|
|
J Proteome Res. Author manuscript; available in PMC 2009 January 1. Published in final edited form as: | PMCID: PMC2538956 NIHMSID: NIHMS62875 |
Characterization of global yeast quantitative proteome data generated from the wild-type and glucose repression Saccharomyces cerevisiae strains: the comparison of two quantitative methods Renata Usaite,1,2 James Wohlschlegel,2 John D. Venable,2 Sung K. Park,2 Jens Nielsen,1 Lisbeth Olsson,1 and John R. Yates, III2* 1 BioCentrum-DTU, Technical University of Denmark, Kgs. Lyngby, Denmark 2 The Scripps Research Institute, La Jolla, California The quantitative proteomic analysis of complex protein mixtures is emerging as a technically challenging but viable systems-level approach for studying cellular function. This study presents a large-scale comparative analysis of protein abundances from yeast protein lysates derived from both wild-type yeast and yeast strains lacking key components of the Snf1 kinase complex. Four different strains were grown under well-controlled chemostat conditions. Multidimensional protein identification technology followed by quantitation using either spectral counting or stable isotope labeling approaches was used to identify relative changes in the protein expression levels between the strains. A total of 2388 proteins were relatively quantified and more than 350 proteins were found to have significantly different expression levels between the two strains of comparison when using the stable isotope labeling strategy. The stable isotope labeling based quantitative approach was found to be highly reproducible among biological replicates when complex protein mixtures containing small expression changes were analyzed. Where poor correlation between stable isotope labeling and spectral counting was found, the major reason behind the discrepancy was the lack of reproducible sampling for proteins with low spectral counts. The functional categorization of the relative protein expression differences that occur in Snf1-deficient strains uncovers a wide range of biological processes regulated by this important cellular kinase. Keywords: CenSus, MudPIT, Snf4, AMP-activated kinase, 15N, mass spectrometry, proteomics In the field of systems biology, the integration of different –omics technologies is critical for understanding the cell as a whole. Compared to transcriptomics, proteomics continues to remain elusive in terms of generating and integrating experimental data into interpretive and predictive models. Since proteome analysis provides complementary information about biological systems and pathways, it is essential that new proteomic methodologies are developed in order to provide a comprehensive look at a biological process. The proteome complexity is overwhelming and based on Human Proteome Project findings there are far fewer protein-coding genes in the human genomes than proteins in the human proteome, which is estimated to reach 1,000,000 protein isoforms. 1 Diversity of post-translation modifications, alternative splicing and dynamics of protein complexes generate this complexity, which holds important biological information. Detection and understanding of the complete proteome will therefore allow uncovering of global signaling and kinetic traits. However, the complexity of the proteome remains one of the major issues preventing us from having analytical tools that allow comprehensive detection and/or quantification of the ‘complete’ proteome. Various tools combining mass spectrometry and liquid chromatography or gel-based protein separation, are being developed and used to improve complex proteome identification. 2 Multidimensional protein identification technology (MudPIT) has emerged as a sensitive tool for the separation and identification of proteins from complex mixtures. 3, 4 MudPIT, in combination with the metabolic or in vitro labeling of proteins, generates accurate and abundant quantitative proteome results. 5, 6 Other proteome identification and quantification tools based on protein chip technology or other mass spectrometry-based strategies find their niche in specific proteome areas as reviewed in Patterson et al.7Accurate quantitative global proteomics data is critical for studying global cellular processes and for integrating the data with other –omics data. Different approaches have been and are being developed to quantify protein expression differences using mass spectrometry. 8-10 Relatively few comparative studies of different quantitative proteome approaches have been performed (when real biological samples were used) in order to evaluate accuracy and reproducibility of quantitative proteome results. 11, 12 Zybailov and co-authors have compared quantitative proteome outputs generated by spectral counting and stable isotope labeling approach, using RelEx. 13 They argued that a strong correlation between two proteome quantification methods can be obtained and that spectral counting was more reproducible and capable of quantifying proteins over a wider dynamic range. In this study we used the yeast Saccharomyces cerevisiae as a model system. In order to evaluate the effect of a genetic perturbation, we evaluated three deletion strains, where Snf1 kinase, its regulatory subunit Snf4 or both Snf1 and Snf4 complex subunits were deleted, in comparison to the wild-type strain. The Snf1 protein kinase complex plays a role in the glucose repression signaling cascade and regulates carbon metabolism through phosphorylation of transcription factors such as Mig1 and Cat8. 14 Snf1 has been found to directly regulate the activity of signaling proteins such as Hog85, metabolic proteins such as Acc1, and directly phosphorylates histone H3. 15, 16 Various small scale studies on the Snf1 protein imply that Snf1, as well as its mammalian homolog AMPK, might be a global regulator and energy balancer in the cell. 17 It is therefore interesting to perform a global scale quantitative proteome studies on the Snf1-deficient strains, as we expect to affect the expression of a large number of proteins when the Snf1 kinase complex is dysfunctional. In this study a global quantitative proteome dataset was generated, analyzed using two different quantitative strategies and its quality was evaluated using classical statistics, and prior biological knowledge. Four yeast strains ( Δsnf1, Δsnf4, Δsnf1Δsnf4 and wild-type strain) were metabolically 14N- and 15N- labeled and pre-grown in biological triplicates in well-controlled steady state carbon limited chemostat cultivations, which are highly reproducible and suitable for global scale comparative studies 18. Based on previous studies, 14, 19 it was expected that the Snf1-deficient strains had different transcript and consequently altered protein expression profiles (compared to the wild-type strain) due to lasting glucose repression effect under carbon limitation. Global protein extracts were generated from the yeast culture samples and analyzed using online multidimensional fractionation coupled with tandem mass spectrometry using an LTQ-Orbitrap mass spectrometer 20. Stable isotope labeling and spectral counting approaches were used to calculate protein expression differences and the two quantitative proteomic outputs were statistically evaluated and compared. Yeast strains All S. cerevisiae strains used in this study were generated from the CEN.PK 113-7D laboratory strain (Scientific Research & Development GmbH, Oberursel, Germany) ( Table 1). | Table 1Yeast Saccharomyces cerevisiae strains used in this study |
Chemostat Cultivation of cells Steady state aerobic chemostat cultures were grown at 30 °C in 2 L bioreactors (Applikon) using a working volume of 0.5 L and a dilution rate of D = 0.100 (± 0.005) h -1. Cultures were fed with a modified minimal medium containing 75 mM of nitrogen and 250 mM of carbon (calculated on a per atom basis), which ensured that growth was limited by the carbon source and nitrogen was in excess. The sole difference in growth conditions among the strains used was that the wild-type strain was grown in >99% ammonium- 15N sulfate, >99 atom % 15N, manufactured by Spectra Stable Isotopes (Columbia, MD) while the deletion Δsnf1, Δsnf4 and Δsnf1Δsnf4 strains were grown in 99% ammonium- 14N sulfate, 99 atom % 14N, manufactured by ISOTEC INC (Miamisburg, OH). The pH was measured on-line and kept constant at 5.0 by automatic titration with 4 M KOH using an Applikon (ADI1030) bio-controller. The stirring speed was set to 800 rpm and the dry airflow rate was 0.5 L/min. The exhaust gas from chemostat cultivation was led through a condenser and the mole-% of carbon dioxide and oxygen was measured on-line using a PC-controlled acoustic gas analyzer (Brüel & Kjær, Denmark). After batch cultivation, dry weight, metabolite concentrations, and gas profiles were monitored until they reached steady state (constant over at least five-residence times) after which samples for protein extraction were taken. All samples were collected no later than 10-15 residence times from the start of continuous operation: (i) to avoid any strain adaptation that generally occurs over long-term cultivation, 21 and (ii) to ensure that metabolic labeling was enriched to > 99%. Media The carbon limited minimal medium composition was based on the study described by Verduyn et al.22 The amounts per liter of the compounds were: (NH 4) 2SO 4, 5 g; KH 2PO 4, 3 g; MgSO 4·7H 2O, 0.5 g; D-glucose, 7.5 g; Antifoam 289 (A-5551, Sigma-Aldrich), 0.05 ml; EDTA (Titriplex III ®), 15.0 mg; ZnSO 4·7H 2O, 4.5 mg; MnCl 2·2H 2O, 0.82 mg; CoCl 2·6H 2O, 0.3 mg; CuSO 4·5H 2O, 0.3 mg; Na 2MoO 4·2H 2O, 0.4 mg; CaCl 2·2H 2O, 4.5 mg; FeSO 4·7H 2O, 3.0 mg; H 3BO 3, 1.0 mg; KI, 0.1 mg, biotin, 0.05 mg; p-benzoic acid, 0.2 mg; nicotinic acid, 1.0 mg; Ca-pantothenate, 1.0 mg; pyridoxin, HCl, 1.0 mg; thiamin, HCl 1.0 mg; m-inositol, 25.0 mg. Isolation and extraction of yeast total protein pool Samples for total protein isolation were taken from the chemostat cultivations by rapidly sampling of 2 × 200 mL of culture into a 500 mL centrifuge tube containing 200 mL of crushed ice. Cells were quickly pelleted (4500 rpm at 0 °C for 2 min), instantly frozen by dropwise addition into liquid nitrogen, and stored at − 80 °C until further analysis. Cells were lysed in a buffer containing 7.4 mL/L of β-mercaptoethanol and 7.4 g/L of NaOH while incubated at 4 °C for 20 minutes. The proteins were precipitated by using 25% TCA (v/v) and ice-cold acetone. The pellet was air-dried and the protein fraction was obtained by extracting the pellet with a 1 × Invitrosol™ (Invitrogen) and 8M urea mixture. The protein concentration was determined using the BCA™ Protein Assay Kit (PIERCE). Lysates from 14N-labeled and 15N-labeled samples were mixed 1:1 by protein weight and subjected to protein digestion. Digest of extracted protein pool 200 μg of total protein (100 μg of 14N-labeled and 100 μg of 15N-labeled) was reduced by adding Tris(2-Carboxyethyl) phosphine (TCEP) to 5 mM and incubating at 20 °C for 30 min. The reduced sample was then carboxyamidomethylated by adding iodoacetamide (IAA) to 10 mM and incubating at 20 °C for 20 min in the dark. Endoproteinase LysC was added at an enzyme/substrate ratio of 1:100 and the samples were incubated for 4 h at 37 °C. The samples were subsequently diluted to 2M urea with 100 mM Tris-HCl, pH 8.5, brought to 1 mM CaCl2, and further digested by adding trypsin at enzyme/substrate ratio of 1:20 and incubating overnight at 37 °C. The digestion reaction was quenched by adding formic acid to 5% (v/v) to reduce the pH to 2-3. The peptide mixture was purified by solid-phase extraction using SPEC-PLUS PTC18 cartridges (Ansys Diagnostics, Lake Forest, CA). Samples were freeze-dried and resuspended in 5% formic acid solution. Samples not immediately analyzed were stored at -80 °C. MudPIT analysis The protein pool digest was pressure-loaded onto a 250 μM ID fused silica capillary column with a filtered union (UpChurch Scientific, Oak Harbor, WA) that was previously packed with 3 cm of 5-μm Partisphere strong cation exchanger (Whatman, Clifton, NJ) followed by 3 cm 5-μm Aqua C18 material (Phenomenex, Ventura, CA). After loading, this trapping column was washed with buffer containing 5% acetonitrile/0.1% formic acid. After desalting, a 100 μm i.d. capillary with a 5-μm pulled tip packed with 18 cm 3-μm Aqua C18 material (Phenomenex, Ventura, CA) was attached to the filter union and the entire split-phase column (desalting column–filter union–analytical column) was placed inline with an Eksigent nanoLC-2D HPLC (Dublin, CA) and analyzed using a 7-step separation modified from that described previously 3, 4. A further optimization of MudPIT was performed in this study. It showed that extending the length of the analytical column from 10 to 18 cm and reducing the number of chromatography steps from 12 to 7, while keeping total analysis time the same (24 hours), improved peptide detection and resulted in identification of 30% more protein IDs. The buffers used were 5% acetonitrile/0.1% formic acid (buffer A), 80% acetonitrile/0.1% formic acid (buffer B), and 500 mM ammonium acetate/5% acetonitrile/0.1% formic acid (buffer C). Steps 1 and 7 had a 120 min lasting buffer B gradient profile 2-90%. Steps 2-6 had the following profile: 10 min of 98% buffer A, 5 min of X% salt pulse, a 10 min gradient from 2-10% buffer B, a 180 min gradient from 10-50% buffer B, and a 15 min gradient from 50-90% buffer B. The 5 min salt pulse percentages (X) were 0, 10, 20, 30, 50, 70, 100%, respectively, for the 7-step analysis. As peptides eluted from the microcapillary column, they were electrosprayed directly into a linear ion trap/Orbitrap (LTQ-Orbitrap) hybrid mass spectrometer (Thermo Electron Corp., Bremen, Germany) with the application of a distal electrospray voltage of 2.5 kV versus the inlet of the mass spectrometer. A cycle of one full-scan mass spectrum (150-2000 m/z) collected in the orbitrap mass analyzer followed by 4 data-dependent MS/MS spectra collected in the LTQ was repeated continuously throughout each step of the multidimensional separation. More detailed description of LTQ-Orbitrap settings is described by Yates et al.20 Application of mass spectrometer scan functions and HPLC solvent gradients were controlled by the Xcalibur datasystem. Analysis of Tandem Mass Spectra MS/MS were analyzed using the following software analysis protocol. MS/MS remaining after filtering were searched with the SEQUEST™ algorithm 23 against a database of Saccharomyces cerevisiae ORFs downloaded from the Saccharomyces Genome Database (SGD) on March 3, 2005. This database was concatenated to a decoy database in which the sequence for each entry in the original database was reversed in order to estimate the false positive rate from the search 24. No enzyme specificity was considered for any search. Two separate SEQUEST parameter files were prepared and SEQUEST™ was run twice on each of ms2 files to separately sequence peptides that were either 14N- or 15N-labeled. SEQUEST results were assembled and filtered using DTASelect2.0 program, an improved version of DTASelect 25 that uses a linear discriminant analysis to dynamically set XCorr and DeltaCN thresholds for the entire dataset to achieve a user-specified false positive rate (5% in this analysis). The false positive rates are estimated by the program from the number and quality of spectral matches to the decoy database. Quantification of the relative protein abundances DTASelect2.0 output files were submitted to CenSus as described in Venable et al.9 to calculate the relative protein abundance differences based on reconstructed ion chromatograms (i.e. 14N/ 15N CenSus Ratio). The same DTASelect2.0 output files were also used to calculate normalized spectral counts (NSpC) for each protein k: in which the total number of tandem MS spectra matching peptides from protein k ( SpC) was divided by the protein's length ( L), then divided by the sum of SpC/L for all N proteins identified. Spectral counts serve as a parameter for estimating protein abundances and have been used for calculating relative protein abundance differences. 8, 26 Normally spectral counts are merged among all experiments in order to average variation and increase the number of significantly detected peptides. In this study, relative protein abundance differences based on normalized 14N- and 15N- spectral counting were calculated in each MudPIT experiment separately and compared to relative quantification of the extracted ion chromatograms from each labeled and unlabeled peptide pair using the Census algorithm. For the purpose of this paper, this is an effective way to compare spectral counting versus stable isotope labeling since both methods are applied to the same dataset which eliminates run to run variation and offers insight into the strengths and weaknesses of both quantitation methods. Normalization and statistical data analysis Calculated spectral counts in each experiment were normalized based on protein length and a total number of spectra detected per experiment. In addition, linear normalization was used in those cases, when the median of the protein abundance ratios (calculated using normalized spectral counts) was not equal to one. In stable isotope labeling based quantification, normalization is performed using the CenSus algorithm (Venable et al, 2007). A CenSus algorithm was used to correct inaccurate 15N- and 14N ratios- based on errors in sample mixing. This correction is based on the assumption that the natural log of all ratios will form a Gaussian distribution and the median of all ratios in a properly mixed sample should be equal to zero. These computational steps assured that quantified relative protein abundance ratios between mass spectrometry runs could be fairly compared. Outputs generated by spectral counting and stable isotope labeling were subjected to comparative analysis. A t-test using a threshold of P < 0.05 was used to describe biological variance. A hypergeometric distribution test was used to define biological processes, in which the proteins, with significantly (P < 0.05) changed abundances, were enriched (hypergeometric test: P < 0.01) in response to the disruption of Snf1 kinase complex. Physiological profile of strains used in this study Three S. cerevisiae yeast deletion strains, Δsnf1, Δsnf4, Δsnf1Δsnf4, and the wild-type strain were cultivated in biological triplicates in glucose-limited chemostat cultivations at a dilution rate of 0.1 h-1. The only nitrogen source used in these cultivations was ammonium sulphate -- 15N-labeled ammonium sulphate for the wild-type strain and 14N-labeled ammonium sulphate for the deletion strains. At steady state, yeast performed only respiratory metabolism and biomass yield on glucose was found to be 0.50 g g-1 for the wild-type strain and 0.45 g g-1 for the Δsnf1, Δsnf4 and Δsnf1Δsnf4 strains. It was assumed that total protein was constant in the deletion as well as wild-type strains since no major differences in morphology or distribution throughout cell-cycle phases were observed while performing microscopy screening. The Δsnf1, Δsnf4 and Δsnf1Δsnf4 strains were compared to the wild-type strain by analyzing them in 1:1 total protein mixtures derived from one of the 99%-14N-labeled deletion strains and 99%-15N-labeled wild-type strain (details in materials and methods). Characterization of identified proteome One sample of each yeast lysate was generated and analyzed using MudPIT. The total number of proteins identified per each strain was calculated by merging spectral counting data from three biological replicates ( Table 2). On average, 1600 proteins based on a 2-peptide threshold (and 2300 proteins based on 1-peptide threshold) were identified per one 24-hour MudPIT experiment. The number of identified proteins in a given time frame found in this study was compatible with the latest achievements that were generated using other global protein identification technologies like GelC-MS/MS or three-dimensional LC-MS/MS. 27, 28 A previous immunoblotting TAP Western study (referred to as the ‘control study’ in this report) 29 has estimated that over 4000 proteins are expressed in asynchronously growing yeast cultures. This suggested that using a 2-peptide per protein threshold, we identified less than half of the possible proteins present in the global proteome mixture analyzed. Two factors that may limit our ability to identify all possible proteins in the sample are: (i) our analytical set-up, which includes protein extraction, protein digestion, and peptide fractionation methods, failed to identify the subsets of proteins with unusual chemical properties, and (ii) the dynamic range of the lysate is too large to be comprehensively analyzed in one run using our current two dimensional chromatography strategy. | Table 2The number of proteins identified in the study a |
In addition to SEQUEST and the DTASelect2.0 algorithm used in this study, we also examined the effects of considering only unique peptide identifications. All peptides that were found to match the sequences of two or more proteins were discarded from the analysis and not used to define the list of identified or quantified proteins. This additional filtering step reduced the list of identified proteins by 5-14% per experiment ( Table 2), but offered significant benefits in data quality as discussed below. In prior studies, dynamic range, sensitivity and sequencing speed were discussed as being the major parameters limiting the complete identification of the yeast proteome. 27, 30 In our study, assuming that the abundance of proteins was distributed over the same range in the cells throughout their exponential 29 and steady state (chemostat cultivations) growth phases, we determined the sensitivity and the dynamic range of MudPIT analysis performed in this study (). A total of 100 μg of total protein per sample per strain was used in each MudPIT run. In agreement with de Godoy calculations, 27 it was assumed that the lowest abundance proteins were present at a concentration of about 20 fmol in every sample analyzed. Results showed that 40% of the proteins were detected in each range of protein abundances from 40 to 2000 molecules per cell (). A total of 77% of proteins were detected in each range spanning protein abundances from 4000 to 512000 molecules per cell. A default 2-peptide minimum per protein was used to derive these numbers of identified proteins. If a 1-peptide per protein threshold was used, 84% of proteins were identified in the protein abundance range from the lowest up to 512000 molecules per cell. A 1-peptide threshold dramatically increased the number of detected proteins for the lower range of protein abundances (lowest up to 8000 molecules/cell) - this fraction of the proteome, based on the control study, 29 covers 80% of total yeast proteome. The identified proteins were also analyzed to look for trends in other characteristics such as protein pI or protein mass. These parameters did not appear to have a significant impact as the identified proteins had a wide range of masses and pIs (data not shown). An additional pool of 650 proteins, beyond the list of proteins quantified in the control study, 29 was identified indicating that a differently expressed protein pool might have been present when cells were pre-grown in the set growth rate chemostat cultivations compared to the exponential growth cultivations. The main limitation for the reliable identification of the complete proteome was found to be the inability to identify more than one peptide from the lowest abundance proteins. Overall, the result of this study showed that by using the Orbitrap platform for generation of high mass accuracy data and DTASelect2.0, we could begin to sample the lowest abundance proteins. Relative protein expression differences were quantified using two algorithms Two quantitative approaches, spectral counting and stable isotope labeling, were used to assess relative protein expression differences between the Δsnf1, Δsnf4 and Δsnf1Δsnf4 strains and the wild-type strain. While spectral counting uses all identified unique-peptide spectra for a given protein to calculate a relative protein abundance, the stable isotope labeling approach uses individual 14N- and 15N-labeled unique-peptide pairs and quantitates relative peptide abundances to infer average relative abundance of a corresponding protein (performed by CenSus) 9. These two features constitute the main differences between the CenSus and the spectral counting calculated outputs. The ‘total quantified’ proteome was determined by merging relative protein abundance differences calculated for each of the three biological replicates using stable isotope labeling or spectral counting. One third of the translated yeast proteome was quantified using the stable isotope labeling approach ( Table 3). Using the same 4-peptide/protein threshold for both of the quantitative approaches, stable isotope labeling (compared to spectral counting) quantified more proteins in the study. Table 3 shows that twice as many significant changes in protein expression could be found using isotope labeling compared to spectral counting. Additional evidence supporting the idea that stable isotope labeling outperformed spectral counting under these study conditions can be seen in . The spectra detected in the same two biological replicate MudPIT experiments were used for a comparative relative protein abundance quantification study using stable isotope and spectral counting approaches. Only proteins, which were quantified based on 2-peptide pairs (4-peptide/protein) per MudPIT run, were used in the graphical representation of the comparison (). These results indicated that in complex protein mixtures, where protein sequence coverage was frequently low and relatively small differences in protein expression were expected, the calculated relative protein expression differences appeared to be more accurate and reproducible when stable isotope labeling was used compared to spectral counting. | Table 3Results of quantitative proteome analysis |
Protein quantification based on spectral counting was found to be largely improved when spectra generated from multiple experiments were averaged and merged so that abundance values could be determined from a larger number of spectra per protein. Larger numbers of proteins were quantified using stable isotope labeling (2373 proteins for the wild type, 2057 for the Δsnf1, 2081 for the Δsnf4 and 1763 for the Δsnf1Δsnf4) as compared to the use of spectral counting ( Table 3) and averaged spectral counts from biological replicates generated more accurate measurement of relative protein abundance then those of each individual replicate. Merging of the spectral counting results was not used in this study since it did not allow statistical evaluation of the quantitative data. A detailed comparison between our quantitative analysis and the absolute proteins levels determined in the control study showed that ~ 37% of the proteins were quantitated in each range of protein abundances from lowest up to 4000 molecules per cell, and ~ 74% of the proteins were quantitated in each range spanning from 4000 up to 512000 molecules per cell. Similar profiles, ~ 25% of low abundance proteins and ~ 70% of high abundance proteins, were relatively quantified when spectral counting was used. The result of this study presents the largest number of quantified global yeast proteome by mass spectrometry to date. Stable isotope labeling accurately and reproducibly quantified proteins even when only a small number of spectra per protein (generated from a single mass spectrometry experiment) was identified. The stringent statistical evaluation of biological variance derived among single MudPIT experiments was suitable to use in this study and it strengthened the presence of true biological changes substantially. Elimination of non-unique spectra improves quantitative proteome dataset Our results showed that by removing non-unique peptides from the experiment, the number of identified proteins was reduced by 5-14% per MudPIT run. Importantly, this led to a significant improvement in the ability to identify statistically meaningful differences in protein expression among proteins identified only by unique peptides. The data from the same mass spectrometry experiment was analyzed with and without the inclusion of non-unique peptides using both stable isotope labeling and spectral counting. A total of 3% of proteins quantified by stable isotope labeling and a total of 12% of proteins quantified by spectral counting were found to have different relative protein expression levels depending on whether non-unique spectra were included in the analysis (data not shown). Significant improvement of quantitative data after non-unique peptides were excluded was illustrated with the following example (, ). This example focused on the relative quantitation of the two highly homologous proteins, Adh1 and Adh2 (89% sequence similarity). After digestion, Adh1 and Adh2 yielded a combination of unique and shared peptides that were identified and relatively quantified. presents the ratios of quantified Adh1 and Adh2 peptides when the spectra of the Δsnf1 strain were compared to the spectra of the wild-type strain. From this figure, it could be clearly seen that the unique peptides clustered at a specific ratio for each protein while the shared, non-unique peptides, representing a mixed population of both Adh1- and Adh2-derived peptides, clustered at a different, inaccurate ratio. The ultimate result of including or excluding these non-unique peptides in the quantitative analysis can be seen in . The Adh1 and Adh2 averaged protein abundance ratios were inaccurate (showed a large standard deviation) when non-unique peptides were included, and were more accurate (displayed a lower standard deviation) when non-unique peptides were excluded. To interpret these results in a biological context, quantitative data based on only unique peptides clearly showed that acetaldehyde-ethanol conversion was shifted towards ethanol production when the Δsnf1 strain was compared to the wild-type strain. This result correlated with prior data and the fact that the Δsnf1 strain cannot grow on ethanol. 31 In contrast, when non-unique peptide spectra (causing the large standard deviation) were included in the quantitative datasets for the Adh1 and Adn2 proteins, no meaningful biological conclusions could be derived about ethanol oxidation. Overall, when non-unique peptides were discarded from the datasets, the calculation of averaged protein expression differences and their standard deviations were much improved: 3% more proteins (for the CenSus case) were found to have significantly changed ( P < 0.05) expression in the mutant strains compared to the wild type strain and could be included in further biological analyses. Correlation between stable isotope labeling and spectral counting analyses We next examined how well the proteins quantitated by either spectral counting or stable isotope labeling could be correlated with each other. For this analysis, we focused on the quantitative proteomics analysis of the Δsnf1 strain. A pool of significantly changed proteins ( P < 0.05) based on spectral counting and stable isotope labeling was used in this comparison ( Table 3). A total of 79 proteins were found to have changed expression ( P < 0.05) levels when quantified by both approaches and only 3 of these proteins (Ybr078w, Ydr505c, Ygl103w) correlated poorly. The quantitation of these 3 proteins proved to be difficult by either method due to the low number (~ 7) of spectra detected per protein per strain. The poor spectra quality for these 3 outliers prevented us from concluding whether stable isotope labeling or spectra counting was more successful in these cases. Among the remaining 302 proteins, which expression was determined to be significantly changed based on CenSus output, 65 were not quantified by spectral counting due to too few (1 or 2) spectra identified per protein per strain in each experiment. 242 out of the 302 proteins were quantified using less than 10-peptide pairs per protein, further highlighting the ability of stable isotope labeling to quantify proteins based on fewer peptides than spectral counting. Several biologically meaningful and differentially expressed proteins were found in this subset. Some of these proteins, such as the long chain fatty acyl-CoA synthetase Faa1 (Yor317w), hexokinase Hxk2 (Ygl253w) and glycogen synthase Gsy1 (Yfr014c) were identified as having significantly changed expression (in the Δsnf1 strain compared to the wild-type strain) using the stable isotope labeling approach. However, these proteins were not found to be reproducibly quantified among biological replicates using spectral counting due to the low number of identified spectra per each of these proteins. The Faa1, Hxk2 and Gsy1 were predicted to be affected in the Snf1-kinase complex disrupted strains. 14, 31 These findings demonstrate that by using the stable isotope labeling approach, the list of quantified proteins was enhanced and more biologically important findings were revealed, even from low numbers of spectra. A total of 96 proteins were found to be significantly changed when quantified by spectral counting alone. 77 (out of 96) proteins were quantified based on less than 10-peptide pairs per protein. 29 (out of 96) proteins were quantified based on 5-peptide pairs per proteins and these were found to have opposite abundance ratios compared to the stable isotope labeling approach. Overall, the results indicated that there is a good correlation between stable isotope labeling and spectral counting, except when spectral counting is based on low numbers of spectra. Biological evaluation of the quantified proteome Our results suggested that for the dataset used in this study, stable isotope labeling was more effective and reproducible than spectral counting for discerning protein expression differences in the mutant yeast strains. Based on this observation, we focused on the CenSus-produced quantitative results, emphasizing the subset of proteins that had significant (P < 0.05), thus biologically-relevant, protein expression changes when the proteome of the three mutants was compared to the proteome of the wild-type strain. A reproducible protein expression change across multiple biological replicates strongly supported the conclusion that a true biological phenomenon was being observed instead of a stochastic, insignificant event. The Census-generated list of 2388 proteins was categorized by GO annotations to determine their connections to different biological processes, molecular functions and localizations. We found that the sub-cellular distribution of the quantified (and identified) proteins was not significantly different from the sub-cellular distribution of the entire yeast proteome. This suggests that the protein extraction method using the urea-invitrosol mixture based protocol was unbiased and contained soluble and membrane-associated proteins. The hypergeometric distribution analysis test was applied to determine whether the enrichment of proteins with significantly changed expression was present among certain GO biological processes and GO molecular function categories. The test was performed in relation to the total pool of 2388 quantified proteins and the results were summarized in the Table 4. | Table 4List of significantly changed proteins in the Δsnf1, Δsnf4 and Δsnf1Δsnf4 strains and the GO process categories, in which these proteins were found to be enricheda |
Enriched protein expression changes (hypergeometric test: P < 0.01) within carbon metabolism and respiration GO biological process categories were expected, since the list of genes within these processes are known to be transcriptionally regulated by Snf1. 31 Enriched protein expression changes within the nucleic acid metabolic process group were also expected as there is an increasing number of reports describing the role of Snf1 in transcription regulation through histone modification and chromatin remodeling. 16, 32 Our global scale proteomics study also indicated potentially new areas of Snf1-mediated regulation. For example, little was known about the Snf1's involvement in the regulation of amino acid metabolism. Table 4 shows that the enriched number of proteins with significant expression changes was found within the amino acid and derivative metabolic process group for all three mutant strains. This indicated that amino acid metabolism was highly linked to the Snf1 kinase's function under these experimental conditions. Further studies are required to determine the links. The oxidoreductase activity GO molecular function category was found to be enriched (hypergeometric distribution test: P < 0.01) with significantly changed expression having proteins, in the three mutant strains compared to the wild-type strain. These proteins distributed among the GO biological process categories, such as: generation of precursor metabolites and energy (Adh2, Idp2, Tdh1), electron transport, amino acid (Glt1), lipid, carbohydrate, and vitamin (Gut2) metabolic processes ( Table 4). These findings indicated that Snf1 was clearly playing a role as global regulator in the central metabolism of yeast. 33Overall, the study identified protein expression changes that were expected based on prior literature, confirming the high quality of the generated global quantitative proteome dataset. It also indicated Snf1 involvement in the regulation of metabolic processes for which Snf1 has not been explicitly discussed before, indicating its broader regulatory role in yeast. Elimination of non-unique peptides improved quality of quantitative proteome dataset substantially A significant improvement in the ability to identify statistically meaningful protein expression differences among proteins identified just by unique peptides was illustrated by . One major advantage of using CenSus was the ability to visualize the data. Graphical evaluation of the data (relatively quantified peptide expression differences) offered an opportunity to identify unusual distributions of peptide ratios that might limit the ability to accurately determine the protein abundance ratio. This was clearly demonstrated in where the difficulties caused by the presence of redundant (non-unique) peptides in the analysis were readily identified. Thus, manual evaluation of quantitative data, where possible, could lead to a significant improvement in the quality of the quantitative results. For spectral counting, on the other hand, the high variability in spectral counts for different peptides of a protein limited their utility. Meaningful quantitative information required summing up all peptide spectra corresponding to a protein across an entire experiment to generate one abundance value for the protein. In this case, it was not possible to identify unusual distributions of peptide ratios to assess their affects on the quality of the overall quantitative measurements. In conclusion, the elimination of non-unique peptides improved the quality of the identified and quantified global proteome datasets. This was particularly crucial when the quantification was performed using spectral counting since no information regarding the distribution of the peptide ratios could be obtained. Benefits found when stable isotope labeling or spectral counting were used The results of the comparative analysis between stable isotope labeling and spectral counting showed distinct benefits in each method. Stable isotope labeling based quantification by CenSus was found to be more sensitive, highly reproducible between biological replicates, and generated higher precision data for low abundance peptides and proteins compared to spectral counting (). Based on these characteristics and the ability to statistically evaluate the distribution of peptide ratios for a protein to assess the precision of the expression level measured for each protein in that experiment, the isotope labeling approach was better able to identify small, but still biologically significant, changes in protein expression levels between the mutants and the wild-type strain. In the case of spectral counting, when a large number of spectra was identified for each of the proteins, the results of the quantified relative protein expression differences correlated well with stable isotope labeling. The high spectral coverage required by spectral counting could be readily achieved when simple protein mixtures were analyzed or repetitive MudPIT analysis was performed on the same sample and spectra from multiple analyses could be summed up. In those cases where redundant spectral coverage of proteins could be achieved, spectral counting becomes advantageous since it is easy to use and does not require prior isotopic labeling of the sample. In conclusion, the study examined the effectiveness of two different quantitative proteomic strategies applied to the same proteomic dataset. Significant protein expression changes were found and confirmed using both analysis methods. In those cases where a discrepancy was found between the methods, the discrepancy could typically be explained by too few identified spectra for that protein. Unusual distributions of peptide ratios between samples due to the inclusion of peptides that map to more than one protein (non-unique peptides) or modified peptides was another potential factor capable of skewing calculated protein abundance ratios when either of quantitative approaches was used. 1File001: Supporting Information Available Complete lists of proteins, which abundances were found to be significantly ( P < 0.05) changed (in the Δsnf1, Δsnf4 and Δsnf1Δsnf4 versus wild-type strain) based on stable isotope labeling approach, the number of peptide pairs used for quantification, normalized relative protein abundance differences and t-test based significance of calculated abundance differences are included as supplementary data. This material is available free of charge at http://pubs.acs.orgWe thank Akira Motoyama for valuable discussions regarding MudPIT setup and suggestions on how to improve chromatography based separation of peptides. This work was supported by the Danish Research Agency for Technology and Production and National Institutes of Health grants 5R01 MH067880 and P41 RR11823. 1. Humphery-Smith I. A human proteome project with a beginning and an end. Proteomics. 2004;4:2519–2521. [PubMed] 2. Aebersold R, Mann M. Mass spectrometry-based proteomics. Nature. 2003;422(13):198–207. [PubMed] 3. Washburn MP, Wolters D, Yates JR., 3rd Large-scale analysis of the yeast proteome by multidimensional protein identification technology. Nature Biotechnology. 2001;19(3):242–7. 4. Delahunty C, Yates JR., 3rd Protein identifcation using 2D-LC-MS/MS. Methods. 2005;35:248–255. [PubMed] 5. Washburn MP, Ulaszek R, Deciu C, Schieltz DM, Yates JR., 3rd Analysis of Quantitative Proteomic Data Generated via Multidimensional Protein Identification Technology. Anal Chem. 2002;74:1650–1657. [PubMed] 6. Gygi SP, Rist B, Gerber SA, Turecek F, Gelb MH, Aebersold R. Quantitative analysis of complex protein mixtures using isotope-coded affinity tags. Nat Biotechnol. 1999;17(10):994–999. [PubMed] 7. Patterson SD, Aebersold RH. Proteomics: the first decade and beyond. Nat Genetics. 2003;33(Suppl):311–323. [PubMed] 8. Pang JX, Ginanni N, Dongre AR, Hefta SA, Opitek GJ. Biomarker discovery in urine by proteomics. J Proteome Res. 2002;1(2):161–169. [PubMed] 9. Venable JD, Wohlschlegel J, McClatchy DB, Park SK, Yates JR., 3rd Relative quantification of stable isotope labeled peptides using a linear ion trap-orbitrap hybrid mass spectrometer. Anal Chem. 2007;79(8):3056–3064. [PubMed] 10. Saito A, Nagasaki M, Oyama M, Kozuka-Hata H, Semba K, Sugano S, Yamamoto T, Miyano S. AYUMS: an algorithm for completely automatic quantitation based on LC-MS/MS proteome data and its application to the analysis of signal transduction. BMC Bioinformatics. 2007;8:15. [PubMed] 11. Old WM, Meyer-Arendt K, Aveline-Wolf L, Pierce KG, Mendoza A, Sevinsky JR, Resing KA, Ahn NG. Comparison of label-free methods for quantifying human proteins by shotgun proteomics. Mol Cell Proteomics. 2005;4(10):1487–1502. [PubMed] 12. Kolkman A, Dirksen EH, Slijper M, Heck AJ. Double standards in quantitative proteomics: direct comparative assessment of difference in gel electrophoresis and metabolic stable isotope labeling. Mol Cell Proteomics. 2005;4(3):255–266. [PubMed] 13. Zybailov B, Coleman MK, Florens L, Washburn MP. Correlation of relative abundance ratios derived from peptide ion chromatograms and spectrum counting for quantitative proteomic analysis using stable isotope labeling. Anal Chem. 2005;77(19):6218–6224. [PubMed] 14. Carlson M. Glucose repression in yeast. Curr Opin Microbiol. 1999;2(2):202–207. [PubMed] 15. Shirra MK, Patton-Vogt J, Ulrich A, Liuta-Tehlivets O, Kohlwein SD, Henry SA, Arndt KM. Inhibition of acetyl coenzyme A carboxylase activity restores expression of the INO1 gene in a snf1 mutant strain of Saccharomyces cerevisiae. Mol Cell Biol. 2001;21(17):5710–5722. [PubMed] 16. Lo WS, Duggan L, Emre NC, Belotserkovskya R, Lane WS, Shiekhattar R, Berger SL. Snf1--a histone kinase that works in concert with the histone acetyltransferase Gcn5 to regulate transcription. Science. 2001;293(553):1142–1146. [PubMed] 17. Carling D. AMP-activated protein kinase: balancing the scales. Biochimie. 2005;87(1):87–91. [PubMed] 18. Usaite R, Patil KR, Grotkjær T, Nielsen J, Regenberg B. Global Transcriptional and Physiological Responses of Saccharomyces cerevisiae to Ammonium, L-Alanine, or L-Glutamine Limitation. Appl Microbiol Biotechnol. 2006;72(9):6194–6203. 19. Usaite R, Nielsen J, Olsson L. Physiological characterization of glucose repression in the strains with SNF1 and SNF4 genes deleted. J Biotechnol. 2008;133:73–81. [PubMed] 20. Yates JR, 3rd, Cociorva D, Liao L, Zabrouskov V. Performance of a linear ion trap-Orbitrap hybrid for peptide analysis. Anal Chem. 2006;78(2):493–500. [PubMed] 21. Ferea TL, Botstein D, Brown PO, Rosenzweig RF. Systematic changes in gene expression patterns following adaptive evolution in yeast. Proc Natl Acad Sci USA. 1999;96(17):9721–9726. [PubMed] 22. Verduyn C, Postma E, Scheffers WA, Van Dijken JP. Effect of benzoic acid on metabolic fluxes in yeasts: a continuous-culture study on the regulation of respiration and alcoholic fermentation. Yeast. 1992;8(7):501–517. [PubMed] 23. Eng J, McCormack A, Yates J. An Approach to Correlate Tandem Mass Spectral Data of Peptides with Amino Acid Sequences in a Protein Database. J Am Soc Mass Spectrom. 1994;5:976–989. 24. Peng J, Elias JE, Thoreen CC, Licklider LJ, Gygi SP. Evaluation of multidimensional chromatography coupled with tandem mass spectrometry (LC/LC-MS/MS) for large-scale protein analysis: the yeast proteome. J Proteome Res. 2003;2(1):43–50. [PubMed] 25. Tabb DL, McDonald WH, Yates JR., 3rd DTASelect and Contrast: tools for assembling and comparing protein identifications from shotgun proteomics. J Proteome Res. 2002;1(1):21–26. [PubMed] 26. Liu H, Sadygov RG, Yates JR., 3rd A model for random sampling and estimation of relative protein abundance in shotgun proteomics. Anal Chem. 2004;76(14):4193–4201. [PubMed] 27. de Godoy LM, Olsen JV, de Souza GA, Li G, Mortensen P, Mann M. Status of complete proteome analysis by mass spectrometry: SILAC labeled yeast as a model system. Genome Biol. 2006;7(6):R50. [PubMed] 28. Wei J, Sun J, Yu W, Jones A, Oeller P, Keller M, Woodnutt G, Short JM. Global proteome discovery using an online three-dimensional LC-MS/MS. J Proteome Res. 2005;4(3):801–808. [PubMed] 29. Ghaemmaghami S, Huh WK, Bower K, Howson RW, Belle A, Dephoure N, O'Shea EK, Weissman JS. Global analysis of protein expression in yeast. Nature. 2003;425(6959):737–741. [PubMed] 30. McDonald WH, Yates JR., 3rd Shotgun proteomics and biomarker discovery. Dis Markers. 2002;18(2):99–105. [PubMed] 31. Schuller HJ. Transcriptional control of nonfermentative metabolism in the yeast Saccharomyces cerevisiae. Curr Genet. 2003;43:139–160. [PubMed] 32. Verdone L, Wu J, van Riper K, Kacherovsky N, Vogelauer M, Young ET, Grunstein M, Di Mauro E, Caserta M. Hyperacetylation of chromatin at the ADH2 promoter allows Adr1 to bind in repressed conditions. EMBO J. 2002;21(5):1101–1111. [PubMed] 33. Polge C, Thomas M. SNF1/AMPK/SnRK1 kinases, global regulators at the heart of energy control? Trends in Plant Science. 2007;12(1):20–28. [PubMed] 34. Van Dijken JP, Bauer J, Brambilla L, Duboc P, Francois JM, Gancedo C, Giuseppin ML, Heijnen JJ, Hoare M, Lange HC, Madden EA, Niederberger P, Nielsen J, Parrou JL, Petit T, Porro D, Reuss M, van RN, Rizzi M, Steensma HY, Verrips CT, Vindelov J, Pronk JT. An interlaboratory comparison of physiological and genetic properties of four Saccharomyces cerevisiae strains. Enzyme Microb Technol. 2000;26(910):706–714. [PubMed]
|
The publisher's final edited version of this article is available at 
