HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) All Versions of this Article: T500009-MCP200v1 4/8/1180    most recent Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Glossary Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Savitski, M. M. Articles by Zubarev, R. A. Search for Related Content PubMed PubMed Citation Articles by Savitski, M. M. Articles by Zubarev, R. A. Social Bookmarking                      What's this?

---

Technology

New Data Base-independent, Sequence Tag-based Scoring of Peptide MS/MS Data Validates Mowse Scores, Recovers Below Threshold Data, Singles Out Modified Peptides, and Assesses the Quality of MS/MS Techniques^*

Mikhail M. Savitski, Michael L. Nielsen and Roman A. Zubarev

From the Laboratory for Biological and Medical Mass Spectrometry, Uppsala University, S-75123 Uppsala, Sweden

	ABSTRACT

TOP ABSTRACT EXPERIMENTAL PROCEDURES RESULTS DISCUSSION REFERENCES

 
The Mascot score (M-score) is one of the conventional validity measures in data base identification of peptides and proteins by MS/MS data. Although tremendously useful, M-score has a number of limitations. For the same MS/MS data, M-score may change if the protein data base is expanded. A low M-value may not necessarily mean poor match but rather poor MS/MS quality. In addition M-score does not fully utilize the advantage of combined use of complementary fragmentation techniques collisionally activated dissociation (CAD) and electron capture dissociation (ECD). To address these issues, a new data base-independent scoring method (S-score) was designed that is based on the maximum length of the peptide sequence tag provided by the combined CAD and ECD data. The quality of MS/MS spectra assessed by S-score allows poor data (39% of all MS/MS spectra) to be filtered out before the data base search, speeding up the data analysis and eliminating a major source of false positive identifications. Spectra with below threshold M-scores (poor matches) but high S-scores are validated. Spectra with zero M-score (no data base match) but high S-score are classified as belonging to modified sequences. As an extension of S-score, an extremely reliable sequence tag was developed based on complementary fragments simultaneously appearing in CAD and ECD spectra. Comparison of this tag with the data base-derived sequence gives the most reliable peptide identification validation to date. The combined use of M- and S-scoring provides positive sequence identification from >25% of all MS/MS data, a 40% improvement over traditional M-scoring performed on the same Fourier transform MS instrumentation. The number of proteins reliably identified from Escherichia coli cell lysate hereby increased by 29% compared with the traditional M-score approach. Finally S-scoring provides a quantitative measure of the quality of fragmentation techniques such as the minimum abundance of the precursor ion, the MS/MS of which gives the threshold S-score value of 2.

The reliability problem is effectively addressed by a variety of different search engines, e.g. Mascot (4), Sequest (5), etc., through the use of a scoring technique that evaluates the probability of a false positive identification. Although the evaluation methods might be sophisticated, they have some limitations. One limitation is that for the same MS/MS data the score may significantly change if the content of the protein data base is altered (protein addition to and deletion from the data base are everyday phenomena); thus the score is dependent on whether the data base accurately represents all occurring peptides. Another important limitation is that some poor quality data that should not be trusted can give by pure chance a nearly perfect match, and thus corresponding peptides are wrongly identified with a very high score. Thus, even above threshold identifications call for confirmation by search engine-independent techniques (6–8). Conversely a low score may arise because of two very different reasons: poor matching and poor MS/MS data quality. Thus when an extremely high quality, informative in terms of fragmentation, MS/MS spectrum returns zero or a below threshold score, it is most often discarded, whereas the actual reason for the poor score is that the peptide in question is not present in the data base. Alternatively an MS/MS spectrum of a peptide definitely present in the data base will receive a low score and be discarded because of the poor quality resulting from the presence of noise spikes, distorted isotopic distribution, missing fragments, etc. For instance, Mascot search engine uses M-score (9) that takes into account the number of mismatched fragments and their relative abundances. The user does not usually know what caused the poor M-score unless a time-consuming manual inspection of the spectrum is performed. Thus the automatic routine cannot make an intelligent decision, which should be in the first case to search the data in question allowing for modifications/mutations (10, 11). In the second case, the decision should be to repeat the analysis or look for supporting information, e.g. other peptides belonging to the same protein, or to invoke the retention time (12). Thus a large fraction, perhaps more than 50%, of potentially useful MS/MS data is currently discarded; this aggravates the efficiency problem.

Complementing the M-score with another data base-independent score (13) that would evaluate primarily the quality of the MS/MS data provides a means for distinguishing between the reasons for the poor M-score and for making the above intelligent decision, rendering the tedious and time-consuming manual data inspection superfluous. This has recently been realized by several groups who have designed data base-independent scoring principles. For instance, Bern et al. (13) assessed the quality of tandem MS data obtained on a low resolution instrument and managed to filter out 75% of the unidentifiable spectra while losing only 10% of the identifiable spectra. They found that the number of peaks and their abundances (the guidelines for manual data quality inspection) had in fact little classification power compared with the number of peak pairs separated by an amino acid mass. In this study, we introduce and evaluate a new scoring principle (S-score) that differs from the scoring suggested by Bern et al. (13) in several aspects. First, the S-score is based on just one parameter, the maximum length of peptide sequence tag, which simplifies the interpretation of the S-score value. Second, it utilizes high mass accuracy afforded by FTMS (14). Finally S-score uses the MS/MS information that comes not only from the traditional collision activated dissociation (CAD)¹ but also from electron capture dissociation (ECD) (15, 16) performed on the same peptide.

The application of the S-score goes beyond the mere filtering out of the "bad" spectra and includes salvaging some of the below threshold data. Moreover a further verified sequence tag used for the S-score is compared with the search engine sequence assignment, revealing cases of false positive identifications. In some cases, the sequence tag immediately reveals the presence of a modified sequence, removing the need for a separate de novo sequencing program.

An important issue also pertaining to this discussion is the instrument performance, including also the performance of a given fragmentation technique. Traditionally fragmentation efficiency has been measured as a ratio of the total abundance of fragmentation products and the abundance of the precursor ion before fragmentation (17). However, this is a general approach that is silent about the information quality of the data. For instance, the same efficiency could be assigned to two different MS/MS peptide spectra, one containing a single but very intense peak corresponding to the NH₃ loss from the precursor ion and the other containing low intensity but extensive backbone fragmentation. Clearly the information content of these two MS/MS spectra would be different. We demonstrate the applicability of the S-score for quantitative assessment of the information content in peptide MS/MS spectra.

	EXPERIMENTAL PROCEDURES

TOP ABSTRACT EXPERIMENTAL PROCEDURES RESULTS DISCUSSION REFERENCES

 
Mass Spectrometry—
All experiments were performed on an Agilent 1100 nanoflow system coupled to a 7-tesla hybrid linear ion trap Fourier transform mass spectrometer (LTQ FT, Thermo Electron Corp., Bremen, Germany) equipped with a nanoelectrospray ion source (Proxeon Biosystems, Odense, Denmark) as described previously (18). Briefly 70 µg of an Escherichia coli whole cell lysate was loaded onto a one-dimensional gel, and the protein bands were visualized with colloidal Coomassie Blue. The gel lane was excised into 20 equally sized fractions, and finally the proteins were reduced, alkylated, and in-gel digested with modified sequence grade trypsin (Promega, Madison, WI) as described previously in the literature (19). Samples were vacuum-centrifuged to remove all organic solvents and prior to mass spectrometric analysis were reconstituted into 20 µl of HPLC water containing 0.1% TFA. Mass spectrometric experiments were performed using unattended data-dependent acquisition in which the mass spectrometer automatically switches between a high resolution survey scan (r = 100.000) followed by consecutive ECD and CAD fragmentation (r = 25.000) of the two most abundant multiply charged peptides eluting at this moment from the nano-LC column. A total number of 20 samples of an E. coli lysate were analyzed, and prior to data mining all acquired dta files were combined for more thorough analysis.

In the previous analysis (18) the numbers of uniquely identified proteins for each gel band (sample) were calculated and then added together. Here instead the number of unique proteins from the combined gel strips was derived. Thus the number of proteins identified by using the complementary pairs approach was 224 in this case compared with 256 reported earlier.

S-score Description—
The S-score was calculated from the so called dta files that contain the mass and the charge state of the precursor as well as the m/z values and intensities of all the fragment ion peaks in the spectrum above a certain cutoff intensity value. For every precursor, two dta files are present, one representing CAD and the other representing ECD fragmentation.

To build a sequence tag that serves as the basis for the S-score, the program takes deisotoped, neutral CAD fragment masses (potentially true fragments (PTFs)), adds the molecular mass and the mass of a water molecule to the peak list, and builds a sequence ladder (tag) between them, fitting masses of amino acid combinations.

To create a PTF list the data in the CAD and ECD dta files were deisotoped and charge-deconvoluted (20) to the neutral state. Ion fragments of monoisotopic mass <800 Da appear often without their heavier isotopes due to the noise cutoff. In these cases, the neutral masses were derived assuming that the charge state of the peak cannot exceed the charge state of the precursor ion if the peak originates from a CAD dta file and will be less than the charge state of the precursor if the peak originates from an ECD dta file (due to charge reduction in ECD). Thus a peak without isotopes located at m/z = 500.2 in a CAD spectrum of a 2+ precursor will be allowed to have a neutral mass of 500.2 and 1000.4 Da. Both these masses are considered for the construction of complementary pairs.

A fragment ion M_c, I_c with neutral mass M_c and intensity I_c in a CAD dta file of a peptide with neutral mass M will qualify for the PTF set if it passes at least one of the following tests. (a) It is an element of a "golden pair" (21). That means there is a corresponding fragment M_e, I_e in the ECD dta file that together with the CAD fragment ion satisfies one of the following equations.

(Eq. 1)

(Eq. 2)

(Eq. 3)

(Eq. 4)

To reduce the computational redundancy and uncertainty, the algorithm primarily targets y-ion sequence tags. Thus if M_c is identified as a b-ion (satisfies Equation 1 or Equation 4), the mass of the complementary y-ion is calculated using the peptide mass and added to the list instead of M_c. Note that the total amount of information in the fragment list remains unaltered. (b) It is an element of a "complementary pair." That means there is a corresponding fragment ion M_c', I_c' in the same CAD dta file satisfying the following equation.

(Eq. 5)

Here of course y-ions are indistinguishable from b-ions. (c) It is the first peak in an isotopic cluster. Furthermore all masses of fragment ions have to pass through the peptide mass window (18) to qualify.

After adding to the PTF set of qualified masses the mass of H₂O and the neutral mass of the peptide M (the two additional masses are added to improve the length of the y-ion sequence tag), the algorithm proceeds to find the longest possible amino acid sequence tag that can be constructed from these masses. The number of masses used for the construction of this tag minus one is the S-score value. The maximum allowed mass difference between adjacent masses was M_s. The choice of 575 Da for the M_s value will be explained in the next section.

The masses in the PTF set are sorted in an increasing order {m₁, ..., m_N}. The algorithm assigns to each value m_i, 1 i N, a value of a running tag length TL_i, which is initially zero for all i. Following that, the masses m_i are selected in an increasing order (i = 1, 2, ..., N), and the corresponding values TL_i are determined for each mass m_i according to the following recipe. First the differences _j = m_i – m_j are calculated for all j < i. Then the program goes through all _j values and, if the current _j is equal to any combination of one or several amino acids and does not exceed the maximum allowed mass difference M_s, then TL_i accepts the value of TL_j + 1 unless its current value is higher. After determining all TL_i values, the largest TL_i is selected, and the value of this TL_i is the S-score for this dta file.

Reliable Tag—
Besides the sequence tag that gives rise to the S-score, a "reliable" sequence tag (RST) was constructed for each spectrum. The objective was to design the most reliable sequence tag available in mass spectrometry to date and to use this tag to produce the most reliable sequence identification. The RST is constructed as follows. Only fragment ions satisfying conditions a and b simultaneously are selected for the construction of this tag. The PTF masses of the tag are thus doubly confirmed. The maximum step difference for the reliable tag was chosen to be 398 Da. The justification for this will be given in the following section.

	RESULTS

TOP ABSTRACT EXPERIMENTAL PROCEDURES RESULTS DISCUSSION REFERENCES

 
To evaluate the significance of the S-score for characterizing the quality of MS/MS data, histograms of S-scores obtained for 2+ precursor ions were built (Fig. 1). The expectation was that the S-score could provide a means of differentiation between good peptide spectra on one hand and poor or non-peptide spectra on the other hand. The 2+ charge state was selected because it is the predominant charge state in the precursor population (75%) and also because we wanted to obtain the clearest possible picture by untangling the effect of different charge states.

View larger version (30K): [in this window] [in a new window]   FIG. 1. a–e, distributions of S-scores of the acquired data for different Ms values. e, Ms = 700 Da, but the peptide mass is not included in the PTF set for each dta file. f, the figure-of-merit curves of the two different criteria for Ms estimation. The solid line corresponds to criterion 1, and the dashed line corresponds to criterion 2. arb., arbitrary.

Selection of M_s—

(Eq. 6)

where Max₁ and Max₂ were the two local maxima of the histogram, and Min was the height of the lowest point between them. The step difference that yielded the highest B was 600 Da (Fig. 1f, solid line). Criterion 2 measured the difference between the mean values of the S-score distributions of assumed "good" and bad spectra. The criterion for good spectra was the presence of complementary pairs or golden pairs; the spectra that did not meet this criterion were deemed bad. We should mention that this was a rather crude separation but one thought to be sufficiently good for optimizing the M_s value. This approach gave a maximum at 550 Da (Fig. 1f, dashed line). A compromise between the two criteria was chosen, M_s = 575 Da. In Fig. 1, a–d, the evolution of bimodality is shown along with an abnormality in the abundance of the S = 2 peak that originates at M_s = 500 Da and increases with higher masses. This behavior is explained by the fact that, given a sufficiently small peptide mass, e.g. 900 Da, a single cleavage site located in the middle of the peptide can give S = 2, an increment of two from the previous value S = 0. When the peptide mass is removed from consideration, this abnormality disappears as seen in Fig. 1e (compare with Fig. 1d).

Classification of MS/MS Spectra—
The prime purpose of introducing the S-score was to partition the acquired MS/MS data into three classes: A, B, and C. Class A was reserved for data that has been identified by Mascot and whose credibility was proven either by the significance of the Mowse score, by the RST, or by both. Elements of class B have not fulfilled the criteria for qualifying as class A data, and in some cases Mascot has not even suggested a sequence for them, but according to the S-score they are most likely peptides with decent MS/MS spectra and should be worth pursuing further identification. Finally class C consists of MS/MS data that according to the S-score either belongs to non-peptides or peptides with such poor MS/MS spectra that reliable identification is impossible, and thus an attempt of identification would be counterproductive.

Ranking of A-data—
In the conventional approach to proteomics, the peptide identification quality is synonymous to the M-score value. With the introduction of parallel S-scoring, this is no longer so. With this in mind, the following ranking procedure was devised. The acquired MS/MS data were processed according to the scheme depicted in Fig. 2.

View larger version (31K): [in this window] [in a new window]   FIG. 2. Flowchart for the data processing procedure. The MS/MS data are sorted into three classes: A (identified by Mascot), B (good spectra, not identified by Mascot), and C (bad spectra, no identification). The class A data is ranked according to its validity.

The average number of peptides per protein N_pep increases each time when the peptide set is extended by adding the peptides of the lower rank. This is an indication of the validity of peptide IDs in the lower ranks because if the added peptides were false positives, they would likely be distributed among unrelated proteins, and N_pep would decrease.

Validation of S-score Threshold—
As can be seen from the flowchart (Fig. 2), a total of four different filters were applied at various stages. The first one is simply a requirement for the S-score to have an above threshold value of 2. The justification for this value is deduced from Fig. 3. In Fig. 3, the S-score distribution is presented for MS/MS files of charge state 2+. The distribution (black columns) is clearly bimodal and has a distinct valley at S = 1. The second distribution (light columns) in Fig. 3 is that of MS/MS files for which complementary pairs exist. Note the relatively low abundances of S = 0 and S = 1 that support the choice of S = 2 as a threshold value. This distribution has a mean of 6.7. In Fig. 4, the distribution of Mascot-identified (meaning Mascot suggested a sequence with any score) peptides (light columns) is plotted against the complementary pair distribution (black columns). The mean of the Mascot-identified distribution is clearly shifted toward higher values (mean of 7.4) and also supports the choice of S = 2 as a threshold. No peptides with scores above M = 34 were lost using the S = 2 cutoff. The highest scoring Mascot identification that was discarded had an M-score of 21. We should note here that Mascot searches were made against the non-redundant data base and with oxidation of methionine chosen as the only viable modification, and so peptides with other modifications have no chance of being correctly identified. These peptides could account for the difference between the "complementary pairs exist" and the "Mascot found" distributions. The fact that this difference (data not shown) has a significantly lower mean (4.7) could be due to the generally inferior fragmentation of modified peptides.

View larger version (26K): [in this window] [in a new window]   FIG. 3. Distribution of S-scores of all dta files (black columns) versus the distribution of S-scores for dta files from which complementary pairs were derived according to Equations 1–5 (light columns).

View larger version (24K): [in this window] [in a new window]   FIG. 4. Distribution of S-scores for dta files from which complementary pairs were derived (black columns) versus the distribution of S-scores for dta files for which Mascot found a peptide sequence (light columns).

Fig. 5, a and b, shows the same two distributions plotted differently. The first is the total distribution of peptides for which Mascot suggested a sequence (black columns), and the second is the distribution of peptides for which a sequence was suggested and an RST existed (light columns). Here the RST existed in 40% of the cases. Note that although the average S-scores were similar for both distributions (Fig. 5a), the average M-score for the RST-backed distribution was much higher (38 versus 27). Only 41% of the total distribution is above the Mascot-suggested threshold of 34, whereas 69% of RST-containing spectra gave hits above this threshold. Thus the mere presence of RST increased the probability of positive ID by Mascot by more than 50%.

View larger version (17K): [in this window] [in a new window]   FIG. 5. a, distribution of S-scores for dta files for which Mascot found a peptide sequence versus the distribution of S-scores for dta files from which an RST was derived. b, distribution of Mascot scores for dta files for which Mascot found a peptide sequence versus the distribution of Mascot scores for dta files from which an RST was derived.

View larger version (19K): [in this window] [in a new window]   FIG. 6. Distribution of Mascot scores for dta files for which the Mascot-suggested peptide sequence was confirmed by the RST (black columns) versus the distribution of Mascot scores for which the Mascot-suggested peptide sequence was rejected by the RST (light columns).

Revealing Modifications with RST—
As an example, the peptide LFSVVADDR was identified by Mascot with a below threshold M-score of 13. The S-score, however, was rather high (S = 7), and the spectra contained 12 complementary pairs of fragments. Moreover RST existed and gave possible sequences ([AY]||[HP]||[FS]||[SMo])(I)(V)(A)(D)(D)([RV]||[AAI]||[GVV]) where Mo stands for oxidized methionine. In this notation, the amino acids in square brackets do not have a defined order, e.g. [AY] could in reality be both AY and YA. The || sign means "or," i.e. ([RV]||[AAI]||[GVV]) means that at the C-terminal sequence could be [RV], [AAI], or [GVV] with I being either leucine or isoleucine. Thus this RST is consistent with many possible sequences, for instance PHLVADDAIA or YAIVADDVR. However, the Mascot-suggested sequence was not among them, thus conflicting with the RST.

In the conflict between RST and Mascot, RST is likely to be more trustworthy, and thus such spectra must be searched in an extended data base allowing for modifications/mutations. Because the RST suggests the presence of at least two aspartic acids, it is logical to assume the possibility of deamidation (24). Indeed a Mascot search allowing for this modification gave a positive identification with an M = 72, and the sequence YAIVANDVR successfully fitted all 12 complementary fragment masses. This peptide is present in an E. coli protein that was additionally identified by at least two unmodified peptides. As additional evidence of modification/mutation, the unmodified sequence was also identified from a different dta file with M = 62. This example shows the potential of using the RST to detect modifications and mutations.

Connection between S-score and M-score—
The S-score is data base-independent and reflects how successful the peptide backbone fragmentation was. Mascot tries to fit the fragmentation pattern to a tryptic peptide in the data base, and the score reflects the probability of achieving a match of a given quality at random when searching the data base.

The S-score and M-score, however, are not completely independent because both increase when the number of true backbone fragments in an MS/MS spectrum increases. Thus one can expect a statistical correlation between M- and S-scores, which is interesting to derive as it could be used for a priori prediction of the M-score for individual spectra before a data base search. A straightforward correlation between the S-score and M-score is low with r = 0.37 (25) (Fig. 7a). However, this correlation is significant given the large amount of data points, >14000 (26). There appears to be a bifurcation of the data that is visibly detectable for high S-values. In Fig. 7b, the dta files with S-score above 7 were selected, and their M-score distribution was plotted (black columns). Clearly, the distribution is bimodal. The distribution with the lower mean (A) could arise if Mascot fits data to the wrong peptide. This can happen if the true peptide is either not present in the data base or modified; in either case the fit Mascot makes would most likely be a poor one utilizing only part of the available data and thus resulting in a low M-score. To test this hypothesis we looked at the distribution of dta files for which an RST was found (Fig. 7b, light columns) and then identified the files (Fig. 7c) for which the RST affirmed the Mascot-suggested sequence (black columns) and for which it conflicted with the suggested sequence (light columns). As predicted, the low M-score distribution consisted mostly of false positives, whereas the high M-score distribution consisted mostly of correct identifications. Here false positive IDs were likely due to mutated and/or modified peptides. The ratio between the two distributions in Fig. 7c is 1:3. We believe that this ratio reflects the relative content of mutated and/or modified peptides in the test mixture. The relatively large content (24.7%) is not surprising given the high rate of mutations in E. coli.

View larger version (37K): [in this window] [in a new window]   FIG. 7. a, a scatter plot showing the correlation between the S-score values and Mascot score values of all dta files for which Mascot found a peptide sequence. b, the Mascot score distribution of all dta files with S > 7 and for which Mascot found a sequence (black columns) versus the subset of this distribution for which an RST existed (light columns). c, the part of the light columns distribution in b for which the RST confirmed the Mascot-suggested sequence (black columns) versus the complementary part to the same distribution for which the RST rejected the Mascot-suggested sequence (light columns). d, a scatter plot showing the correlation between the S-score values and Mascot score values of all dta files for which Mascot found a peptide sequence and from which an RST was derived and confirmed the Mascot sequence.

Connection between S-score and Precursor Abundance—
Naturally one would expect that the extent of detectable fragmentation of a peptide should relate to its precursor abundance. Indeed for an abundant precursor, more fragment ions will have a chance of making it over the noise level. Thus for higher S-scores, a higher average precursor abundance should be expected. Fig. 8a shows the distribution of precursor abundances in our data. The distribution scales to 1/abundance, although the automatic gain control (AGC) mode was used (this was supposed to accumulate the same number of precursors in each scan). Without AGC, the slope would be much steeper (with an ideal AGC, the distribution would be flat). Fig. 8b shows how the average precursor abundance increases with the S-score. The increase from S = 0 to S = 2 is comparatively slow followed by a steep almost linear increase from S = 2 to S = 5. At S = 5 a saturation occurs. The curve again suggests that S = 2 is a logical threshold value. At S = 2, the average intensity of precursor ions is 30,000. It is tempting to suggest simplifying the analysis by using this intensity value as a threshold instead of S = 2. This, however, is not a good idea. In Fig. 8, c and d, the distribution of S-values is plotted for the abundance intervals of precursor ions between 20,000 and 30,000 as well as 30,000 and 40,000. The bimodality of these distributions implies that a simple threshold for intensity values is not enough to discriminate between good and bad spectra. The bimodality is likely due to the presence of non-peptide precursors for which the abundances do not have to be low to produce low S-scores.

View larger version (28K): [in this window] [in a new window]   FIG. 8. a, the distribution of precursor abundances for all dta files. b, relation between the S-score of peptides and their average precursor abundances. c, the distribution of S-scores for dta files with precursor abundances between 20,000 and 30,000. d, the distribution of S-scores for dta files with precursor abundances between 30,000 and 40,000.

	DISCUSSION

TOP ABSTRACT EXPERIMENTAL PROCEDURES RESULTS DISCUSSION REFERENCES

	ACKNOWLEDGMENTS

 
Christopher M. Adams, Frank Kjeldsen, Thomas Köcher, and Oleg Silivra are acknowledged for fruitful discussions.

	FOOTNOTES

 
Received, April 19, 2005, and in revised form, May 9, 2005.

Published, MCP Papers in Press, May 22, 2005, DOI 10.1074/mcp.T500009-MCP200

¹ The abbreviations used are: CAD, collisionally activated dissociation; AGC, automated gain control; ECD, electron capture dissociation; ID, identification; LTQ, linear trap quadrupole; M-score, Mascot score; PTF, potentially true fragment; RST, reliable sequence tag.

^* This work was supported by Wallenberg Consortium North Grant WCN2003-UU/SLU-009 (to R. A. Z.). The purchase of the LTQ FT instrument was supported by a Knut och Alice Wallenbergs Stiftelse grant (to R. A. Z. and Carol Nilsson). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

To whom correspondence should be addressed: Laboratory for Biological and Medical Mass Spectrometry, Uppsala University, Box 583, S-75123 Uppsala, Sweden. Tel.: 46-18-471-5729; Fax: 46-18-471-5729; E-mail: Mikhail.Savitski{at}bmms.uu.se

	REFERENCES

TOP ABSTRACT EXPERIMENTAL PROCEDURES RESULTS DISCUSSION REFERENCES

 

1. Pandey, A., and Mann, M. (2000) Proteomics to study genes and genomes. Nature 405, 837 –846[CrossRef][Medline]

2. Aebersold, R., and Mann, M. (2003) Mass spectrometry-based proteomics. Nature 422, 198 –207[CrossRef][Medline]

3. Venable, J. D., Dong, M. Q., Wohlschlegel, J., Dillin, A., and Yates, J. R., III (2004) Automated approach for quantitative analysis of complex peptide mixtures from tandem mass spectra. Nat. Methods 1, 39 –45[CrossRef][Medline]

4. Perkins, D. N., Pappin, D. J. C., Creasy, D. M., and Cottrell, J. S. (1999) Probability-based protein identification by searching sequence databases using mass spectrometry data. Electrophoresis 20, 3551 –3567[CrossRef][Medline]

5. Eng, J. K., McCormack, A. L., and Yates, J. R. (1994) An approach to correlate tandem mass-spectral data of peptides with amino-acid-sequences in a protein database. J. Am. Soc. Mass Spectrom. 5, 976 –989[CrossRef]

6. Keller, A., Nesvizhskii, A. I., Kolker, E., and Aebersold, R. (2002) Empirical statistical model to estimate the accuracy of peptide identifications made by MS/MS and database search. Anal. Chem. 74, 5383 –5392[Medline]

7. Fenyo, D., and Beavis, R. C. (2003) A method for assessing the statistical significance of mass spectrometry-based protein identifications using general scoring schemes. Anal. Chem. 75, 768 –774[Medline]

8. Olsen, J. V., and Mann, M. (2004) Improved peptide identification in proteomics by two consecutive stages of mass spectrometric fragmentation. Proc. Natl. Acad. Sci. U. S. A. 101, 13417 –13422[Abstract/Free Full Text]

9. Pappin, D. J. C., Hojrup, P., and Bleasby, A. J. (1993) Rapid identification of proteins by peptide-mass fingerprinting. Curr. Biol. 3, 327 –332[CrossRef][Medline]

10. Sunyaev, S., Liska, A. J., Golod, A., and Shevchenko, A. (2003) MultiTag: multiple error-tolerant sequence tag search for the sequence-similarity identification of proteins by mass spectrometry. Anal. Chem. 75, 1307 –1315[Medline]

11. Mann, M., and Wilm, M. (1994) Error tolerant identification of peptides in sequence databases by peptide sequence tags. Anal. Chem. 66, 4390 –4399[Medline]

12. Kristensen, D. B., Brond, J. C., Nielsen, P. A., Andersen, J. R., Sorensen, O. T., Jorgensen, V., Budin, K., Matthiesen, J., Veno, P., Jespersen, H. M., Ahrens, C. H., Schandorff, S., Ruhoff, P. T., Wisniewski, J. R., Bennett, K. L., and Podtelejnikov, A. V. (2004) Experimental Peptide Identification Repository (EPIR)—an integrated peptide-centric platform for validation and mining of tandem mass spectrometry data. Mol. Cell. Proteomics 3, 1023 –1038[Abstract/Free Full Text]

13. Bern, M., Goldberg, D., McDonald, W. H., and Yates, J. R., III (2004) Automatic quality assessment of peptide tandem mass spectra. Bioinformatics 20, i49 –54[Abstract]

14. Marshall, A. G., and Hendrickson, C. L. (2002) Fourier transform ion cyclotron resonance detection: principles and experimental configurations. Int. J. Mass Spectrom. 215, 59 –75[CrossRef]

15. Zubarev, R. A., Kelleher, N. L., and McLafferty, F. W. (1998) Electron capture dissociation of multiply charged protein cations. A nonergodic process. J. Am. Chem. Soc. 120, 3265 –3266[CrossRef]

16. McLafferty, F. W., Fridriksson, E. K., Horn, D. M., Lewis, M. A., and Zubarev, R. A. (1999) Biochemistry. Biomolecule mass spectrometry. Science 284, 1289 –1290[Free Full Text]

17. Zubarev, R. A., Horn, D. M., Fridriksson, E. K., Kelleher, N. L., Kruger, N. A., Lewis, M. A., Carpenter, B. K., and McLafferty, F. W. (2000) Electron capture dissociation for structural characterization of multiply charged protein cations. Anal. Chem. 72, 563 –573[Medline]

18. Nielsen, M. L., Savitski, M. M., and Zubarev, R. A. (March 16,2005) Improving protein identification using complementary fragmentation techniques in Fourier transform mass spectrometry. Mol. Cell Proteomics 4, 835 –845[Abstract/Free Full Text]

19. Wilm, M., Shevchenko, A., Houthaeve, T., Breit, S., Schweigerer, L., Fotsis, T., and Mann, M. (1996) Femtomole sequencing of proteins from polyacrylamide gels by nano-electrospray mass spectrometry. Nature 379, 466 –469[CrossRef][Medline]

20. Gentzel, M., Kocher, T., Ponnusamy, S., and Wilm, M. (2003) Preprocessing of tandem mass spectrometric data to support automatic protein identification. Proteomics 3, 1597 –1610[CrossRef][Medline]

21. Horn, D. M., Zubarev, R. A., and McLafferty, F. W. (2000) Automated de novosequencing of proteins by tandem high-resolution mass spectrometry. Proc. Natl. Acad. Sci. U. S. A. 97, 10313 –10317[Abstract/Free Full Text]

22. Peng, J. M., Elias, J. E., Thoreen, C. C., Licklider, L. J., and Gygi, S. P. (2003) Evaluation of multidimensional chromatography coupled with tandem mass spectrometry (LC/LC-MS/MS) for large-scale protein analysis: the yeast proteome. J. Proteome Res. 2, 43 –50[CrossRef][Medline]

23. Shevchenko, A., Sunyaev, S., Loboda, A., Shevehenko, A., Bork, P., Ens, W., and Standing, K. G. (2001) Charting the proteomes of organisms with unsequenced genomes by MALDI-quadrupole time-of-flight mass spectrometry and BLAST homology searching. Anal. Chem. 73, 1917 –1926[Medline]

24. Wright, H. T. (1991) Nonenzymatic deamidation of asparaginyl and glutaminyl residues in proteins. Crit. Rev. Biochem. Mol. Biol. 26, 1 –52[Medline]

25. Pearson, K. (1896) Mathematical contributions to the theory of evolution: III. regression, heredity, and panmixia. Philos. Trans. R. Soc. Lond. A Math. Phys. Sci. 187, 253 –318

26. Bewick, V., Cheek, L., and Ball, J. (2003) Statistics review 7: correlation and regression. Crit. Care 7, 451 –459[CrossRef][Medline]

CiteULike

Complore

Connotea

Del.icio.us

Digg

Reddit

Technorati

This article has been cited by other articles:

				M. Brosch, S. Swamy, T. Hubbard, and J. Choudhary Comparison of Mascot and X!Tandem Performance for Low and High Accuracy Mass Spectrometry and the Development of an Adjusted Mascot Threshold Mol. Cell. Proteomics, May 1, 2008; 7(5): 962 - 970. [Abstract] [Full Text] [PDF]

				D. A. Stead, N. W. Paton, P. Missier, S. M. Embury, C. Hedeler, B. Jin, A. J. P. Brown, and A. Preece Information quality in proteomics Brief Bioinform, March 1, 2008; 9(2): 174 - 188. [Abstract] [Full Text] [PDF]

				D. Bouyssie, A. G. de Peredo, E. Mouton, R. Albigot, L. Roussel, N. Ortega, C. Cayrol, O. Burlet-Schiltz, J.-P. Girard, and B. Monsarrat Mascot File Parsing and Quantification (MFPaQ), a New Software to Parse, Validate, and Quantify Proteomics Data Generated by ICAT and SILAC Mass Spectrometric Analyses: Application To the Proteomics Study of Membrane Proteins from Primary Human Endothelial Cells Mol. Cell. Proteomics, September 1, 2007; 6(9): 1621 - 1637. [Abstract] [Full Text] [PDF]

				M. Lerner, M. Corcoran, D. Cepeda, M. L. Nielsen, R. Zubarev, F. Ponten, M. Uhlen, S. Hober, D. Grander, and O. Sangfelt The RBCC Gene RFP2 (Leu5) Encodes a Novel Transmembrane E3 Ubiquitin Ligase Involved in ERAD Mol. Biol. Cell, May 1, 2007; 18(5): 1670 - 1682. [Abstract] [Full Text] [PDF]

				N. Bandeira, D. Tsur, A. Frank, and P. A. Pevzner Protein identification by spectral networks analysis PNAS, April 10, 2007; 104(15): 6140 - 6145. [Abstract] [Full Text] [PDF]

				M. L. Nielsen, M. M. Savitski, and R. A. Zubarev Extent of Modifications in Human Proteome Samples and Their Effect on Dynamic Range of Analysis in Shotgun Proteomics Mol. Cell. Proteomics, December 1, 2006; 5(12): 2384 - 2391. [Abstract] [Full Text] [PDF]

				Q. C. Ru, L. A. Zhu, J. Silberman, and C. D. Shriver Label-free Semiquantitative Peptide Feature Profiling of Human Breast Cancer and Breast Disease Sera via Two-dimensional Liquid Chromatography-Mass Spectrometry Mol. Cell. Proteomics, June 1, 2006; 5(6): 1095 - 1104. [Abstract] [Full Text] [PDF]

				M. M. Savitski, M. L. Nielsen, and R. A. Zubarev ModifiComb, a New Proteomic Tool for Mapping Substoichiometric Post-translational Modifications, Finding Novel Types of Modifications, and Fingerprinting Complex Protein Mixtures Mol. Cell. Proteomics, May 1, 2006; 5(5): 935 - 948. [Abstract] [Full Text] [PDF]

				A. I. Nesvizhskii, F. F. Roos, J. Grossmann, M. Vogelzang, J. S. Eddes, W. Gruissem, S. Baginsky, and R. Aebersold Dynamic Spectrum Quality Assessment and Iterative Computational Analysis of Shotgun Proteomic Data: Toward More Efficient Identification of Post-translational Modifications, Sequence Polymorphisms, and Novel Peptides Mol. Cell. Proteomics, April 1, 2006; 5(4): 652 - 670. [Abstract] [Full Text] [PDF]

				J. Salmi, R. Moulder, J.-J. Filen, O. S. Nevalainen, T. A. Nyman, R. Lahesmaa, and T. Aittokallio Quality classification of tandem mass spectrometry data Bioinformatics, February 15, 2006; 22(4): 400 - 406. [Abstract] [Full Text] [PDF]

				J. V. Olsen, L. M. F. de Godoy, G. Li, B. Macek, P. Mortensen, R. Pesch, A. Makarov, O. Lange, S. Horning, and M. Mann Parts per Million Mass Accuracy on an Orbitrap Mass Spectrometer via Lock Mass Injection into a C-trap Mol. Cell. Proteomics, December 1, 2005; 4(12): 2010 - 2021. [Abstract] [Full Text] [PDF]

This Article Abstract