Am J Pharmacogenomics 2005; 5 (5): 281-292

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 1. Mass Spectrometers: Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 1.1 Ion Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 1.2 Mass Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 1.3 Ion Detectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 1.4 Tandem Mass Spectrometry (MS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 2. Analysis of MS Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 2.1 Baseline Correction and Noise Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 2.2 Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 2.3 Spectra Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 2.4 Peak Finding and Deconvolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 2.5 Peak Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 2.6 Peptide Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 3. Support Vector Machine (SVM)-Recursive Feature Elimination (RFE) Feature Selection and Classification . . . . . . . . . . . . . . . . . . . . . 286 3.1 SVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286 3.2 SVM-RFE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 3.3 Numerical Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288 4. Discussions and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 The ultimate goal of cancer proteomics is to adapt proteomic technologies for routine use in clinical Abstract laboratories for the purpose of diagnostic and prognostic classification of disease states, as well as in evaluating drug toxicity and efficacy. Analysis of tumor-specific proteomic profiles may also allow better understanding of tumor development and the identification of novel targets for cancer therapy. The biological variability among patient samples as well as the huge dynamic range of biomarker concentrations are currently the main challenges facing efforts to deduce diagnostic patterns that are unique to specific disease states. While several strategies exist to address this problem, we focus here on cancer classification using mass spectrometry (MS) for proteomic profiling and biomarker identification. Recent advances in MS technology are starting to enable high-throughput profiling of the protein content of complex samples. For cancer classification, the protein samples from cancer patients and noncancer patients or from different cancer stages are analyzed through MS instruments and the MS patterns are used to build a diagnostic classifier. To illustrate the importance of feature selection in cancer classification, we present a method based on support vector machine-recursive feature elimination (SVM-RFE), demonstrated on two cancer datasets from ovarian and lung cancer. 282 Rajapakse et al. The proteome is a highly dynamic entity whose variation value to the physician, and the clinical laboratory has to process reflects changes in physiological states in response to various the raw data to facilitate its interpretation. Clearly, the underlying stimuli. As such, the comparison of the appropriate proteomes has presumption is that the methods used by the clinical chemist are been increasingly useful in identifying diagnostic, prognostic, and accurate, consistent, and validated. The biological variability therapeutic markers of disease and infection,[1,2] as well as in among patient samples as well as the huge dynamic range of evaluating drug toxicity and efficacy,[3] leading to more effective biomarker concentrations are currently the main challenges facing and better tailored treatments.[4,5] Proteomics also enables the efforts to deduce diagnostic patterns that are unique to specific characterization of various subtypes of disease, uncovering a disease states.[20] number of novel potential drug targets.[6] Cancer proteomics[7,8] is Many strategies are available for the identification of cancer an important subset of proteomics, involving the identification and biomarkers, including: quantitative analysis of differentially expressed proteins relative to 1. differential display of proteins, whereby normal and tumor healthy tissue counterparts at different stages of disease, from lysates are compared and upor downregulated proteins are identipre-neoplasia to neoplasia.[9] Tumor-specific proteomic profiles fied[21] are generated to better understand tumor development and pro2. comparative analysis of secreted proteins and membrane fracgression while novel targets for therapy and novel markers for tions of different tumor cell lines to yield potential biomarkers[6] early diagnosis can be identified.[10] 3. generation of unique protein profiles pertaining to specific Generally, the earlier cancer is detected, the more favorable the tumors by using mass spectrometry (MS)[22-28] (see figure 1). prognosis for the patient.[11,12] Unfortunately, in many cases, canOur focus in this review is on the classification of cancer by cer is not diagnosed until cancer cells have metastasized and using MS data. curative treatment is limited. This is especially applicable for 1. Mass Spectrometers: Fundamentals cancers that present vague or no symptoms, or those that are relatively inaccessible to physical examination including breast, MS[29] involves the measurement of the mass-to-charge (m/z) ovarian, liver, and lung cancer.[5,11,13,14] Biomarkers that are specifratio of ions and is increasingly being applied to sift through ic and sensitive for a particular cancer type and detectable in complex protein mixtures to find biomarker patterns that can be high-risk patients or patients with early-stage cancer will be invalused for diagnosis, prognosis, or monitoring of disease.[20] Two uable for detecting, staging, monitoring, and controlling the disdifferent types of instruments are mostly used for the majority of ease. With respect to cancer, protein biomarkers refer to subproteomics work: the matrix-assisted laser desorption ionization stances or proteomic patterns that highlight the presence of cancer (MALDI)-time of flight (TOF) instruments and the electro-spray in the body; they can be compounds secreted by the tumor itself or ionization (ESI)-tandem MS instruments. as a result of a specific response of the body to the presence of Mass spectrometers contain at least three major parts: an ion cancer. Generally, biomarkers are measurable in tissues, cells or source, a mass analyzer, and an ion collection/detection system. A fluids, but to minimize the trauma and cost of screening, while sample is introduced into the mass spectrometer by using a direct maximizing utility, biomarkers that are measurable in serum,[15] insertion probe, direct infusion, or chromatographic separation urine,[16] or saliva[17,18] are preferred. interfaced with the MS instrument, which converts the compoAlthough the current state of cancer proteomics application is nents of a sample mixture to ions and then analyzes them on the still largely limited to research, the ultimate goal is to adapt these basis of their m/z ratio. applications for routine use in clinical laboratories. Apart from being used for the early detection of asymptomatic patients and 1.1 Ion Source diagnosis of symptomatic patients, biomarkers may potentially be used for the surveillance of individuals who have an increased The analysis of substances by mass spectrometers requires the chance of developing cancer. Also, biomarkers may be used to formation of either positive or negative gas phase ions by a device monitor patients with a previous medical history of cancer for referred to as the ‘ion source’ which produces ions by protonation recurrence.[19] The clinical chemist provides patient-derived labo(M+H+ → MH+), cationization (M+Cat+ → MCat+), electron ratory data to the physician who then incorporates information ejection (M → M+•+e–), electron capture (M+e– →M–), from various other assessments to make a diagnosis of the padeprotonation (MH → M–+H+), or the transfer of a charged tient’s condition. However, the raw laboratory data are of limited molecule from the condensed to the gas phase. The ESI and  2005 Adis Data Information BV. All rights reserved. Am J Pharmacogenomics 2005; 5 (5) Proteomic Cancer Classification with Mass Spectrometry Data 283 In MALDI, samples are co-crystallized on the target plate after mixing with a matrix solution (an ultra-violet-absorbing compound). The co-crystal is irradiated by a pulsed laser beam, causing high-density energy to accumulate within it. The accumulated energy induces the samples and matrix molecules to vaporize. This results in proton transfer between the sample and matrix. This ionization process produces both positive and negative ions, depending on the nature of the sample. For peptides and proteins, the positive ions are formed by accepting protons as they are ejected from the matrix. Most of the peptide ions from MALDI are singlycharged because each peptide molecule tends to pick up a single proton. Surface-enhanced laser desorption ionization (SELDI)-TOF is a technology involved in quantitative analysis of protein mixtures, which uses chemical or biological trapping surfaces that allow differential capture of proteins based on the intrinsic properties of the proteins themselves. Sample fluids are directly applied to the trapping surfaces, and any proteins in the sample fluids which have an affinity for the trapping surface will bind to it. The unbound proteins are removed with a series of washes. The bound proteins are laser desorbed and ionized for mass spectral analysis. The role of SELDI-TOF in biomarker discovery primarily involves comparing the protein profiles from control and test samples and, thereafter, focusing on statistically significant differences.[30] Although SELDI-TOF is able to rapidly deduce proteomic patterns from complex samples via differential comparison, this technology is still at an early stage of development. Therefore, SELDI-TOF also faces a number of limitations. One of the main limitations is the reproducibility of SELDI-TOF protein profiling experiments.[31-34] Secondly, SELDI-TOF is limited in terms of sample resolution. Although it is particularly suited to proteins with a molecular mass below 30 kDa,[35] SELDI-TOF poorly Sample collection and preparation Mass spectrometry analysis