Optimization and Use of Peptide Mass Measurement Accuracy in Shotgun Proteomics

Transcription

1 MCP Papers in Press. Published on April 23, 26 as Manuscript M5339-MCP2 Optimization and Use of Peptide Mass Measurement Accuracy in Shotgun Proteomics Wilhelm Haas, Brendan K. Faherty, Scott A. Gerber, Joshua E. Elias, Sean A. Beausoleil, Corey E. Bakalarski, Xue Li, Judit Villén, and Steven P. Gygi Department of Cell Biology, Taplin Biological Mass Spectrometry Facility, Harvard Medical School, Boston, MA 2115 Corresponding author: Telephone: (617) Fax: (617) Running Title High mass accuracy in shotgun proteomics Abbreviations The abbreviations used are: ACN, acetonitrile; Cn, delta cross-correlation; AGC, automatic gain control; CAI, codon adaptation index; CID, collision-induced dissociation; DTT, dithiothreitol; FA, formic acid; FP, false positive (peptide/protein assignment/identification); FT-ICR MS, Fourier transform ion cyclotron resonance mass spectrometry; HPLC, high-performance liquid chromatography; IMAC, Immobilized metal-affinity chromatography; LC-MS/MS, liquid chromatography tandem mass spectrometry; MS, mass spectrometry; MS/MS, tandem mass spectrometry; MS n, multiplestage mass spectrometry; ORF, open reading frame; PEEK, polyetheretherketone; peptide/protein ID, peptide/protein identification; SD, standard deviation; SDS, sodium dodecylsulfate; SGD, Saccharomyces Genome Database at Stanford University; SIM scan, selected ion monitoring scan; S/N, signal-to-noise ratio; TIC, total ion current; TOF, time-offlight; TP, true positive (peptide/protein assignment/identification); XCorr, cross-correlation score. 1 Copyright 26 by The American Society for Biochemistry and Molecular Biology, Inc.

2 Summary Mass spectrometers that provide high mass accuracy such as Fourier transform ion cyclotron resonance (FT-ICR) instruments are increasingly used in proteomics studies. Although the importance of accurately determined molecular masses for the identification of biomolecules is generally accepted, its role in the analysis of shotgun proteomics data has not been thoroughly studied. To gain insight into this role, we used a hybrid linear quadrupole ion trap/ft-icr mass spectrometer (LTQ FT) for liquid chromatography tandem mass spectrometry (LC-MS/MS) analysis of a highly complex peptide mixture derived from a fraction of the yeast proteome. We applied three data-dependent MS/MS acquisition methods. The FT-ICR part of the hybrid mass spectrometer was either not exploited, used only for survey MS scans, or also used for acquiring selected ion monitoring (SIM) scans to optimize mass accuracy. MS/MS data were assigned with the SEQUEST algorithm and peptide identifications were validated by estimating the number of incorrect assignments using the composite target/decoy database search strategy. We developed a simple mass calibration strategy exploiting polydimethylcyclosiloxane background ions as calibrant ions. This strategy allowed us to substantially improve mass accuracy without reducing the number of MS/MS spectra acquired in an LC-MS/MS run. The benefits of high mass accuracy were greatest for assigning MS/MS spectra with low signal-to-noise ratio and for assigning phosphopeptides. Confident peptide identification rates from these data sets could be doubled by the use of mass accuracy information. It was also shown that improving mass accuracy at a cost to the MS/MS acquisition rate substantially lowered the sensitivity of LC- MS/MS analyses. The use of FT-ICR SIM scans to maximize mass accuracy reduced the number of protein identifications by 4%. 2

3 Introduction Mass spectrometry (MS) has become the method of choice for the characterization of complex protein mixtures from cells or tissues (1). The highest throughput and most comprehensive efforts to catalog protein mixtures have so far been achieved using a strategy known as shotgun proteomics (2). Proteins are proteolytically digested, and the resultant peptide mixture is usually separated by reversed-phase chromatography (LC) coupled on-line to a mass spectrometer. Peptides are ionized and subjected to sequencing by tandem mass spectrometry (MS/MS). Peptide ion sequences are derived based on fragment ions predominately formed through backbone cleavage at amide-bonds. Several algorithms such as SEQUEST (3) and Mascot (4) allow automated assignment of MS/MS spectra by matching acquired data with spectra predicted on the basis of protein sequence databases (5). The protein composition of the analyzed sample is ultimately inferred from identified peptides (6). Recent MS technologies facilitate the acquisition of many thousands of MS/MS spectra over the course of one LC-MS/MS analysis. In general, only a portion of these spectra are correctly assigned to peptide ions using current database search algorithms. Low quality MS/MS data, non-peptide ions selected for fragmentation, as well as peptide ions where the corresponding sequence is not predicted from the searched database may contribute to this phenomenon. Commonly, scores reflecting the quality of the match of acquired MS/MS data to predicted spectra are used as filter criteria to remove incorrect assignments. Manual validation of MS/MS spectra can help distinguish between true and false assignments but becomes infeasible with the ever-increasing size of data sets. Only recently, statistical tools have been developed, which support the validation of peptide and protein assignments (7-1). Furthermore, extending the information from mass spectrometric data by using multiple MS stages (MS n ) (11, 12) or by applying fragmentation techniques alternative to the commonly used collision-induced dissociation (CID) (13) was proposed for increasing the specificity of the peptide identification process. Also, new scoring algorithms incorporating additional information from common CID- MS/MS spectra such as fragment ion intensities have been presented as useful tools for improved differentiation between correct and false peptide assignments (14, 15). The confidence in the identification of a peptide can also be enhanced by accurately measuring the mass of the peptide ion. The importance of high mass accuracy in characterizing peptides has been described in numerous studies (16-21). Mass spectrometers that provide high 3

4 mass accuracy include time-of-flight (TOF) (22) and Fourier transform ion cyclotron resonance (FT-ICR) (23) instruments. Their increasing usage in shotgun proteomics studies demands a thorough study of the role of high mass accuracy in such studies where peptide identifications were until recently based primarily on the rich information given by MS/MS spectra. To accomplish this task, we used a hybrid linear quadrupole ion trap/ft-icr mass spectrometer (LTQ FT, Thermo Electron) (24) for the characterization of a complex peptide mixture. FT-ICR instruments provide the highest mass accuracy of present MS technologies (for reviews see (23) and (25)), but challenges in their use when on-line coupled to separation techniques restricted their application in large-scale proteomics experiments. Foremost among these challenges were relatively large time scales for acquiring MS/MS data as well as practical limitations in achieving high mass accuracy based on the highly variable production of ions across a chromatographic separation. These difficulties have been addressed with the development of the hybrid LTQ FT mass spectrometer, since the linear ion trap (26) allows highspeed performance of MS/MS experiments and the adjustment of the number of ions analyzed in the ICR cell through automatic gain control (AGC). In the present study, we used a fraction of yeast whole-cell lysate tryptic digest as a model complex peptide mixture and analyzed the sample using different data-dependent MS/MS acquisition strategies. To ensure a high acquisition rate, the FT-ICR mass spectrometer was used only for accurate determination of peptide masses while MS/MS experiments were entirely performed in the linear ion trap. The different acquisition strategies produced data sets providing peptide mass accuracies in either the low ppm range or a much wider range typical of traditional ion trap mass spectrometers often used in proteomics experiments. We applied the composite target/decoy database approach to validate the peptide assignments achieved from these data sets (8, 1). Ensuring a similar false positive rate of the identified peptides allowed a fair comparison of the used MS/MS data acquisition methods and the role of mass accuracy in the peptide identification process. We also developed a simple mass calibration strategy for FT-ICR MS data from shotgun proteomics experiments which helped to avoid compromises between mass accuracy and the number of MS/MS spectra acquired in an LC-MS/MS run. The calibration procedure included the exploitation of polydimethylcyclosiloxane ions commonly detected as background ions in microcapillary LC-MS experiments as calibrant ions (27). To extend our understanding of the role of mass accuracy in the peptide profiling of samples different from that 4

5 analyzed in the present study we artificially varied the size of the protein sequence database used for MS/MS spectra assignment as well as the quality of the MS/MS data. We also studied the role of peptide mass accuracy in the interpretation of MS/MS spectra from a 1-fold dilution of the starting sample and from phosphopeptides. Experimental Procedures S. cerevisiae Growth Conditions, Lysis, Lysate Fractionation, Protein Digest, and Enrichment of Phosphopeptides Using Immobilized Metal-Affinity Chromatography. S. cerevisiae (strain DMY1737) was grown and lysed as described in (28). The protein content of the cleared lysate was determined using a Bradford protein assay (Bio-Rad, Hercules, CA) with BSA as an external standard. Disulfide bonds were reduced by adding DTT to a final concentration of 3 mm and incubating the solution at 56 ºC followed by derivatization of cysteine residues with iodoacetamide (approximately 5 mm excess in relation to concentration of reducing agents, 2-mercaptoethanol and DTT) at room temperature in the dark for 2 min. SDS, glycerol, and bromophenol blue were added to the protein solution to a total concentration of 2, 1, and.1 %, respectively, and 1 mg of protein was loaded on a hand-poured preparative 1% polyacrylamide gel (2.6% bis-acrylamide, Bio-Rad). Following electrophoresis, and staining with colloidal Coomassie (Pierce, Rockford, IL) proteins of a molecular weight of approximately 6-11 kda were subjected to in-gel digestion with trypsin (modified sequencing grade porcine trypsin, Promega, Madison, WI) as described previously (29). Briefly, cutting the corresponding gel-region into 1 mm cubes was followed by destaining with 5 mm NH 4 HCO 3 /5% acetonitrile, dehydration of the gel-pieces, 45 min incubation with a solution of 12.5 µg/ml trypsin in 5 mm NH 4 HCO 3 on ice, overnight digestion at 37ºC, and extraction of the peptides with 5% acetonitrile/5% formic acid. Peptides were subjected to C 18 solid phase extraction (Vydac, Hesperia, CA) and dissolved in 5% ACN/5% FA. For preparing a phosphopeptide sample 2 µg of the described reduced and alkylated yeast lysate were separated by SDS gel electrophoresis, and the digest of proteins of a molecular weight of approximately 8-12 kda were subjected to immobilized metal-affinity chromatography using PHOS-Select Iron Affinity Gel (Sigma, St. Louis, Missouri) following the manufacturer s instructions. 5

6 Nanoscale Microcapillary Liquid Chromatography Electrospray Ionization Tandem Mass Spectrometry (LC-MS/MS). LC-MS/MS experiments were performed in triplicate on an LTQ FT mass spectrometer (Thermo Electron, San Jose, CA) equipped with a Finnigan Nanospray II electrospray ionization source (Thermo Electron), an Agilent 11 Series binary high-performance liquid chromatography (HPLC) pump (Agilent Technologies, Palo Alto, CA), and a Famos autosampler (LC Packings, San Francisco, CA). Peptide mixtures were separated on a fused silica microcapillary column with an internal diameter of 125 µm and an in-house prepared needle tip with an internal diameter of approximately 5 µm. Columns were packed to a length of 18 cm with a C 18 reversed-phase resin (Magic C18AQ, particle size 5 µm, pore size 2 Å, Michrom Bioresources, Auburn, CA). 4 µl of sample solution (approximately 1 µg/µl or.1 µg/µl for dilution experiments) were loaded onto the column and separation was achieved by using a mobile phase from 2.5% ACN/.15% FA (buffer A) and 97.5% ACN/.15% FA (buffer B) and applying a linear gradient from 3 to 37% buffer B for 9 min (6 min for the analysis of the phosphopeptide sample) at a flow rate of 3 nl/min provided across a flow splitter by the HPLC pumps. An electrospray voltage of 2.1 kv was applied via a gold electrode through a PEEK junction at the inlet of the microcapillary column. The LTQ FT mass spectrometer was operated in the data-dependent mode using three different acquisition strategies as follows. With the SIM3 method (21) (Fig. 1A) a scan cycle was initiated with a full-scan survey MS experiment (m/z 35-17) performed with the FT-ICR mass spectrometer. The three most abundant ions detected in this scan were subjected to an FT-ICR selected ion monitoring (SIM) scan followed by an MS/MS experiment in the LTQ mass spectrometer. Accumulation of ions for both MS and MS/MS scans was performed in the linear ion trap and the AGC target values were set to 1x1 7 ions for survey MS, 5x1 4 ions for SIM, and 1x1 4 ions for MS/MS experiments. The maximum ion accumulation time was 25 ms and 15 ms in the FT-ICR and MS/MS mode, respectively. The resolution at 4 m/z was set to 2.5x1 4 for the survey MS and to 5x1 4 for the SIM scans. Isolation widths were ±5 m/z for SIM and ±1.25 m/z for MS/MS experiments, and ions were selected for MS/MS when their intensity exceeded a minimum threshold of 12 counts. Singly-charged ions were not subjected to MS/MS. The normalized 6

7 collision energy was set to 35% and 1 microscan was acquired per spectrum. Ions subjected to MS/MS were excluded from further sequencing for 3 s. When using the FT1 method (Fig. 2A) the acquisition of SIM scans was omitted. A full-scan survey MS experiment (m/z range as above, AGC target 1x1 6 ions, resolution 1x1 5, maximum ion accumulation time 1 ms) was acquired in the FT-ICR mass spectrometer followed by MS/MS experiments on the ten most abundant ions detected in the full-ms scan. MS/MS spectra were collected in the LTQ and the settings were identical to those described for the SIM3 method. With the LTQ1 method only the linear quadrupole ion trap (LTQ) and not the FT-ICR mass spectrometer was used (Supplementary Fig. 3). A full-scan MS was again followed by 1 MS/MS experiments on the ten most abundant ions detected in the full-scan MS. The survey MS AGC target was set to 2x1 4 ions, its maximum ion accumulation time was 15 ms. MS/MS settings were as described above with the exception that MS/MS was also performed on singlycharged peptides. Data Processing, Database Searching, and Mass Calibration Procedures. Instrument control and primary data processing were done using the Xcalibur software package, version 1.4 SR1 (Thermo Electron). Data were originally stored in the.raw format. LTQ1 MS/MS data including no information from FT-ICR data were converted into the.dta format the input data format for SEQUEST searches (see below) by the program ExtractMS version 2.11 (Thermo Electron, which separates singly-charged peptide MS/MS spectra from those of multiply-charged peptide. As the unit mass resolution of MS data acquired with the LTQ did not allow the determination of the charge state of multiple charged peptide ions, three.dta files were created for each corresponding MS/MS spectrum assuming potential charge states of +2, +3, and +4. Data from analyses including the use of the FT-ICR mass spectrometer were extracted into the OpenRaw format, using the program xr2or, written in Visual C++ (downloaded from (3). Based on this data structure, in-house Perl scripts were used to extract the exact measured monoisotopic m/z as well as the charge-state of peptide ions selected for MS/MS experiments, and together with MS/MS fragment ion m/z and intensity 7

8 data, this information was included into files created in the.dta format for the acquired MS/MS spectra (Beausoleil, S. A. et al., manuscript in preparation) (31). Using SEQUEST (version 27, revision 12) on a Linux cluster with 17 dual GHz processor nodes or the Sequest-Sorcerer platform ( MS/MS data in the.dta format were searched against a composite target/decoy protein sequence database in which the target component was comprised of protein sequences derived from the known S. cerevisiae open reading frames (ORFs, downloaded 9/1/24 from the Saccharomyces Genome Database (SGD) at Stanford University, CA, ftp://genomeftp.stanford.edu/pub/yeast/data_download/sequence/genomic_sequence/orf_protein/) and protein sequences of known contaminant proteins, such as porcine trypsin and human keratins (6427 entries total). This component was followed in the database by a decoy component composed of the reversed sequences of all proteins in the target component. For simulating the search of the acquired MS/MS spectra against a larger database, the size of the yeast ORF protein sequence database was extended by a factor of nine (57851 entries) using a Markov chain model (fourth order) based on the protein sequences in the original database. A target/decoy version of this database was created as described above. The signal-to-noise ratio of MS/MS spectra was modified by increasing the intensity of the lowest 5% of signals from the original spectra by a factor of 3. Both database and data set manipulations were accomplished with in-house Perlscripts. Database searches were performed by applying precursor ion m/z tolerances of 3 ppm (SIM3), 6 ppm (FT1), or ±2 Da (LTQ1), and fragment ion m/z tolerances of ±1 Da. Cysteine residues were searched as carbamidomethylated (mass increment of Da), and methionine residues were allowed to be oxidized ( Da). When assigning MS/MS spectra from phosphopeptide analyses, serine, threonine, and tyrosine residues were allowed to be phosphorylated ( Da). For the search of phosphopeptide MS/MS data only tryptic peptides were considered for matching with the acquired MS/MS spectra in the database search. For the assignment of MS/MS spectra from unmodified peptides no enzyme specificity constraints were applied and peptide assignments were only accepted when both peptide termini were consistent with trypsin specificity. Three assignments were made for most LTQ1 MS/MS spectra as they were considered as doubly-, triply-, or quadruply-charged peptides (see above). Two assignments for each spectrum were removed from the data set before further data analysis. 8

9 For non-specific searches, this was done by using the tryptic state of the peptide as a primary filter criterion in the presence of tryptic assignments these were preferred over non-tryptic ones and the XCorr score (see below) of the assignment as secondary criteria (primary for tryptic searches). False-positive (FP) peptide assignments were removed by filtering on the basis of two metrics calculated by the SEQUEST program that express the quality of the match between an experimental and a database predicted MS/MS spectrum. The first was the cross-correlation score (XCorr), which reflects the quality of the match of the two spectra. The second was the delta cross-correlation ( Cn) which is the normalized difference between the XCorr values for the two best peptide sequence matches to an acquired MS/MS spectrum (3, 32). XCorr and Cn thresholds were applied to obtain a peptide identification false-positive (FP) rate of ~1%. The FP rate was estimated using the target/decoy database approach: assuming true-positive (TP) peptide identifications were exclusively assigned to peptides from the target database component and FP identifications were equally distributed between target and decoy components, the number of FP peptide identifications was estimated by doubling the number of decoy database assignments and dividing the result by the number of total peptide identifications in the data set (1). The fraction of FP identifications was expressed as a percentage (FP rate). The estimation of protein identification false positive rates was based on the same calculation (for examples see Supplementary Fig. 4 and 5). XCorr and Cn filtering were applied separately for assignments to doubly-, triply-, and quadruply-charged peptide ions. Assignments to singly-charged peptide ions as well as to peptide ions with fewer than eight amino acids were not considered in this study and removed from the data set before further analysis. When the data from triplicate analyses were studied, XCorr and Cn thresholds required to achieve a 1% peptide identification FP rate were determined on the basis of combined data sets. Initial external mass calibration for the LTQ FT was done as recommended by the manufacturer using singly-protonated ions of caffeine, a peptide with the sequence MRFA, and Ultramark 1621 (a mixture of fluorinated phosphazines). Recalibration of FT-ICR peptide ion m/z values was done by using the following calibration function: m/z = a/f + b/f 2 + c TIC/f 2 (1) 9

10 where f is the detected cyclotron frequency, a, b, and c are calibration coefficients, and TIC is the total ion current measured for an MS spectrum (33-36). For re-calibration calibration coefficients recorded in the.raw file format were used to recalculate f from m/z values using the calibration function described in (33). TIC values stored in the.raw format had been normalized accounting for varying ion accumulation times during an analysis. For use in the equation (1), this normalization was reversed by multiplying the TIC values by the ion accumulation time in seconds. Singly-protonated polydimethylcyclosiloxane ions, (Si(CH 3 ) 2 O) n, which are commonly detected as background ions in nanoscale LC-MS/MS experiments (27) were exploited as calibrants. Ions containing 5 (m/z ), 6 (m/z ), 7 (m/z ), 8 (m/z ), and 9 (m/z ) dimethylsiloxane monomers were used when their intensity exceeded 1 counts. The Levenberg-Marquardt algorithm implemented in the SigmaPlot software package, version 9 (Systat Software, Point Richmond, CA) was used for optimizing calibration coefficients for the calibration function (1). Peptide ion signal-to-noise (S/N) ratios were determined using an in-house developed software package for automated peptide quantification on the basis of MS data (VISTA, Bakalarski et al., manuscript in preparation). Results Optimization of Mass Accuracy on the LTQ FT The instrument platform used in this study was an LTQ FT mass spectrometer (Thermo Electron) (24). FT-ICR mass spectrometers provide the highest mass accuracy of present mass spectrometry (MS) technologies (23), but there are practical limitations to the performance of these instruments. In FT-ICR MS, ion m/z values are determined based on the frequency of the cyclotron motion ions perform under the influence of a magnetic field in the ICR cell. However, this frequency is also affected by electric fields including those produced by the analyzed ions themselves. These phenomena are termed space-charge effects, and they evoke errors in mass measurement in relationship to the number of ions analyzed in the mass spectrometer. Although various methods have been developed to correct for these errors (25), space-charge effects are still a substantial problem especially when FT-ICR MS is coupled to separation techniques, where highly variable numbers of ions may be introduced into the mass spectrometer over the course of an analysis. With the development of the LTQ FT, this problem has been almost 1

11 completely addressed as the linear trap allows for fine control over the number of ions analyzed in the ICR cell. This feature is termed automatic gain control (AGC) and has recently been exploited by Olsen et al. (21) to optimize the mass accuracy for LTQ FT large-scale proteomics applications. The described method involved the determination of peptide ion m/z values by FT- ICR SIM scans recorded over a narrow m/z range around the analyzed ions. The small m/z range allowed high peptide ion S/N ratios even when only a small number of ions was analyzed in the ICR cell (37). By using the AGC to ensure that a constantly low number of ions was subjected to SIM scans, space-charge effects were avoided, and an excellent mass accuracy was achieved. An absolute mass error of less than 2 ppm was reported for 163 peptides identified from the protein carbamoyl-phosphate synthase identified in a high protein molecular weight fraction of a mouse liver proteome tryptic digest (21). We applied the described method, here denoted as SIM3, to analyze a complex peptide mixture produced by using trypsin to digest a fraction of yeast whole-cell lysate. After recalibration of peptide ion masses (see below) MS/MS data were searched against a yeast protein sequence database using the SEQUEST algorithm. The mass accuracy distribution for the identified peptide ions is shown in Fig. 1B. We were able to confirm the excellent performance of the SIM3 method. The average measured peptide mass error was -.41 ppm and the mass accuracy distribution showed a standard deviation (SD) of.44 ppm. Although providing high mass accuracy, we found the acquisition of SIM scans to be time-consuming (Fig. 1A). Each MS/MS spectrum acquired in an LC-MS/MS peptide profiling experiment might potentially lead to the identification of a peptide. We therefore intended to improve the LTQ FT MS/MS acquisition rate by analyzing the sample with the FT1 method (Fig. 2). By omitting the recording of SIM scans the number of acquired MS/MS spectra was substantially increased. Within an average cycle time of approximately 3.5 seconds, ten MS/MS spectra were recorded, in contrast to only three in a slightly shorter cycle time of 2.6 seconds with the SIM3 method. However, with the FT1 method, peptide ion m/z values had to be determined from the survey MS scan where, in comparison to the SIM scans, an immensely higher number of ions (1,,) were analyzed. These scans provided an average peptide ion mass accuracy of 4.99±2.42 ppm (Fig. 2C3) which was lower than that achieved with the SIM3 method. However, we considered the higher MS/MS acquisition rate of the FT1 strategy as a potentially critical feature for large-scale proteomics studies. To avoid a compromise between 11

12 acquisition rate and mass accuracy we sought to improve the accuracy provided by this method through mass re-calibration of the acquired data. A prerequisite for mass calibration is the presence of ions with known masses to be used as calibrants. Singly-protonated polydimethylcyclosiloxanes, a group of air contaminants, are well-known background ions commonly detected in nano-scale LC-MS experiments (27). We exploited five highly abundant species of these ions with m/z values of 371, 445, 519, 593 and 667 (see Experimental Procedures for exact values) as calibrants for post-acquisition mass calibration for LTQ FT MS data. Polydimethylcyclosiloxane ions were detected above background noise level in only a portion of the approximately 15 FT-ICR survey MS spectra acquired in an FT1 run (Supplementary Fig. 1B), which was ascribed to ionization suppression effects. To amend mass measurement errors in all MS spectra an external calibration strategy had to be applied where MS data from spectra showing no calibrant ion signals were corrected based on signals observed in other spectra. We modified a widely used FT-ICR MS calibration equation developed by Ledford et al. (33) to address differences in the number of ions analyzed in individual MS survey spectra during an FT1 analysis (see Experimental Procedures and Supplementary Fig. 1C) (34-36). By using this calibration strategy, we substantially improved the mass accuracy of FT1 data. When compared to uncalibrated data, the average mass accuracy was reduced from 4.99 to -.25 ppm and the SD of the mass accuracy distribution from 2.42 to 1.46 ppm (Fig. 2C3 and 2C4). We observed that the peptide mass errors could be approximately fitted with a Gaussian distribution (Fig. 2). Thus, a minimum peptide mass search tolerance of three SD of the mass accuracy distribution was required in an MS/MS spectra database search to allow correct assignment of more than 99% of the acquired spectra. Recalibration of FT1 survey MS data lowered the minimum peptide tolerance requirement for a database search of the MS/MS data from approximately 13 ppm to about 5 ppm. The relative high tolerance for the un-calibrated data resulted partly from the fact that the SEQUEST program does not allow adjustments for a shift in mass accuracy distributions. However, the overall mass accuracy of -.25±1.46 ppm of the recalibrated FT1 data was still worse than the -.41±.44 ppm achieved with the SIM3 method. It has to be noted here that the SIM3 mass accuracy distribution displayed in Fig. 1B was also achieved by mass recalibration. About 5% of the SIM scans included polydimethylcyclosiloxane background ions that were exploited applying the mass calibration procedure described above. The mass accuracy distribution for the uncalibrated 12

13 data showed a similar SD but an average mass accuracy shifted to 1.6 ppm (Supplementary Fig. 2). The comparison of the SIM3 and the FT1 method showed that the latter provided a slightly reduced mass accuracy, but at the same time allowed more MS/MS spectra to be acquired in an LC-MS/MS run. Therefore, we next sought to evaluate the role these two features mass accuracy and MS/MS acquisition rate in mass spectrometry based proteomics studies by performing a detailed comparison of MS/MS data acquired with both methods. Benefits and Costs of High Mass Measurement Accuracy in Shotgun Proteomics Experiments In shotgun proteomics experiments, method-dependent differences in peptide mass accuracy achievable with the LTQ FT present a challenge in determining the optimum peptide mass search tolerance for database searching of the acquired MS/MS data. Although narrowing the tolerance was reported to enhance the peptide identification process by decreasing the number of peptides which can be matched to the acquired MS/MS data (21), undershooting the mass accuracy provided by the mass spectrometer would prevent correct assignment of spectra. The smallest applicable tolerance may be determined by searching the data twice. Starting with a high mass tolerance, the settings for the second search would be adjusted in response to the results from the first. However, this strategy substantially increases data analysis time. We also observed that setting the tolerance based on results from data sets acquired with the same method is complicated by slight run-to-run differences in mass accuracy (data not shown). We addressed this problem by exploiting the above described external calibration procedure with polydimethylcyclosiloxane as calibrant ions. It was found that the mass accuracy distribution of background ions closely resembled that of peptide ions (Fig. 2C). Thus, after extracting background ions from the acquired MS survey or SIM spectra and mass recalibration of the MS data, the SD of the error observed for the recalibrated background ion m/z values can be used for defining the peptide mass search tolerance in subsequent database searching of MS/MS data (Fig. 2B). As we observed slightly narrower mass accuracy distributions for background ions than for peptide ions, we used five SD of background ion distributions as peptide mass search tolerance. The highest values for this SD in three runs acquired in this study were 1.2 ppm for FT1, and.63 ppm for SIM3 analyses. Therefore, we used 6 and 3 ppm as peptide mass search 13

14 tolerances for a SEQUEST database search of MS/MS spectra acquired with the FT1 and the SIM3 method, respectively. Searches were done against a composite target/decoy yeast protein sequence database, which allowed the estimation of incorrect assignments in the final data sets as described in Experimental Procedures (8, 1). We did the searches without using enzyme specificity constraints but filtered MS/MS spectra assignments for those to peptides with both termini consistent with trypsin specificity. When compared to matching MS/MS spectra only to fully-tryptic peptides, this search and filter strategy was recently shown to increase the number of peptide identifications with a defined false-positive rate from SEQUEST database searches of MS/MS spectra from tryptic digests (38). We confirmed these findings in the present study (data not shown). We applied XCorr and Cn cutoffs (see Supplementary Table 1) to remove incorrect assignments from the filtered tryptic data set such that the FP identification rate in the final data set was approximately 1%. Peptide assignments with this FP rate are defined here as confident peptide identifications. The closely identical FP rates in the final peptide identification data sets from both methods, FT1 and SIM3, allowed a fair comparison of the two MS/MS acquisition strategies. In order to study the effect of high mass accuracy on the peptide identification process from MS/MS spectra beyond the small differences observed for the accuracy achieved with the FT1 and the LTQ SIM TOP3 methods, we acquired a third MS/MS data set. The sample was analyzed in triplicate using the LTQ1 method. This produced a control data set resembling the quality of MS and MS/MS typical for ion trap mass spectrometers often used in large-scale proteomics studies. The workflow of this method is depicted in Supplementary Fig. 3. By omitting the acquisition of FT-ICR mass spectra the cycle time was slightly reduced to three seconds when compared with that observed for the FT1 method. LTQ1 MS/MS data were assigned with the SEQUEST algorithm in a similar fashion as described for the FT1 and the SIM3 data sets. The peptide mass tolerance was set to ±2Da, which is a typical value for the database searches of ion trap MS/MS data. XCorr and Cn thresholds ensured a FP-rate of approximately 1% for tryptic peptide assignments afforded from a search with no enzyme specificity constraints (Supplementary Table 1). Table 1 and Fig. 3 display average numbers of confident peptide assignments and inferred protein identifications from triplicate analyses of the studied yeast proteome fraction by each of the described LC-MS/MS methods. Strikingly, the FT1 data set including high mass 14

15 accuracy information produced only approximately 1% more confident peptide identifications than the LTQ1 data set acquired using only the LTQ. The number of protein identifications was nearly identical for both methods. The SIM3 data set provided the highest mass accuracy but gave the smallest number (3% fewer peptide and 4% fewer protein identifications) compared to the other two methods. This is best explained by the substantially reduced number of MS/MS spectra acquired with the SIM3 compared to that produced with the other two methods (Table 1, Fig. 3A). SIM experiments acquired with the SIM3 experiment reduced the number of MS/MS experiments relative to the LTQ1 and the FT1 methods by 7 and 6%, respectively (Table 1). Fig. 3B shows that peptide ions missed with the SIM3 method were mainly from proteins predicted to be of low abundance as they were encoded by genes with codon adaptation indices (CAI) of.2 and lower. In addition, a generally lower number of unique peptides identified for a given protein (Fig. 3C) accompanied this effect. We sought for a more detailed analysis of both phenomena by studying peptide ion S/N as well as CAI of genes encoding for the corresponding proteins for peptides missed with the SIM3 method. We extracted both values for peptides identified with the FT1 method and plotted their relationship for ions identified with both methods, FT1 and SIM3, and for ions correctly assigned only based on FT1 data (Fig. 4). It was observed that most of the peptide ions identified with both methods had a S/N of 1 or higher while a high portion of ions identified only with the FT1 method showed a S/N of lower than 1 (Fig. 4B and 4C). These low S/N values correlated well with a low CAI of corresponding genes for the derived proteins. The FT1 method also allowed increased sequence coverage for identified proteins by confident identification of peptides producing only small ion signals (Fig. 4D). The described costs of sacrificing a high MS/MS acquisition rate for obtaining a slightly enhanced mass accuracy were less surprising than the small general benefits high mass accuracy provided in the assignments of the acquired MS/MS spectra. FT1 data produced only 1% more confident peptide identifications than that acquired with the LTQ1 method. Furthermore, these additional peptide assignments primarily increased the sequence coverage for the inferred protein identifications but not their number (Fig. 3). The peptide mass search tolerances applied in the assignment of MS/MS spectra were ±2 Da for LTQ1 and only 6 ppm for FT1 data. This difference reduced the mass range for peptides matched in a SEQUEST database search of an FT1 MS/MS spectrum for a 1 Da peptide by a factor of approximately 3. The small 15

16 benefit provided by the applied narrow peptide mass search tolerance emphasizes the primary role of the information given by MS/MS fragment ions in the peptide identification process. It must be noted that the low resolution of LTQ survey MS data did not allow the determination of the charge states of peptide ions, and each MS/MS spectra of a potentially multiply charged peptide ion was considered here to be of either a doubly, triply, or quadruply charged ion. Therefore, the number of searched MS/MS spectra was almost three times the number of acquired spectra. This not only greatly increased the database search time but also increased the number of misassigned MS/MS spectra as only one charge state could possibly be correct. However, the number of confidently-assigned spectra showed that the tryptic state of the assigned peptides as well as XCorr scores (see Experimental Procedures) allowed effective discrimination of incorrect and correct assignments. We next sought to estimate the role of high mass accuracy for the analysis of samples different from those analyzed in this study. Although yeast is a well established model organism for proteomics studies, the size of the yeast ORF protein sequence database is relatively small (6427 entries) when compared to databases for other species (57478 entries in the Human International Protein Index sequence database, 9/5/25, To simulate assignment of MS/MS spectra against a larger database, we artificially extended the number of entries in the original yeast protein sequence database by a factor of nine using a Markov chain model. LTQ1 and FT1 data were searched against the new database as described above. We also aimed to study the influence of the quality of MS/MS spectra on the role peptide mass accuracy has on their assignment. Therefore, we reduced the S/N of the acquired MS/MS data by multiplying the intensity of lowabundance signals in the spectra by a factor of thirty before searching the manipulated data set against the original yeast protein sequence database. Because limited sample amount is an assumed reason for low S/N MS/MS spectra, we analyzed approximately.4 µg of the above described fraction of yeast proteome tryptic digest in triplicate by LC-MS/MS using the FT1 and the LTQ1 methods. The analyzed sample amount was 1/1 of the original sample amount. Results from these searches are summarized in Fig. 5 and Supplementary Table 2. Assigning MS/MS spectra using a larger database size slightly reduced the number of confident peptide identifications, a phenomenon that has been described previously (39). However, in 16

17 comparison to the search against the original target/decoy database, the effect of high mass accuracy information did not substantially change. FT1 data produced only 12% more confident peptide identifications than LTQ1 data. Reducing MS/MS spectrum quality caused a large decrease in the number of identified peptides from both data sets but the effect was more severe for the assignment of LTQ1 spectra. Only approximately half the number of confident peptide identifications was found. Analyzing a 1-fold decreased sample amount reduced the number of confidently-identified peptides by a factor of only approximately two. Comparing LTQ1 and FT1 methods showed that accurately measured peptide masses generated 2% more confidently-identified peptides. Considering the substantial decrease in analyzed sample amount, it was intriguing to observe a relative small differences in identified peptides (5%) compared to the original sample (Fig. 5D). We therefore analyzed another sample dilution of 25-fold by both methods with similar results (3% average decrease in identifications compared to original sample and only 1% difference in identified peptides by both methods data not shown). These data suggest the AGC function of the LTQ FT effectively compensated for the 1-fold decrease in the analyzed sample amount. Examples of relatively noisy MS/MS spectra are those of phosphopeptides. These spectra are often dominated by ions resulting from the neutral loss of phosphoric acid, while sequence specific fragment ions formed through cleavage of the peptide backbone amide bonds can be of low intensity showing a small S/N (4, 11). We sought to examine if our results for the assignment of artificially altered MS/MS data also applied to phosphopeptide MS/MS spectra. We used immobilized metal-affinity chromatography (IMAC) to enrich for phosphorylated peptides (41, 42) from the digest of a fraction of the yeast proteome and analyzed the sample using 6 min LC-MS/MS methods using the FT1 and the LTQ1 data acquisition strategies. In the database search of the acquired MS/MS data, we allowed serine, threonine, and tyrosine residues to be phosphorylated. We found that applying no enzyme specificity for the ±2Da peptide mass tolerance search of LTQ1 data followed by filtering for tryptic peptides lowered the number of confident peptide identifications when compared to a search only considering tryptic peptides (data not shown). The number of peptides potentially matching an MS/MS spectrum is greatly increased through allowing a variable modification on three different residues. This complicates the automated assignment of MS/MS spectra, especially when they may be of reduced quality (39). The non-specific database search slightly increased the number 17

18 of peptide assignments from the 6 ppm peptide mass tolerance of the acquired FT1 phosphopeptide MS/MS data. However, to use similar database search strategies we reduced the database size for the search of both data sets by considering only tryptic peptides. XCorr cutoffs were used to remove false positive assignments (Supplementary Table 1). Assignments to the same peptide modified at different positions usually show very similar XCorr values. Cn cutoffs would therefore remove a significant fraction of correct phosphopeptide identifications. Fig. 6 shows the results from the database searches at a 1% FP rate. Omitting the use of the FT- ICR part of the LTQ FT produced 2422 MS/MS spectra with the LTQ1 method, which were substantially more than the 1642 spectra acquired with the FT1 method. The smaller number of acquired MS/MS spectra for these samples was ascribed to the fact that the phosphopeptide sample was less complex, available in a smaller amount than the above described yeast lysate digest, and only analyzed using a 1-hr gradient. The high peptide mass accuracy provided by the FT1 method allowed the confident assignment of 396 phosphopeptide MS/MS spectra (1% FP rate). This was more than twice the number (19 spectra) assigned from the LTQ1 run. These results highlight the importance of mass accuracy for large-scale phosphorylation studies. Discussion MS/MS data are currently the main source of information for identifying large numbers of peptides generated by digestion of protein mixtures in so-called bottom-up mass spectrometry-based proteomics experiments. Recently, mass spectrometers with the capability to provide exceptionally high mass accuracy and resolution of precursor ions along with rapid MS/MS collection capabilities have become available for widespread use (43, 22, 24). However, the measured mass accuracies are currently not sufficient to overcome the necessity of acquiring MS/MS information for the identification of peptides (18). This becomes even more apparent when one considers the as yet incompletely understood level of complexity added to the proteome through post-translational modifications (44). However, accurately determined peptide masses are being increasingly used to support the automated assignment of MS/MS spectra to enhance the confidence in peptide identification (21, 45). Our study showed complementary roles for peptide precursor ion mass accuracy and the quality of information provided by MS/MS data in the identification of peptide ions. We found that the benefit of obtaining accurate masses for precursor ions would appear to be greatly 18

19 overestimated for the analysis of complex mixtures of unmodified peptides (Fig. 5A) even with highly diluted samples (Fig. 5D). Triplicate analyses of the yeast sample acquiring precursor ion masses with either the FT-ICR or a linear ion trap mass spectrometer (LTQ) gave surprisingly similar results (Fig. 5A). Although the FT-ICR detection allowed a database search of the acquired MS/MS data with a peptide mass tolerance of 6 ppm while ±2 Da was applied for searching LTQ data, the high mass accuracy only provided the confident identification of 1% more peptides and no observable difference in the number of protein identifications (Fig. 3A). We did find however that the role of mass accuracy in the peptide identification process was largely increased for the assignment of MS/MS spectra with low quality. This was shown for the assignment of MS/MS data with artificially lowered S/N ratio (Fig. 5C) and confirmed for the automated interpretation of phosphopeptide MS/MS spectra (Fig. 6). In the latter, b- and y- type ions often show low S/N as spectra from peptides phosphorylated at serine or threonine residues tend to be dominated by intense peaks from the neutral losses of phosphoric acid and water. For the interpretation of these MS/MS data, a 6 ppm peptide mass search space doubled the number of confidently-identified peptides. High mass accuracy data collected in the MS but not in the MS/MS mode support peptide identification through MS/MS database searches by reducing the number of peptides from the searched database which are potentially matching the acquired MS/MS data. It has been described previously that low quality spectra and large database sizes complicate the correct assignment of MS/MS data (39), which was confirmed in this study by showing the different roles of accurately determined peptide masses in the assignment of MS/MS data of different quality. Here, we did not study the role of high mass accuracy in MS/MS spectra. Although accurately measured fragment ion masses are expected to considerably increase the certainties associated with each individual peptide identification, using the FT-ICR mass spectrometer for analyzing fragment ions in large-scale proteomics experiments with the LTQ FT substantially decreases the number of MS/MS spectra acquired in an LC-MS/MS run (46, 13). Disadvantages of improving mass accuracy in an LC-MS/MS run at the cost of MS/MS acquisition rate were shown by a comparison of the SIM3 and the FT1 methods (Fig. 4). Although SIM scans provided a slightly better mass accuracy than the FT-ICR survey MS spectra acquired with the FT1 method, omitting SIM scans nearly tripled the number of acquired MS/MS scans resulting in a 1.5-fold increase for confident peptide and protein 19

20 identifications (Fig. 3A). We developed a simple mass recalibration procedure involving the use of commonly-detected polydimethylcyclosiloxane background ions as calibrants. Using this procedure, we substantially improved the mass accuracy in FT-ICR survey MS scans, which avoided a compromise between mass accuracy and MS/MS acquisition rate in LTQ FT largescale proteomics experiments. Acknowledgements This work was supported partially by National Institutes of Health grants GM67945 and HG3456 (S.P.G.). The authors also wish to thank D. Moazed, Department of Cell Biology, Harvard Medical School for providing yeast lysate. Note After submission of this work the use of a polydimethylcyclosiloxane ion as calibrant ion for mass recalibration in proteomics studies was published in a study by 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 (Mol. Cell. Proteomics 25, 4, ). 2

21 References 1. Aebersold, R. and Mann, M. (23) Mass spectrometry-based proteomics. Nature McCormack, A. L., Schieltz, D. M., Goode, B., Yang, S., Barnes, G., Drubin, D. and Yates, J. R., 3rd (1997) Direct analysis and identification of proteins in mixtures by LC/MS/MS and database searching at the low-femtomole level. Anal. Chem Eng, J. K., McCormack, A. L. and Yates, J. R., 3rd (1994) An approach to correlate tandem mass spectral data of peptides with amino acid sequences in a protein database. J. Am. Soc. Mass Spectrom Perkins, D. N., Pappin, D. J., Creasy, D. M. and Cottrell, J. S. (1999) Probability-based protein identification by searching sequence databases using mass spectrometry data. Electrophoresis Sadygov, R. G., Cociorva, D. and Yates, J. R., 3rd (24) Large-scale database searching using tandem mass spectra: looking up the answer in the back of the book. Nat Methods Nesvizhskii, A. I. and Aebersold, R. (25) Interpretation of shotgun proteomics data: the protein inference problem. Mol. Cell. Proteomics Keller, A., Nesvizhskii, A. I., Kolker, E. and Aebersold, R. (22) Empirical statistical model to estimate the accuracy of peptide identifications made by MS/MS and database search. Anal. Chem Moore, R. E., Young, M. K. and Lee, T. D. (22) Qscore: an algorithm for evaluating SEQUEST database search results. J. Am. Soc. Mass Spectrom Nesvizhskii, A. I., Keller, A., Kolker, E. and Aebersold, R. (23) A statistical model for identifying proteins by tandem mass spectrometry. Anal. Chem Peng, J., Elias, J. E., Thoreen, C. C., Licklider, L. J. and Gygi, S. P. (23) Evaluation of multidimensional chromatography coupled with tandem mass spectrometry (LC/LC- MS/MS) for large-scale protein analysis: the yeast proteome. J. Proteome Res Beausoleil, S. A., Jedrychowski, M., Schwartz, D., Elias, J. E., Villen, J., Li, J., Cohn, M. A., Cantley, L. C. and Gygi, S. P. (24) Large-scale characterization of HeLa cell nuclear phosphoproteins. Proc. Natl. Acad. Sci. U.S.A Olsen, J. V. and Mann, M. (24) Improved peptide identification in proteomics by two consecutive stages of mass spectrometric fragmentation. Proc. Natl. Acad. Sci. U.S.A Nielsen, M. L., Savitski, M. M. and Zubarev, R. A. (25) Improving protein identification using complementary fragmentation techniques in Fourier transform mass spectrometry. Mol. Cell. Proteomics Havilio, M., Haddad, Y. and Smilansky, Z. (23) Intensity-based statistical scorer for tandem mass spectrometry. Anal. Chem Elias, J. E., Gibbons, F. D., King, O. D., Roth, F. P. and Gygi, S. P. (24) Intensitybased protein identification by machine learning from a library of tandem mass spectra. Nat. Biotechnol Zubarev, R. A., Hakansson, P. and Sundqvist, B. (1996) Accuracy requirements for peptide characterization by monoisotopic molecular mass measurements. Anal. Chem

22 17. Clauser, K. R., Baker, P. and Burlingame, A. L. (1999) Role of accurate mass measurement (±1 ppm) in protein identification strategies employing MS or MS/MS and database searching. Anal. Chem Conrads, T. P., Anderson, G. A., Veenstra, T. D., Pasa-Tolic, L. and Smith, R. D. (2) Utility of accurate mass tags for proteome-wide protein identification. Anal. Chem Lipton, M. S., Pasa-Tolic, L., Anderson, G. A., Anderson, D. J., Auberry, D. L., Battista, J. R., Daly, M. J., Fredrickson, J., Hixson, K. K., Kostandarithes, H., Masselon, C., Markillie, L. M., Moore, R. J., Romine, M. F., Shen, Y., Stritmatter, E., Tolic, N., Udseth, H. R., Venkateswaran, A., Wong, K. K., Zhao, R. and Smith, R. D. (22) Global analysis of the Deinococcus radiodurans proteome by using accurate mass tags. Proc. Natl. Acad. Sci. U.S.A He, F., Emmett, M. R., Hakansson, K., Hendrickson, C. L. and Marshall, A. G. (24) Theoretical and experimental prospects for protein identification based solely on accurate mass measurement. J. Proteome Res Olsen, J. V., Ong, S. E. and Mann, M. (24) Trypsin cleaves exclusively C-terminal to arginine and lysine residues. Mol. Cell. Proteomics Chernushevich, I. V., Loboda, A. V. and Thomson, B. A. (21) An introduction to quadrupole-time-of-flight mass spectrometry. J. Mass Spectrom Marshall, A. G., Hendrickson, C. L. and Jackson, G. S. (1998) Fourier transform ion cyclotron resonance mass spectrometry: a primer. Mass Spectrom. Rev Syka, J. E., Marto, J. A., Bai, D. L., Horning, S., Senko, M. W., Schwartz, J. C., Ueberheide, B., Garcia, B., Busby, S., Muratore, T., Shabanowitz, J. and Hunt, D. F. (24) Novel linear quadrupole ion trap/ft mass spectrometer: performance characterization and use in the comparative analysis of histone H3 post-translational modifications. J. Proteome Res Zhang, L.-K., Rempel, D., Pramanik, B. N. and Gross, M. L. (25) Accurate mass measurements by Fourier transform mass spectrometry. Mass Spectrom. Rev Schwartz, J. C., Senko, M. W. and Syka, J. E. P. (22) A two-dimensional quadrupole ion trap mass spectrometer. J. Am. Soc. Mass Spectrom Schlosser, A. and Volkmer-Engert, R. (23) Volatile polydimethylcyclosiloxanes in the ambient laboratory air identified as source of extreme background signals in nanoelectrospray mass spectrometry. J. Mass Spectrom Verdel, A. and Moazed, D. (25) Labeling and characterization of small RNAs associated with the RNA interference effector complex RITS. Methods Enzymol Shevchenko, A., Wilm, M., Vorm, O. and Mann, M. (1996) Mass spectrometric sequencing of proteins silver-stained polyacrylamide gels. Anal. Chem Leptos, K. C., Sarracino, D. A., Jaffe, J. D., Krastins, B. and Church, G. M. (26) MapQuant: Open-source software for large-scale protein quantification. Proteomics Senko, M. W., Beu, S. C. and McLafferty, F. W. (1995) Determination of monoisotopic masses and ion populations for large biomolecules from resolved isotopic distributions. J. Am. Soc. Mass Spectrom

23 32. Sadygov, R. G., Cociorva, D. and Yates, J. R., 3rd (24) Large-scale database searching using tandem mass spectra: looking up the answer in the back of the book. Nat. Methods Ledford, E. B., Jr., Rempel, D. L. and Gross, M. L. (1984) Space charge effects in Fourier transform mass spectrometry. II. Mass calibration. Anal. Chem Brown, C. E. and Smith, M. J. C. (199) The present status and prospects for Fourier transform-ion cyclotron resonance/mass spectrometry. Spectroscopy World Masselon, C., Tolmachev, A. V., Anderson, G. A., Harkewicz, R. and Smith, R. D. (22) Mass measurement errors caused by "local" frequency perturbations in FTICR mass spectrometry. J. Am. Soc. Mass Spectrom Muddiman, D. C. and Oberg, A. L. (25) Statistical evaluation of internal and external mass calibration laws utilized in Fourier transform ion cyclotron resonance mass spectrometry. Anal. Chem Guan, S., Marshall, A. G. and Scheppele, S. E. (1996) Resolution and chemical formula identification of aromatic hydrocarbons and aromatic compounds containing sulfur, nitrogen, or oxygen in petroleum distillates and refinery streams. Anal. Chem Elias, J. E., Haas, W., Faherty, B. K. and Gygi, S. P. (25) Comparative evaluation of mass spectrometry platforms used in large-scale proteomics investigations. Nat. Methods Resing, K. A., Meyer-Arendt, K., Mendoza, A. M., Aveline-Wolf, L. D., Jonscher, K. R., Pierce, K. G., Old, W. M., Cheung, H. T., Russell, S., Wattawa, J. L., Goehle, G. R., Knight, R. D. and Ahn, N. G. (24) Improving reproducibility and sensitivity in identifying human proteins by shotgun proteomics. Anal. Chem DeGnore, J. P. and Qin, J. (1998) Fragmentation of phosphopeptides in an ion trap mass spectrometer. J. Am. Soc. Mass Spectrom Andersson, L. and Porath, J. (1986) Isolation of phosphoproteins by immobilized metal (Fe 3+ ) affinity chromatography. Anal. Biochem Ficarro, S. B., McCleland, M. L., Stukenberg, P. T., Burke, D. J., Ross, M. M., Shabanowitz, J., Hunt, D. F. and White, F. M. (22) Phosphoproteome analysis by mass spectrometry and its application to Saccharomyces cerevisiae. Nat. Biotechnol Makarov, A. (2) Electrostatic axially harmonic orbital trapping: a high-performance technique of mass analysis. Anal. Chem Peng, J. and Gygi, S. P. (21) Proteomics: the move to mixtures. J. Mass. Spectrom Yates, J. R., Cociorva, D., Liao, L. and Zabrouskov, V. (26) Performance of a mass analyzer with orbital trapping for peptide analysis. Anal. Chem Macek, B., Olsen, J. V., Zhang, Y. and Mann, M. (25) Assesment of high versus low mass accuracy MS/MS for complex mixture analysis in proteomics. Proc. 53rd ASMS Conf. on Mass Spectrom., San Antonio, TX, June 5-9, Sharp, P. M. and Li, W. H. (1987) The codon adaptation index - a measure of directional synonymous codon usage bias, and its potential applications. Nucleic Acids Res Ghaemmaghami, S., Huh, W. K., Bower, K., Howson, R. W., Belle, A., Dephoure, N., O'Shea, E. K. and Weissman, J. S. (23) Global analysis of protein expression in yeast. Nature

24 Figure Legends Figure 1. The SIM3 data-dependent MS/MS acquisition strategy (21). (A) Workflow and average cycle time: the acquisition of an FT-ICR survey MS scan was followed by an FT-ICR selected ion monitoring (SIM) scan and an LTQ MS/MS experiment on the three most abundant ions detected in the survey scan (periods of ion accumulation are shown in blue, those of ion analysis in green; R, resolution at m/z 4). (B) Peptide ion mass accuracy distribution after recalibration. A Gaussian distribution (orange) was fitted to the measured mass accuracy distribution (blue dots). The pre-calibration distribution for this data set is shown in Supp. Fig. 2). Figure 2. The FT1 data-dependent MS/MS acquisition strategy. (A) Workflow and average cycle time: an FT-ICR survey MS scan was followed by the acquisition of linear ion trap (LTQ) MS/MS experiments on the ten most abundant ions detected in the survey MS experiment (R, resolution at m/z 4). (B) Mass recalibration and MS/MS spectra assignment: survey MS data were recalibrated using polydimethylcyclosiloxane ions as calibrants. Mass accuracy distributions for these ions before and after recalibration are shown in (C1) and (C2). Yeast proteome tryptic digest MS/MS spectra including unmodified and recalibrated peptide ion masses were assigned using the SEQUEST algorithm. Mass accuracy distributions of the assigned peptide ions are shown in (C3, unmodified) and (C4, recalibrated). Peptide ion mass tolerances for the SEQUEST database searches were set according the mass accuracy distribution of polydimethylcyclosiloxane ions (approximately five standard deviations, 2 ppm for unmodified and 6 ppm for recalibrated data). Figure 3 Peptide and protein identifications from three different data-dependent MS/MS acquisition strategies. A fraction of yeast whole-cell lysate tryptic digest was analyzed using three data-dependent MS/MS acquisition strategies LTQ1, FT1, and SIM3. (A) The number of acquired MS/MS spectra, confidently assigned peptides (1% FP rate), and inferred protein identifications for the three methods (Table 1). (B) Codon adaptation index (CAI) distributions for genes encoding the identified proteins. CAI is a measure of protein abundance with low abundance proteins predicted to be encoded by genes with values <.2 (47, 48). (C) 24

25 Number of unique peptides identified per protein using different MS/MS acquisition strategies. Values are means +/- the SD from triplicate LC-MS/MS analyses. Figure 4. Comparison of peptide assignments from FT1 and SIM3 data. (A) 472 and 274 unique peptides were identified using the FT1 and the SIM3 methods, respectively, considering only proteins derived from at least two unique peptide assignments in three analyses. The overlap of these data sets is shown. The overlap of these data sets is shown (Venn diagrams showing the overlap of peptide and protein identifications from all three methods are shown in Supplementary Figs. 4-7). (B) Peptide ion signal-to-noise ratio (S/N) versus CAI of genes encoding for the inferred proteins relationship for peptides identified by both methods. (C) The same relationship for peptide ions identified by the FT1 but not the SIM3 method. For the depiction in (B2) and (C2) peptides were binned according their S/N (1-1; step-sizes 1 for 1-1, 1 for 1-1, 1 for 1-1, and 1 for 1-1) and CAI values (-1, step-size.25) and the number of peptide identifications was plotted for each bin. (D) S/N of peptide ions identified from phosphoglucomutase with both methods (blue) or only with the FT1 method (orange). Figure 5. The benefits of high mass accuracy on the peptide identification process from different sample amounts, data set compositions, and database sizes. (A) Confident peptide identifications from FT1 and LTQ1 data at a 1% FP rate. MS/MS spectra were searched with the SEQUEST algorithm against a target/decoy composite database including all known yeast protein sequences. (B) The same data sets were searched against a database generated by increasing the size of the yeast protein sequence database by a factor of nine using a Markov chain model (Experimental Procedures). (C) The S/N ratios from MS/MS spectra of the same data sets were artificially decreased, and the spectra were searched against the original yeast protein sequence database. (D) The sample amount analyzed by both methods was decreased 1-fold and the acquired MS/MS spectra were searched against the original yeast protein sequence database. Values are mean +/- SD from triplicate analyses. Figure 6. The benefits of high mass accuracy for assigning phosphopeptide MS/MS spectra. Phosphopeptides were enriched from a fraction of yeast whole-cell lysate by 25

26 immobilized metal affinity chromatography (IMAC) (Experimental Procedures). Identical sample amounts were analyzed using the FT1 and the LTQ1 strategies in 6 min LC-MS/MS runs. The numbers of acquired MS/MS spectra and confident phosphopeptide identifications from both data sets are shown. 26

27 Tables Table 1. Number of MS/MS spectra, confident peptide identifications, and inferred protein identification from analyzing a fraction of yeast proteome digest using three different datadependent MS/MS acquisition strategies in 9 min LC-MS/MS analyses (mean and SD from three analyses). Peptide and protein identification false positive rates are given in parentheses. LTQ1 FT1 SIM3 Acquired MS/MS spectra 17627± ±17 592±37 Peptide IDs 3663±29 (.9±.3%) 434±36 (1.±.1%) 275±28 (1.±.3%) Protein IDs 394±7 (9±3) 397±5 (9±) 236±6 (1±3) 27

28 Figure 1. A LTQ IonAccumulationfor Ion Accumulation SIM Scans (5x1 4 ions) for Full MS (1x17 ions) MS/MS MS/MS MS/MS FTICR Full MS Scan (R = 2.5x14) SIM MS Scans (R = 5x14) B Peptide IDs Time [ms] SIM3 Peptide Ions Mass Accurcay -.41 ±.44 ppm Mass accuracy (ppm) Figure 1. The SIM3 data-dependent MS/MS acquisition strategy (21). (A) Workflow and average cycle time: the acquisition of an FT-ICR survey MS scan was followed by an FT-ICR selected ion monitoring (SIM) scan and an LTQ MS/MS experiment on the three most abundant ions detected in the survey scan (periods of ion accumulation are shown in blue, those of ion analysis in green; R, resolution at m/z 4). (B) Peptide ion mass accuracy distribution after recalibration. A Gaussian distribution (orange) was fitted to the measured mass accuracy distribution (blue dots). The pre-calibration distribution for this data set is shown in Supp. Fig. 2). 28

29 Figure 2. A Ion Accumulation for Full MS (1x16 ions) MS/MS LTQ FTICR Full MS Scan (R = 1x15) Time [ms] B Raw FT survey MS data Mass calibration based on polydimethylcyclosiloxane background ions (BGIs) C Background Ions Background ions Calibr. FT survey MS data LTQ MS/MS data Background ions only (pre-calibration) Mass accuracy 3.82± 1.65 ppm Mass accuracy (ppm) Background ions only (calibrated) Mass accuracy.4± 1.1 ppm Mass accuracy (ppm) Peptide IDs Peptide IDs SEQUEST MS/MS database search Peptide ion mass tolerance defined by BGI mass accuracy Peptide ions (pre-calibration) Mass accuracy 4.99± 2.42 ppm Mass accuracy (ppm) Peptide ions (calibrated) Mass accuracy -.25± 1.46 ppm Mass accuracy (ppm) Figure 2. The FT1 data-dependent MS/MS acquisition strategy. (A) Workflow and average cycle time: an FT- ICR survey MS scan was followed by the acquisition of linear ion trap (LTQ) MS/MS experiments on the ten most abundant ions detected in the survey MS experiment (R, resolution at m/z 4). (B) Mass recalibration and MS/MS spectra assignment: survey MS data were recalibrated using polydimethylcyclosiloxane ions as calibrants. Mass accuracy distributions for these ions before and after recalibration are shown in (C1) and (C2). Yeast proteome tryptic digest MS/MS spectra including unmodified and recalibrated peptide ion masses were assigned using the SEQUEST algorithm. Mass accuracy distributions of the assigned peptide ions are shown in (C3, unmodified) and (C4, recalibrated). Peptide ion mass tolerances for the SEQUEST database searches were set according the mass accuracy distribution of polydimethylcyclosiloxane ions (approximately five standard deviations, 2 ppm for unmodified and 6 ppm for recalibrated data). 29

30 Figure 3. LTQ1 FT1 SIM3 A MS/MS spectra Peptide IDs Protein IDs B C Number of unique peptides Number of genes Number of unique peptides CAI Protein number Protein number Figure 3 Peptide and protein identifications from three different data-dependent MS/MS acquisition strategies. A fraction of yeast whole-cell lysate tryptic digest was analyzed using three data-dependent MS/MS acquisition strategies LTQ1, FT1, and SIM3. (A) The number of acquired MS/MS spectra, confidently assigned peptides (1% FP rate), and inferred protein identifications for the three methods (Table 1). (B) Codon adaptation index (CAI) distributions for genes encoding the identified proteins. CAI is a measure of protein abundance with low abundance proteins predicted to be encoded by genes with values <.2 (47, 48). (C) Number of unique peptides identified per protein using different MS/MS acquisition strategies. Values are means +/- the SD from triplicate LC-MS/MS analyses. 3

31 Figure 4. A FT1, SIM3, 274 B FT1/SIM3 overlap C FT1 unique S/N 1 S/N D CAI 2 2 Pept. IDs CAI S/N FT1 unique S/N Pept. IDs FT1/SIM overlap CAI CAI Phosphoglucomutase (PGM2, YMR15C, CAI.38) S/N Figure 4. Comparison of peptide assignments from FT1 and SIM3 data. (A) 472 and 274 unique peptides were identified using the FT1 and the SIM3 methods, respectively, considering only proteins derived from at least two unique peptide assignments within three analyses. The overlap of these data sets is shown. The overlap of these data sets is shown (Venn diagrams showing the overlap of peptide and protein identifications from all three methods are shown in Supplementary Figs. 4-7). (B) Peptide ion signal-to-noise ratio (S/N) versus CAI of genes encoding for the inferred proteins relationship for peptides identified by both methods. (C) The same relationship for peptide ions identified by the FT1 but not the SIM3 method. For the depiction in (B2) and (C2) peptides were binned according their S/N ratios (1-1; step-sizes 1 for 1-1, 1 for 1-1, 1 for 1-1, and 1 for 1-1) and CAI values (-1, step-size.25) and the number of peptide identifications was plotted for each bin. (D) S/N of peptide ions identified from phosphoglucomutase with both methods (blue) or only with the FT1 method (orange). 31

32 Figure 5. A Original data set B Increased database size C Peptide IDs Peptide IDs % Decreased spectrum quality (S/N) FT1 LTQ1 +9% Peptide IDs D Peptide IDs FT1 LTQ1 1-fold decreased sample amount FT1 LTQ1 FT1 LTQ1 +12% +2% Figure 5. The benefits of high mass accuracy on the peptide identification process from different sample amounts, data set compositions, and database sizes. (A) Confident peptide identifications from FT1 and LTQ1 data at a 1% FP rate. MS/MS spectra were searched with the SEQUEST algorithm against a target/decoy composite database including all known yeast protein sequences. (B) The same data sets were searched against a database generated by increasing the size of the yeast protein sequence database by a factor of nine using a Markov chain model (Experimental Procedures). (C) The S/N ratios from MS/MS spectra of the same data sets were artificially decreased, and the spectra were searched against the original yeast protein sequence database. (D) The sample amount analyzed by both methods was decreased 1-fold and the acquired MS/MS spectra were searched against the original yeast protein sequence database. Values are mean +/- SD from triplicate analyses. 32

33 Figure 6. MS/MS Spectra Phosphopeptide IDs FT1 LTQ LTQ1 +18% Figure 6. The benefits of high mass accuracy for assigning phosphopeptide MS/MS spectra. Phosphopeptides were enriched from a fraction of yeast whole-cell lysate by immobilized metal affinity chromatography (IMAC) (Experimental Procedures). Identical sample amounts were analyzed using the FT1 and the LTQ1 strategies in 6 min LC-MS/MS runs. The numbers of acquired MS/MS spectra and confident phosphopeptide identifications from both data sets are shown. FT1 33

34 Supplementary Data From Optimization and Use of Peptide Mass Measurement Accuracy in Shotgun Proteomics Wilhelm Haas, Brendan K. Faherty, Scott A. Gerber, Joshua E. Elias, Sean A. Beausoleil, Corey E. Bakalarski, Xue Li, Judit Villén, and Steven P. Gygi Department of Cell Biology, Taplin Biological Mass Spectrometry Facility, Harvard Medical School, Boston, MA

35 Tables Supplementary Table 1. Applied SEQUEST score (XCorr and Cn) cutoffs, peptide search mass tolerances, and enzyme specificity constraints (N.S., non-specific search, assignments were filtered for tryptic peptides; T, tryptic search). Separate XCorr and Cn cutoffs were applied for each ion charge state +2, +3, and +4. XCorr Cn Pept. mass Enz. spec. search search tolerance constraint Unmodified peptides LTQ1 Original sample ±2 Da N.S. Increased database size ±2 Da N.S. Decreased spectrum quality ±2 Da N.S. 1-fold decreased sample amount ±2 Da N.S. FT1 Original sample ppm N.S. Increased database size ppm N.S. Decreased spectrum quality ppm N.S. 1-fold decreased sample amount ppm N.S. SIM3 Original sample ppm N.S. Phosphopeptides LTQ ±2 Da T FT ppm T 35

36 Supplementary Table 2. Influence of data set composition, database size and sample amount on the number of confident peptide identifications from linear ion trap (LTQ) and LTQ FT MS/MS spectra (mean and SD from three analyses). Peptide and protein identification false positive rates are given in parentheses. LTQ1 peptide IDs FT1 peptide IDs Original data set 3663±29 (.9±.3%) 434±17 (1.±.1%) Increased database size 3176±15 (.9±.2%) 356±48 (1±.1%) Decreased spectrum quality (S/N) 327±11 (.8±.4%) 62±45 (.7±.3%) 1-fold decreased sample amount 184±53 (.9±.3%) 2167±4 (.9±.3%) 36

37 A Relative Abundance B C TIC No. of background calibrant ions x1 6 3.x x1 6 2.x x Retention time (min) Retention time (min) 1.x1 6 5.x Retention time (min) Supplementary Figure 1. (A) FT-ICR survey MS base peak chromatogram from analyzing the tryptic digest of a fraction of the yeast proteome using the FT1 method. (B) Distribution of polydimethylcyclosiloxane background ions over the FT1 analysis presented in (A). The blue dots indicate the number of calibrant ions per survey MS spectrum. A moving average of this number over ten MS spectra is shown by the orange line. (C) Total ion current (TIC) for the FT-ICR survey MS spectra acquired over the course of same run. Blue dots represent individual values for each spectrum, the orange line is a moving average over ten spectra. TIC values were calculated by multiplying the normalized TIC values (see Experimental Procedures) stored in Xcalibur RAW files by the ion accumulation time in seconds. 37

38 Peptide IDs SIM3 Peptide ions Mass accurcay 1.68 ±.42 ppm Mass accuracy (ppm) Supplementary Figure 2. Peptide ion mass accuracy distribution from a SIM3 run. The data were not subjected to mass recalibration. A Gaussian distribution (orange) was fitted to the measured mass accuracy distribution (blue dots). Full MS MS/MS Time [ms] Supplementary Figure 3. Workflow of the LTQ1 method. Survey MS as well as MS/MS spectra were acquired only using the linear ion trap (LTQ) part of the LTQ FT (periods of ion accumulation are shown in blue, those of ion analysis in green). 38

39 Supplementary Figure 4. Venn diagram showing the overlap of unique peptide identifications from analyzing a trypsinized fraction of the yeast proteome using the FT1, the LTQ1, and the SIM3 methods. Overlapping subsets are defined by the set operators (union), (intersection), and / (difference). Only peptides from proteins derived from at least two unique assignments within three analyses were considered. In the absence of any decoy database match in a subset of identifications, the FP-rate was estimated to be smaller than that calculated for the hypothetical presence of one decoy database sequence. The FP rate was calculated based on the number of decoy database matches. For example in the FT (total) group, 2 decoy matches were identified amongst 472 total matches. 2 x #decoy/total x 1% = (2x2/472)x1%=.9%. Venn diagrams showing the peptide identification overlap for the FT1 replicates are shown in Supplementary Figures 6 and 7. Redundant peptides were defined as having identical amino acid sequence, modification state, and charge state in any analysis. Unique peptides were defined as having different amino acid sequence, modification state, or charge state across all analyses. 39 Downloaded from by guest on November 21, 218

40 Supplementary Figure 5. Venn diagram showing the overlap of protein identifications from analyzing a trypsinized fraction of the yeast proteome using the FT1, the LTQ1, and the SIM3 methods. Overlapping subsets are defined by the set operators (union), (intersection), and / (difference). Only proteins derived from at least two unique assignments within three analyses were considered (see Supplementary Fig. 4). In the absence of any decoy database match in a subset of identifications, the FP-rate was estimated to be smaller than that calculated for the hypothetical presence of one decoy database sequence. The FP rate was calculated based on the number of decoy database protein matches. For example in the FT1 (total) subset, 1 decoy protein was identified amongst 318 total proteins. 2 x #decoy/total x 1% = (2x1/318)x1%=.63%. 4 Downloaded from by guest on November 21, 218

41 Supplementary Figure 6. Venn diagram showing the overlap of peptide identifications from analyzing a trypsinized fraction of the yeast proteome using the FT1 method in triplicate. Overlapping subsets are defined by the set operators (union), (intersection), and / (difference). Only unique peptide identifications were considered. An average number of 434 ± 36 identifications (including redundancy) was identified per analysis (Table 1). Redundant peptides were defined as having identical amino acid sequence, modification state, and charge state in any analysis. Unique peptides were defined as having different amino acid sequence, modification state, or charge state across all analyses. 41 Downloaded from by guest on November 21, 218

42 Supplementary Figure 7. Venn diagram showing the overlap of peptide identifications from analyzing a trypsinized fraction of the yeast proteome using the FT1 method in triplicate. Overlapping subsets are defined by the set operators (union), (intersection), and / (difference). Only unique peptides from proteins derived from at least two unique peptide assignments were considered. In the absence of any decoy database match in a subset of identifications, the FP-rate was estimated to be smaller than that calculated for the hypothetical presence of one decoy database sequence. A comparison of the 472 total unique peptides from three analyses to peptide identifications from LTQ1 and SIM3 runs is shown in Fig. 4 and in Supplementary Fig Downloaded from by guest on November 21, 218