Development and Evaluation of Normalization Methods for Label-free Relative Quantification of Endogenous Peptides

The performances of 10 different normalization methods on data of endogenous brain peptides produced with label-free nano-LC-MS were evaluated. Data sets originating from three different species (mouse, rat, and Japanese quail), each consisting of 35–45 individual LC-MS analyses, were used in the study. Each sample set contained both technical and biological replicates, and the LC-MS analyses were performed in a randomized block fashion. Peptides in all three data sets were found to display LC-MS analysis order-dependent bias. Global normalization methods will only to some extent correct this type of bias. Only the novel normalization procedure RegrRun (linear regression followed by analysis order normalization) corrected for this type of bias. The RegrRun procedure performed the best of the normalization methods tested and decreased the median S.D. by 43% on average compared with raw data. This method also produced the smallest fraction of peptides with interblock differences while producing the largest fraction of differentially expressed peaks between treatment groups in all three data sets. Linear regression normalization (Regr) performed second best and decreased median S.D. by 38% on average compared with raw data. All other examined methods reduced median S.D. by 20–30% on average compared with raw data. The performances of 10 different normalization methods on data of endogenous brain peptides produced with label-free nano-LC-MS were evaluated. Data sets originating from three different species (mouse, rat, and Japanese quail), each consisting of 35–45 individual LC-MS analyses, were used in the study. Each sample set contained both technical and biological replicates, and the LC-MS analyses were performed in a randomized block fashion. Peptides in all three data sets were found to display LC-MS analysis order-dependent bias. Global normalization methods will only to some extent correct this type of bias. Only the novel normalization procedure RegrRun (linear regression followed by analysis order normalization) corrected for this type of bias. The RegrRun procedure performed the best of the normalization methods tested and decreased the median S.D. by 43% on average compared with raw data. This method also produced the smallest fraction of peptides with interblock differences while producing the largest fraction of differentially expressed peaks between treatment groups in all three data sets. Linear regression normalization (Regr) performed second best and decreased median S.D. by 38% on average compared with raw data. All other examined methods reduced median S.D. by 20–30% on average compared with raw data. Peptidomics is defined as the analysis of the peptide content within an organism, tissue, or cell (1Clynen E. Baggerman G. Veelaert D. Cerstiaens A. Van der Horst D. Harthoorn L. Derua R. Waelkens E. De Loof A. Schoofs L. Peptidomics of the pars intercerebralis-corpus cardiacum complex of the migratory locust, Locusta migratoria.Eur. J. Biochem. 2001; 268: 1929-1939Crossref PubMed Scopus (143) Google Scholar, 2Schulz-Knappe P. Zucht H.D. Heine G. Jürgens M. Hess R. Schrader M. Peptidomics: the comprehensive analysis of peptides in complex biological mixtures.Comb. Chem. High Throughput Screen. 2001; 4: 207-217Crossref PubMed Scopus (200) Google Scholar, 3Verhaert P. Uttenweiler-Joseph S. de Vries M. Loboda A. Ens W. Standing K.G. Matrix-assisted laser desorption/ionization quadrupole time-of-flight mass spectrometry: an elegant tool for peptidomics.Proteomics. 2001; 1: 118-131Crossref PubMed Scopus (111) Google Scholar). The proteome and peptidome have common features, but there are also prominent differences. Proteomics generally identifies proteins by using the information of biologically inactive peptides derived from tryptic digestion, whereas peptidomics tries to identify endogenous peptides using single peptide sequence information only (4Svensson M. Sköld K. Nilsson A. Fälth M. Nydahl K. Svenningsson P. Andrén P.E. Neuropeptidomics: MS applied to the discovery of novel peptides from the brain.Anal. Chem. 2007; 79 (18–21): 15-16Crossref PubMed Scopus (52) Google Scholar). Endogenous neuropeptides are peptides used for intracellular signaling that can act as neurotransmitters or neuromodulators in the nervous system. These polypeptides of 3–100 amino acids can be abundantly produced in large neural populations or in trace levels from single neurons (5Svensson M. Sköld K. Nilsson A. Fälth M. Svenningsson P. Andrén P.E. Neuropeptidomics: expanding proteomics downwards.Biochem. Soc. Trans. 2007; 35: 588-593Crossref PubMed Scopus (51) Google Scholar) and are often generated through the cleavage of precursor proteins. However, unwanted peptides can also be created through post-mortem induced proteolysis (6Sköld K. Svensson M. Kaplan A. Björkesten L. Aström J. Andren P.E. A neuroproteomic approach to targeting neuropeptides in the brain.Proteomics. 2002; 2: 447-454Crossref PubMed Scopus (104) Google Scholar). The later aspect complicates the technical analysis of neuropeptides as post-mortem conditions increase the number of degradation peptides. The possibility to detect, identify, and quantify lowly expressed neuropeptides using label-free LC-MS techniques has improved with the development of new sample preparation techniques including rapid heating of the tissue, which prevents protein degradation and inhibition of post-mortem proteolytic activity (7Sköld K. Svensson M. Norrman M. Sjögren B. Svenningsson P. Andrén P.E. The significance of biochemical and molecular sample integrity in brain proteomics and peptidomics: stathmin 2–20 and peptides as sample quality indicators.Proteomics. 2007; 7: 4445-4456Crossref PubMed Scopus (100) Google Scholar, 8Svensson M. Sköld K. Svenningsson P. Andren P.E. Peptidomics-based discovery of novel neuropeptides.J. Proteome Res. 2003; 2: PubMed Scopus Google Scholar). has by (4Svensson M. Sköld K. Nilsson A. Fälth M. Nydahl K. Svenningsson P. Andrén P.E. Neuropeptidomics: MS applied to the discovery of novel peptides from the brain.Anal. Chem. 2007; 79 (18–21): 15-16Crossref PubMed Scopus (52) Google Scholar, M. Sköld K. Nilsson A. Fälth M. Svenningsson P. Andrén P.E. Neuropeptidomics: expanding proteomics downwards.Biochem. Soc. Trans. 2007; 35: 588-593Crossref PubMed Scopus (51) Google Scholar) and K. B. Baggerman G. J. Schoofs L. Peptidomics: the approach of techniques and for PubMed Scopus Google Scholar) that the peptidome between of and to the discovery of biologically peptides of or However, differences in peptide only from biological differences but also from and the of induced is common to the raw data. This is a in the of using M. A of normalization methods for data on and 2003; PubMed Scopus Google Scholar, W. A. A. M. applied to data and to the of 2002; PubMed Scopus Google Scholar, S. P. J. for a method single and Res. 2002; PubMed Scopus Google Scholar). a methods for data have also for peptide data produced with LC-MS techniques M. A of normalization methods for data on and 2003; PubMed Scopus Google Scholar, W. A. A. M. applied to data and to the of 2002; PubMed Scopus Google Scholar, S. P. J. for a method single and Res. 2002; PubMed Scopus Google Scholar, M. S. K. for analysis for protein of PubMed Scopus Google Scholar, for with mass and label-free Proteome Res. PubMed Scopus Google Scholar, of proteins and by mass 2007; PubMed Google Scholar, W. S. S. P. M. of proteins and by mass or Chem. 2003; PubMed Scopus Google Scholar). the for techniques is that the or be different in this between LC-MS Global normalization methods to all are used to a single normalization between M. S. K. for analysis for protein of PubMed Scopus Google Scholar, of proteins and by mass 2007; PubMed Google Scholar, W. S. S. P. M. of proteins and by mass or Chem. 2003; PubMed Scopus Google a of peaks to be between W. S. S. P. M. of proteins and by mass or Chem. 2003; PubMed Scopus Google Scholar) is or peptides are used as a by for with mass and label-free Proteome Res. PubMed Scopus Google normalization methods for tryptic peptide data were The that or regression best in but also that the best procedure be for each data set used for LC-MS data have for with mass and label-free Proteome Res. PubMed Scopus Google Scholar, a of peaks and PubMed Scopus Google Scholar, J. A. and methods for using mass 4: PubMed Scopus Google but to only for with mass and label-free Proteome Res. PubMed Scopus Google Scholar) have used data sets to of have data of endogenous peptides. this the of 10 different normalization methods were on data produced by a to an or quadrupole used to the treatment normalization in the regression normalization of regression normalization of in of regression regression followed by analysis order on mass methods that were for data were and novel regression followed by analysis order normalization is The normalization methods were using three data sets of endogenous brain peptides originating from three different species (mouse, rat, and Japanese quail), each consisting of 35–45 individual LC-MS Each data set contained both technical and biological quadrupole average biological to the treatment normalization in the regression normalization of data regression normalization of data in median normalization of technical normalization normalization regression normalization regression followed by analysis order normalization normalization on This on normalization methods for label-free of endogenous peptides from brain on data from three different data sets with different The data set from of a of with and treatment A. Fälth M. K. Sköld K. Svenningsson P. Andrén P.E. of and peptides in an of PubMed Scopus Google Scholar). The data set from from of different and with or Nilsson A. Fälth M. K. M. Andren P.E. A analysis of peptides to the endogenous and in of Proteome Res. PubMed Scopus Google Scholar). The data set from of Japanese from and different with or A. A. M. M. K. K. B. P. E. and L. in information the different data sets can be found in Data The were by or the were by rapid heating using a and to post-mortem degradation (7Sköld K. Svensson M. Norrman M. Sjögren B. Svenningsson P. Andrén P.E. The significance of biochemical and molecular sample integrity in brain proteomics and peptidomics: stathmin 2–20 and peptides as sample quality indicators.Proteomics. 2007; 7: 4445-4456Crossref PubMed Scopus (100) Google Scholar). The were as A. Fälth M. K. Sköld K. Svenningsson P. Andrén P.E. of and peptides in an of PubMed Scopus Google Scholar, Nilsson A. Fälth M. K. M. Andren P.E. A analysis of peptides to the endogenous and in of Proteome Res. PubMed Scopus Google All three data sets were in a randomized block fashion. the procedure a sample of followed by from each treatment by The procedure all individual The biological in block were technical of the in block for a single sample that in block the procedure a sample of 10 followed by from each treatment by The procedure all individual the procedure a sample from each treatment by and a technical of a single sample the of each The procedure all individual the and data sets each of to in the block in the data and in the data set to were a block in the were the number of in each treatment and in each of LC-MS of the number of LC-MS of all biological were of 10 single in a new The analyses of the peptide were performed on a to a mass and or a mass Data were a information the nano-LC-MS conditions can be found in Data peptide for each data set can be found in the the biological of the data sets A. Fälth M. K. Sköld K. Svenningsson P. Andrén P.E. of and peptides in an of PubMed Scopus Google Scholar, Nilsson A. Fälth M. K. M. Andren P.E. A analysis of peptides to the endogenous and in of Proteome Res. PubMed Scopus Google The A. M. D. Nilsson A. Fälth M. Sköld K. Svensson M. S. Svenningsson P. Andrén P.E. Björkesten L. method for LC-MS data sets for peptides and including and Proteome Res. 2007; PubMed Scopus Google Scholar) used for of peaks J. L. Björkesten L. J. analysis of proteins using a novel for of PubMed Scopus Google Scholar). were using a of and all data sets the to the number of and peaks to all normalization methods and the data were that only peaks that were analyses in each data set were of normalization peaks were on different for each data set and of all analyses all LC-MS of three in each biological in all biological of in each biological in a new The of normalization normalization and normalization on normalization the of the peaks in each sample were used for normalization J. L. Björkesten L. J. analysis of proteins using a novel for of PubMed Scopus Google Scholar). the and the with is is and is a of These peptides were used for normalization The of data for each analysis by a regression to the peptides in each analysis and the average regression all analyses from each The produced with the were for both normalization This method to be on the of The approach is the of an A is the in and A is the average is used to differences in of analyses, and differences are by the to the of a of analyses of a single which used the method S. P. J. for a method single and Res. 2002; PubMed Scopus Google Scholar). The procedure differences were from all This as for the in data M. A of normalization methods for data on and 2003; PubMed Scopus Google Scholar). can be for large data sets. The method has used for peptide data for with mass and label-free Proteome Res. PubMed Scopus Google Scholar) produced with The fraction of peaks used for the regression each set and the the between the of all from the and This the as used by for with mass and label-free Proteome Res. PubMed Scopus Google Scholar). This method to be on the of and also by for with mass and label-free Proteome Res. PubMed Scopus Google Scholar). Linear regression normalization is performed by regression to the A and are by from the regression for method the procedure differences were from all The the between the of all from the and normalization methods and were to the normalization the for analysis using the median or normalization has often used for data S. P. J. for a method single and Res. 2002; PubMed Scopus Google Scholar) and has applied to peptide data produced with for with mass and label-free Proteome Res. PubMed Scopus Google Scholar). This method for the normalization between M. A of normalization methods for data on and 2003; PubMed Scopus Google Scholar). The method is used with this type of and has also used for peptide data for with mass and label-free Proteome Res. PubMed Scopus Google Scholar, label-free method for the of proteins from biological Proteome Res. 4: PubMed Scopus Google Scholar). The of the method is to the of the all This method is a type of normalization analysis is as a and all other are to the A single normalization for each analysis is from the median of the of for all peaks between the analyses in and the analysis M. S. K. for analysis for protein of PubMed Scopus Google Scholar, of proteins and by mass 2007; PubMed Google Scholar, W. S. S. P. M. of proteins and by mass or Chem. 2003; PubMed Scopus Google Scholar). the analysis in each data set with the peaks as This method for single and data. The through and of and the to a the is of the W. A. A. M. applied to data and to the of 2002; PubMed Scopus Google Scholar). This method also to be on the of but of of of analyses as for method a single analysis by the median for all Linear regression normalization performed on this by regression to each individual on the regression new were for each both and of the regression that the analysis order of the LC-MS to in the data were only on analysis order will correct for as differences in of sample compared with the of the with analysis order normalization to the different of A regression regression and Scopus Google using in the Linear and methods for in PubMed Scopus Google Scholar, of 2003; PubMed Scopus Google Scholar) for each the analysis and the all analyses for each to the each a of by to the used for the of the the A and a of to of for all of and were used for the of A applied to each data set using H.D. R. of information block are Scopus Google Scholar). were in a randomized block order there be differences or only interblock differences. the fraction of peaks with block differences in each data were whereas biological and technical were the fraction of peaks with LC-MS order differences in the data the were the and the The with and including the block the fraction of peaks with biologically biological were whereas block and technical were were with and including the block of for in the A and for for Scholar) W. for Scholar) B. M. S. B. L. J. K. W. S. R. M. G. G. L. J. development for and PubMed Google Scholar) used to and were using A of from the that some were this only a on the in of peaks and for analysis the median in block from all analysis the with in the A and for for Scholar) of the largest were the between technical and the including both technical and biological biological to the treatment were on average in median S.D. and of of compared with raw data. the and data sets the only of different biological whereas the data set to three technical of biological the data which only methods were in a in the average the different data but this will only have a on the a of the different normalization methods decreased median S.D. and compared with the raw the average for the three data sets The performances of 10 different normalization methods on three different data sets of endogenous peptide data produced with label-free mass were The in all three data sets were derived from brain (mouse, rat, and quail), and each data set of 35–45 individual LC-MS a new set of the The in the and were in a randomized block and each data set produced within a of and different of peaks were for analysis in the and data of peaks is to the number of an peaks of all analyses in each data set there a the of the analysis in for the data whereas the found for the data set analysis order-dependent found for the data and there were differences in between LC-MS analyses The for this type of data is that the biological and different only a fraction of the peaks that normalization all analyses have the All methods for the method bias. The methods in a to that of the and data. the data for a analysis with a decreased for that The for this is that lowly expressed peaks will have an to the A but as also found in the and data sets. S.D. and of raw data for the between and and between and the three data sets. all three data sets the largest found using the and RegrRun methods with an average decreased median S.D. of and 43% and on average a and in all other examined methods the generally and between and and between and on average in median S.D. and all data sets the replicates, using the and the S.D. and were found to be compared with using the normalization using the RegrRun method the median S.D. for the for both the and data and in the data set data normalization the between technical be a with a of using the RegrRun method the regression to for the technical the data set the technical of the different biological were also The raw and data produced regression from that S.D. and set set set in a new the including biological compared with the The median S.D. and of the raw data between and and between and all three data sets the largest found using the and RegrRun methods with an average in median S.D. of and The largest in also found for the and RegrRun and on the data set only methods and RegrRun in a of median S.D. and found for the all other examined methods generally produced of median S.D. and and of the different normalization methods were by using derived from the in median S.D. and The method with the largest the of and to in the data both S.D. and the RegrRun method the best followed by and and all performed with in the of and the and and The method the a of the different normalization decreased median S.D. and compared with raw the average for the three data set The methods and RegrRun in a and average median S.D. and a and The methods for which only in a in average median S.D. and were in a randomized block order there be differences or interblock differences. All normalization methods RegrRun in large or and of peaks with interblock differences. in a regression of the analysis all three data sets a in interblock the of in the in in peaks with interblock differences. in an increase of interblock that to A of also found the in within to the treatment type different normalization there be or only differences in between the in the data set were groups of the and the sample The differences between the groups were tested using data or of the block and data. is that both the fraction and the of the of peaks order-dependent is for data data only of the peaks were between the groups of the of the data of peaks differences. This fraction decreased to including the block in the a analysis performed on the data set to the to and normalization the contained data with analysis order whereas contained data that or all three data a large data or a fraction and data of examined peaks analysis order-dependent all three data sets and for a of all peaks there a of the of the LC-MS Only the RegrRun which and analysis order analysis order-dependent bias. the of and data in of a for each data including or the block in the The RegrRun method the block the largest fraction of in all three data sets the and including the block in the the fraction of in the data set there is a of the differentially expressed peaks found using the different normalization including the block in the for the and methods the to the RegrRun method the block in an increase of of peaks as differentially expressed with all methods compared with the block in the A also found in the and data sets The of LC-MS data is to or to differences in peptide in and or other bias. this the performances of 10 different normalization methods for the of endogenous brain peptides were The normalization methods is that the differences in of all or a of peaks be However, this be to in in peptide data P. J. M. in mass Scholar). an to this the data were that only peaks of the data set were in each some analyses peaks be to The normalization for analyses can peaks are in the normalization procedure or with have for with mass and label-free Proteome Res. PubMed Scopus Google Scholar, A. and of of the on and of D. Scholar) also found that the on the of This the of regression for peptide data. a a and new to each individual analysis both and of the regression the median S.D. and a as a in However, new from regression also a the regression is to The used to this type of The endogenous peptides were through However, that the of the of different normalization methods found in the also be to peptides derived from tryptic for with mass and label-free Proteome Res. PubMed Scopus Google Scholar) have compared the of different normalization for of peptides originating from proteins from and technical in three different data a protein a and a brain system. The analyses for some were performed a of and also of the The data were in a block each block of analyses whereas in all analyses within each data set were for with mass and label-free Proteome Res. PubMed Scopus Google Scholar) that normalization by regression to in and to in performed but also that the between different data sets. The peptides in the are endogenous brain peptides. All LC-MS analyses for each individual and were performed within a of and the between and within an the the RegrRun method performed the best in the three data sets evaluated. that the that some of is and between MS analyses all is regression followed by a to each on the analysis order in a in median S.D. and compared with all other normalization The method performed second best in median S.D. and whereas the other normalization methods performed data in the data of the peaks in between the and the there be differences between the This fraction decreased including the block in the data only of the peaks were different in the of or the method in different of analysis order-dependent of all peaks in the different data This type of the RegrRun order-dependent is to The LC-MS of to different treatment groups have to be in a randomized order-dependent with treatment will be but has of for 2002; PubMed Scopus Google Scholar, J. The of in mass some PubMed Scopus Google Scholar, and analysis of with biological PubMed Scopus Google Scholar, and analysis in proteomics 2007; PubMed Scopus Google Scholar). all three data the for the of were used and the is set there will be a of This can be by interblock as the LC-MS are performed in a randomized an approach is to median of to the treatment different all three data using a The and were with peptides. However, the normalization the in both data sets. is that a number of peptides the of the However, on the of peptides for normalization or for the of normalization methods be with the display a different type of compared with the of the peaks in the data The of analysis order-dependent be to as of the in the in the or a of and other The that different were in the and data sets and in the data that the the of a of peaks between LC-MS analyses, be to an increase a of the some individual peaks there were of which has also found by R. for LC-MS proteomics Proteome Res. 7: PubMed Scopus Google whereas for other peaks there an in some peaks a of found a A sample between each block of as a for a between each The in be with in the data set there were large differences in the between compared with the and data sets. is to that normalization will but in the and data the of the technical on other of only the content of the a single technical of each sample the number of peaks for each biological The be as as has used to protein each peptide is and to protein R. for LC-MS proteomics Proteome Res. 7: PubMed Scopus Google Scholar, analysis of mass data from complex using Proteome Res. 7: PubMed Scopus Google Scholar). used to the fraction of differentially expressed peaks using and data. A normalization method on the fraction of differentially expressed peaks is that the fraction of is However, in the of this is a common approach for the of different normalization methods P. M. methods for of and Res. 2007; 35: PubMed Scopus Google Scholar, J. normalization for the 2007; 1: PubMed Google Scholar). The largest fraction of differentially expressed peaks for all three data set found using the method and data the in differentially expressed peaks compared with data including the block in the the fraction of differentially expressed peaks found to be for and data compared with data. the to be within each in this type of data the analysis order is The of also to the but the this for using the RegrRun method there will be a of large LC-MS All three different data sets in this large differences in the fraction of with the number of peaks the and data sets whereas the data set large differences This is also be from a biological of the data set different were The that a large fraction of all peaks have an on the of the normalization procedure only a fraction of the peaks is to be differentially expressed between this the performances of 10 different normalization methods on endogenous peptide data generated with LC-MS were The methods and are for method and The novel method RegrRun for both differences between LC-MS analyses and analysis order-dependent and decreased the median S.D. by and 38% compared with raw both and including biological The levels of peaks in a set of LC-MS analyses display analysis order-dependent bias. Only the RegrRun method this type of bias. normalization methods on a only reduced the median S.D. by