Four mouse reference plasma pools were each subjected to extensive fractionation and separate liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis of digests from individual fractions. The combined four experiments yielded 800,507 spectra with a PeptideProphet 10 probability (P) score of ≥0.9 from a total of 568 LC-MS/MS runs. The overall false discovery rate for spectrum assignment was calculated to be 1.2% based on PeptideProphet cutoffs. Of the 13,779 distinct peptides with P ≥0.9 encompassed in the analysis, 13,461 peptides were successfully mapped to the Ensembl Mouse 11 release 43, which was built on the NCBI m36 mouse assembly. Of the set of 13,779 distinct peptides, 9,170 (67%) were identified with at least 2 spectra. Cys-containing, Trp-containing, and Lys-containing peptides represented 3,709 (27%), 2,067 (15%), and 8,392 (61%) of the total peptides, respectively (Table 1).

The mean peptide length for the 13,779 peptide set was 16, with a range of 6-49 amino acids. There was a bias in the distribution of peptide length, which favored relatively long peptides (Figure 1). A similar finding from human proteome studies was previously reported 12. The under-representation of short peptides may result from losses due to reduced sequence-specific fragment ions, or difficulty in distinguishing them from noise due to low m/z values. The mean molecular weight of this set was approximately 1,750 Da with a range of 640-4,100 Da. The majority of peptides identified were either neutral or acidic, with an average pI of approximately 6. There were 5958, 8874, 7753, and 5368 distinct peptides identified for the 4 reference sets (Table 1). The number of peptides identified may relate to variability in protein levels between reference sets, particularly for abundant proteins, which affects mass spectrometer peptide sampling and variability in protein recovery with sample processing.

A combined search of data from all four reference sets was done using ProteinProphet 13. This yielded 2,982 distinct International Protein Index (IPI) identifications corresponding to 2,631 known genes plus 281 hypothetical proteins with an error rate less than 5%. Among these, 2,131 (71%) proteins were identified with at least 3 unique peptides, 2,600 (87%) with at least 2 unique peptides, and 382 with only one unique peptide (singlets, 13%). Among these singlets, 140 (37%) were observed only once within the whole study, and thus are likely the major source of false identifications. Cytoplasmic proteins contributed the most, at approximately 29%; extracellular, nucleus, and plasma membrane proteins accounted for 17%, 17%, and 14%, respectively, based on ingenuity pathway analysis 14. The limited contribution of secreted and surface membrane proteins to the overall total may be the result of release through non-secreted pathways and cell turnover. The tissue distribution and gene expression levels of this set of proteins was investigated based on the mouse SymAtlas (Novartis Research Foundation) 15. The tissue with the maximum expression for a given gene was assigned to that particular gene. Approximately 8% of identified proteins had the highest expression of their corresponding genes represented by liver (mRNA per gram tissue), which is considered a major source of abundant plasma proteins. The range of MS/MS events corresponding to high confidence protein identifications based on two peptides or more varied between approximately 50,000 and 2. We previously observed a significant correlation between the number of MS/MS events for a given protein and its plasma protein concentration (log(MS/MS events) = (0.623 × log(Protein concentration)) + 0.0625) (unpublished data). We infer, therefore, that abundance of mouse plasma proteins identified in this study may span a logarithmic range of 7.

Proteins undergo numerous post-translational modifications, notably cleavages in the case of proteins shed into the circulation. The mouse PeptideAtlas may be queried to determine the distribution of peptides identified for a given protein and the occurrence of cleaved forms in particular fractions. This is illustrated for the epidermal growth factor receptor (EGFR), which is a single-pass type I membrane protein 16, as an example of the depth of analysis achieved and of the capabilities of the mouse PeptideAtlas. Detection of over-expressed EGFR is of relevance to a number of disease processes 17. The trans-membrane region is located at amino acids 646-668, and the extracellular domain is between amino acids 25-645 (Figure 2). A total of over 4,000 MS2 events in mouse PeptideAtlas corresponding to EGFR matched 34 distinct peptides that spanned exclusively the extracellular domain of the protein, resulting from cleavage and release of this domain. The PeptideAtlas provides a graphic interface for protein fragmentation and can be used as a tool for comparisons between different species. Of interest, one peptide derived from mouse EGFR, PAp00148806 (IPLENLQIIR amino acids 99-108, lab 34), was also identified in the PeptideAtlas analysis of the Human Proteome Organization (HUPO) human plasma samples as the mouse and the human share the same sequence for this peptide 9.

Approximately 3,000 distinct peptides were identified that spanned exon boundaries, of which 1,717 were observed at least twice. Among these, 180 peptides were observed in all 4 mouse plasma pools/experiments, and mapped to proteins with a unique genome location. The database represents a useful resource for validation of splice isoforms predicted in the Ensembl mouse genome. For example, PAp00024736, with a peptide sequence DQGSCGSCWAFGAVEAISDR, was mapped to a single protein cathepsin B precursor with a unique genome location on mouse chromosome 14. This peptide was observed a total of 39 times in multiple fractions and in all 4 experiments. The first nine amino acids (DQGSCGSCW) covered one exon from genome location 62089806 to 62089832, while the rest (AFGAVEAISDR) were located on another exon at genome location 62090517 to 62090549. Cathepsin B regulates the hydrolysis of proteins with broad specificity for peptide bonds 18.

A list of approximately 950 peptides that did not map to any annotated gene in the mouse genome was developed, with an overall 1.7% false discovery rate for peptide identification. These represent putative novel open reading frames. The majority (61%) had at least two observations. One example is PAp00438183, with a sequence of RPQMVEGDHGDEIFSVFGAPLLK, which was identified over 300 times and in all 4 experiments. The peptide was mapped to a single protein, IPI00138342, the liver carboxylesterase N precursor. Its sequence matched to the coding region of this protein on chromosome 8 with 91% identity.

Alternative splicing plays a major role in protein diversity without significantly increasing genome size. Aberrations in alternative splice variants are known to contribute to a number of diseases 19. As highlighted recently in the Encode project 20, the extent of transcript structural variation in mammalian genomes has been under-appreciated. To identify novel splice isoforms that are translated into protein products, we interrogated the intact protein analysis system (IPAS; see Materials and methods) data sets using a protein sequence collection containing the products of both known and hypothetical transcripts.

A target database with over 10 million sequences was built upon the ECgene 21 and Ensembl mouse 11 databases as described in Materials and methods and in Fermin et al. 22. An extensive computational analysis for a reference set was done to identify novel forms. Using a X!Tandem expect value of <1e-3 as a threshold, we identified a total of 12,461 proteins and 8,154 distinct peptides matching 147,051 spectra in the target database. Among these, 7,291 distinct peptides (90%) were in multi-peptide sets and 863 (10%) in single peptide identification sets. At this threshold, 81 distinct reversed sequence proteins (0.65%) and 53 reversed sequence peptides (0.65%) matched 69 distinct spectra.

The splice isoforms derived from a gene share exons with each other. Further, specific peptides may occur in several members of a paralogous family of proteins. To obtain a measure of the number of independent protein identifications, it is necessary to integrate protein identifications into covering sets (see Materials and methods). To be included in the integrated protein list, peptide spectral matches need to be unique and not explained by another protein in the list. The integrated list for the set of protein identifications including novel splice isoform translation products contained a total of 1,324 distinct proteins. Multiple splice isoforms were identified for a number of proteins that have been suggested as potential disease biomarkers in previous studies: Cpn1 (carboxypeptidase N, polypeptide 1), Pzp (pregnancy zone protein), Fabp5 (fatty acid binding protein 5) and Mbl2 (mannose binding lectin) 23. In selecting proteins to be members of the covering sets, our algorithm gives preference to annotated protein sequences. Of the integrated 1,324 protein sets, 1058 (80%) were annotated gene products (that is, proteins in the Ensembl protein collection) and 199 (15%) were found only in the ECGene collection of novel transcripts. Note that in many cases, the set of identifications 'covered' by an annotated protein will also contain proteins derived from previously un-annotated transcripts.

We examined mouse splice variants identified as homologues of human counterparts. Similarities were uncovered with human peptides. A case in point is mouse splice variant M13C2563_1_s386_e8960_1_rf2_c1_n0, which was identified with 13 distinct peptides from 19 spectra through the genomic database search. Peptides LLEAQIATGGIIDPK, GFFDPNTEENLTYLQLK, and LNDSILQATEQR were identified by four, three and two spectra, respectively. However, when all 13 peptides were searched against the NCBI NR database using the blastp program, 12 peptides matched to a predicted mouse protein similar to desmoplakin protein. In contrast, all 13 peptides were found to be homologous to the human desmoplakin sequence. The alignment of peptide LLEAQIATGGIIDPK using the UCSC Blast program is shown in Figure 3. This peptide aligns to the coding sequence of human desmoplakin, but not to the annotated mouse desmoplakin gene. Therefore, our data clarify similarities between the mouse and the human coding sequences for this gene. Such detailed analysis of splice variants may identify novel alternative splicing relevant to disease.

When the protein products of alternative splice isoforms differ in structure, we expect that the physical properties of these proteins, in particular their fraction location, may vary. Finding evidence for non-neighboring fractions with distinct peptide content for the same protein supports the identification of multiple products from the same gene. Figure 4 illustrates this analysis for the major histocompatibility complex (MHC) H2 K1 K region antigen. Intact proteins were subjected to anion exchange (AX) and reverse phase (RP) chromatography fractionation in this study. Each fraction was then subjected to tryptic digestion and LC-MS/MS analysis. Figure 4 shows the distribution of peptides derived from this gene among the AX and RP chromatography fractions. The unique peptides identified in the major splice isoform and alternative splice variants were found in separate fractions, consistent with translation of the different splice isoforms yielding distinct protein products with distinct physical properties. The MHC H2 K1 K region gene encodes one of the MHC class 1 antigens, which may be altered in tumors 24.

To determine whether sufficient depth of analysis was achieved to identify proteins that are relevant to disease, we performed ingenuity pathway analysis 14 for 1,058 proteins that were fully annotated to known proteins. A number of known genes may be candidate disease biomarkers based on splice variants. For example, CD44 25 is a cell-surface glycoprotein that participates in a variety of cellular functions, including tumor metastasis. Alternative splicing generates a diverse collection of structurally and functionally distinct protein products from this gene 142627. CD44 gene products, consisting of three splice isoforms, are found in the integrated set of proteins defined by the lead peptide YGFIEGNVVIPR. This group of 3 splice isoforms was identified with 3 distinct peptides from 14 spectra in our data set. All members of this group included all three peptides.

The extracellular matrix protein 1 is another example where the major splice isoform and two novel alternative splice isoforms were found in the integrated list. Extracellular matrix protein 1 is a secreted protein that has been implicated in cell proliferation, angiogenesis and differentiation. This protein is preferentially expressed in epithelial tumors and has been suggested as a potential cancer biomarker 28.

The data for the mouse plasma PeptideAtlas was produced from an initial set of four reference plasma samples. Each represented a mixture of plasma from a mouse model of cancer and its matched control as presented in Table 2. All experiments were carried out using the IPAS methodology 29. Pooled plasma samples, immunodepleted of the abundant proteins albumin, IgG, and transferrin using an Ms-3 column (Agilent, Wilmington, DE, USA) were subjected to an extensive two-dimensional fractionation schema using AX chromatography followed by RP chromatography, yielding approximately 150 fractions per plasma pool. Each fraction was subjected to in-solution digestion with trypsin followed by LC-MS/MS analysis on either an LTQ-FT or LTQ-Orbitrap mass spectrometer (Thermo-Finnigan, Waltham, MA, USA) coupled to a nanoAcquity nanoflow chromatography system (Waters, Milford, MA, USA).

Acquired data were automatically processed through the Computational Proteomics Analysis System 35 pipeline. The database search was initiated using the X!Tandem algorithm 12 with the comet k-score module plug-in 36. The mouse IPI database 37, version 3.13, was used throughout the project. All results of sequence searching were subsequently processed through PeptideProphet 10 for peptide validation and quality control, using the Trans Proteomic Pipeline 38, version 2.9.9. All identifications with a PeptideProphet probability (P) 0.9 and above were combined to form a master list of observed peptides across all four reference sets. All experimental data were loaded into a proteomics analysis database module under the Systems Biology Experiment Analysis Management System 39. The identified peptide sequences were then mapped to the Ensembl mouse release 43 proteome based on NCBI mouse genome build 36, and the results were loaded into the PeptideAtlas relational database 4041.

ProteinProphet 13 was run on these reference data sets in a combined mode for protein identification. Proteins with less than 5% error rate from ProteinProphet were further characterized with respect to tissue expression. Mouse GNF 1M GeneAtlas, developed at the Genomics Institute of the Novartis Research Foundation 15, was downloaded, and the tissue with the maximum expression for a given gene was assigned to that particular gene/protein. Subcellular location and major functional network categorizations were based on ingenuity pathway analysis 14.

A target database, which had a total of 10,381,156 protein sequences, was generated 22 as follows: cDNA sequences from the ECgene database 21 and from the Ensembl mouse database 11 were obtained in FASTA format; each sequence set was translated separately in 3 reading frames and the first instance of every protein sequence longer than 14 amino acids was recorded; the sequences were combined and then filtered for redundancy (preference was given to protein sequences originating from an Ensembl transcript); and a collection of common protein contaminants and reverse protein sequences was appended to the sequence database. The formatted peak list files from each dataset were searched against the target database using X!Tandem; peptides identified with an X!Tandem expect value of <0.001 were extracted for further analyses.

Many transcription units produce transcripts with multiple splice isoforms that share exons. This often precludes assigning a peptide as the translation product of a specific splice isoform. To avoid redundancy in analysis, protein identifications were integrated to determine the minimum number of covering sets. A covering set is defined as a set of proteins containing all of the identified peptides. The minimal covering set is the covering set with the smallest number of members.

Defining a minimal covering set is a nondeterministic polynomial-time hard (NP hard) calculation, so exact definition of the minimal covering set is not feasible for data sets of the size found in protein/peptide identifications. Instead, we employ a heuristic approach to find approximately optimal solutions. Each member of the covering set of proteins is defined by a 'lead peptide'. The minimal covering set of proteins needed to explain the observed peptide identification data was generated as follows: step 1, a list of peptides identified in all experiments was created; step 2, the peptide list was ordered by the number of spectra matching the peptide; step 3, the peptide matching the largest number of spectra was selected (the 'lead peptide'); step 4, a list of all proteins containing this peptide was generated; step 5, the protein list was sorted by the number of spectra matching peptides derived from the protein; step 6, the Ensembl protein having the largest number of matching spectra was selected (in the event of ties, the protein having the largest number of distinct peptides was selected; if there were no Ensembl proteins in the list, the splice variant with the largest number of spectra and distinct peptides was selected); step 7, all peptides contained in this protein were removed from the peptide list; step 8, the procedure was repeated beginning with step 3 until no unassigned peptides remained, thus generating a covering set of proteins that contained all of the identified peptides. Each protein in the covering set contains at least one peptide that is unique to this protein among all of the members of the covering set.