Relative Protein Quantification by Isobaric SILAC with Immonium Ion Splitting (ISIS)*S

Metabolic labeling techniques have recently become popular tools for the quantitative profiling of proteomes. Classical stable isotope labeling with amino acids in cell cultures (SILAC) uses pairs of heavy/light isotopic forms of amino acids to introduce predictable mass differences in protein samples to be compared. After proteolysis, pairs of cognate precursor peptides can be correlated, and their intensities can be used for mass spectrometry-based relative protein quantification. We present an alternative SILAC approach by which two cell cultures are grown in media containing isobaric forms of amino acids, labeled either with 13C on the carbonyl (C-1) carbon or 15N on backbone nitrogen. Labeled peptides from both samples have the same nominal mass and nearly identical MS/MS spectra but generate upon fragmentation distinct immonium ions separated by 1 amu. When labeled protein samples are mixed, the intensities of these immonium ions can be used for the relative quantification of the parent proteins. We validated the labeling of cellular proteins with valine, isoleucine, and leucine with coverage of 97% of all tryptic peptides. We improved the sensitivity for the detection of the quantification ions on a pulsing instrument by using a specific fast scan event. The analysis of a protein mixture with a known heavy/light ratio showed reliable quantification. Finally the application of the technique to the analysis of two melanoma cell lines yielded quantitative data consistent with those obtained by a classical two-dimensional DIGE analysis of the same samples. Our method combines the features of the SILAC technique with the advantages of isobaric labeling schemes like iTRAQ. We discuss advantages and disadvantages of isobaric SILAC with immonium ion splitting as well as possible ways to improve it.

Methods for quantitative proteomics based on stable isotope labeling have become in the last decade very powerful tools to investigate cellular processes (1)(2)(3). The main appli-cations of such methods have been the analysis of changes in protein expression (4) as well as the elucidation of networks of molecular (protein-protein (5) or DNA-protein) interactions or the study of post-translational modifications (6,7). Isotope labeling methods can be classified in two broad classes. Those based on residue-specific chemical derivatization with labeled reagents have the great advantage of offering greater flexibility in the choice of the chemistry and are universally applicable. However, they can suffer from the complexity of the steps involved and from the risk of side reactions. Metabolic labeling approaches, on the other hand, are only possible for organisms whose cells can be cultured in strictly controlled conditions. They offer, however, the advantage that no additional labeling steps are involved and that the proteins maintain all their native properties. Such techniques are especially attractive when complex biochemical purifications are necessary to obtain the samples to be studied because extracts of cells can be mixed at the very beginning of the procedure, thereby eliminating artifacts due to slight variations in the purification conditions for the different samples.
Many chemical labeling schemes have been devised. In turn, they can be subdivided in two groups based on the approaches needed for quantification. Most techniques rely on measuring the intensity of light/heavy parent ions in MS survey scans to establish intensity ratios. One technical issue with such an approach is the need to establish an unambiguous correlation (based on m/z and retention times in liquid chromatography) between tandem MS spectra used for identification and their precursor peaks in MS survey scans. ICAT (8) and a multitude of other labeling schemes (9 -12) belong to this class.
The second group is formed by techniques that exploit quantitative information embedded in tandem MS spectra rather than in survey scans (13,14). Earlier examples of this principle used protease-mediated 16 O/ 18 O labeling (15)(16)(17). Such approaches, however, require co-fragmentation of the precursors, and this necessitates very close or identical masses, such as in the iTRAQ 1 scheme. The latter is a special case that uses isobaric tagging reagents that fragment to give, for every peptide, diagnostic low mass ions used for quantification. Compared with methods with distinct mass precursors, these techniques present the advantages that no precursor mass splitting results in higher signal intensity and data handling is easier because only MS/MS data are needed, thus eliminating the need to integrate peaks along chromatographic runs. Appropriate design of the reagent has allowed the reporter ion mass to be varied, and this has been used to implement 4-and 8-fold multiplexing work flows.
By contrast, classical SILAC (18) is a metabolic labeling method similar, from the point of view of the quantification procedure, to non-isobaric precursor methods. To achieve accurate quantification, it needs high resolution mass data with low noise levels together with a powerful software able to correlate MS/MS identifications with full scan MS data and extract integrated ion intensities for the precursors and cognate, but often unidentified labeled analogues.
We propose an alternative SILAC scheme that combines some useful features of iTRAQ-like work flows with the convenience of SILAC for studying samples from cultured cells. The technique is based on the incorporation of isobaric labeled amino acids that generate distinct residue immonium ion fragments. We show that reliable quantification is possible purely on MS/MS spectra using very simple data extraction tools.

MATERIALS AND METHODS
Cell Culture and Labeling-Unlabeled cell extracts were obtained from BJAB human B cells grown in standard RPMI 1640 medium supplemented with 10% fetal bovine serum and penicillin-streptomycin (all from Invitrogen).
Metabolic labeling was performed on two human melanoma cell lines: SBCL2, derived from an radial growth phase melanoma (19) and kindly provided by Dr. Giovanella (Stehlin Foundation for Cancer Research, Houston, TX), and SKMel28, derived from skin metastasis (20) and obtained from the ATCC collection (number HTB-72, LGC Promochem). SBCL2 and SKmel28 cells were grown in high glucose Dulbecco's modified Eagle's medium deficient in L-valine, L-leucine, L-isoleucine, and L-glutamine (Cell Culture Technologies, Gravesano, Switzerland). The medium was reconstituted from concentrated stocks according to the manufacturer's instructions, sterile filtered, and stored at 4°C. L-Isoleucine, L-leucine, and L-valine labeled either with 1-13 C (99% 1-13 C enrichment) or 15 N (98% 15 N enrichment) were obtained from Cambridge Isotope Laboratories (Andover, MA). The powdered amino acids were diluted in PBS to obtain stock solutions of 40 mg/ml for valine and 15 mg/ml for leucine and isoleucine that were sterile filtered and stored at Ϫ80°C. SKMel28 and SBCL2 cells were grown in the base medium supplemented with 4 mM L-glutamine (Amimed, Allschwil, Switzerland), 10% dialyzed fetal bovine serum (Invitrogen), penicillin-streptomycin (Invitrogen), and the amino acids 1-13 C-(for SKMel28) or 15 N-labeled (for SBCL2 and SKMel28 as specified under "Results") L-valine, L-leucine, and L-isoleucine. The final concentrations of Val, Leu, and Ile were 94, 105, and 105 mg/ liter, respectively, corresponding to the standard composition of the Dulbecco's modified Eagle's medium. Cell lines were grown for 3 weeks to achieve complete labeling.
Preparation of Protein Samples-Cells at about 80% confluence were washed twice with PBS to remove serum proteins. Cells were detached by trypsinization and washed twice in PBS. Pellets of cells were resuspended in ice-cold hypotonic buffer (10 mM HEPES, 1.5 mM MgCl 2 , and 10 mM KCl) and sonicated for three cycles of 5 s each on ice. The lysates were centrifuged twice (10 min at 13,000 rpm at 4°C) to pellet cellular debris; each time the supernatant was recovered. The protein concentration of the second supernatant was determined using the Bradford protein assay (Bio-Rad); the relative protein concentration of samples to be mixed was verified by densitometry on whole lanes of SDS-PAGE gels after Coomassie staining. For mixing experiments, lysates were combined in a known ratio equal to 1:3 for 15 N: 13 C SKMel28 extracts or 1:1 for 13 C SKMel28: 15 N SBCL2. For the co-elution experiment, 15 N: 13 C SKMel28 lysates were combined in 1:2 ratio.
2D Gel Electrophoresis-2D DIGE experiments were performed according to the manufacturer's instructions (GE Healthcare) (21). Briefly 300 g of protein from 15 N SBCL2 and 13 C SKMel28 cell lysates were precipitated with acetone overnight at Ϫ20°C; the pellet was dissolved in 25 l of DIGE Cell Lysis Buffer (30 mM Tris/HCl, pH 8.5, 7 M urea, 2 M thiourea, 4% (w/v) CHAPS). 50 g of protein from 15 N SBCL2 and 13 C SKMel28 lysates were minimally labeled with 800 pmol of reconstituted Cy3 and Cy5 dyes, respectively. 25 g of the two samples were mixed together, labeled with Cy2, and used as internal standard. The Cy2-, Cy3-, and Cy5-labeled samples were combined, supplemented with 2% pH 3-11 ampholytes (IPG Buffer, GE Healthcare), and applied by cup-loading on a 13-cm, pH 3-11 non-linear Immobiline DryStrip (GE Healthcare) previously rehydrated in DIGE Cell Lysis Buffer overnight. Isoelectric focusing was performed on an Ettan IPGphor (GE Healthcare) system to attain a total of 22,500 V-h. Prior to the second dimension, strips were equilibrated for 15 min in a reducing buffer containing 6 M urea, 2% (w/v) SDS, 30% (v/v) glycerol, 32 mM DTT, 100 mM Tris, pH 8. This was followed by a 15-min alkylation in a buffer containing 6 M urea, 2% (w/v) SDS, 30% (v/v) glycerol, 240 mM iodoacetamide, 100 mM Tris, pH 8. Second dimension migration was carried out on a 13 ϫ 9-cm 8 -16% Criterion precast gel (Bio-Rad) at a constant voltage of 80 V for 4 h. The gel was rinsed in water and then scanned using a Molecular Imager FX (Bio-Rad) scanner. Gel images were analyzed using Im-ageMaster 2D Platinum DIGE Software version 5.0 (GE Healthcare). The gel was Coomassie-stained overnight and destained with 10% acetic acid. Spots of interest were manually excised from the gel.
1D Gel Electrophoresis for Protein Digestion-For 1D gel electrophoresis, 60 g of protein from either pure or mixed lysates were subjected to limited electrophoretic separation on a 10% SDS-PAGE minigel, i.e. the migration was stopped after the front had moved by about 2.5 cm into the separating gel at which point the bands of a prestained size marker were visible and separated in the 20 -250-kDa range. After Coomassie staining, each lane was cut into four fractions corresponding to regions of different molecular weights.
In-gel Protein Digestion with Trypsin-Slices excised from 1D or 2D gels were transferred to 96-well plates. In-gel proteolytic cleavage with sequencing grade trypsin (Promega, Madison, WI) was performed automatically in a ProGest robotic work station (Genomic Solutions, Ann Arbor, MI) according to a described protocol (22) The liquid supernatant from the digestion was recovered and concentrated by evaporation. The final volume was adjusted to 60 l with 0.1% TFA in 2% acetonitrile.
LC-MS/MS-Digests were desalted using C 18 StageTips (Proxeon, Odense, Denmark), which have an estimated loading capacity equal to 10 g. The peptides were eluted from the microcolumns with 80% acetonitrile after which the eluate was dried and resuspended in 7 l of solvent A (2% acetonitrile, 0.5% formic acid). 2.5 l of purified sample were injected on a reversed-phase C 18 column (PepMap100, 3 m, 100 Å, LC Packings) and separated by nanoflow liquid chromatography on an Ultimate (LC Packings) system on line with an electrospray quadrupole-time-of-flight mass spectrometer (API QSTAR Pulsar i, Applied Biosystems/SCIEX, Concord, Ontario, Canada). The gradient used for separation was from 2 to 40% acetonitrile in water with 0.5% formic acid at a flow rate of 200 nl/min.
The mass spectrometer was controlled by the Analyst QS 1.1 software set to operate in information-dependent acquisition mode to automatically switch between MS and collision-induced dissociation MS/MS. Survey full-scan MS spectra were acquired from 400 to 1200 m/z in 1 s after which the two most intense ions with charge 2ϩ to 4ϩ were isolated for fragmentation. In a second method (method B), the three most intense ions with 2ϩ to 4ϩ charge were fragmented.
Each parent ion underwent two different MS/MS acquisitions. The first scan (1-s duration) spanned the 50 -1200 m/z mass range, used a collision energy proportional to precursor mass, and was intended to collect sequence information. The second MS/MS scan (0.2-s duration) was focused on (iso)leucine and valine immonium ions and therefore covered narrow ranges of masses (71-74 and 85-88 m/z for method "A" or a single scan at 64 -95 m/z for method "B") to achieve a better sampling of the reporter ions. The collision energy for this narrow range scan was fixed at 60 eV to promote a higher degree of fragmentation, and the masses were acquired in the "enhanced" mode that allows a considerable gain in signal intensity by specifically pulsing ions of a narrow mass range (23). Three (method A) or two (method B) MS/MS spectra were cumulated for each selected peptide; former target peptides and their isotopes were dynamically excluded for 90 s (with a mass tolerance of 50 milli-mass units).
All runs were performed in triplicate for the 15 N SKMel28: 13 C SK-Mel28 1:3 mixture fraction using method A. The 15 N SBCL2: 13 C SKMel28 1:1 mixture was analyzed in quadruplicate using both methods A and B in duplicate.
For the analysis of 2D DIGE spots by LC-MS a shorter gradient from 2 to 40% acetonitrile in 45 min was used. All MS parameters were the same as those described above.
For testing peptide co-elution, a special acquisition method was used in which two precursors were fragmented for 10 consecutive scans, giving a total cycle time of 21 s. The elution profile of the peak was then followed to monitor immonium ion fragment intensities during the same cycle.
Protein Identification and Quantification-The Mascot.dll script (version 1.6b21) supplied by Matrix Science was used to extract spectra of each run using the following parameters: 2ϩ to 4ϩ charge state, peak centroiding, no deisotoping, spectra rejected if containing less than eight peaks, no minimum peak intensity required, report peak area. For mixing experiments, Mascot generic format flat text (.mgf) files deriving from different fractions of the same mixture were pooled before performing the database search. Proteins were identified using Mascot version 2.1 (Matrix Science, London, UK) and searching the UniProt database, restricted to human taxonomy. Database releases used were 8.7 of September, 19 2006 for experiments with unlabeled samples (71,233 sequences after taxonomy filter) and release 11.0 of May, 29 2007 (70,004 sequences after taxonomy filter) for experiments with labeled samples. One trypsin missed cleavage was allowed, cysteine carbamidomethylation was set as fixed modification, and methionine oxidation was defined as variable modification. Unless otherwise specified, peptide mass tolerance was set to 1.2 Da (0.6 Da for unlabeled samples), and the fragment mass tolerance was set to 0.3 Da. We used a broad peptide mass tolerance (1.2 Da) to include in Mascot version 2.1 searches the ϩ1-Da 13 C precursor peptides that can be incorrectly chosen during peak detection. For labeled samples, 13 C labeling and 15 N labeling on valine, isoleucine, and leucine were also set as fixed peptide modifications unless specified. The Mascot multidimensional protein identification technology (MudPIT) scoring system was used for all analyses of mixtures (the unlabeled sample was analyzed with standard scoring) with significance threshold at p ϭ 0.05 and minimum ion score equal to 14. The same .mgf files used for protein identification mentioned above were searched against a decoy database automatically generated by Mascot version 2.2 and containing random sequences having the same length and average amino acid composition of the forward "normal" database to compute the false discovery rate (FDR) of protein identification. Significance threshold was set at p ϭ 0.05, and minimum ion score was set at 14 as in the previous search used for protein identification; the resulting FDR values are reported under "Results" for the different shotgun analyses. Proteins spanning the same set of peptides, or a subset, were collapsed into a single entry on the hit list according to the principle of parsimony. For these cases, only one representative protein is listed. Proteins with matched peptides in common with other sequences were validated only if matched by at least one unique ("bold red") discriminating peptide. No further attempt was made to discriminate protein isoforms that were indistinguishable with the available mass spectrometry data.
To  15 N](iso)leucine with a tolerance of Ϯ0.020 amu. For each identified protein, these intensities were associated to the corresponding valine-and (iso)leucine-containing peptides. In the case of peptides containing both valine and (iso)leucine, the quantification was considered as independent. Only Val-containing peptides with non-zero values on both 72.081 and 73.078 m/z and Ile/Leu-containing peptides with non-zero values on both 86.097 and 87.094 m/z were used for estimation of abundance ratios. R software (24) (2.3.1 version) was used for all statistical calculations. For each protein presenting at least four quantified peptides, a straight line was fitted by least squares through the points represented by the 13 C and 15 N immonium ion intensities for each peptide; the corresponding slope is an estimate of the abundance ratio (25). The standard error of the estimate, the t value of the test for significance of the slope, and the corresponding p value as well as the 95% confidence interval were also calculated from the linear regression. Outlier points, which could result either from experimental variability or from the presence of differentially expressed isoforms of the same protein matched by the same set of peptides, were not removed before linear regression. Thus no attempt was made to discriminate such protein isoforms. The regression line was fitted twice using in turn the 13 C and 15 N immonium ion intensities as response variable; the two estimations of the abundance ratio were extremely close, and the one that produced the smallest residuals was used. Other methods were considered in particular orthogonal regression, which produced extremely similar results (data not shown). Because the ratio 15 N/ 13 C should in theory remain constant for all peptides regardless of the intensities of the individual ions, the regression line was constrained to pass through the origin. Examination of the plots for proteins with a large number of peptides (the first 30 Mascot protein matches in the 1:1 13 C SKMel28: 15 N SBCL2 data set) suggested that this assumption is correct. Unless otherwise specified, the reported ratios are defined heavy/light (H/L).

RESULTS
Protein Labeling Strategy-Isobaric SILAC with immonium ion splitting (ISIS) differential labeling is based on the incorporation of isobaric analogues of specific amino acids capable of providing distinct immonium ions fragments whose intensity can be relatively quantified in MS/MS spectra. Essential amino acids containing the stable 1-13 C (culture A) or ␣-15 N (culture B) isotope are added to a cell culture medium lacking these same amino acids but containing all others in unlabeled form (Fig. 1A). Cells are then grown for a time sufficient to obtain a near complete (98% or more) protein labeling, i.e. six or more cell divisions. Every labeled amino acid residue (be it 1-13 C-or ␣-15 N-labeled) in a peptide or protein increases its nominal mass by 1 amu. Corresponding peptides originating from the cultures A and B should therefore be isobaric, but they are expected to generate upon collision-induced fragmentation different immonium ions. These residue ions are produced by the simultaneous cleavage of both the peptide bond N-terminal to the residue of interest as well as of the C␣-CO bond within the residue (26). Residues labeled with 13 C on the carbonyl carbon (position 1) thus produce "light" immonium ions identical to those generated by naturally occurring (unlabeled) residues because the labeled C-1 atom is lost during MS/MS fragmentation. On the other hand, amino acid residues labeled on the ␣ carbon or 15 N-labeled on the backbone nitrogen are expected to produce immonium ions 1 amu heavier than the natural ones because the label is retained after CID.
To perform relative protein quantification, extracts from differentially labeled cells can be mixed directly after lysis, preserving in this way the ratio of 1-13 C-to ␣-15 N-labeled proteins during the next steps of sample preparation (Fig. 1B). The protein mixture is then digested, and the resulting peptides are analyzed by LC-MS/MS. The intensity values of the couples of immonium ions deriving from differentially labeled amino acids detectable in MS/MS spectra are then used to calculate the relative amount of a given protein in the two samples. Co-elution of 1-13 C-and 15 N-labeled precursor peptides, necessary to obtain a correct ratio estimate, is expected because unlabeled peptides have been reported previously to co-elute with their 13 C-and 15 N-labeled analogues (27,28). The quantification therefore takes place at the level of MS/MS spectra and is basically simultaneous to the protein identification step.
Preliminary Analysis on Unlabeled Sample-Among the 10 amino acids essential for mammalian cells, we selected Val, Leu, and Ile for labeling. These three amino acids are among the most abundant in proteins, and their frequency in the Swiss-Prot database is 6.72, 9.66, and 5.88%, respectively. In addition, they all generate characteristic intense immonium ions upon CID. We choose to use the combination of these three amino acids to increase the likelihood of obtaining at least one labeled residue in each peptide generated by digestion. The immonium ions of these amino acids in their natural or 1-13 C-labeled form are localized at 72.081 m/z for valine and at 86.097 m/z for leucine and isoleucine. 15 N-labeled analogues of the same amino acids are expected to give signals at 73.078 and 87.094 m/z, respectively.
Before trying to exploit immonium ion intensities for quantification purposes, it was necessary to verify that (i) ions at 72.081 and 86.097 m/z are detected in MS/MS spectra of unlabeled peptides and are reliable diagnostic signals for the presence of Val, Ile, and Leu and that (ii) no naturally occurring residue produces strong signals at 73.078 and 87.094 amu so that these "channels" can be used for the specific quantification of heavy labeled peptides. To this end, an unlabeled complex mixture of proteins from human cells was digested and analyzed by LC-MS/MS, generating a collection of 4759 spectra. Immonium ion intensities were then extracted from the resulting centroided spectra with an in-house built peak matching script; proteins identified by at least two unique peptides were analyzed. Intense signals at 72.081 (Val) and 86.097 m/z (Ile/Leu) were found to correlate with the presence of the corresponding amino acid (supplemental Table 1 15 N-labeled amino acids produce in MS/MS spectra light and heavy immonium ions, respectively. R ϭ CH(CH 3 ) 2 (valine), CH 2 CH(CH 3 ) 2 (leucine), or CH(CH 3 )CH 2 CH 3 (isoleucine). During immonium ion formation, 1-13 C-labeled amino acids lose the label and produce an ion 1 amu lighter than the one deriving from 15 N-amino acids. b, ISIS analytical steps. Cell lines are metabolically labeled using 1-13 C and 15 N isobaric variants of the three essential amino acids valine, leucine, and isoleucine. The samples are mixed together after lysis, digested, fractionated by SDS-PAGE, and analyzed by LC-MS/MS. After a survey scan, each precursor undergoes two different MS/MS scans, a broad m/z range scan for protein identification and a narrow range scan focused on the immonium ion region, to detect light and heavy peak intensities. After database search, reporter immonium ion intensities are extracted from CID spectra and used to calculate heavy/light protein ratios. 73.078 and 87.094 m/z, and their intensity was proportional to those at 72.081 and 86.097 m/z, respectively (supplemental Table 1 and supplemental Fig. 2). In fact, the correlation analysis by linear regression of the intensities of the peaks pairs localized at 72.081/73.078 m/z and at 86.097/87.094 m/z showed that, for each peptide, the heavier peaks were on average 5.6 and 6.5% (R 2 ϭ 0.97) of the light peaks, respectively, thus very close to the percentage expected from the theoretical isotope distribution (4.8% for Val and 6.0% for Leu/Ile). This suggests that the peaks at 73.078 and 87.094 m/z were produced by the naturally occurring heavier (mostly 13 C 1 ) isotopic forms of the Val and Ile/Leu immonium ions.
We also explored the eventuality that the natural immonium ion of asparagine (which has a calculated mass of 87.056 amu) could interfere with the quantification based on the [ 15 N](iso)leucine immonium ion because two peaks are localized at close m/z values. To this end, we repeated the correlation analysis relaxing from Ϯ0.02 to Ϯ0.2 amu the tolerance for the extraction of intensities at 87.094 m/z to include Asn immonium ion signals. As a result, a linear regression with a poor correlation coefficient was obtained (R 2 ϭ 0.63) (supplemental Fig. 3); when Asn-containing peptides were removed from the data set, the correlation improved dramatically (R 2 ϭ 0.96), and the regression line was almost identical to the one obtained using a narrow mass tolerance (supplemental Fig.  2B). This demonstrates that, at least in some sequences, Asn can generate strong immonium ion signals that could bias the quantification; this was avoided using a reduced tolerance (Ϯ0.02 amu) for peak matching, which effectively filtered out Asn immonium ions. Analysis of a 15 N-labeled sample showed that at our working resolution (4200 at this mass) two distinct peaks were detected at 87.055 and 87.100 m/z, confirming that it was possible to discriminate between these two species (Fig. 2B).
We concluded that unlabeled Val and Ile/Leu could provide specific immonium ions at 72.081 and 86.097 m/z and that most probably at 73.078 and 87.094 m/z only the expected naturally occurring isotopes were most probably found. Furthermore there were no intense peaks produced by other naturally occurring amino acids. Therefore 15 N-labeled valine, leucine, and isoleucine were considered to be suitable for quantification according to our scheme.
Labeling Efficiency and Isobaricity-To test protein labeling, SKMel28 and SBCL2 cells were grown in medium containing either the 1-13 C-or the ␣-15 N-labeled forms of the three amino acids (Val, Ile, and Leu), respectively. The two cultures were grown in identical condition for 3 weeks, a time considered sufficient to achieve the complete incorporation of labeled analogues into proteins (18). A third culture was maintained in unlabeled medium as negative control. Lysates for labeled and unlabeled cells were fractionated by SDS-PAGE, digested, and analyzed by LC-MS/MS on a QQ-TOF instrument. Tandem mass spectra were then submitted to database search by Mascot. By comparing the molecular weights of identical peptides identified from labeled versus unlabeled cells, it was possible to verify that each 1-13 C-or 15 N-amino acid analogue added 1 amu to the original mass and that 1-13 C-and 15 N-labeled peptides were isobaric ( Fig. 2A). However, we noticed a difference in the isotope distribution of peptide signals in 15 N samples with a higher abundance of heavier isotope species; this phenomenon and its implications are addressed under "Discussion." No unlabeled peaks were noticed during manual inspection of MS spectra. To comprehensively check labeling efficiency, a database search was performed on a large data set (mixture of SKMel28 and SBCl2 cells, see below) without selecting ISIS labels as fixed modification, i.e. with native peptide masses and a precursor mass tolerance of 0.3 Da. Over a total of 23,784 spectra submitted to Mascot, only 267 were matched, identifying 54 proteins (FDR above identity threshold equal to 17%). The majority of these matched spectra (193, 72%) contained no Val, Ile, or Leu and were thus expected to have unmodified masses. All the remaining 74 matched spectra were either false positives or contaminants (keratin and trypsin). On the other hand, when ISIS labels were used as fixed modification in Mascot, 12,796 spectra were matched using the same search parameters, identifying 604 proteins (FDR above identity threshold equal to 2.9%); the percentage of sequences not containing Val, Ile, and Leu was in this case much smaller (only 2.7%). This confirmed that the labeling was highly efficient.
1-13 C-and 15 N-labeled peptides were isobaric within the limits of measurement error. Their MS/MS spectra showed identical series of b and y ions (supplemental Fig. 4). In fact, the nature of the labeling is such that the only fragments expected to be split in ESI spectra after CID fragmentation are immonium and a ions; as expected, split light/heavy a ions of significant intensity were observed only for peptides containing the labeled amino acids localized within the first two N-terminal residues. Finally the masses of peptides that did not contain Val, Ile, or Leu were unchanged after labeling ( Fig. 2A).
To highlight occurrence of immonium ions in a complex peptide mixture, we cumulated data by adding several hundreds of MS/MS spectra obtained from the correspondent fraction of unlabeled and [ 13 C]Val/Ile/Leu-and [ 15 N]Val/Ile/ Leu-labeled mixture of proteins (Fig. 2B). In the immonium ion region of MS/MS spectra, intense peaks at 72.081 and 86.097 m/z were visible in the 1-13 C-labeled sample, whereas equally intense signals at 73.078 and 87.094 m/z were visible in 15 N-labeled extracts (Fig. 2B). Two peaks were detected at m/z 87.055 and 87.100, corresponding to the natural immonium ion of Asn and the "heavy" immonium ion of Leu/Ile. Weaker peaks at 72.081 and 86.097 m/z were visible in the 15 N-labeled extracts probably due to residual amounts of unlabeled Val, Ile, and Leu amino acids in the peptides.
The co-elution of 1-13 C-and 15 N-labeled precursor peptides, a prerequisite for the accurate estimate of light/heavy ratios, was confirmed using a 2:1 mixture of completely labeled SKMel28 lysates. Because the precursors are isobaric, their elution profiles are indistinguishable in MS mode and can only be discriminated upon fragmentation. Therefore the sample was analyzed using a modified LC-MS/MS method that acquired CID spectra of the same precursor for 20 s per cycle; this cycle time covered the typical chromatographic peak width, which was estimated to be between 20 and 30 s. Although this led to the identification of a limited number of peptides, the intensity profiles of the light and heavy immonium ions could be monitored and plotted. An example is shown in supplemental Fig. 5 and demonstrates co-elution for all precursors detected in this analysis.
Protein Quantification: Method Validation with a Normalized Mixture-We then tested the ability of the method to reliably quantify proteins in a complex environment by analyzing a fraction composed of 1-13 C-and 15 N-labeled lysates of SK-Mel28 cells mixed in a known ratio (3:1) before SDS-PAGE fractionation. The fraction of proteins with a molecular mass between 40 and 80 kDa was selected and analyzed as described under "Materials and Methods." After data submission to Mascot, 127 proteins were identified with a total of 2035 spectra matched. Of these, 1978 (97%) were matched to sequences containing at least one of the three amino acids used for labeling. The false positive rate calculated for peptide matches above identity threshold was 3.6%. Intensity values for the 72.081/73.078 and 86.097/87.094 m/z peak pairs were extracted from the Mascot .dat file (containing the matched spectra in centroided form) and associated to the corresponding peptides using an in-house programmed script; intensity pairs where one of the values was zero were discarded as specified under "Material and Methods." This yielded a total of 1149 peptides; 537 were quantified on valine, 1040 were quantified on (iso)leucine, and 428 peptides were quantified on both residues. Only the proteins presenting a minimum of four measurements were quantified to increase the confidence level of data (supplemental Fig. 6). Over a total of 127 proteins, it was thus possible to compute 80 protein quantifications (63%) as shown in Fig. 3A. Using a linear model, the mean heavy/light estimated ratio was equal to 0.361, thus very close to the expected value of 0.333. Confidence intervals were particularly small for proteins characterized by a large number of peptides and thus showing higher Mascot scores. The ratio distribution was spread from a minimum value of 0.155 to a maximum of 0.472; 50% of the ratios were between 0.307 and 0.379. The deviation from the mean ratio could reflect both the biological variability in the protein content of the cell (the two samples originated from two distinct cultures) and the experimental error; the latter was likely larger for the proteins identified with a lower Mascot score, which usually correlates with low signal intensity. Ratio values were transformed to a log base 2 scale to express -fold changes and to normalize the magnitude of under-and overexpressed proteins (Fig. 3A). This transformation yielded an approximately normally distributed data set.
The data of this validation set were then used to define criteria for acceptance of quantification measurements, which were later applied to a sample analysis. To discriminate the high confidence protein ratios from the ones with large confidence intervals, we set cutoffs for statistical values. We decided to fix minimum t and maximum p values equal to 9 and 0.001, respectively. Overall 65 ratios of 80 passed this filter and were considered "high confidence" (Fig. 3B).
Differential Analysis-The ISIS method was applied to the differential analysis of proteins in SKMel28 and SBCL2 human cell lines, both derived from epithelial melanoma but at different stages of tumor progression. Because of their similar background, we expected an overall similar pattern of protein expression. On the other hand, these two cell lines display some visible differences in morphology and in adhesion behavior, so we speculated that some differences should be clearly detectable in their proteomes. SBCL2 and SKMel28 cells were differentially labeled and lysed. Extracts were mixed in 1:1 ratio and fractionated by 1D PAGE into four fractions, which were digested by trypsin and analyzed in replicates with ISIS-specific scan cycles as described under "Materials and Methods." We identified 582 proteins (13,504 peptides) with a false positive rate of identification of 2.5% (calculated for peptide matches above Mascot identity threshold). Over 582 proteins, we obtained 527 heavy/light ratios (91%). After applying the statistical cutoff values described in the previous section, 502 high confidence estimates were found (Fig. 4). The mean value was 0.921, thus close to the expected value of 1.00, whereas 50% of the ratios were between 0.761 and 1.009, likely reflecting the normal biological variability in the protein content of the cells. The box plot highlights the presence of both high and low "outlier" ratios deviating from the expected normal distribution, indicating proteins differentially expressed in the two cell lines. We ar- bitrarily decided to consider as significant only the proteins at least 2-fold over-or underexpressed, therefore presenting heavy/light ratios larger than 1.858 (more abundant in SBCL2) or smaller than 0.465 (more abundant in SKMel28). With these thresholds we identified 13 proteins more abundant in SBCL2 and 13 other proteins more abundant in the SKMel28 cell line. These proteins are listed in Table I.
Comparison with 2D DIGE-We subsequently compared the results obtained with the ISIS method with those obtained using 2D DIGE (21), an established technology for relative protein quantification. To this end, 1-13 C-labeled SKMEL and 15 N-labeled SBCL2 cell lysates already analyzed by shotgun ISIS were minimally labeled with Cy5 and Cy3 dyes, mixed, and separated by 2D PAGE. After fluorescence-based quantification by DIGE, 32 spots with varying ratios were cut, trypsinized, and analyzed by LC-MS/MS with ISIS-specific scan mode (supplemental Fig. 7 and supplemental Table 2). In 11 cases more than one protein was identified from a single spot, especially when they were cut from a crowded area. Although the ISIS method was able to compute distinct protein ratios, it was difficult in such cases to compare the results with those obtained by DIGE. For this reason, we decided to focus our comparison on the data resulting from spots con-taining only one protein (21 spots). The results for these 21 spots are presented in Fig. 5. Overall the quantification data obtained using both methods appeared to be linearly correlated (R 2 ϭ 0.90) and to provide similar values (DIGE/ISIS values mean ratio ϭ 0.74). The biggest differences were found when spot boundaries were uncertain, i.e. spots adjacent to others or faint ones. To correctly compare ISIS and DIGE quantifications, the ratio values obtained from eight well isolated and intense spots, containing only one protein, were compared (Fig. 5). The correlation between DIGE and ISIS was confirmed to be linear in the range analyzed with a very high R 2 (0.99), demonstrating that the results given by the two quantification methods were very similar and implying that ISIS quantification was accurate.
13 proteins found differentially expressed by DIGE and ISIS in 2D gel spots were also identified and quantified in the shotgun ISIS experiment. Fig. 6 shows the result of the comparison of the SKMel28:SBCl2 ISIS ratios obtained from the excised spots and from the complete lysates mixture. Roughly half of the proteins showed similar values, whereas the remaining proteins were clearly diverging. In several cases, such as cofilin-1, which was quantified in four different spots, this can be explained by the fact that the spots highlighted by DIGE probably represent protein isoforms pro- Only proteins with H/L ratios at least 2-fold higher or lower than the average value (0.929) were considered: ratios above 1.858 and below 0.465 identify proteins more abundant in SBCL2 and SKMel28, respectively. Each protein is reported with the accession number, the number of matched spectra used for quantification, the H/L (SBCL2/SKMel28) ratio, and the standard error. Similarly to other peptide-centric MS-based quantification techniques, the ISIS approach was able to provide either the total estimation of the relative protein amount or a more distinct quantification of different isoforms, depending on the fractionation techniques used. DISCUSSION We implemented a new methodology to identify and quantify protein changes in complex mixtures. This approach, which we named ISIS, is based on two sets of isobaric amino acids metabolically incorporated into proteins during cell culture and able to produce specific reporting immonium ions. In this study, we verified that precursor peptides bearing the different labels were isobaric, co-eluted during the LC-MS run, and yielded the expected reporter immonium ions whose specificity was assessed.
One intriguing phenomenon we observed was that peptides from [ 15 N]Val/Ile/Leu-labeled cells consistently displayed isotope distribution with higher abundances of the ϩ1, ϩ2, and ϩ3 isotope peaks. This fact was also observed in peptides not containing Val/Ile/Leu ( Fig. 2A) but was not detected in samples labeled with 1-13 C-amino acids. This suggests that the amino nitrogen from Val, Ile, or Leu can be utilized by cells to be incorporated (although probably with low frequency) in other amino acids. Essential amino acids like Val, Ile, and Leu cannot be synthesized by mammalian cells, which, however, possess enzymes for their degradation. The first step in the catabolism of these three amino acids is transamination that transfers the NH 2 group to glutamate, which in turn can be used to synthesize other amino acids, thus probably leading to a low rate of incorporation of 15 N elsewhere in the peptide sequence. This phenomenon, however, should not affect quantifications obtained with the ISIS method because only fragments produced by the supplied essential amino acids are measured. In our experiments the masses of precursors were not significantly modified, and we found that the only precaution to be taken was to use a sufficiently large (at least 2.5 amu) isolation window to capture all isotopic peaks for MS/MS.
One of the critical issues during development of this method was sensitivity. Immonium ions are not always strong peaks in MS/MS spectra acquired under CID conditions set to fragment peptides for protein identification. As a comparison, the chemical structure of the iTRAQ reporter ions has been designed to provide efficient fragmentation and high ionization efficiency for a higher sensitivity of detection. However, we found that the intensity of the immonium ion peaks can be significantly increased using a fast "zoom" scan with higher collision energy (60 eV) and by exploiting the capabilities of a pulsing instrument (QSTAR) to maximize the transmission efficiency of small fragment ions. The quantification scans of narrow range obtained were then combined with the CID spectra of full range before database search and quantification.
The data analysis work flow was simple; after protein identification by Mascot, an in-house built script extracted immonium ion peak intensities from the Mascot .dat file and associated them to the corresponding peptides. The relative protein quantification together with the associated statistics The * highlights proteins whose spot quantification differs less than 20% from the shotgun quantification. Cofilin-1 was detected in four spots on the 2D gel; all spot ISIS ratios are compared with the unique global ratio for cofilin-1 obtained by shotgun ISIS. PEBP, phosphatidylethanolamine-binding protein; TPIS, triosephosphate isomerase; HNRPD, heterogeneous nuclear ribonucleoprotein D.
was computed with the freeware software R, using a linear regression method, for the proteins presenting at least four quantified peptides. Because the quantification in ISIS work flow is in principle identical to that performed with iTRAQ data, both freeware and recent commercial software packages (e.g. iTRACKER and Mascot 2.2) can be easily adapted to process ISIS data.
The ISIS strategy was tested on a complex mixture of fully labeled proteins with known mixing ratio, showing good specificity and sensitivity and providing reliable protein quantification. 97% of the peptide sequences were found to contain at least one of the amino acids used for labeling, allowing the quantification of 63% of the identified proteins. The confidence intervals for the computed ratios were small, especially for the protein characterized by intense peaks or a high number of peptides. Using this data set, p and t cutoff values were set to detect high confidence ratios and discard the "low confidence" ratios. The implemented method was then used for a preliminary analysis of the proteomes of two different melanoma cell lines at different stages of tumor progression and allowed the identification and the quantification of differentially expressed proteins. Over a total of 582 proteins identified from replicate analyses, it was possible to obtain 502 high confidence ratios (86%). 26 proteins were found to be differentially expressed in these two cell lines. Interestingly a number of proteins involved in cytoskeletal modeling and organization (plectin-3, myosin-9, annexin-A1, calpain-2, and A-and B-filamins) were found to be preferentially expressed in SBCL2 cells. The differential expression of this group of proteins could at least partially explain some differences in morphology that can be observed between SBCL2 and SK-Mel28 cells.
To further validate our method, we performed a DIGE analysis of the same lysate mixtures, and we compared the protein ratios obtained using DIGE with those from ISIS quantification. Although some difficulties were experienced due to the identification of more proteins in one spot and to the uncertainty of spot boundaries in crowded regions of the gel, the two approaches generally provided similar results. Whenever a 2D gel spot contained more than one protein, the ISIS quantification could determine two distinct ratios, providing more accurate quantitative data. On the other hand, the ISIS shotgun experiment without two-dimensional separation provided information for the total level of identified proteins, without distinction for the possible isoforms produced by post-translational modifications, as usual for shotgun mass spectrometry-based techniques. The ISIS quantification obtained in the shotgun analysis detected a maximum difference in protein expression equal to 4.5-fold (Table I). On the other hand, the ratio values obtained from ISIS quantification of 2D gel spots spanned a broader range, detecting a maximum ratio value of 15.2 (Figs. 5 and 6). This indicates that the ISIS quantification method can potentially detect changes in protein expression of at least 15-fold.
Comparison with Available Techniques-The ISIS methodology potentially combines some typical advantages of metabolic labeling schemes with the ease of MS/MS-based quantification techniques such as iTRAQ for studying samples from cultured cells. Metabolic labeling allows mixing the differentially labeled samples just after cell lysis, thus eliminating sample-to-sample variations due to subsequent manipulations. This is especially important whenever complex multistep protein purification must be carried out and explains why SILAC and similar techniques are widely applied for the study of subcellular compartments or protein complexes. Although the number of amino acid analogues required for ISIS is high (six for the scheme described here), the reagents are significantly cheaper (about half the price for the same volume of medium) than those commonly used for classical SILAC ([ 13 C 6 ]lysine and [ 13 C 6 ]arginine). One distinctive advantage of the ISIS scheme is that when Val, Ile, and Leu are used for labeling, a considerable fraction of all peptides (37%) can be quantified twice through both channels, thus significantly increasing the number of data points and the confidence of quantification. Also in comparison with classical SILAC, the ISIS scheme (like iTRAQ) has the advantage that the spectra complexity is not increased by doubling of all peaks, which also results in a gain in sensitivity. The isobaric nature of the ISIS tags allows MS/MS-based quantification and very simple data analysis independent of native instrument data formats. Finally in contrast with the commercial SILAC, ISIS labeling should be easily applicable to analyze non-tryptic peptides and could be a viable solution to study fractions enriched with membrane or acidic proteins, which necessitate cleavage with proteases different from trypsin or chemical cleavage.
One main disadvantage of the ISIS method relative to the iTRAQ technique is the difficulty to modify the technique to perform higher order (Ͼ2 samples) multiplexing. One general limitation is the need of an instrument that is able to produce high levels of precursor fragmentation and transmission of very low mass ions.
Perspectives-Some features of ISIS method can be still improved; for example the data extraction procedure can be further explored to obtain a better determination of peak areas by optimizing centroiding parameters. Furthermore natural isotopic distributions ( 13 C content of natural amino acids) and isotopic impurities of reagents should be taken into account during quantification, and this should lead to less biased calculations especially for high ratios. Moreover we observed that, as expected, the peptide ratios obtained with low 13 C and 15 N ion intensities were much more variable than ratios obtained from intense immonium peaks. This indicates that a weighted linear regression may be appropriate to downweight the least precise measurements (the results described here were produced using an unweighted regression). Also the use of orthogonal regression, instead of the least squares, could produce a more accurate estimate of the ratios and the corresponding standard errors. Finally as for the iTRAQ tech-nique, an implementation of the ISIS protocol could benefit from the use of the latest generation of instruments, such as Orbitrap with an additional octopole collision cell for accurate measurement of low mass ions or QQ-TOF with intelligent data acquisition routines for the optimization of data acquisition as a function of signal intensity and signal/noise ratio. * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.