Bayesian Proteoform Modeling Improves Protein Quantification of Global Proteomic Measurements

As the capability of mass spectrometry-based proteomics has matured, tens of thousands of peptides can be measured simultaneously, which has the benefit of offering a systems view of protein expression. However, a major challenge is that, with an increase in throughput, protein quantification estimation from the native measured peptides has become a computational task. A limitation to existing computationally driven protein quantification methods is that most ignore protein variation, such as alternate splicing of the RNA transcript and post-translational modifications or other possible proteoforms, which will affect a significant fraction of the proteome. The consequence of this assumption is that statistical inference at the protein level, and consequently downstream analyses, such as network and pathway modeling, have only limited power for biomarker discovery. Here, we describe a Bayesian Proteoform Quantification model (BP-Quant) 1BP-QuantBayesian proteoform quantificationLC-MSliquid chromatography mass spectrometryNETnormalized elution time (NET)PQPQprotein quantitation by peptide quality control.1BP-QuantBayesian proteoform quantificationLC-MSliquid chromatography mass spectrometryNETnormalized elution time (NET)PQPQprotein quantitation by peptide quality control. that uses statistically derived peptides signatures to identify peptides that are outside the dominant pattern or the existence of multiple overexpressed patterns to improve relative protein abundance estimates. It is a research-driven approach that utilizes the objectives of the experiment, defined in the context of a standard statistical hypothesis, to identify a set of peptides exhibiting similar statistical behavior relating to a protein. This approach infers that changes in relative protein abundance can be used as a surrogate for changes in function, without necessarily taking into account the effect of differential post-translational modifications, processing, or splicing in altering protein function. We verify the approach using a dilution study from mouse plasma samples and demonstrate that BP-Quant achieves similar accuracy as the current state-of-the-art methods at proteoform identification with significantly better specificity. BP-Quant is available as a MatLab® and R packages. As the capability of mass spectrometry-based proteomics has matured, tens of thousands of peptides can be measured simultaneously, which has the benefit of offering a systems view of protein expression. However, a major challenge is that, with an increase in throughput, protein quantification estimation from the native measured peptides has become a computational task. A limitation to existing computationally driven protein quantification methods is that most ignore protein variation, such as alternate splicing of the RNA transcript and post-translational modifications or other possible proteoforms, which will affect a significant fraction of the proteome. The consequence of this assumption is that statistical inference at the protein level, and consequently downstream analyses, such as network and pathway modeling, have only limited power for biomarker discovery. Here, we describe a Bayesian Proteoform Quantification model (BP-Quant) 1BP-QuantBayesian proteoform quantificationLC-MSliquid chromatography mass spectrometryNETnormalized elution time (NET)PQPQprotein quantitation by peptide quality control.1BP-QuantBayesian proteoform quantificationLC-MSliquid chromatography mass spectrometryNETnormalized elution time (NET)PQPQprotein quantitation by peptide quality control. that uses statistically derived peptides signatures to identify peptides that are outside the dominant pattern or the existence of multiple overexpressed patterns to improve relative protein abundance estimates. It is a research-driven approach that utilizes the objectives of the experiment, defined in the context of a standard statistical hypothesis, to identify a set of peptides exhibiting similar statistical behavior relating to a protein. This approach infers that changes in relative protein abundance can be used as a surrogate for changes in function, without necessarily taking into account the effect of differential post-translational modifications, processing, or splicing in altering protein function. We verify the approach using a dilution study from mouse plasma samples and demonstrate that BP-Quant achieves similar accuracy as the current state-of-the-art methods at proteoform identification with significantly better specificity. BP-Quant is available as a MatLab® and R packages. Bayesian proteoform quantification liquid chromatography mass spectrometry normalized elution time (NET) protein quantitation by peptide quality control. Bayesian proteoform quantification liquid chromatography mass spectrometry normalized elution time (NET) protein quantitation by peptide quality control. The application of MS-based proteomics has resulted in large-scale studies in which the set of measured, and subsequently identified, peptides is often used to estimate protein abundance. In particular, label-free MS-based proteomics is highly effective for identification of peptides and measurement of relative peptide abundances (1Baker E.S. Liu T. Petyuk V.A. Burnum-Johnson K.E. Ibrahim Y.M. Anderson G.A. Smith R.D. Mass spectrometry for translational proteomics: progress and clinical implications.Genome Med. 2012; 4: 63Crossref PubMed Scopus (58) Google Scholar, 2Gillette M.A. Carr S.A. Quantitative analysis of peptides and proteins in biomedicine by targeted mass spectrometry.Nat. Methods. 2013; 10: 28-34Crossref PubMed Scopus (359) Google Scholar), but it does not directly yield protein quantities. The importance of accurate protein quantification cannot be understated; it is the essential component of identifying biomarkers of disease or defining the relationship between gene regulations, protein interactions, and signaling networks in a cellular system (3Smith L.M. Kelleher N.L. Proteoform: a single term describing protein complexity.Nat. Methods. 2013; 10: 186-187Crossref PubMed Scopus (884) Google Scholar, 4Waters K.M. Pounds J.G. Thrall B.D. Data merging for integrated microarray and proteomic analysis.Brief Funct. Genomic Proteomic. 2006; 5: 261-272Crossref PubMed Scopus (88) Google Scholar). The major challenge is that protein abundance depends not only on transcription rates of the gene but also on additional control mechanisms, such as mRNA stability, translational regulation, and protein degradation. Moreover, the functional activity of proteins can be altered through a variety of post-translational modifications or proteolytic processing and alternative splicing, events which selectively alter the abundance of some selected peptides while leaving others unchanged (4Waters K.M. Pounds J.G. Thrall B.D. Data merging for integrated microarray and proteomic analysis.Brief Funct. Genomic Proteomic. 2006; 5: 261-272Crossref PubMed Scopus (88) Google Scholar). This complexity of the proteome, in addition to issues associated with the measurement and identification errors, presents a significant challenge to accurate relative protein quantification (5Goh W.W. Lee Y.H. Chung M. Wong L. How advancement in biological network analysis methods empowers proteomics.Proteomics. 2012; 12: 550-563Crossref PubMed Scopus (51) Google Scholar). Smith et al. (3Smith L.M. Kelleher N.L. Proteoform: a single term describing protein complexity.Nat. Methods. 2013; 10: 186-187Crossref PubMed Scopus (884) Google Scholar) recently described the importance of capturing protein variation in all forms (e.g. post-translational modifications, splice variants), all of which are collectively referred to as proteoforms. To date, little has been described in respect to automated identification of proteoforms. Most work on improving protein quantification for label-free data focuses on removing variation from the data through peptide filtering, such as removing shared peptides or those that do not meet frequency and coefficient of variation (CV) thresholds (6Lai X. Wang L. Tang H. Witzmann F.A. A novel alignment method and multiple filters for exclusion of unqualified peptides to enhance label-free quantification using peptide intensity in LC-MS/MS.J. Proteome Res. 2011; 10: 4799-4812Crossref PubMed Scopus (43) Google Scholar). Several recent methodologies have described approaches to deal with shared peptides either through linear programming (7Dost B. Bandeira N. Li X. Shen Z. Briggs S.P. Bafna V. Accurate mass spectrometry based protein quantification via shared peptides.J. Comput. Biol. 2012; 19: 337-348Crossref PubMed Scopus (29) Google Scholar), hierarchical modeling (8Blein-Nicolas M. Xu H. de Vienne D. Giraud C. Huet S. Zivy M. Including shared peptides for estimating protein abundances: A significant improvement for quantitative proteomics.Proteomics. 2012; 12: 2797-2801Crossref PubMed Scopus (19) Google Scholar), or peptide detectability (9Tang H. Arnold R.J. Alves P. Xun Z. Clemmer D.E. Novotny M.V. Reilly J.P. Radivojac P. A computational approach toward label-free protein quantification using predicted peptide detectability.Bioinformatics. 2006; 22: e481-e488Crossref PubMed Scopus (143) Google Scholar) to improve protein-level quantification. However, these approaches do not identify or quantify distinct proteoforms of a protein. The current state-of-the-art method for proteoform discovery uses a combination of correlation and clustering to identify distinct patterns (10Forshed J. Protein quantification by peptide quality control (PQPQ) of shotgun proteomics data.Methods Mol. Biol. 2013; 1023: 149-158Crossref PubMed Scopus (3) Google Scholar). Protein quantification by peptide quality control (PQPQ) to date has only been applied to labeled data but is generic and can be applied to label-free experiments as well. Generating a parsimonious protein list is a well-known practice in protein inference. The procedure involves assembling the most concise set of proteins across the assigned peptide sequences observed in an experiment (11Zhang B. Chambers M.C. Tabb D.L. Proteomic parsimony through bipartite graph analysis improves accuracy and transparency.J. Proteome Res. 2007; 6: 3549-3557Crossref PubMed Scopus (264) Google Scholar). We present the novel approach of statistically informed peptide selection using a Bayesian proteoform quantification (BP-Quant) approach for parsimonious, relative protein quantification (Fig. 1). Research objectives are at the foundation of BP-Quant. Peptides to be used for protein quantification are selected using a peptide-specific signature vector that is determined by statistical hypothesis tests relevant to the research objective(s) of the experiment as well as the directionality of the effect(s) (i.e. upward or downward abundance). The statistical signature-based approach does not omit shared peptides, although it does not directly account for the fact that a peptide is shared. BP-Quant-based protein quantification is a multi-step process that (1) generates the peptide signature vectors based on the research hypotheses, (2) assigns probabilities to potential proteoform configurations, and (3) selects peptides with an over-representation of a similar signature to be used for relative protein abundance estimates. Discovery of candidate plasma biomarkers is of interest to many applications, such as disease specificity, drug toxicity, drug response, and fundamental research (12Senapati S. Barnhart K.T. Biomarkers for ectopic pregnancy and pregnancy of unknown location.Fertil. Steril. 2013; 99: 1107-1116Abstract Full Text Full Text PDF PubMed Scopus (29) Google Scholar, 13Chung L. Baxter R.C. Breast cancer biomarkers: proteomic discovery and translation to clinically relevant assays.Expert Rev. Proteomics. 2012; 9: 599-614Crossref PubMed Scopus (22) Google Scholar, 14Galasko D. Golde T.E. Biomarkers for Alzheimer's disease in plasma, serum and blood - conceptual and practical problems.Alzheimers Res. Ther. 2013; 5: 10Crossref PubMed Scopus (49) Google Scholar, 15Pin E. Fredolini C. Petricoin 3rd, E.F. The role of proteomics in prostate cancer research: biomarker discovery and validation.Clin. Biochem. 2013; 46: 524-538Crossref PubMed Scopus (51) Google Scholar). Plasma samples collected from standard inbred mice under the National Institutes of Health National Institute of Environmental Health Sciences Biomarkers of Exposure project (http://www.niehs.nih.gov/health/topics/science/biomarkers/) were used for a dilution study. Although the focus of this larger project was to understand the development and progression of complex disease due to exposure to environmental stressors in the presence of risk factors such as obesity or exposure to inhaled endotoxins (e.g. lipopolysaccharide) (16Tilton S.C. Waters K.M. Karin N.J. Webb-Robertson B.J. Zangar R.C. Lee K.M. Bigelow D.J. Pounds J.G. Corley R.A. Diet-induced obesity reprograms the inflammatory response of the murine lung to inhaled endotoxin.Toxicol. Appl. Pharmacol. 2013; 267: 137-148Crossref PubMed Scopus (13) Google Scholar), here we constructed an experiment that yields expected protein ratios and for which known concentrations can be used to define datasets with positive examples (proteins with multiple proteoforms) and negative examples (proteins with a single proteoform). Mouse plasma samples (n = 16) were previously depleted of the seven most abundant proteins using an IgY7 depletion column and digested using the approach described in Zhou et al. (17Zhou J.Y. Petritis B.O. Petritis K. Norbeck A.D. Weitz K.K. Moore R.J. Camp D.G. Kulkarni R.N. Smith R.D. Qian W.J. Mouse-specific tandem IgY7-SuperMix immunoaffinity separations for improved LC-MS/MS coverage of the plasma proteome.J. Proteome Res. 2009; 8: 5387-5395Crossref PubMed Scopus (15) Google Scholar). For this dilution the digested mouse plasma samples were with an of digested to a peptide for of the depleted and digested mouse plasma peptide samples and S. peptide were with to the peptide The mouse plasma samples were to (1) S. (2) S. (3) S. and S. was and to a of using samples were with to peptide samples = samples by in were and across a system to a mass by of an as previously described Tang K. M.A. Shen D.J. Moore R.J. Smith R.D. automated system for in proteomic PubMed Scopus Google Scholar). were using as previously described Moore R.J. D.J. Tang K. Smith R.D. and for mass 2006; PubMed Scopus Google Scholar). were by into a using for D. J. for in proteomics.Proteomics. 5: PubMed Scopus Google Scholar). of in and in and were using a the system was at with for of peptides the the was at for elution was with an column of and the was from to using a by a increase to for to the The of the and the were and Data were collected the mass Quantitative data were using the Data system N. E.S. N. Anderson G.A. Smith R.D. a data system for proteomics.Proteomics. 2006; 6: PubMed Scopus Google Scholar), which is a of (e.g. N. K. Anderson G.A. Smith R.D. for automated processing and of mass spectrometry 2009; 10: PubMed Scopus Google Scholar) and N. N. Smith R.D. an to peptide 2007; PubMed Scopus Google Scholar). The of the data to the and intensity of the major in mass N. K. Anderson G.A. Smith R.D. for automated processing and of mass spectrometry 2009; 10: PubMed Scopus Google Scholar). The data were in a to identify of mass observed in using an that a in for of N. N. Smith R.D. an to peptide 2007; PubMed Scopus Google Scholar). to and referred to as an has a normalized elution time and abundance estimate by the of the that the The peptide of in a is to a single was determined by measured and to the and observed of of the peptides in a mouse accurate mass and time Qian W.J. Smith R.D. in proteomics data analysis and using an accurate mass and time Rev. 2006; PubMed Scopus Google Scholar) of and for mass and elution The peptides from this process were as a for data analysis and are available in in data and at abundance data were to the and data were as to Peptides with an of data to a or quantification statistical (e.g. or of across the set of biological were B.J. Waters K.M. Pounds J.G. statistical of peptide and peptide improves identification of significant peptides from MS-based proteomics Proteome Res. 9: PubMed Scopus Google Scholar). The of mouse plasma and S. peptides were for of peptide abundance across the of Waters K.M. Pounds J.G. Webb-Robertson B.J. quality control processing of proteomics 2011; PubMed Scopus Google Scholar). The peptides associated with the mouse plasma were were and additional peptide was to data across the of The mouse plasma data were normalized using a the samples of only mouse peptides = 16) were using a peptide B.J. Pounds J.G. Waters K.M. A statistical selection for in proteomics experiments through of 2011; PubMed Scopus Google Scholar). The dilution and were normalized as a of the expected dilution to the normalized peptide set of only mouse presents an of the BP-Quant approach to protein quantification. the research for identifying proteins with statistically significant differential (i.e. between control and and peptide as an of and using an statistical such as an with a to for the multiple the peptide L. M. to and Data Scholar). In the statistical are into signatures based on the or and the has or the and statistical in peptide abundance. In particular, the significant are for such as (1) control (2) control and (3) In the signature vector is based on the of quantitative peptide abundance across are defined by the research A signature vector can also be defined based on the of the frequency of response (i.e. across which is highly relevant for proteomics data the fraction of the signature vector a which data for a quantitative statistical are using the signature vector that are based on The the of all possible proteoform configurations, identifying many proteoforms are present and the peptides associated with The protein-level using standard approaches M.A. Pounds J.G. Smith R.D. Waters K.M. Webb-Robertson B.J. A analysis of computational approaches to relative protein quantification using peptide in label-free proteomics 2013; PubMed Scopus Google Scholar). The of the Bayesian inference is to a proteoform the for protein We that peptide signature can in a and signature is defined as a = proteoform is present for protein and For protein we a for signature and similar to work in protein inference Arnold R.J. Li Radivojac P. Tang H. A Bayesian approach to protein inference in shotgun Comput. Biol. 2009; PubMed Scopus (58) Google Scholar, A of statistical methods for protein identification using tandem mass 2012; 5: PubMed Google Scholar), the is to the a proteoform which is the observed signature The proteoform is by the of all are described in The that we a of peptides signature for protein is the proteoform for that the assumption that signature is we can In the assumption of of peptides will not this assumption well for model development similar to protein inference. We model the that we a of for a signature as a For we have a frequency of for signature of protein is only a that we this signature or for a protein the proteoform is is we the probabilities from the of a the of by is the in the will be We this to the of signatures that are of a is the is as a The computational time to is the possible set of for is by the observed signatures protein we only a of all possible proteoform is used to the probabilities of the possible proteoform in and the uses the peptides assigned to proteoform to a protein-level The inference of proteoform is the expected frequency of signature A approach is used to this by defining with the as the data to be but defined to have statistical changes between is by data for peptide in the data as a with a of and the estimate of from the observed for the peptide are at to yield the of from the peptide as observed in the all peptides have been of the BP-Quant are For the we define as the of peptides to have a single proteoform by the of We the process to identify a estimate of as the across all The estimate of is based and the of signatures as defined in and To the capability of BP-Quant to identify the of proteoforms associated with a protein we the dilution which of proteins protein from to We these to a standard approach (PQPQ) that uses defined from correlation to define proteoforms (10Forshed J. Protein quantification by peptide quality control (PQPQ) of shotgun proteomics data.Methods Mol. Biol. 2013; 1023: 149-158Crossref PubMed Scopus (3) Google Scholar, J. M. J. from shotgun proteomics data by protein quantification and peptide quality control Proteomics. 2011; Full Text Full Text PDF PubMed Scopus Google Scholar). We used the available of which in with all the is for labeled we defined the dilution as control set and to the analysis on the ratios to this For the peptide to proteins and proteoforms were from the and protein quantification was to BP-Quant using a M.A. Pounds J.G. Smith R.D. Waters K.M. Webb-Robertson B.J. A analysis of computational approaches to relative protein quantification using peptide in label-free proteomics 2013; PubMed Scopus Google Scholar, Qian N. Petyuk V. Camp Anderson Smith a statistical for quantitative analysis of PubMed Scopus Google Scholar). The dilution in native has biological and protein have only a single BP-Quant a single proteoform as the most for protein. of the proteins as To in respect to accuracy and datasets for which proteins are defined the of proteoforms is known are This can be constructed with from the dilution As an a protein in the dilution with The peptides the of the pattern is a from the The peptides have the dilution and the pattern is an increase from by a for the In this (Fig. these are the peptides and associated abundance from protein for which the peptides were to the dilution for this we can define that we have distinct proteoforms. For protein in the dilution we a defined of proteoforms, identify which peptides will to proteoform of and the dilution for To in the we the the were on proteins with multiple proteoforms, and the of proteoforms for these proteins from of For of the the predicted of proteoforms to the known of proteoforms was and the accuracy was The accuracy was defined based on positive and and are the the of proteins that the of such as can be used but similar not a of BP-Quant with respect to accuracy for of the datasets The are of and with standard of and for BP-Quant and In of the BP-Quant a which was statistically significant by a The of accuracy in does not directly the of the the of a protein with a single proteoform is as multiple proteoforms, a positive or a negative we also the which is the of and the of the will be and will as the of and BP-Quant has a significantly based on a larger in of the the for the for approach as well as a of the in the to the of data with proteoform we used the dilution data to negative by the data for peptide and leaving the dilution In this be statistical between and those for which this is the be The accuracy of BP-Quant on this negative set is in The accuracy of BP-Quant on the negative set was with a standard of while resulted in an accuracy of only with a standard of This is similar to the for the dilution with of the proteins were to have multiple proteoforms only proteoform is The of proteomic for biomarker discovery in a of that, by to biological is benefit to a statistical of the data in the context of the research hypothesis and the peptide and protein available for biological informed peptide selection is essential for of the relative protein quantification approaches with complex such as those used in or clinical approaches are for abundances which are often of interest in biomarker discovery In BP-Quant is a signature-based peptide selection it can be used as a to protein quantification method (e.g. or linear Proteoform identification is of the major of the protein quantification and is a to biomarker discovery and improve fundamental of biological systems at the pathway (3Smith L.M. Kelleher N.L. Proteoform: a single term describing protein complexity.Nat. Methods. 2013; 10: 186-187Crossref PubMed Scopus (884) Google Scholar). The BP-Quant approach the discovery of significant biological patterns in the presence of and also the discovery of multiple significant biological patterns (i.e. possible by peptides that patterns of expression. In many such proteoforms, such as those from post-translational modifications, have been as associated with such as Alzheimer's and J. Li L. modifications and of associated with and Biol. Full Text Full Text PDF PubMed Scopus Google Scholar). BP-Quant and similar accuracy to a clustering approach to proteoform The current of BP-Quant that peptides are which is not the will using protein-level such as or peptide or such as to identify these and model in the Bayesian In the of the BP-Quant have been with a of of the peptides to a as peptides have been significantly as a response to BP-Quant has the to the peptides from the altered peptides, can to the functional of proteoforms a biological This is a for systems studies in which the of is as as changes in abundance. It also an additional of on biomarker discovery improving the to identify