. As a result of pooling the specimens, it is possible to pick up the different types of systematic biases that exist in the original data set, which may be unrelated to the disease, thus increasing the possibility of false discovery due to site-specific systematic bias resulting from pre-analytical variables. An alternative approach that is employed in our laboratories is to use each sites data set separately [6,23]. For example, potential biomarkers unearthed in a discovery data set from one site would then be cross-compared and validated using other sites data sets. This type of model mimics the multicenter validation process that all clinically useful biomarkers must conform with prior to clinical use. An example of the success of this approach can be found in the work of Zhang et al. who created two different discovery and validation cohorts using specimens from four different test sites [23]. Candidate biomarkers were cross-validated between discovery and validation cohort sets, as well as against a currently used cancer biomarker CA 125. Using such an approach, the authors successfully identify three candidate proteins that have the potential to distinguish ovarian cancer from healthy controls [23]. In addition to study design, proper specimen handling and processing should be utilized to reduce preanalytical error. Because error introduced during processing can be next to impossible to trace once the experiment is completed, it is important to rigorously control sample processing in order to minimize the introduction of variation at this phase of the experiment. Some sources of pre-analytical error have been addressed. Additionally, using SELDI as an example, other sources include the lot-to-lot variation of SELDI chips, reagents, or chemicals used throughout the experiment. In addition, differences in the way samples are applied to SELDI chip surfaces, or any MS chip surface, may also introduce error. The use of single lot reagents, chemicals, and SELDI chips should help reduce pre-analytical variation and error. An automated robot can also reduce variability during sample processing and transfer. To improve reproducibility and to reduce the variability observed with most mass spectrometers, control material should be analyzed after calibration of the

D.A. Colantonio, D.W. Chan / Clinica Chimica Acta 357 (2005) 151–158

instrument, to ensure that the MS is performing optimally. For example, external instrument calibration of the spectra should be performed using peptide calibrators containing several peptides of varying masses. Standardization of output requires the use of control material that contains proteins whose peaks can be consistently detected. Once the instrument is calibrated, standardization of the output can be accomplished by running a control sample and adjusting parameters such as laser intensity, detector voltage, and detector sensitivity to ensure that the spectra are consistent from run-to-run. Furthermore, as a means of assessing the performance of the MS during an experiment, control material should also be analyzed throughout an experimental run. A recent study by Semmes et al. demonstrated that implementation of such strategies can yield reproducible results [28]. The inter-laboratory reproducibility of SELDI-TOF output and the ability to differentiate cancer from health cases were assessed. Seven different academic institutions analyze the same control serum and, using three identifiable protein peaks found in the control serum, the inter-laboratory variation in mass accuracy, resolution, signal-to-noise ratio, and normalization intensity of m/z peaks was determined. The across-laboratory measurement revealed a CV for mass accuracy of 0.1%, a signal-to-noise ratio of ~ 40%, and a normalized intensity of 15–36% [28]. Furthermore, each site preformed extremely well in differentiating samples originating from prostate cancer cases and control cases. The use of control material after calibration and throughout an experimental run ensures that the instrument is performing optimally and thus ensures that the data are reliable. As a result of the copious amount of data produced from proteomic analysis, use of bioinformatics tools is required to sift through the data. The statistical tools employed and the type of analysis conducted will depend on the study design. One approach used by our laboratory, and which was developed in-house, is the use of unified maximum separability analysis (UMSA) algorithm [23]. This algorithm bincorporates information from the traditional multivariate statistical classification methods into the support vector machine algorithm [29] to provide a robust approach to analyzing high-dimensional expression dataQ [23]. In a study described

157

earlier, Zhang et al. applied this algorithm, together with other statistical methods, to identify three potential proteins that discriminated between early stage ovarian cancer and healthy controls [23]. Although a comprehensive review of all currently available proteomic bioinformatics tools is beyond the scope of this review, one of the primary applications of proteome informatics analysis is to identify proteins of interest from mass spectrometric data.

6. Conclusion Like any new technology when it is first introduced, proteomics has been touted with much hope and promise. Proteomics holds the promise of being capable of btaking a snapshotQ of the total protein compliment of a cell, identifying both normal and aberrant proteins, and doing so with high-throughput capabilities. The application of proteomic techniques and strategies to the field of medicine is slowly transforming the way biomarker discovery is conducted. However, the complexity of serum is the source of both its promise to clinical applications and its challenge to proteomic analysis. Although the clinical application of proteomics offers much promise for biomarker discovery, further work is required to enhance the performance and reproducibility of established proteomic tools, and issues regarding pre-analytical variables, analytical variability, and biological variation must be addressed if further progress is to be gained. Although the current reality may not have kept pace with previous expectations, research into improving the robustness of high-throughput technologies, coupled with the growing realization of the many factors that can lead to spurious results, is propelling the area of clinical proteomics forward in the right direction.

References [1] Huber LA. Is proteomics heading in the wrong direction? Nat Rev Mol Cell Biol 2003;4(1):74 – 80. [2] Herbert B. Advances in protein solubilisation for two-dimensional electrophoresis. Electrophoresis 1999 (April–May); 20(4–5):660 – 3. [3] Rabilloud T. Solubilization of proteins for electrophoretic analyses. Electrophoresis 1996 (May);17(5):813 – 29.

158

D.A. Colantonio, D.W. Chan / Clinica Chimica Acta 357 (2005) 151–158

[4] Gorg A, Weiss W, Dunn MJ. Current two-dimensional electrophoresis technology for proteomics. Proteomics 2004;12(4): 3665 – 85. [5] Shevchenko A, Wilm M, Vorm O, Mann M. Mass spectrometric sequencing of proteins silver-stained polyacrylamide gels. Anal Chem 1996;68:850 – 8. [6] White NC, Chan DW, Zhang Z. Bioinformatic strategies for proteomic profiling. Clin Biochem 2004;37:636 – 41. [7] Stasyk T, Huber LA. Zooming in: fractionation strategies in proteomics. Proteomics 2004;12(4):3704 – 16. [8] Aebersold R, Mann M. Mass-spectrometry-based proteomics. Nature 2003 (March);422(13):198 – 207. [9] Mann M, Hendrickson RC, Pandey A. Analysis of proteins and proteomes by mass spectrometry. Annu Rev Biochem 2001;70:437 – 73. [10] Merchant M, Weinberger SR. Recent advancements in surfaceenhanced laser desorption/ionization time of flight-mass spectrometry. Electrophoresis 2000 (April);21(6):1164 – 77. [11] Tang N, Tornatore P, Weinberger SR. Current developments in SELDI affinity technology. Mass Spectrom Rev 2004 (January–February);23(1):34 – 44. [12] Diamandis EP. Proteomic patterns in biological fluids: do they represent the future of cancer diagnostics? Clin Chem 2003; 49(8):1272 – 8. [13] Anderson NL, Anderson NG. The human plasma proteome: history, character, and diagnostic prospects. Mol Cell Proteomics 2002 (November);1(11):845 – 67. [14] Bishopm ML, Duben-Engelkirkm JL, Fody EP. Clinical chemistry: principles, procedures, correlations. 4th ed. Lippincott Williams & Wilkins; 2000. Chapter 18. [15] Lathrop JT, Anderson NL, Anderson NG, Hammond DJ. Therapeutic potential of the plasma proteome. Curr Opin Mol Ther 2003 (June);5(3):250 – 7. [16] Gianazza E, Arnaud P. A general method for fractionation of plasma proteins. Dye–ligand affinity chromatography on immobilized Cibacron blue F3-GA. Biochem J 1982 (January 1);201(1):129 – 36. [17] Georgiou HM, Rice GE, Baker MS. Proteomic analysis of human plasma: failure of centrifugal ultrafiltration to remove albumin and other high molecular weight proteins. Proteomics 2001 (December);1(12):1503 – 6. [18] Wang YY, Cheng P, Chan DW. A simple affinity spin tube filter method for removing high-abundant common proteins or

[19]

[20]

[21]

[22]

[23]

[24]

[25]

[26] [27]

[28]

[29]

enriching low-abundant biomarkers for serum proteomic analysis. Proteomics 2003 (March);3(3):243 – 8. Govorukhina NI, Keizer-Gunnink A, van der Zee AG, de Jong S, de Bruijn HW, Bischoff R. Sample preparation of human serum for the analysis of tumor markers. Comparison of different approaches for albumin and gamma-globulin depletion. J Chromatogr A 2003 (August 15);1009(1–2):171 – 8. Colantonio DA, Dunkinson C, Bovenkamp DE, Van Eyk JE, Effective removal of albumin from serum. Proteomics [in press]. Petricoin E, Wulfkuhle J, Espina V, Liotta LA. Clinical proteomics: revolutionizing disease detection and patient tailoring therapy. J Proteome Res 2004 (March–April);3(2):209 – 17. Rai AJ, Zhang Z, Rosenzweig J, Shih IeM, Pham T, Fung ET, Sokoll LJ, Chan DW. Proteomic approaches to tumor marker discovery. Arch Pathol Lab Med 2002 (December);126(12): 1518 – 26. Zhang Z, Bast Jr RC, Yu Y, Li J, Sokoll LJ, Rai AJ, et al. Three biomarkers identified from serum proteomic analysis for the detection of early stage ovarian cancer. Cancer Res 2004 (August 15);64(16):5882 – 90. Rai AJ, Gelfand CA, Haywood BC, Warunek D, et al., Human Proteome Organization—plasma proteome project specimen collection and handling: towards the standardization of parameters for plasma proteome samples. Proteomics [in press]. Marshall J, Kupchak P, Zhu W, Yantha J, Vrees T, Furesz S, et al. Processing of serum proteins underlies the mass spectral fingerprinting of myocardial infarction. J Proteome Res 2003 (July–August);2(4):361 – 72. Fraser CG. Biological variation: from principles to practice. AACC Press; 2001. Drake SK, Bowen RA, Remaley AT, Hortin GL. Potential interferences from blood collection tubes in mass spectrometric analyses of serum polypeptides. Clin Chem 2004 (December);50(12):2398 – 401. Semmes OJ, Feng Z, Adam BL, Banez LL, Bigbee WL, Campos D, et al. Evaluation of serum protein profiling by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry for the detection of prostate cancer: I. Assessment of platform reproducibility. Clin Chem 2005;51:102 – 12. Vapnik VN. Statistical learning theory. Wiley-Interscience; 1998. p. 736.