Molecular & Cellular Proteomics · 2005 · 111 citations · 31 references
Life SciencesBiochemistryMolecular BiologyGene Expression AnalysisMonoclonal AntibodiesProteomicsEnzyme-linked Immunosorbent AssayMolecular Tumor ProfilingProtein MicroarraysProtein QuantificationProteomic ProfilingQuantitative ProteomicsAntibodiesAntigen ArraysBiomarker NormalizationMicroarray AnalysisMultiparametric AnalysisNormalizationProtein Abundance AnalysisProtein Abundance Profiling
The measurements of coordinated patterns of protein abundance using antibody microarrays could be used to gain insight into disease biology and to probe the use of combinations of proteins for disease classification. The correct use and interpretation of antibody microarray data requires proper normalization of the data, which has not yet been systematically studied. Therefore we undertook a study to determine the optimal normalization of data from antibody microarray profiling of proteins in human serum specimens. Forty-three serum samples collected from patients with pancreatic cancer and from control subjects were probed in triplicate on microarrays containing 48 different antibodies, using a direct labeling, two-color comparative fluorescence detection format. Seven different normalization methods representing major classes of normalization for antibody microarray data were compared by their effects on reproducibility, accuracy, and trends in the data set. Normalization with ELISA-determined concentrations of IgM resulted in the most accurate, reproducible, and reliable data. The other normalization methods were deficient in at least one of the criteria. Multiparametric classification of the samples based on the combined measurement of seven of the proteins demonstrated the potential for increased classification accuracy compared with the use of individual measurements. This study establishes reliable normalization for antibody microarray data, criteria for assessing normalization performance, and the capability of antibody microarrays for serum-protein profiling and multiparametric sample classification. The measurements of coordinated patterns of protein abundance using antibody microarrays could be used to gain insight into disease biology and to probe the use of combinations of proteins for disease classification. The correct use and interpretation of antibody microarray data requires proper normalization of the data, which has not yet been systematically studied. Therefore we undertook a study to determine the optimal normalization of data from antibody microarray profiling of proteins in human serum specimens. Forty-three serum samples collected from patients with pancreatic cancer and from control subjects were probed in triplicate on microarrays containing 48 different antibodies, using a direct labeling, two-color comparative fluorescence detection format. Seven different normalization methods representing major classes of normalization for antibody microarray data were compared by their effects on reproducibility, accuracy, and trends in the data set. Normalization with ELISA-determined concentrations of IgM resulted in the most accurate, reproducible, and reliable data. The other normalization methods were deficient in at least one of the criteria. Multiparametric classification of the samples based on the combined measurement of seven of the proteins demonstrated the potential for increased classification accuracy compared with the use of individual measurements. This study establishes reliable normalization for antibody microarray data, criteria for assessing normalization performance, and the capability of antibody microarrays for serum-protein profiling and multiparametric sample classification. Antibody microarrays may be very effective for the discovery and application of molecular diagnostics tools. A key feature of microarrays is multiplexing, the ability to measure multiple proteins in low volumes, consuming small amounts of both precious clinical samples and expensive antibodies. Because many antibodies can be tested in parallel, many candidate molecular markers may be efficiently screened. In addition, the relationships between multiple analytes may be observed so that the use of combinations of proteins in disease diagnostics may be assessed. The use of combinations of proteins for disease diagnostics may produce fewer false positive and false negative results relative to tests based on single proteins. Antibody microarrays can be run efficiently in parallel, enabling studies on the large populations of samples that are necessary for marker discovery and validation. Additional advantages of the platform are good reproducibility, high sensitivity, and quantitative accuracy over large concentration ranges (1Haab B.B. Dunham M.J. Brown P.O. Protein microarrays for highly parallel detection and quantitation of specific proteins and antibodies in complex solutions.Genome Biol. 2001; 2: 1-13Google Scholar). We and others have demonstrated the experimental feasibility of antibody and protein microarrays in applications such as protein profiling of cancer tissue (2Knezevic V. Leethanakul C. Bichsel V.E. Worth J.M. Prabhu V.V. Gutkind J.S. Liotta L.A. Munson P.J. Petricoin E.F.I. Krizman D.B. Proteomic profiling of the cancer microenvironment by antibody arrays.Proteomics. 2001; 1: 1271-1278Google Scholar, 3Sreekumar A. Nyati M.K. Varambally S. Barrette T.R. Ghosh D. Lawrence T.S. Chinnaiyan A.M. Profiling of cancer cells using protein microarrays: discovery of novel radiation-regulated proteins.Cancer Res. 2001; 61: 7585-7593Google Scholar), autoimmune diagnostics (4Joos T.O. Schrenk M. Hopfl P. Kroger K. Chowdhury U. Stoll D. Schorner D. Durr M. Herick K. Rupp S. Sohn K. Hammerle H. A microarray enzyme-linked immunosorbent assay for autoimmune diagnostics.Electrophoresis. 2000; 21: 2641-2650Google Scholar), protein interaction screening (5MacBeath G. Schreiber S.L. Printing proteins as microarrays for high-throughput function determination.Science. 2000; 289: 1760-1763Google Scholar, 6Ge H. Upa, a universal protein array system for quantitative detection of protein-protein, protein-DNA, protein-rna and protein-ligand interactions.Nucleic Acids Res. 2000; 28: e3Google Scholar, 7Lueking A. Horn M. Eickhoff H. Buessow K. Lehrach H. Walter G. Protein microarrays for gene expression and antibody screening.Anal. Biochem. 1999; 270: 103-111Google Scholar, 8Zhu H. Bilgin M. Bangham R. Hall D. Casamayor A. Bertone P. Lan N. Jansen R. Bidlingmaier S. Houfek T. Mitchell T. Miller P. Dean R.A. Gerstein M. Snyder M. Global analysis of protein activities using proteome chips.Science. 2001; 293: 2101-2105Google Scholar), and antibody-based detection of multiple antigens (1Haab B.B. Dunham M.J. Brown P.O. Protein microarrays for highly parallel detection and quantitation of specific proteins and antibodies in complex solutions.Genome Biol. 2001; 2: 1-13Google Scholar, 9deWildt R.M.T. Mundy C.R. Gorick B.D. Tomlinson I.M. Antibody arrays for high-throughput screening of antibody-antigen interactions.Nat. Biotechnol. 2000; 18: 989-994Google Scholar, 10Arenkov P. Kukhtin A. Gemmell A. Voloshchuk S. Chupeeva V. Mirzabekov A. Protein microchips: use for immunoassay and enzymatic reactions.Anal. Biochem. 2000; 278: 123-131Google Scholar, 11Huang R.-P. Huang R. Fan Y. Lin Y. Simultaneous detection of multiple cytokines from conditioned media and patient’s sera by an antibody-based protein array system.Anal. Biochem. 2001; 294: 55-62Google Scholar, 12Mendoza L.G. McQuary P. Mongan A. Gangadharan R. Brignac S. Eggers M. High-throughput microarray-based enzyme-linked immunosorbent assay (ELISA).BioTechniques. 1999; 27: 778-788Google Scholar). The routine application of antibody microarrays to biological and marker-based research requires establishing optimized experimental and analysis methods. Experimental optimization can help to improve the accuracy and reproducibility of measurements, but the analysis methods must be properly developed and applied to ensure the proper interpretation of the data. A common data processing procedure applied to microarray data is normalization, which adjusts the data from each microarray to account for possible systematic experimental variation in factors such as sample labeling efficiency, scanner readout efficiency, and microarray quality (13Park T. Yi S.G. Kang S.H. Lee S. Lee Y.S. Simon R. Evaluation of normalization methods for microarray data.BMC Bioinformatics. 2003; 4: 33Google Scholar, 14Yang Y.H. Dudoit S. Luu P. Lin D.M. Peng V. Ngai J. Speed T.P. Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation.Nucleic Acids Res. 2002; 30: e15Google Scholar). Several normalization procedures have been developed for DNA microarrays. A method developed early on for DNA microarrays is global normalization, which normalizes each array by the median or mean of the intensity log ratios on the array. Other normalization methods purport to correct for systematic errors that may affect arrays non-globally when not all of the spots on an array have the same bias. Intensity-dependent normalization adjusts two-color microarray ratios to account for intensity-based bias in the ratios (14Yang Y.H. Dudoit S. Luu P. Lin D.M. Peng V. Ngai J. Speed T.P. Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation.Nucleic Acids Res. 2002; 30: e15Google Scholar, 15Workman C. Jensen L.J. Jarmer H. Berka R. Gautier L. Nielser H.B. Saxild H.H. Nielsen C. Brunak S. Knudsen S. A new non-linear normalization method for reducing variability in DNA microarray Biol. 2002; or normalization has been used to account for bias by variation with in the used for (14Yang Y.H. Dudoit S. Luu P. Lin D.M. Peng V. Ngai J. Speed T.P. Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation.Nucleic Acids Res. 2002; 30: e15Google Scholar). normalization the that the of the of log ratios be the same for all (14Yang Y.H. Dudoit S. Luu P. Lin D.M. Peng V. Ngai J. Speed T.P. Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation.Nucleic Acids Res. 2002; 30: e15Google Scholar). of microarray data have been developed for normalization L. Normalization and analysis of DNA microarray data by and Biol. 2002; Scholar). The two-color comparative fluorescence detection method that we and others have used for antibody microarray is to the two-color labeling used for cDNA microarray so the normalization methods for cDNA arrays may be for two-color antibody microarray the in antibody microarray such as a and of and a different labeling may mean that the optimal normalization methods may be A and of the normalization for antibody microarrays have not yet been the of procedure for data analysis and an analysis of normalization methods for antibody microarray data is necessary large we studies to normalization methods for antibody microarray data. of antibody microarray measurements from serum samples of patients with pancreatic cancer and of control subjects were and we seven different normalization methods. The methods a of major classes of normalization of but by methods from a of classes we could the effects of normalization on the data. developed methods with were not normalization method use of of data and the and of normalization methods must be of of DNA normalization methods have used the criteria of reproducibility between data (13Park T. Yi S.G. Kang S.H. Lee S. Lee Y.S. Simon R. Evaluation of normalization methods for microarray data.BMC Bioinformatics. 2003; 4: 33Google Scholar, R.A. M. Speed T.P. A of normalization methods for high array data based on and 2003; Scholar, D. Yi S.G. T. normalization using in microarray data.BMC Bioinformatics. Scholar), the of from R.A. M. Speed T.P. A of normalization methods for high array data based on and 2003; Scholar), and the of in data (13Park T. Yi S.G. Kang S.H. Lee S. Lee Y.S. Simon R. Evaluation of normalization methods for microarray data.BMC Bioinformatics. 2003; 4: 33Google Scholar). In we different to a of the of The criteria for and the methods were reproducibility between data accuracy in with and the of trends in the data In addition, using data, we the potential of using combinations of measurements for the classification of the samples were collected from samples from patients with pancreatic and samples from patients with such as or were collected at the of control samples from were collected at the in The samples were in at and each been samples were collected by for human subjects were from for the of antibodies and that were in or were using protein A Protein A to the The antibodies were at concentrations of in antibodies and one antibody were to the and were using from for the detection of and and from for the detection of serum protein concentrations were in using the assay were each with a using a the were for each in of by and at for microarray using a The were at in a to of the antibodies to the were for and in for at in and in from each of the serum samples with and with serum into a at and a of or in This labeling a of relative to the serum proteins an serum protein molecular of The and of the labeling were to each serum protein but not to in of proteins with antibody-antigen The with used of as a normalization the for on a of to each to the and the were to for The by each a at for The samples were and amounts of the were to each of the protein with to a concentration of to a concentration of and to a serum of of each serum sample on a microarray with at for the were in with to the sample and for each in with The were for fluorescence at and using a microarray scanner The used to the data. intensity for each antibody by the is the median of each and is the of variation by the of all the on the array. that not the intensity in both a between the of the of or of the in were from The of median fluorescence to median fluorescence and the ratios from antibody measurements the same array were using the and were using the and were log and median by that not have measurements in at least of the samples were from the normalization methods were applied to the microarray data. The of each are The ratios were by a normalization for each array that by is the of the antibody spots on the array. The ratios were by a normalization for each array that by is the of the antibody spots on the is the IgM concentration of the serum sample on that and is the mean IgM concentration of all of the The ratios were by a normalization for each array that by is the of the antibody spots on the array. The ratios were by a normalization for each array that by is the mean of all of the antibody spots on the array. each the ratios of the antibody measurements were with to the of the spots over both the and and a as by the for the R. R. A for data analysis and Scholar). The ratios of the individual spots were so that the (14Yang Y.H. Dudoit S. Luu P. Lin D.M. Peng V. Ngai J. Speed T.P. Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation.Nucleic Acids Res. 2002; 30: e15Google Scholar). The ratios were using the method The array were by the method as Forty-three different serum samples from patients with pancreatic from patients with other of and from were on a microarray containing 48 different antibodies serum proteins and cancer The of protein to each antibody from each serum sample relative to the protein from a common using two-color comparative fluorescence (1Haab B.B. Dunham M.J. Brown P.O. Protein microarrays for highly parallel detection and quantitation of specific proteins and antibodies in complex solutions.Genome Biol. 2001; 2: 1-13Google Scholar, H. J. R. J. B.B. Antibody microarray profiling of human cancer Antibody screening and of potential 2003; Scholar). The of samples in each on a different using of microarrays that been on different The antibodies on the microarrays and a of the of each are in The variation the samples in relative protein to the antibodies can be using Brown P.O. D. analysis and of expression U. S. A. Scholar). The ratios of to fluorescence from the were and of the the data for a serum sample over all the antibodies and and each the measurements from a antibody in a set. The antibodies that measurements between at least of the data by a for H. K. M. C. D. J. B.B. on antibody microarrays for serum-protein Biol. Scholar). the 48 antibodies, the using data. The the and by or to each antibody measurements, such as from and and to each the high reproducibility and of measurements. Other antibody measurements, such as from and are reproducibility or from the other collected measurements from seven of the proteins were in the and to by to the microarray measurements. measurement to microarray measurements. This between the and microarray measurements of the and accuracy of the microarray measurements for proteins. We the effects of seven different normalization procedures on measurement reproducibility, measurement accuracy, and trends in the data normalizes each array by from a in to a normalizes each array by an in to the to normalizes each array by an in to a the mean of the ratios in each array to a intensity-based to account for in the data that may from in the ratios at (14Yang Y.H. Dudoit S. Luu P. Lin D.M. Peng V. Ngai J. Speed T.P. Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation.Nucleic Acids Res. 2002; 30: e15Google and intensity-based by normalization to the of an the method in of the methods for the methods for factors that affect the arrays such as labeling or scanner methods were not tested each antibody by all the and the spots were data variability by systematic between The of normalization on the reproducibility between data by both the of variation and the between the The of each antibody by between the triplicate measurements from each serum sample for each normalization The for each antibody were compared between the data and each of data using a The from to Normalization by mean the method that a in with the data Normalization by and resulted in an that We the of individual antibodies that or in the data compared with the data. A used to the between the data and each of data for each by mean or by IgM to an abundance of antibodies with a in the data and of antibodies with and relative to the data. by or by a high of antibodies with a and a low with a the of antibodies in a new A for reproducibility is to a between were between the of arrays normalization by each method for each The from each of the data were compared with the from the data In the of with normalization by and IgM to many antibodies and that and antibodies and that the data. of the antibodies and that increased by or over the data. In normalization by or resulted in a antibodies with and many with the data. Normalization by mean not to the antibody in with the The between all of the the the reproducibility of the data compared with the data normalization by mean IgM to or of with with with with IgM IgM IgM in a new the effects of normalization on measurement accuracy, measurements for seven of the proteins were used as a which the microarray measurements were The between the microarray measurements over the and the measurements over the samples were with each normalization The between the microarray and data for all seven of the normalization methods is in a Normalization by of the seven methods between and Normalization based on the IgM to a with the for of the data by normalization a in with for of the antibodies in with the data. on with measurements, normalization by an to a to affect measurement accuracy, and measurement accuracy may be normalization by we the of normalization on trends in the data to determine normalization methods trends effects were by and the proteins that the sample normalization by each of the data the from of the normalization methods methods with the reproducibility and accuracy, as in the are in and the other are in the data. The may be to the effects of normalization on trends in the data In all the control samples and the other classes are In all the to their microarray that the are between the the and the data The samples to into or in one has high of most of the proteins samples and is low in most of the proteins samples and and is low in proteins and high in and samples and The patterns are in the mean data the of patients that are high in most proteins or low in most proteins are not as are in the other Because a major use for antibody microarray is to between sample we the of normalization on the between the sample The antibody measurements that between the sample classes using data between the were by a and the of antibodies that were in common between methods the data, antibody measurements the serum samples from patients with pancreatic cancer and the samples from control antibodies, were common to the analysis from data and were common from data. The other normalization methods between and common with the data. the trends observed normalization by and are to each other and to the data, but normalization by the other methods resulted in a of antibodies that the Because of the high accuracy and reproducibility of normalization by and of to the trends in the data normalization method used for the analysis between normalization in a new A major of the ability to measure multiple proteins is the of using combinations of proteins for sample classification relative to the use of single proteins. We application of the data using the data. The data were over the using the antibodies that measurements as over the between at least of the data of the 48 antibodies the reproducibility antibodies measurements with different between the samples from patients with pancreatic cancer and from control The of the measurements from each of antibodies in the cancer and control samples are in antibodies, and in the pancreatic cancer and the in the pancreatic cancer and the that all of the between the cancer and control The potential from using combined antibodies to the samples using a of the T. R. a of 2000; 28: Scholar, Y. M. Y. M. M. A for protein profiling of data for cancer 2003; 4: Scholar). is an method for individual into a combined and to be to T. R. a of 2000; 28: Scholar), a by other used classification A the optimal of antibodies in a combined of the samples were used as a to a for of the samples were as a to determine the of the This each using a different of for classification. The when of an antibody into the increased the The the in classification of samples and the of the combined from the data set. This the of used to each of the cancer and control samples as to one of the The antibodies used in the multiparametric classification and the of the is in The and were over the the using the antibodies at and that using all seven antibody measurements antibody used the results are sensitivity, and or which not possible using single classification using multiple in a new The major of to determine the optimal normalization procedure for antibody microarray data. This necessary systematic of normalization methods for antibody microarray yet been analysis between the normalization methods in the data and a of the methods which normalization method is study to an for assessing and normalization methods. of the normalization methods in one but not very in the of using multiple criteria for the by IgM to data the data. the accuracy of the measurements as by with the Because IgM concentrations from sample to to a is not and the accuracy of the measurements. be to a protein in the serum that could be used as a that all serum proteins are to between as by by IgM to the trends in the data in with the other normalization methods Other methods in but not in normalization data, but the reproducibility the data as by the normalization to the data not improve The of proteins that cancer from and the of the The method developed for DNA microarray data and on a large of data to produce an of intensity-based in the fewer data as with data, such may be and may to the data. Other normalization methods that use of trends in data may not on which the in of not on arrays the in small of proteins could and normalization good accuracy, and the trends in the data relative to the data, but the reproducibility normalization Normalization by mean in reproducibility and accuracy, the normalization to trends in the data normalization by or Normalization by mean is the concentration of the proteins is between Because the concentration may a small of normalization by mean may produce results that the trends in the data all of the the normalization of the methods to have the by the of an such as IgM is are to the A is not to the so the not correct for of bias that the The accuracy of a is of the or of the of the and the of proteins. of normalization by an are that highly for that protein must be for sample and that one on the quality of the microarray measurements for that in the normalization method are possible and necessary for antibody microarrays. The high reproducibility of the mean data that by many proteins may be We are on the mean In addition, different such as or proteins that have to human may A of or highly specific proteins may produce results the use of a single Other antibody or protein array may have other optimal normalization the methods a for which is In addition, other sample such as from tissue or may and the normalization to be In to for normalization data the of the of combined measurements for sample classification. markers may be to improve the markers The of the sample classification when using the multiparametric compared with the use of single the potential of antibody microarray data for This is not to be specific for pancreatic cancer most of the proteins used in the between the cancer and other disease The of a specific for pancreatic cancer measurements from proteins that are with pancreatic A detection such as the two-color method demonstrated H. K. M. C. D. J. B.B. on antibody microarrays for serum-protein Biol. Scholar), the measurement of abundance proteins that may to a specific for pancreatic using that are on the of specific serum protein in pancreatic cancer can be from data of the small sample and potential bias between the and control but the observed between the cancer and control samples were with the high of with pancreatic The of protein and in the disease samples a positive in and markers in pancreatic 1999; Scholar, J.S. T. cancer as a and cancer J. 1999; Scholar), and a in the of and as observed is observed in an T. A. of the in patients with pancreatic 1999; Scholar). of serum and IgM have been observed in cancer S.L. A of serum from patients with and Scholar, of serum and tissue in Scholar), and has been with pancreatic cancer J.S. A.M. expression of and in human pancreatic J. 2002; Scholar). In reliable methods for antibody microarray data and criteria for assessing normalization methods. we that many different proteins in serum samples can be using antibody microarrays and that capability is for multiparametric sample classification. the for studies that could to diagnostics for pancreatic cancer and other We for and and for the of antibodies. with
31