Covariation of Peptide Abundances Accurately Reflects Protein Concentration Differences

Most implementations of mass spectrometry-based proteomics involve enzymatic digestion of proteins, expanding the analysis to multiple proteolytic peptides for each protein. Currently, there is no consensus of how to summarize peptides' abundances to protein concentrations, and such efforts are complicated by the fact that error control normally is applied to the identification process, and do not directly control errors linking peptide abundance measures to protein concentration. Peptides resulting from suboptimal digestion or being partially modified are not representative of the protein concentration. Without a mechanism to remove such unrepresentative peptides, their abundance adversely impacts the estimation of their protein's concentration. Here, we present a relative quantification approach, Diffacto, that applies factor analysis to extract the covariation of peptides' abundances. The method enables a weighted geometrical average summarization and automatic elimination of incoherent peptides. We demonstrate, based on a set of controlled label-free experiments using standard mixtures of proteins, that the covariation structure extracted by the factor analysis accurately reflects protein concentrations. In the 1% peptide-spectrum match-level FDR data set, as many as 11% of the peptides have abundance differences incoherent with the other peptides attributed to the same protein. If not controlled, such contradicting peptide abundance have a severe impact on protein quantifications. When adding the quantities of each protein's three most abundant peptides, we note as many as 14% of the proteins being estimated as having a negative correlation with their actual concentration differences between samples. Diffacto reduced the amount of such obviously incorrectly quantified proteins to 1.6%. Furthermore, by analyzing clinical data sets from two breast cancer studies, our method revealed the persistent proteomic signatures linked to three subtypes of breast cancer. We conclude that Diffacto can facilitate the interpretation and enhance the utility of most types of proteomics data. Most implementations of mass spectrometry-based proteomics involve enzymatic digestion of proteins, expanding the analysis to multiple proteolytic peptides for each protein. Currently, there is no consensus of how to summarize peptides' abundances to protein concentrations, and such efforts are complicated by the fact that error control normally is applied to the identification process, and do not directly control errors linking peptide abundance measures to protein concentration. Peptides resulting from suboptimal digestion or being partially modified are not representative of the protein concentration. Without a mechanism to remove such unrepresentative peptides, their abundance adversely impacts the estimation of their protein's concentration. Here, we present a relative quantification approach, Diffacto, that applies factor analysis to extract the covariation of peptides' abundances. The method enables a weighted geometrical average summarization and automatic elimination of incoherent peptides. We demonstrate, based on a set of controlled label-free experiments using standard mixtures of proteins, that the covariation structure extracted by the factor analysis accurately reflects protein concentrations. In the 1% peptide-spectrum match-level FDR data set, as many as 11% of the peptides have abundance differences incoherent with the other peptides attributed to the same protein. If not controlled, such contradicting peptide abundance have a severe impact on protein quantifications. When adding the quantities of each protein's three most abundant peptides, we note as many as 14% of the proteins being estimated as having a negative correlation with their actual concentration differences between samples. Diffacto reduced the amount of such obviously incorrectly quantified proteins to 1.6%. Furthermore, by analyzing clinical data sets from two breast cancer studies, our method revealed the persistent proteomic signatures linked to three subtypes of breast cancer. We conclude that Diffacto can facilitate the interpretation and enhance the utility of most types of proteomics data. Mass spectrometry-based proteomics is the preferred technology for quantitative and comprehensive analysis of proteins in complex biological mixtures (1.Ong S.E. Mann M. Mass spectrometry-based proteomics turns quantitative.Nat. Chem. Biol. 2005; 1: 252-262Crossref PubMed Scopus (1317) Google Scholar). Because a typical experiment involves proteolytic digestion, the actual analytes measured by liquid chromatography-tandem mass spectrometry (LC-MS/MS) 1The abbreviations used are: LC-MS/MS, liquid chromatography coupled to tandem mass spectrometry;ANOVA, analysis of variance;CV, coefficient of variation;CPTAC, Clinical Proteomic Tumor Analysis Consortium;DDA, data-dependent acquisition;ERPR, estrogen or progesterone receptor;FARMS, Factor Analysis for Robust Microarray Summarization;FDR, false discovery rate;FQR, false quantification rate;HER2, human epidermal growth factor receptor 2;iTRAQ, isobaric tag for relative and absolute quantitation;LFQ, label-free quantification;LOQ, limit of quantification;MC, Monte Carlo method;FDRMC, false discovery rate based on sequential Monte Carlo simulation;MPIB, Max Planck Institute of Biochemistry;PECA, Probe-level Expression Change Averaging;PQPQ, protein quantification by peptide quality control;PSM, peptide spectrum match;RT, retention time;SILAC, stable isotope labeling by amino acids in cell culture;S/N, signal-to-noise ratio;SpC, spectral counting;TN, triple-negative;XIC, extracted ion chromatography. 1The abbreviations used are: LC-MS/MS, liquid chromatography coupled to tandem mass spectrometry;ANOVA, analysis of variance;CV, coefficient of variation;CPTAC, Clinical Proteomic Tumor Analysis Consortium;DDA, data-dependent acquisition;ERPR, estrogen or progesterone receptor;FARMS, Factor Analysis for Robust Microarray Summarization;FDR, false discovery rate;FQR, false quantification rate;HER2, human epidermal growth factor receptor 2;iTRAQ, isobaric tag for relative and absolute quantitation;LFQ, label-free quantification;LOQ, limit of quantification;MC, Monte Carlo method;FDRMC, false discovery rate based on sequential Monte Carlo simulation;MPIB, Max Planck Institute of Biochemistry;PECA, Probe-level Expression Change Averaging;PQPQ, protein quantification by peptide quality control;PSM, peptide spectrum match;RT, retention time;SILAC, stable isotope labeling by amino acids in cell culture;S/N, signal-to-noise ratio;SpC, spectral counting;TN, triple-negative;XIC, extracted ion chromatography. are the proteolytic peptides of the analyzed proteins. Inferring the identity of proteins that were present in the original mixture before digestion is problematic, especially when proteins are homologs. This cannot be solved by increasing the mass accuracy of peptide and mass of and of mass Mass Scopus Google Scholar). Currently, there is no consensus how such protein be of for protein identification using tandem mass PubMed Google M. M. for protein false discovery rate estimation in proteomic data PubMed Scopus Google to quantitative proteomic Chem. PubMed Scopus Google Scholar). when relative protein from multiple of peptides. is that the peptide abundances are to their protein's concentration Mann M. Mass spectrometry-based proteomics in cell Biol. PubMed Scopus Google Scholar). is to a protein's concentration by the average or of peptides' abundances Mann M. modified protein abundance for estimation of absolute protein amount in proteomics by the of peptides 2005; PubMed Scopus Google quantification of proteins by a of PubMed Scopus Google quantification of complex mass spectrometry data for proteomic PubMed Scopus Google Scholar). the peptide mixture from protein is in the measured peptide abundances of many can the of peptides be to enzymatic or the of the and the with peptides from other or to be measured in of the experiments M. mass spectrometry in from to the Chem. PubMed Scopus Google Scholar). for many proteins, the quantitative data on peptides are and protein quantification and analysis in label-free experiments with complex PubMed Scopus Google Mann M. label-free quantification by and peptide PubMed Scopus Google M. for analysis of quantitative mass spectrometry-based proteomic PubMed Scopus Google or a analysis to the to quantitative proteomic Chem. PubMed Scopus Google for the analysis of tandem mass spectrometry PubMed Scopus Google proteomics data for PubMed Scopus Google Scholar). most do not of the covariation of peptide abundances measured in peptides that a covariation with the other peptides from the same can of the such other have to the of protein and protein quantification of proteomic Scopus Google using peptide quantification PubMed Scopus Google and in data-dependent quantitative label-free PubMed Scopus Google or a for peptides for quantitative analysis M. from proteomics data by protein quantification and peptide quality control PubMed Scopus Google a for identification and in proteomics PubMed Scopus Google Scholar). have in of quantification or the with and incorrectly peptides as when each in the with peptide in in the of are the technology measures multiple of the protein quantification and analysis in label-free experiments with complex PubMed Scopus Google M. M. Expression by to PubMed Scopus Google M. between and quantitative proteomics PubMed Scopus Google Scholar). in have proteomics to a is to that of Mann M. Mass spectrometry-based proteomics in cell Biol. PubMed Scopus Google Mann M. proteomics the PubMed Scopus Google M. and human analysis by PubMed Scopus Google Scholar). the sets of in a experiment and for Scopus Google the of the is multiple for peptide to be attributed to the protein and be by in a false of as is to that are for the actual between peptide identification and quantification be as in our data PubMed Scopus Google have a to be with peptide In the rate of false between peptide identity and peptide not and is the increasing in proteomics studies, the impact of false a the of the with multiple a in analysis for and a set of have for Scopus Google of for in 2005; PubMed Scopus Google summarization method for PubMed Scopus Google Scholar). We that from analysis can be applied in proteomics to quantification In we a analysis that we factor analysis summarization method for PubMed Scopus Google for the of a for PubMed Scopus Google in to incoherent The factor analysis by the covariation multiple experiments of abundances of a of peptides linked to a to the factor analysis to the between the the the is by In factor analysis the of and is The the protein concentration that is extracted from the of multiple The signal-to-noise is estimated for of peptides attributed to a to is or to The of peptides incoherent peptides of other peptides. peptides are from the before the relative in protein concentration as a weighted of differences in abundance of the peptides. and incoherent peptide Diffacto the quantitative to extract biological from the same proteomics data We that Diffacto is a and method for for proteomics data PubMed Scopus Google Scholar). mass to by by three and analyzed in experiments with the to were with spectrum in a or for data-dependent of tandem mass This of mass using peptide for label-free Chem. PubMed Scopus Google the of between the three peptide identification by of peptides M. for identification of peptides in data-dependent tandem mass PubMed Scopus Google in the three were set to and for set to and The of and were based on M. and human analysis by PubMed Scopus Google and the the of in the from of human cell cell and were The of human peptides reduced the of peptides and the of peptides that of peptides. In each of peptide mixture to a of were three in a experiment peptides and data were to and We peptides using the M. for identification of peptides in data-dependent tandem mass PubMed Scopus Google in the were by the with the using the by M. proteomics PubMed Scopus Google Scholar). with the original and were in a a for protein PubMed Scopus Google protein of human protein of and the of using proteomics for tandem mass PubMed Scopus Google Scholar). of set as a and of as a The applied and and mass were set to and The resulting were by for each identification and quantification were the data PubMed Scopus Google in peptide were from the and with using M. proteomics PubMed Scopus Google Scholar). the applied and mass to to the with the of and to a used to and a consensus The consensus the for the identity peptide were from with to the the of quantitative a ion applied by from were to the the were not by the consensus mass a retention and the consensus and were estimated FDR applied as a quality for data PubMed Scopus Google Scholar). If a consensus linked to with the most abundances were as a of abundances from and of the and by the average of of peptide abundances for each identification and quantification were from the of two clinical mass spectrometry data. The breast cancer data set from the M. M. to in breast PubMed Scopus Google Scholar). This set in a as the original to the of and breast cancer control were by estimation of mixture and by the of the Peptides quantified in no were The breast cancer data set from the original the Max Planck Institute of Mann M. Proteomic of breast cancer PubMed Scopus Google Scholar). of breast cancer with a standard of mixture of breast cancer cell were and by estimation of mixture the original data we the of by the negative the were in with the data. Peptides quantified in were measured ion abundances of a protein's peptide do not directly to the actual amount of the original protein of the of proteolytic digestion and the abundances of peptides be to the protein we are of the peptides. The of peptides' to the in protein the of in complex estimated to of and the of in mass Mass PubMed Scopus Google Scholar). the and the limit of a quantitative of a proteolytic peptide a that is by the the protein concentration and the error of peptides with are and used to the protein concentration in the M. to protein PubMed Scopus Google Scholar). This estimation be when is to and is of the concentration for each of most the peptides. in the of peptides and the error of on the peptide concentration in most proteomics Most the in proteomics is to relative in protein concentration between biological In such the of the actual protein between are of the can be as a of two and the estimation when the error is the factor is The error is to be of a that the for factor analysis and in peptide abundances were by to a the peptide abundances from the mixture were used as the the data for peptide abundances in the data were in relative used factor analysis from summarization method for PubMed Scopus Google as a The in the factor analysis that the peptide digestion were to the protein concentration in the on the for a relative protein concentration in of of abundances to as from peptides be as to the in the the of peptide and the for the by errors of that and are factor analysis can be with as a factor and as Furthermore, the a Here, for the peptides' a is to the a estimation of that the covariation of abundances and to for protein the by the and for the set of peptides or for the of a for PubMed Scopus Google Scholar). can be in the summarization method for PubMed Scopus Google and for the of a for PubMed Scopus Google Scholar). The factor can be as peptides' to protein concentration and used as as as quality control in Peptides weighted of the were or and were from This quality of peptides, the of protein the of the relative abundance of a protein is as a weighted a weighted in of abundances of peptides for the of experiments The of each peptide by the factor from the This method the by peptides' to the in abundances the and when in a for the applied when a of peptide from In that in that were with of the abundance for the set of peptides. on the same as in the factor a protein in with experiments and peptides by the factor analysis we the estimated relative abundances and the average abundance for peptides in The of the of and the of can be as of in of be from the of of in the can be as This in with of be for the having the same peptide such a is the of quantification errors be of the covariation of peptides, the of typical is having by samples. In that the not have the same the In to control the of the protein the protein two types of were applied to the of the Monte Carlo multiple PubMed Scopus Google and the proteomics data for PubMed Scopus Google Scholar). Monte Carlo multiple Monte Carlo multiple PubMed Scopus Google applied when the is to of of of of for each of Monte Carlo The the for each protein based on the with the as the of that or that with the and the estimated abundances from the weighted and as the of for the be as each of a set of estimated based on and a estimation of of Robust estimation of the false discovery PubMed Scopus Google Scholar). The when for a or of proteins are the FDR of the proteomics data for PubMed Scopus Google on the other the for each of the peptides of a protein. on the set the for the peptide The can be by the for the the the of the with a the by the of and the of the for the were estimated using the same as in the we the not be that to the the protein be applied with The each protein by the abundance of three most abundant peptides peptides when having three This is applied as the with the that peptides M. to protein PubMed Scopus Google Scholar). The used the of peptide to of protein is applied in with and using labeling The from the a for identification and in proteomics PubMed Scopus Google by Peptides abundances were with for each peptides were based on the of correlation using the with for method and The from as the representative of the and by the average The in Mann M. label-free quantification by and peptide PubMed Scopus Google Mann M. enables peptide identification mass and protein PubMed Scopus Google the as the data with the and peptide the of between 1% FDR of and FDR of proteins were abundances for each were by the abundances from the three quantified in were the impact of false peptide on the protein the correlation between the concentration and the protein quantification each the of abundance from the were correlation between the and the protein quantification were with negative correlation the were as false quantifications. data of the experiments from were from the and as in data PubMed Scopus Google Scholar). of we used M. for of tandem mass PubMed Scopus Google that the peptide Mass PubMed Scopus Google to peptide directly from the with mass and mass of set as a and of as a The protein a for protein PubMed Scopus Google for the peptides using protein The were by and using of the peptide and the identity quality peptides were as and were to the by the peptides were to be with the same of the of In each to have multiple to the proteins that are from the identification from the of the proteins were and were as the same protein with the identification of a the quantification peptide that to proteins and were quantified in of the Diffacto analysis based on the resulting peptide with FDR based on Monte Carlo and for Diffacto are for data analysis and in Mass PubMed Scopus Google and M. M. M. in Scholar). The of Diffacto is to proteins that have between and to such each to a the of peptide abundances were to a relative The of abundance the of peptides for each the in based on the peptide abundance be a of two the to the relative of protein concentration the by peptide for a each peptide be weighted by a on the covariation between each other peptides and the estimated the between and a factor analysis be with as a factor and as the to the of each peptide and the of the of protein concentration We a factor analysis summarization method for PubMed Scopus Google as a The method for the analysis of we the method to of abundance from a set of peptides that are linked to The factor analysis the of measured peptide abundances to the protein concentration the and the errors of the a This a a estimation of factor to the of peptide the a for set of peptides by the to the We a of to the sets of peptides two and for the of a for PubMed Scopus Google Scholar). sets were from the analysis of proteins. We the factor for the sets of peptides and a between and we applied of for peptides and peptides that are incoherent to the estimated The and not or or a that for with complex using as we the relative protein abundances for each of each by the weighted of the relative peptide abundances. the of of peptides in to protein concentration we a set of label-free experiments with mixtures of and standard in We quantified in peptides that were attributed to proteins proteins by peptide identity in data PubMed Scopus Google quantification rate of The of the peptide quantification in three experiments a we the data set using peptides and proteins that the of quantified in for peptides or for The peptide quantification of three as as in the We the in the based on the abundances of peptides from that of in the mixtures We no limit of quantification for the amount of The from most of the peptides that the not for of the with concentration abundance of the we a that the of the the in the protein concentrations. each we the covariation structure of peptides' abundances and of of the proteins sets of peptides the of the proteins a on the of peptides, that protein the of quantification in the the proteins, two peptides proteins three peptides In the sets of peptides, the covariation extracted by the factor analysis as the quality 11% of the peptides for the quantitative summarization of their incoherent The quantification in Diffacto, weighted the in label-free quantification by are by and quantification a we between the estimated abundances of the proteins in mixtures and the actual of human or cell and for a of the peptides' concentration the of the from the of quantification errors such to the This by of the abundance to the in a of the of a set to be the rate for that of samples. When using peptide the for to to the of in the Diffacto proteins, and the between the abundances and the actual the rate of by weighted of peptide abundances. The and of relative of human proteins and proteins in quantification using weighted to the actual protein concentration in the proteins to have quantification errors with of the proteins, of on average of peptides protein. of three other are in Because the human and the were two in the the two a for to the false quantification rate We as the of protein with the actual and The to be for proteins by such proteins, the estimated the of not the that on average a proteins quantified by two peptides, the to the the proteins with three peptides, we the to be before and We the of Diffacto other quantification of most abundant of quantified and M. from proteomics data by protein quantification and peptide quality control PubMed Scopus Google a method that peptide correlation for and quality proteins based on of their peptides' a that to is in to quantification a for the proteins of using the same set of peptide Diffacto not proteins, and the other with peptide peptides by of correlation a not by the estimation error of reduced the with and especially when having a of peptides. for proteins with peptides, of the quantified In and Diffacto the of false proteins having by the The quantified proteins of 1.6%. This the of multiple peptide abundances protein. the of Diffacto in we two data sets from clinical breast cancer studies, by the Clinical Proteomic Tumor Analysis M. M. to in breast PubMed Scopus Google and by the Max Planck Institute of Mann M. Proteomic of breast cancer PubMed Scopus Google Scholar). The be as a of clinical and peptides on The on the other a in were a standard mixture of breast cancer cell using the we the same as in to the three estrogen or progesterone receptor epidermal growth factor receptor and We the quantified peptides by of protein We in proteins in the data as with the and estimated that protein of the were between the cancer subtypes and This the proteins estimated by using that the FDR control by Monte Carlo In the we proteins of the their as protein This can be with the proteins we in the data proteins of the and the differences between the subtypes of cancer as in the data. the same of Diffacto proteins the proteins by as many as the proteins by the of to the protein concentration we that Diffacto a set of proteins that in the original of the original proteins were as We and the proteins in the that were not by Diffacto, protein that of many proteins of the by were and proteins were estimated not as estimated by the error control by the we the of the protein abundances between the and the of proteins by Diffacto, we proteins in and were by correlation of protein between the and data sets In for the proteins that were to the original the correlation of protein between the data the proteins were We three M. of breast cancer based on PubMed Scopus Google M. M. to in breast PubMed Scopus Google and cancer by proteomic of PubMed Scopus Google with the and and the the by the of and as that protein quantification can be with a suboptimal identification we the data from the M. M. of abundant proteins in label-free quantitative PubMed Scopus Google for In the proteins were with with proteins, and were measured by We no the proteins or the we used peptide by protein the This each of the with multiple and the proteins using such identification process, Diffacto and quantified proteins two of concentration differences and were not quantified in the relative of the protein and quantification a of correlation The signal-to-noise of the proteins the of the quantification based on of the peptide most proteins in the data. protein we used peptide by protein the such a identification Diffacto to and representative protein concentration error rate of relative protein by the average abundances for proteins, or by the peptide abundances for of proteins in the not in the are by a concentration error rate of Diffacto relative protein by the average abundances for proteins, or by the peptide abundances for of proteins in the not in the are by a concentration in a mass spectrometry proteomics by a of and quantitative of proteolytic peptides, with of Currently, protein quantification being by the identification of mass spectrometry and that the to quantitative accuracy is to the identification not be the as we the in in of do not the identification to in PubMed Scopus Google quantitative Furthermore, quantification such as spectral have a limit to their accuracy of the of and the average of peptide we not the of such in we do not using Diffacto with data. The of the in the the the we have in the peptide abundances to The of peptides' abundances by the factor analysis method in Diffacto not a quality control a weighted for peptide and in data-dependent quantitative label-free PubMed Scopus Google In of label-free proteomics and accuracy of PubMed Scopus Google that of peptide be applied in for protein quantification This with a of that the of peptides and the of protein be to Diffacto to summarize protein concentration based on that of peptides, such as we applied in the breast cancer protein and We a in of our the of PubMed Scopus Google to the of using peptide and to the of data the data set is to a comprehensive we and quantitative accuracy for the proteins and In the approach, the of the protein abundance differences is from be by the with the most In the of not of such of data analysis for in proteomics and the is is not that to of proteins in proteomics controlled by from to many false in a analysis for PubMed Scopus Google FDR or have the in identification by M. M. for protein false discovery rate estimation in proteomic data PubMed Scopus Google to in PubMed Scopus Google M. to false discovery in PubMed Scopus Google Scholar). in proteomics are in and to the of multiple in quantitative PubMed Scopus Google in We a clinical data set of cancer M. M. analysis of label-free and for quantitative proteomic PubMed Scopus Google as many proteins of the cancer be by Diffacto from and data and of the we not This a of how many experiments are in to a Diffacto analysis with a FDR In three experiments are the in to are for and are the for a the increasing in we the when be by in quantification is to that we how to the of the false of peptides and proteins. protein quantification such as the in the with FDR and by the of in and data are to and of Diffacto is We of for on of We The the of for on the with with