Online Nanoflow Multidimensional Fractionation for High Efficiency Phosphopeptide Analysis*

Despite intense, continued interest in global analyses of signaling cascades through mass spectrometry-based studies, the large-scale, systematic production of phosphoproteomics data has been hampered in-part by inefficient fractionation strategies subsequent to phosphopeptide enrichment. Here we explore two novel multidimensional fractionation strategies for analysis of phosphopeptides. In the first technique we utilize aliphatic ion pairing agents to improve retention of phosphopeptides at high pH in the first dimension of a two-dimensional RP-RP. The second approach is based on the addition of strong anion exchange as the second dimension in a three-dimensional reversed phase (RP)-strong anion exchange (SAX)-RP configuration. Both techniques provide for automated, online data acquisition, with the 3-D platform providing the highest performance both in terms of separation peak capacity and the number of unique phosphopeptide sequences identified per μg of cell lysate consumed. Our integrated RP-SAX-RP platform provides several analytical figures of merit, including: (1) orthogonal separation mechanisms in each dimension; (2) high separation peak capacity (3) efficient retention of singly- and multiply-phosphorylated peptides; (4) compatibility with automated, online LC-MS analysis. We demonstrate the reproducibility of RP-SAX-RP and apply it to the analysis of phosphopeptides derived from multiple biological contexts, including an in vitro model of acute myeloid leukemia in addition to primary polyclonal CD8+ T-cells activated in vivo through bacterial infection and then purified from a single mouse.

Despite intense, continued interest in global analyses of signaling cascades through mass spectrometry-based studies, the large-scale, systematic production of phosphoproteomics data has been hampered in-part by inefficient fractionation strategies subsequent to phosphopeptide enrichment. Here we explore two novel multidimensional fractionation strategies for analysis of phosphopeptides. In the first technique we utilize aliphatic ion pairing agents to improve retention of phosphopeptides at high pH in the first dimension of a twodimensional RP-RP. The second approach is based on the addition of strong anion exchange as the second dimension in a three-dimensional reversed phase (RP)-strong anion exchange (SAX)-RP configuration. Both techniques provide for automated, online data acquisition, with the 3-D platform providing the highest performance both in terms of separation peak capacity and the number of unique phosphopeptide sequences identified per g of cell lysate consumed. Our integrated RP-SAX-RP platform provides several analytical figures of merit, including: (1) orthogonal separation mechanisms in each dimension; (2) high separation peak capacity ( Reversible phosphorylation plays a central role in the regulation of normal cell physiology. The strong links between aberrant signaling and human disease, along with the potential for specific inhibition of disrupted kinase activity, continue to drive efforts aimed at systematic and largescale analysis of phosphorylation in cells and tissues. Shortly after introduction of immobilized metal affinity chromatography (IMAC) 1 as an enrichment tool prior to mass spectrometry (MS) analysis (1)(2)(3), several laboratories demonstrated the feasibility of phosphopeptide identification and quantitation en masse (4 -7). In the ensuing years, despite widespread proliferation of improved and innovative (8 -14) phosphoproteomics methods, the field struggled with low specificity and poor reproducibility within and across protocols and laboratories. These limitations effectively made dynamic range a secondary issue for the majority of studies. Over the past ca. 5 years, the performance of phosphopeptide enrichment protocols and related methods has stabilized; in fact several groups (15)(16)(17)(18)(19)(20)(21)(22)(23) have successfully coupled phosphopeptide enrichment with online or offline fractionation schemes to achieve, in some cases, over 10,000 phosphopeptide identifications. Although these strategies provide for larger phosphosite catalogs, closer inspection reveals that the analytical efficiency, as measured by the number of phosphopeptide identifications per microgram of biological lysate consumed, has remained surprisingly consistent at Ϸ1-10 phosphopeptides/g across a wide range of sample types (Table I). One explanation is that the physicochemical properties of phosphopeptides render them less amenable to fractionation by commonly used techniques. For example, although the combination of strong cation exchange (SCX) with reversed phase (RP) has been tremendously successful for fractionation of tryptic peptides generally (24 -26), its performance in phosphoproteomics studies has been hampered inpart by poor phosphopeptide binding in the first dimension (SCX) (10,27) and low separation peak capacity (28). These results are consistent with the modest difference in pK a between the binding sites of typical SCX resins and phosphorylated side chains of serine, threonine, and tyrosine amino acids. These limitations have led to the development of numerous other strategies for fractionation of phosphopeptides including strong anion exchange chromatography (17,22), continuous pH gradients (29), ERLIC (30,31), and hydrophilic interaction chromatography (19), all utilizing RP chromatography for the second dimension separation. In this report we describe two approaches for fractionation of phosphopeptides that build upon our recent work (32) in coupling true, nanoflow chromatography with reversed phase-reversed phase (RP-RP) fractionation (28). First we utilize aliphaticfunctionalized quarternary amines as ion pairing agents to improve retention of phosphopeptides in the first dimension RP separation performed at high pH. In a second approach, we use the standard ammonium formate RP-RP buffer system (28), and add anion exchange as a second dimension in a 3-D, RP-SAX-RP configuration. The latter provided for the highest performance in terms of separation peak capacity, orthogonality, and the number of unique phosphopeptide sequences identified per g of cell lysate consumed. We demonstrate the technical reproducibility of RP-SAX-RP at various fractionation depths and input levels. Finally, we use RP-SAX-RP for the quantitative analysis of divergent signaling between two clinically relevant, constitutively active FLT3 mutants in an in vitro model of acute myeloid leukemia, in addition to qualitative identification of phosphopeptides in primary polyclonal CD8 ϩ T-cells activated in vivo through bacterial infection and then purified from a single mouse. Collectively our RP-SAX-RP platform provides significantly improved efficiency (IDs/microgram input) along with multiple analytical figures of merit, including: (1) orthogonal separation mechanisms in each dimension; (2) high separation peak capacity; (3) efficient retention of singly-and multiply phosphorylated pepitdes; (4) compatibility with automated, online LC-MS analysis.

EXPERIMENTAL PROCEDURES
All multiplierz scripts referenced in the manuscript are freely available on our website at:http://blais.dfci.harvard.edu/index.php?id ϭ 64.
K562 Cell Culture and Preparation of Digested Lysate-K562 cells were cultured in RPMI 1640 media supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin at 37°C in 5% CO 2 . Cells (Ϸ5e7) were harvested by centrifugation during log phase. After washing twice with 20 ml phosphate buffered saline, the pellet was lysed with 3 ml of 8 M urea, 100 mM ammonium bicarbonate, and 30 l each of Sigma-Aldrich phosphatase inhibitor cocktails I and II. Protein concentration was determined using the Bradford Assay (Bio- Batch Mode Phosphopeptide Enrichment-Magnetic Ni-NTA agarose beads were washed 3ϫ with 400 l water, and treated with 400 l of 100 mM EDTA, pH 8.0 for 30 min with end-over-end rotation. EDTA solution was removed, and beads were then washed 3ϫ with 400 l water, and treated with 600 l of 10 mM aqueous FeCl 3 solution for 30 min with end-over-end rotation. After removing excess metal ions, beads were washed 3ϫ with 400 l water, and 1ϫ with 400 l 1:1:1 acetonitrile:methanol:0.01% acetic acid. Tryptic peptides (in 80% MeCN/0.1% TFA; typical peptide concentrations were Յ1 g/ l; see Supplemental Experimental Procedures for additional details) were then added to the beads, and phosphopeptide capture proceeded for 30 min with end-over-end rotation. After removing the supernatant, beads were washed 3ϫ with 400 l 80% acetonitrile/ 0.1% TFA. Phosphopeptides were eluted with 50 l 1:20 ammonia/ water for 30 min, dried to Ϸ5 l by vacuum centrifugation, and reconstituted with 20 mM ammonium formate buffer.
Two-dimensional RP-RP and Three-dimensional RP-SAX-RP Systems-The multidimension fractionation platform ( Fig. 1) consisted of a two-pump UPLC system with autosampler (Waters, Milford, MA) and an external valve (Valco, Austin, TX). For RP-RP experiments (Fig.  1A), the first dimension analytical column consisted of a 150 m I.D. capillary packed with 5 cm of 5 m C18 (XBridge, Waters). The weak needle wash and isocratic pump delivered various ion pairing agents all at pH 10.0. To transfer peptides between the first and second dimension columns, the trapping and vent valves were positioned to direct flow to the second dimension precolumn (150 m I.D. capillary packed with 4 cm of POROS10R2, Applied Biosystems, Framingham, MA), the isocratic pump delivered 1 l/min of pH 10 buffer (with sample loop of 20 l in-line), and the binary pump delivered 10 l/min of 0.2 M acetic acid (to dilute organic content and acidify the first dimension column effluent) for 25.5 min. Elution of peptides was accomplished by injection of acetonitrile fractions at pH 10.0 with or without an ion-pairing agent. After trapping, both valve positions were switched, to create a precolumn split and allow for gradient elution (2-30% B in 60 min, A ϭ 0.2 M acetic acid, B ϭ acetonitrile with 0.2 M acetic acid) of peptides to the analytical column (30 m I.D. capillary packed with 12 cm of 5 m Monitor C18, Column Engineering, Ontario, CA containing integrated 1 m emitter tip), and into the mass spectrometer (Fig. 1B) at a flow rate of Ϸ30 nL/min. Threedimensional RP-SAX-RP experiments were conducted in a similar fashion, except that an additional anion exchange column (150 m I.D. capillary packed with 5 cm of POROS10HQ) was connected to the outlet of the first dimension RP column. Elution of peptides trapped on this column was performed using injections of ammonium formate, or potassium chloride at pH 10.0. Experiments were designed to transfer fractions of equivalent peptide complexity to the final dimension column with each injection. Toward this end, organic eluents used to elute peptides from the first to second dimensions typically included a low concentration of salt (70 mM ammonium formate or 10 mM KCl) to simultaneously elute a subset of peptides to the final dimension column in the same trapping event (see Supple-mental Experimental Procedures). After completion of a fractionation experiment, columns were regenerated with 10 l 90% acetonitrile, 10 l of 1 M KCl in 10% acetic acid, 10 l of 50% acetonitrile with 500 mM ammonium formation, and 10 l of 900 mM ammonium formate in 10% acetic acid. In addition, the final dimension column was subjected to a series of three rapid (5 min.) organic gradients.

MS and MS/MS Parameters on the LTQ-Orbitrap XL Mass
Spectrometer-The LTQ-Orbitrap XL mass spectrometer (ThermoFisher Scientific, San Jose, CA) was programmed to operate in data dependent mode, such that the top 10 most abundant precursors in each MS scan were subjected to MS/MS (CAD, electron multiplier detection, collision energy ϭ 35%, isolation width ϭ 2.8 Da, threshold ϭ 20,000). Dynamic exclusion was enabled with a repeat count of 1 and an exclusion duration of 30 s. ESI voltage was 2.2 kV. Additional parameters can be found in Supplemental Experimental Information.
Quantitative Phosphoproteomics Analysis of Acute Myeloid Leukemia (AML) Model System (WT, WT ϩ FL, ITD, D835Y)-Murine BaF3 cells that stably expressed wild-type (WT), or constitutively active mutant forms (ITD and D835Y, respectively) of the receptor tyrosine kinase FLT-3 were cultured as described (33). WT cells were serum starved overnight and were left untreated as a control, or were treated with 50 ng/ml of FLT-3 ligand (FL) for 5 min. Cells were collected, rinsed with cold PBS, and lysed in 8 M urea. Solubilized lysates were diluted to a final urea concentration of 1.8 M with 50 mM ammonium bicarbonate, and protein concentration was measured by BCA assay. Protein lysates (ϳ2E6 cells corresponding to ϳ100 g protein) were digested with trypsin after reduction with 10 mM dithiotreitol and alkylation with 55 mM iodoacetamide. For each biological condition (WT, WTϩFL, ITD, D835Y), 100 g of tryptic peptides were reconstituted with 30 l 1 M triethylammonium bicarbonate, 70 l ethanol, and one tube of iTRAQ reagent (114, WT; 115, WTϩFL; 116, D835Y; 117, ITD). Labeled peptides were mixed, dried by vacuum centrifugation, desalted, and reconstituted in 400 l of 80% MeCN, 0.1% TFA, and phosphopeptides enriched via batch mode enrichment (described above). Phosphopeptides were subjected to a 67 fraction threedimensional RP-SAX-RP MS/MS analysis using a QSTAR Elite mass spectrometer (AB Sciex, Foster City, CA).

MS and MS/MS Parameters on the QSTAR Elite Mass
Spectrometer-MS spectra were acquired for 0.5 s. A minimum threshold of 50 counts was used for MS/MS, with a charge state range of 2 ϩ to 4 ϩ and a multiplier value of 4. Up to 5 precursors could be selected for each cycle, with a 20 s exclusion time for each precursor selected for MS/MS. One broad range collisionally activated dissociation (CAD) (m/z: 140 -2000) and narrow range CAD (m/z: 104 -125) were acquired for each precursor. A fixed collision energy of 60 eV was used for narrow range CAD scans.
Automated Tip Washing and Positioning-Digital PicoView ESI sources (New Objective, Woburn, MA; model DPV-550 for the Orbitrap XL and DPV-400 for the QSTAR Elite) were used to facilitate positioning of the emitter tip at the orifice of the mass spectrometer during each analysis (Fig. 1D). While loading samples or trapping peptides on the final dimension precolumn, the tip was positioned at a wash station and bathed in a continuous stream of water.
Data Processing-Orbitrap data files were directly accessed and converted to .mgf using default parameters of the multiplierz script orbi2mgf.mz. Several scans in the first data file of each experiment were manually summed in XCalibur to derive the mass error of Si(CH 3 ) 2 O)) 6 . This value was used to re-calibrate precursor masses in .mgf files from each set of experiments using the proportional equation described by Wenger et al. (34) (recalibrated mz ϭ mz ϫ(445.120025/observed Si(CH 3 ) 2 O)) 6 m/z; RECAL Orbi data from mgf.mz). QSTAR files were converted to .mgf using Mascot Daemon (2.2.0, settings are described in Supplemental Experimental Information), and .mgf from a given experiment were re-calibrated using a linear equation derived from fitting experimentally observed masses in a high confidence peptide identification (RECAL QSTAR data from mgf.mz). Files were searched using Mascot version 2.2.1 against a forward-reversed human (38190 forward entries) (35) or mouse (34967 forward entries) (36) NCBI refseq database (both downloaded Nov. 2009) with an appended cRAP (common repository of adventitious proteins(37)) database of 752 entries. Precursor and product ion tolerances were 10 ppm and 0.6 Da, or 100 ppm and 0.1 Da for Orbitrap or QSTAR data, respectively. Search parameters included trypsin specificity, up to two missed cleavages, fixed carbamidomethylation (C, ϩ57 Da), variable deamidation (NQ, ϩ1 Da), oxidation (M, ϩ16 Da), and phosphorylation (STY, ϩ80 Da), and, when applicable, fixed iTRAQ modification (N-term, K ϩ145 Da). After searching, an excel spreadsheet containing the mascot search results was generated using Multiplierz software (Geryon edition 0.7.188) (38). Data were processed to remove reverse database hits and forward hits with false discovery rate (FDR) Ͼ1.0%, (forward-reverse_FDR_ filter.mz). A list of unique peptides (FDRϽ1%) from each analysis was generated using the multifile detect feature of multiplierz, and further processed using a script to count the number of unique phosphorylated peptides (39). Unique phosphorylated peptides were defined as unique combinations of sequence and phosphorylation state, i.e. the same sequence phosphorylated at different residues is counted only once, whereas the same sequence appearing with different numbers of phosphorylation sites would be counted according to degree.
Phosphorylation Site Localization-Phosphorylation site localization was evaluated using a multiplierz script that implemented the MD-score method of Savitski et al. (40). Briefly, the MD score of a phosphopeptide is the difference in Mascot score between the highest ranking and the next highest-ranked peptide hit with the same sequence but different assigned phosphorylation site(s). If no such hit exists, the peptide score was used as the MD score. To quantify phosphorylation sites on the protein level, a score threshold was used to filter the peptides at a false localization rate of 1%. This threshold is instrument specific and we employed values recommended by Savistki et al. (40), specifically 11 for Orbitrap data and 10 for QSTAR data.
iTRAQ Reporter Ion Extraction and Processing-For relative quantitation, iTRAQ reporter ion peak intensity values were extracted from narrow range CAD scans (wiff_iTRAQ.mz) and corrected for known isotopic impurities (iTRAQ_software_corrections.mz). A separate LC/ MS/MS analysis of the Fe 3ϩ -NTA resin supernatant was performed, and signal in each quantitation channel (114, 115, 116, and 117) was summed (across all peptides) and used to derive normalization factors to further correct the phosphopeptide data for small variations in source protein concentration. For peptides with multiple MS/MS scans, the associated iTRAQ reporter signal intensity values were summed prior to calculation of ratios.
Analysis of Reproducibility and Fraction-to-fraction Overlap-Extracted ion chromatograms (XICs) were generated for reproducibly detected peptides from replicate three-dimensional RP-SAX-RP experiments using deisotoped MS scan data. For each peptide, an XIC was generated (across every fraction) using a time window corresponding to the earliest RT -0.5 min to the latest RT ϩ 0.5 min and a mass tolerance of 0.015 Da. The precursor signal for a peptide was obtained by summing XIC apex intensities across all fractions in an individual experiment. Fraction to fraction overlap for a given peptide was evaluated by expressing the maximum peptide intensity across all fractions as a percentage of the total signal.
Calculation of Peak Capacity-Peak capacity of a given fraction was estimated by dividing the useful peptide elution range (retention time of last peptide identification -retention time of first identification) by the median peptide peak width at the XIC base. Base peak widths were estimated by multiplying half-height XIC peak widths by a factor of 2. The estimated peak capacity of an experiment was obtained by summation of the peak capacities of the individual fractions.
Orthogonality Plots-To visualize the orthogonality of the threedimensional RP-SAX-RP experiments, we plotted second dimension salt concentration versus first dimension acetonitrile concentration (pH 10.0) and represented the number of phosphopeptides identified in each third dimension LC-MS/MS analysis as circles of proportional diameter. To focus on useful peptide fractionation space, high salt (Ͼ700 mM) and organic (Ͼ25%) fractions were excluded from analysis. Note that each phosphopeptide identification was considered a data point for the linear regression.
Analysis of Amino Acid Frequencies and Permutation Test-A list of unique phosphopeptide sequences (with no missed cleavages) from phosphoELM (release 8.3; March 2010) in text file format was generated using Phosphopeptides_from_ELM.mz. Acidic amino acid frequencies were evaluated by counting the number of each amino acid and dividing by the total number of residues, and were plotted as log ratios (base 2). To evaluate the significance of amino acid biases between specific analyses, we used a permutation test as described previously. (41) Briefly, peptide sequences in each dataset were combined and then randomly permuted so that the number of peptides in each dataset remained constant. After each permutation, amino acid frequencies and log ratios were recalculated for a total of 10,000 times. One-sided p values were calculated by summing the frequency (probability) of all log ratios equal to or more extreme than the observed log ratios and were doubled to yield two-sided p values. Lastly, two-sided p values were Bonferroni corrected based on the number of amino acids. Frequency of occurrence was considered significant for p Յ 0.01.
Infection of Mice-C57BL/6J mice were purchased from The Jackson Laboratory (Bar Harbor, ME) and housed in specific pathogenfree conditions. All animal procedures were approved by the Center for Animal Resources and Comparative Medicine (ARCM, Harvard Medical School, Boston, MA). Recombinant Listeria monocytogenes expressing chicken ovalbumin (Lm-OVA) (30) was provided by Dr. Michael Starnbach (Harvard Medical School). Bacteria were grown to early exponential phase in brain heart infusion broth at 37°C for 1 h, spun down by centrifugation at 3600 rpm for 10 min at 4°C (Allegra 6KR Kneewell Centrifuge, Beckman Coulter, Fullerton, CA) and resuspended in phosphate buffered saline (PBS). Mice were infected at 8 weeks of age by injecting Ϸ1.6 ϫ 10 3 colony forming units of live Lm-OVA i.v. into the lateral tail vein. The inoculum dose was confirmed by duplicate dilution plating on streptomycin-containing Tryptic Soy Agar plates. Bacteria and infected animals and tissues were manipulated in Biosafety Level 2 (BL2) biohazard containment cabinets.
Preparation of CD8 ϩ T-Cell Lysates-Sorted CD8 ϩ T cells (Ϸ1-3 ϫ 10 6 ) were spun down by centrifugation at 1000 rpm for 5 min at 4°C (Allegra 6KR Kneewell Centrifuge, Beckman Coulter, Fullerton, CA); washed by centrifugation at 8000 rpm for 2 min at 4°C (Microcentrifuge 5415R, Eppendorf, Westbury, NY) with cold PBS containing 5 mM sodium fluoride, 1 mM sodium orthovanadate (New England Biolabs, Ipswich, MA) and 1ϫ Halt Phosphatase Inhibitor Mixture (Thermo Scientific, Rockford, IL), and lysed with 150 l of cold 8 M urea in 100 mM ammonium bicarbonate containing phosphatase inhibitors as above. Protein concentrations were determined using the MicroBCA Protein Assay (Thermo Scientific) after diluting an aliquot of the cell lysates with distilled water to Ͻ1 M urea. The total amount of protein obtained from the sorted cells was Ϸ5-10 g/spleen. Proteins were reduced and alkylated as described above, and then desalted using a self-packed C18 column (360 m O.D. ϫ 250 m I.D. with 8 cm of resin). Phosphopeptides were enriched in batch mode as described above and analyzed in a 10 fraction RP-SAX-RP MS/MS experiment using a LTQ-Orbitrap XL (ThermoFisher Scientific, Waltham, MA).
Safety Considerations-Trifluoroacetic acid is corrosive, and should be handled in a fume hood with appropriate protective equipment as described in the manufacturer's material safety data sheet. Recombinant Lm-OVA and infected mice and cells should be handled in a BL2 cabinet.

RESULTS
All native mass spectrometry data and results (mzResults) files, including annotated MS/MS spectra and other metrics as described in the Philadelphia Guidelines are freely accessible online at: pcXKvH75QϩmGbF8do6hCUB0Lvzwq-P2EjY4l04ccANl7zNxLpR6l8NZ3Vv9sAYlrAQdϩ24DWWsQg-LIiLFzϩL4irJPpZMAAAAAAAFkjAϭϭ Multiphosphorylated Peptides Are Poorly Retained on RP at pH 10 -We recently described a true nanoflow, online platform (32) for UHPLC-compatible RP-RP fractionation (28) of tryptic peptides. Fig. 1 (A-D) shows a schematic representation of our two-dimensional nanoflow system and related modifications (Fig. 1E) required for our three-dimensional implementation (see below and Methods). Briefly, our platform is based on a Waters NanoACQUITY UHPLC modified to accept self-packed columns (42) and reconfigured with an additional two-position, six-port valve to enable a precolumn effluent split, in a vented column configuration (43,44), that allows for gradient elution of peptides in the second dimension at low nanoliter per minute flow rates ( Fig. 1B and 1C). A UHPLC binary pump generates the solvent gradient for LC-MS/MS in the usual manner, whereas the second on-board pump is used in isocratic mode to introduce samples or first dimension eluents loaded from autosampler vials. In contrast to previous reports (10, 15-22, 45, 46), our system provides an automated, online platform that leverages the combined advantages of multidimensional fractionation (25,28) and high efficiency electrospray ionization achieved at ultralow effluent flow rates (42). Given the anecdotal evidence that phosphopeptides are particularly susceptible to losses as a result of sample handling procedures, in addition to the predicted (47) and observed (10,27) weak retention of phosphopeptides on SCX resins, we reasoned that our RP-RP platform would provide an improved fractionation system for analysis of phosphopeptides. To evaluate the analytical capabilities of this platform, we used Fe-NTA (8) to enrich phosphopeptides from Ϸ100 g of human myeloid K562 cell lysate. Peptides were loaded through the NanoAQUITY autosampler, onto the first dimension RP column in pH 10 ammonium formate buffer, and analyzed by RP-RP-MS/MS. Data were acquired for four fractions, including the loading flow-through. In total we identified 1356 unique phosphopeptides (Ͻ1% FDR, see supplemental Fig. S1 and supplemental File 1) across all four fractions, corresponding to ϳ14 IDs per microgram of cell lysate. Only 73 nonphosphorylated peptides were observed, and less than 12% of all identified peptides were detected in more than one fraction, indicating high selectivity (Ϸ95%) for Fe-NTA based enrichment and minimal fraction-to-fraction overlap, respectively. Interestingly, we noticed that a significant number (306, or Ϸ23% of the total) of phosphopeptides were detected in the loading flow-through. To explore the physicochemical nature of these peptides, we plotted the percentage of multiply phosphorylated peptides in addition to the frequency of acidic residues, as a function of first dimension acetonitrile concentration ( Fig. 2A). We observed that the flow-through fraction contained a higher percentage of multiply-phosphorylated peptides (37%) as compared with that observed across the total of all peptides detected (21%). At pH ϭ 10 each phosphorylated side chain will carry two negative charges (pK a1 Ϸ1.2 and pK a2 Ϸ6.5) (47,48) and bind weakly to RP resin, in accordance with solvophobic theory (49); poor retention would be further exacerbated for multiplyphosphorylated peptides, in agreement with data in Fig. 2A. Similarly, acidic residues (D, red and E, blue) occurred with greater frequency in the un-retained and 10% acetonitrile fractions (DϩE Ϸ43% and Ϸ22%, respectively) as compared with the 30% acetonitrile fraction (Ϸ8%, Fig. 2A), or more generally as compared with all unique phosphopeptide sequences in the phosphoELM database (Ϸ16%) or the human refseq database (22) (Ϸ12%, supplemental Fig. S2). Next, we applied a permutation test (41) to verify the statistical significance of the apparent bias for D (p ϭ Ͻ0.001) and E (p ϭ Ͻ0.001) residues in phosphopeptides detected in the flowthrough and 30% fractions, or similarly between the 10 and 30% acetonitrile fractions (Fig. 2B).
Based on the above results, we next asked whether alternative organic modifiers could be used to improve phosphopeptide retention on the first dimension column at high pH. In fact, data from previous studies demonstrated that functionalized tertiary and quaternary amines will effectively ion pair with acidic moieties in various biomolecules at high pH to improve retention and separation on RP resins (50 -58). Encouragingly several of these compounds have been used in  ). B, C, An additional six-port, two-position valve is used to configure a vented column geometry in the final dimension and provide efficient capture of peptides on the low-pH RP precolumn. B, First dimension fractions are diluted (10:1) and acidified with RP solvent A (0.1 M acetic acid) introduced by an ultra-high pressure binary pump. C, The same binary pump then delivers an organic gradient to elute peptides from the second dimension column for MS/MS analysis. The precolumn effluent split provided a stable column/ESI flow rate of Ϸ30 nL/min. Active solvent flow paths are highlighted in green. D, A computer-controlled positioning platform (Digital PicoView) automatically moves the emitter tip between "electrospray" and "wash" positions during data acquisition and sample loading, respectively. E, A SAX column is connected to the high pH first dimension to create an online RP-SAX-RP fractionation platform. After loading phosphopeptides, the autosampler is used to inject first-(acetonitrile, orange) or second-(KCl, blue) dimension eluents, respectively. LC-MS applications (50 -54, 56). Toward this end, we first prepared a series of synthetic phosphopeptides, including a triply phosphorylated sequence, and then verified that their retention at high pH generally increased in the presence of ion pairing reagents with greater hydrophobicity (supplemental Table S1). Of the compounds tested, we observed that tetrapropylammonium ion provided the strongest and most reproducible retention of phosphopeptides on RP resin at basic pH, even at concentrations as low as 5 mM. Next we again enriched phosphopeptides from human myeloid K562 cell lysate and performed RP-RP-MS/MS in which 73 fractions were generated in 5 mM ammonium formate buffer with 5 mM tetrapropylammonium ion. In this experiment (supplemental Fig. S3), we detected 4647 unique phosphopeptides (Ͻ1% FDR, Ϸ12 ID/g, see Supplemental File 2) and only 340 nonphosphorylated peptides (selectivity Ϸ93%). Importantly, no phosphorylated peptides were observed in the flow-through fraction, indicating that the tetrapropylammonium ion provided for improved retention of phosphopeptides derived from complex biological extracts on RP resin under alkaline conditions.
To better elucidate the ion pairing properties of this compound, we repeated the analysis described above and observed a positive correlation between relative acidic content and the acetonitrile concentration required for elution from the first dimension column (Fig. 2C) as well as a statistical enrichment of acidic residues in the late-versus early-eluting fractions (Fig. 2D). These results confirm that the overall acidity of phosphopeptides impedes their binding to RP resin at high pH. Despite the capacity of tetrapropylammonium ion to modulate the hydrophobicity of phosphopeptides, we found that only 41 of the 73 fractions contained an appreciable number (Ͼ50) of sequence identifications and that some 52% of phosphopeptides were detected across multiple second dimen- sion fractions. In addition, we observed that Ϸ18% acetonitrile was sufficient to elute the majority of phosphopeptides from the first dimension column (supplemental Fig. S3). Collectively these data, along with those from previous studies that utilized SCX-based fractionation techniques (10,15,27), demonstrate that the acidity of phosphopeptides presents unique challenges with respect to high performance, multidimensional separation strategies for proteomics applications.
Three-dimensional Separation Based on RP-SAX-RP Enables Efficient Fractionation and Identification of Phosphopeptides-Based on the results above, we next asked whether addition of anion exchange as a second dimension, in a three-dimension, RP-SAX-RP geometry (Fig. 1E), would provide an effective fractionation platform. In this configuration, phosphopeptides captured by SAX, either from the loading flow-through or via controlled elution from the first dimension RP column at high pH, are subsequently eluted onto the third dimension RP column with increasing concentrations of salt. To test this strategy, we first injected the phosphopeptide standards prepared in ammonium formate buffer (20 mM, pH ϭ 10) onto a tandem set of RP and anion exchange columns. As expected, the triply phosphorylated peptide bound strongly to the SAX column under these conditions and was only observed in the third dimension LC-MS/MS analysis upon elution with 500 mM KCl; less acidic phosphopeptide standards required a combination of acetonitrile and KCl for elution from the first and second dimension columns, respectively (supplemental Table S2). In essence, this system provides for a three-dimensional analysis of phosphopeptides, with two orthogonal dimensions of RP, and a charge based anion exchange separation.
We next sought to establish several analytical performance metrics for RP-SAX-RP fractionation of phosphopeptides derived from a complex biological matrix. Toward this end we prepared multiple aliquots of tryptic peptides, each derived from 100 g of human myeloid K562 cell lysate. We used batch mode Fe-NTA (8) to enrich phosphopeptides and acquired RP-SAX-RP-MS/MS data at a depth of 11 fractions. This experiment was then repeated, starting with a second aliquot of tryptic peptides, followed by phosphopeptide enrichment, and finally 11 fraction RP-SAX-RP-MS/MS analysis (see supplemental Files 3 and 4). Fig. 3A (left panel) shows an overlay of phosphopeptide identifications from both analyses plotted as a function of fraction number; given the good correlation in total identifications per fraction, we next asked whether individual phosphopeptides eluted across multiple first-and second-dimension fractions or were largely segregated within single, discrete fractions. Using sequence identification as a guide, we observed that Ϸ78% of all phosphopeptides were constrained within individual fractions. However, the use of peptide ID as a surrogate for "detection" is complicated by the stochastic nature of MS/MS. In an effort to mitigate this effect we next used our recently described multiplierz desktop software environment (38,59) to generate extracted ion chromatograms (XICs) for phosphopeptides reproducibly identified in both experiments. (see Experimental Methods). Consistent with the results above, this analysis revealed that on average 85% of all ion current associated with a given phosphopeptide was detected in a single fraction. The neighboring Venn diagram (Fig. 3A, second panel) illustrates that we identified 6276 and 6110 unique phosphopeptides, respectively, with 4376 (54%) identified reproducibly across both RP-SAX-RP analyses. Interestingly, we identified 193 tyrosine phosphorylated peptides containing 202 unique sites of tyrosine phosphorylation, including five unique pY sites on BCR-ABL, the primary oncogenic kinase that drives proliferation in K562 cells. To avoid inflation of IDs we counted all occurrences of the same amino acid sequence and phosphorylation state as a single phosphopeptide sequence (i.e. oxidation, deamidation, and the position of the phosphate group(s) are ignored). Next we plotted the corresponding XIC apex intensity ratios for the set of reproducibly detected phosphopeptides as a histogram (third panel), and found that 95% of these ratios fell within Ϸthreefold of the mean value. To explore the degree of orthogonality across the separation space we represented the number of phosphopeptides identified in each third dimension LC-MS/MS analysis as circles of proportional diameter, and then plotted these as function of both first-and second-dimension eluent concentrations (Fig. 3A, fourth panel). A least squares fit to these data yielded a residual value (R 2 ) of 0.03, indicating a high degree of orthogonality (e.g. poor correlation between first-and second-dimension elution) in the RP-SAX-RP separation.
To further confirm the robustness of our RP-SAX-RP platform we next extended this analysis to fractionation depths of 32 and 53 fractions, respectively ( Fig. 3B and 3C; see supplemental . In each case we observed very similar results in terms of the reproducibility of fractionation (first panel), peptide identification (second panel), and peptide intensity ratios (third panel). In addition we found that the RP-SAX-RP separation platform maintained a high degree of orthogonality at each fractionation depth tested (fourth panel). To test whether the use of an alkali salt would significantly affect system performance, we substituted KCl for ammonium formate as the second dimension eluent and repeated the characterization experiment described above (see supplemental Files 9 -10). Comparison of the data in Fig. 3C and 3D suggest that ammonium formate and KCl provide comparable data, although as a volatile salt, we suspect that the former may provide more robust performance over time (60). Interestingly, at higher fractionation depths, we noted a concomitant increase in the relative number of phosphopeptides contained in the first dimension flow-through (see shaded region in panel one and corresponding column in Table II). This observation highlights the capacity of our RP-SAX-RP configuration to provide controlled fractionation of phosphopeptides that bind poorly to RP at high pH and would otherwise be difficult to detect comprehensively in a single flow-through fraction. Consistent with the data above (Fig. 2), phosphopeptides detected in the first dimension RP flow-through were enriched in acidic residues as compared with the entire fractionation space (on average, 20% versus 16% overall). Table  II lists analytical figures of merit for the RP-SAX-RP analyses described in Fig. 3. As expected, we observed a systematic increase in separation peak capacity as a function of fractionation depth. In addition we achieved a dramatic improvement in efficiency as measured by both the number of phosphopeptide and phosphosite identifications per microgram of input lysate as compared with numerous previous studies (compare Tables I and II, respectively). Collectively, these results indicate that the RP-SAX-RP configuration provides an effective means to fractionate and identify phosphopeptides derived from complex biological matrices and offers multiple analytical figures of merit, including: (1) orthogonal separation mechanisms in each dimension; (2) high separation peak capacity; (3) efficient retention of singly-and multiply-phosphorylated peptides; (4) compatibility with automated, online LC-MS analysis.
Quantification of Oncogenic FLT3 Signaling in Acute Myeloid Leukemia-Aberrant signaling from tyrosine kinases is a well established oncogenic event in hematopoietic malignan- cies (61,62). For example, mutation of the gene encoding Fms-like tyrosine kinase 3 (FLT3) or overexpression of the corresponding protein product represent the most frequent molecular abnormalities in AML, and occur in some 35 and 90% of patients, respectively. Genetic lesions that lead to either in-frame, tandem internal duplications (ITD) near the juxtamembrane domain or point mutations in the kinase activation loop at D835 (typically D835Y) collectively account for ϳ35% of clinical cases. Both mutation classes induce constitutive tyrosine kinase activity and confer IL3-independent growth of hematopoietic cell lines. Interestingly, cumulative evidence to date suggests that ITD and D835Y mutants respond differently to the current suite of targeted therapeutics and that they can activate different downstream targets (63). These observations, along with collective experience to date in the use of therapeutics that target single oncogenes or pathways (64 -67), motivate continued efforts to elucidate oncogenic signaling pathways in greater detail. Toward this end, we used iTRAQ stable isotope labels to encode tryptic peptides derived from cells that express wild-type (WT) or mutant FLT3 kinases (supplemental Fig. S4). Phosphopeptides were enriched via Fe-NTA IMAC and analyzed in a 67 fraction, RP-SAX-RP experiment (Fig. 4). A total of 7787 unique phosphopeptides were detected (supplemental File 11), including 182 that were phosphorylated on tyrosine. These data along with those described in Fig. 3 above, demonstrate that RP-SAX-RP provides for high efficiency fractionation and detection of phosphopeptides derived from different cell lines (human K562 and murine BaF3) and on multiple mass spectrometry platforms (Thermo Oribtrap XL and ABI-SCIEX QSTAR Elite). Moreover, the results in Fig. 4 represent a nearly 10-fold increase in the number of phosphorylation sites detected as compared with a recent study from our lab in which we used an offline RP fractionation scheme (five fractions total) to analyze FLT3 signaling in the same cell system, under the same culture conditions (68). The efficiency of our RP-SAX-RP platform is further evidenced by the fact that our data compare favorably in scale (4586 phosphorylation sites) to that in a recent report from Choudhary et al. (14700 unique phosphorylation sites) (69), while consuming some 25-fold less cell lysate (400 g versus 10 mg).
For the first step in our analysis, we asked what proteins and pathways were activated by stimulation of the WT kinase with FLT3 ligand (FL). We observed regulated (greater than twofold) phosphorylation on members of the MAP kinase pathway (JUN, RPS6KA1, RPS6KA4, MAPK3, and BRAF), as well as proteins involved in lipid (INPP5D and ITPKB) signaling (70) (Fig. 4A, yellow circles). Both of these protein families have been linked to FL-mediated activation of survival and proliferation pathways (71), and serve as a positive control for our approach.
To investigate potential points of divergent signaling, we next compared iTRAQ reporter ion intensities for peptides identified in the context of FLT3-ITD and -D835Y mutants (Fig.  4. RP-SAX-RP analysis of iTRAQ labeled phosphopeptides from an AML model system consisting of wild-type and two oncogenic FLT3 mutants (ITD and D835Y) expressed in murine BaF3 cells. A, Plot of iTRAQ reporter ion intensities (log base 10) for phosphopeptides derived from cells that expressed wild-type FLT3 and were subjected to serum starvation (x axis) or stimulation with FLT3 4B). These data revealed that the point mutant was a potent activator of tyrosine signaling (Fig. 4B, yellow circles); of the 244 phosphotyrosine candidate peptides identified (comprising 182 unique sequence-phosphorylation states), 47 exhibited an iTRAQ ratio consistent with increased phosphorylation via D835Y-mediated signaling. In contrast, only seven peptides were preferentially phosphorylated on tyrosine in the context of FLT3-ITD. In agreement with our previous studies (33) we observed that constitutive signaling in the context of D835Y led to tyrosine phosphorylation near the C terminus of the phosphatase SHP1 (Fig. 4B). Data and results from recent studies (72,73) suggest that these sites may serve as SH-2 domain scaffolds for subsequent assembly of GAB2-SOCS complexes that, in-turn negatively regulate JAK-STAT signaling. In addition, we observed increased phosphorylation on several proteins linked to regulation of signaling in lymphoid cells, including: CD244, PRKC␦, PTK2B, PECAM1, PAG1, PDCD1, CARD11, and PLCG2 (74 -81). Relative phosphorylation of PRKC␦ and STAT5 in the context of FLT3-ITD and -D835Y was confirmed by Western blot of lysates from biological replicates using site specific antibodies (Fig. 4C). Interestingly we observed the opposite situation for serine and threonine phosphorylation. Of the 602 candidate pS/pT peptide sequences that exhibited significant iTRAQ reporter ion ratios (representing 519 unique sequence/phosphorylation states), 479 of these were preferentially phosphorylated in the context of FLT3-ITD signaling, versus only 123 for point mutant. (Fig. 4B, blue circles). KEGG pathway (70) analysis of ITD-mediated phosphoproteins revealed activation of survival and proliferation pathways (insulin, ERBB, MAPK signaling). In addition, we observed phosphorylation on proteins linked to chromatin remodeling (HDAC2, DMT3␤ (82)) and transcription (RUNX1, RUNX3, IKZF2). Western blot against total RUNX1 protein ( Fig. 4D) confirmed that the increase in phosphorylation observed in the context of FLT3-ITD signaling was not an artifact of protein expression; in fact overall RUNX1 protein levels were higher in cells that expressed the point mutant. DMT3␤, RUNX1, and IKZF2 are well-established mediators of hematopoiesis and they are often functionally disrupted in the context of leukemia (83). Moreover, it is now well recognized that oncogenic kinase activity is not sufficient to support complete leukemic transformation (84 -86). Of the two FLT3 mutants studied herein, the ITD is strongly associated with poor clinical prognoses (63,84,86), and our data are consistent with the hypothesis that FLT3-ITD signaling has a strong propensity, relative to D835Y, to drive epigenetic events that disrupt differentiation and lead to a more aggressive AML phenotype. For example, we identified phosphorylation sites (S77, S78, and S433) on the Ikaros family transcription factor IKZF2 that lie in close proximity to SIN3 binding sites (87). It is tempting to speculate that these sites mediate the interaction of IKZF2 and SIN3, given that the latter is known to bind RUNX1, also observed to be preferentially phosphorylated in the context of ITD signaling, and repress transcription through recruitment of HDACs (88).
As noted above, multiple reports demonstrated that FLT3 mutants activate different downstream targets in vitro, and these observations have been corroborated in vivo with murine bone marrow transplant models. For example, Grundler et al. (89) found that FLT3-ITD induced a myeloproliferative disorder whereas transplantation of cells harboring FLT3-D835Y led to aberrant expansion of the lymphoid compartment in recipient mice. These results correlate well with the clinical manifestations observed for leukemia patients who harbor FLT3-ITD or D835Y mutations; the former is most often observed in the context of myeloid malignancies whereas the latter occurs with high frequency in acute lymphoid leukemia (84). In our study, the biological pathways represented by proteins that were identified as being preferentially phosphorylated by either ITD or D835Y are consistent with both in vivo bone marrow transplant models and human clinical phenotypes. Taken together, these data demonstrate that the increased dynamic range of phosphopeptide identification inherent to RP-SAX-RP fractionation provides for identification and quantification of signaling events related to clinically relevant oncogenic kinase activity in complex biological matrices. The relatively modest quantities of cell lysate required for these data (100 g lysate per iTRAQ channel) provide further evidence for the high efficiency of our fractionation scheme.
Analysis of Primary CD8 ϩ T-Cells from a Single Mouse-Although cultured cells lines are an invaluable resource for biomedical research, it is well recognized that, as a result of genetic alternations and selective pressures associated with repeated passage under non-physiologic growth conditions, they may not recapitulate the behavior of cells in vivo (90,91). In principle the use of primary cells will circumvent many of the shortcomings of immortalized cell lines, although in practice this approach is complicated by the frequent mismatch between the availability of biological material and the limited detection and dynamic range provided by a majority of LC-MS platforms. Encouraged by the efficiency gains observed for analysis of cultured cell lines (Figs. 3 and 4), we ligand (y axis), with proteins involved in MAPK signaling (70) highlighted as yellow circles. B, Plot of iTRAQ reporter ion intensities (log base 10) for phosphopeptides derived from cells that stably expressed constitutively active FLT3-ITD (x axis) and -D835Y (y axis) mutants, with phosphotyrosine or phospho-serine/threnine residues displayed as yellow and blue circles, respectively. Selected proteins discussed in the main text are highlighted as large circles with red (pY) and green (pS/pT) outlines, respectively. C, Western blots of PRKC␦-pY311 and STAT5␣/␤-pY694 from lysates of FLT3-D835Y and -ITD expressing cells and mass spectra indicating iTRAQ reporter signals of the tryptic phosphopeptides spanning these phosphorylation sites (D835Y ϭ 116.1; ITD ϭ 117.1). D, Western blot of total RUNX1 protein from lysates of FLT3-D835Y and -ITD expressing cells.
next asked whether our RP-SAX-RP platform would enable reproducible detection of phosphorylation in rare, primary cell populations. As a first step toward this goal, we revisited the experimental strategy described in Fig. 3, but scaled the input K562 lysate down by an order of magnitude to 10 g. Technical replicates were processed as described above and analyzed by RP-SAX-RP at a depth of 10 fractions. Again we observed very good agreement for both analyses in terms of the distribution of phosphopeptides across each fraction (Fig.   5A, first panel; see supplemental Files 12-13), overlap in phosphopeptide identifications (second panel), and precursor intensities (third panel). In addition we observed orthogonal behavior for separation of phosphopeptides in the first-and second-dimensions, respectively (Fig. 5A, fourth panel).
With these results in-hand we next moved on to a samplelimited scenario that would also allow us to test the impact of biological variability on our RP-SAX-RP platform. For this analysis we chose murine primary CD8 ϩ T lymphocytes, which are relatively small (6 -15 m in diameter) specialized cells of the adaptive immune system whose primary function is to destroy host cells infected by viruses or other intracellular pathogens (e.g. bacteria). CD8 ϩ T-cells recognize foreign antigens displayed on the surface of host cells in complex with MHC Class I molecules. Notably, engagement of the T-cell receptor with MHC-antigen complexes drives a diverse array of signaling networks that ultimately results in (1) lysis of infected host cells, (2) growth and clonal expansion of activated T cells, followed some time later by either (3) apoptosis of primed T-cells, or (4) differentiation of activated cells into memory T cells. Although several phosphorylation pathways have been associated with some of these endpoints, the majority of current data is derived from in vitro studies, or through the use of phospho-specific antibodies that collectively target a modest number of phosphoproteins.
To study phosphorylation in primary T cells responding to a bacterial challenge in vivo, we first infected mice with Lm-OVA, and then purified activated CD8 ϩ T cells at the peak of the immune response (8 days postinfection) by MACS and fluorescence-activated cell sorting (FACS) (Fig. 5B). Importantly, it has been demonstrated that the majority of activated T cells in mice at this time point are responding to the ongoing infection in a pathogen-specific manner (92). Approximately 1-3 ϫ 10 6 purified CD44int/high CD8 ϩ T cells (Ͼ94% CD8ϩ/ B220-and Ͼ99% CD44int/high) were obtained from a single spleen (corresponding to Ϸ5-10 g of protein), and lysed immediately upon harvest. Phosphopeptides were isolated by Fe-NTA IMAC, and subjected to a 10 fraction RP-SAX-RP analysis. We profiled phosphopeptides in this manner from three individual mice and identified 2516, 3254, and 2575 unique phosphopeptides, respectively (Table II). Across biological triplicates we observed a high degree of reproducibility in terms of the distribution of phosphopeptides across each fraction (Fig. 5C, top left panel, and supplemental Files 14 -16) and unique phosphopeptide identifications (Fig. 5C, top right panel). Finally as in our previous analyses ( Fig. 3 and Fig. 5A) we again observed orthogonal behavior for separation of phosphopeptides in the first-and second-dimensions, respectively (Fig. 5C, bottom panel).
Among these data we observed phosphorylation sites previously identified in T-cells stimulated in vitro (supplemental Table S3). In addition, we identified phosphorylation on many proteins with known functional roles in T cells, although the specific sites that we detected have not, to the best of our knowledge, been described in activated T cells (supplemental Table S4). Interestingly we also observed phosphorylation on several proteins whose function is largely unexplored in CD8 ϩ T cells. These included regulators of (1) chromatin organization and remodeling, (2) transcription, (3) RNA processing (including pre-mRNA splicing factors), and (4) protein translation. Importantly, the primary CD8 ϩ T lymphocytes used in our analysis were not further manipulated after FACS purification; as a result, the phosphorylation events observed likely reflect physiologic CD8 ϩ T cell responses within the mouse. Collectively, these data provide multiple avenues for future studies designed to decipher the mechanistic role of these, in many cases novel, signaling events in CD8 ϩ T cell activation and fate determination in vivo.
General Recommendations-We conclude with general recommendations and observations that will facilitate continued development and deployment of RP-SAX-RP in other laboratories.
System Capacity-Based on unpublished studies for shotgun analysis of tryptic (e.g. unmodified) peptides, we estimate that the total loading capacity of our first dimension RP column (150 m I.D. ϫ 5 cm of 5 m) is currently Ϸ10 g. Comparisons of total ion chromatograms from MS analyses of total lysates and phosphopeptides indicate that Ϸ1% of tryptic peptides are phosphorylated and retained on the Fe-NTA column during enrichment. These data suggest that our current RP-SAX-RP configuration can accommodate phosphopeptides enriched from up to Ϸ1 mg of biological lysate. Work is ongoing to evaluate the relative performance of larger first dimension columns and to define the point at which unretained peptides from the first dimension begin to exceed the capacity of the second dimension column. We anticipate that careful calibration of these variables will enable robust analysis of phosphopeptides enriched from larger quantities of biological input material.
First and Second Dimension Elution Conditions-We have found that the three-dimensional RP-SAX-RP platform is easily tailored to a variety of sample types and desired fractionation depths. Generally, eluent concentrations in a range of 7 to 25% acetonitrile and 15 to 700 mM ammonium formate represent the useful peptide elution "windows" for first and second dimensions, respectively. Within these limits, the total number of desired fractions can be evenly divided between first and second dimensions, with minimal fraction-to-fraction overlap of peptides. Importantly, our experience to date suggests that these boundary conditions are robust with respect to biological input, obviating the need to run repeated pilot experiments for every sample.
Instrument Time-Total time of analysis is another important consideration when using multidimension fractionation techniques. The analyses described herein required ϳ24 h per 10 fractions. For example experiments performed at a depth of 53 fractions required Ϸ5 days to complete. Work is ongoing to refine third dimension gradient conditions to determine whether total system peak capacity as a function of fractionation depth (supplemental Fig. S5) can be improved in terms of (1) analysis depth (e.g. more fractions) and (2) efficiency (e.g. higher peak capacity per unit of instrument time).
System Reliability-System reliability is particularly important given that a single, multidimension analysis can require several days of instrument time. Consistent with our experience to date using 25 m I.D. analytical columns with integrated electrospray emitters in a one dimension nanoflow configuration (42), or coupled with high-pH RP in an automated RP-RP configuration (32), our RP-SAX-RP platform has proven to be very robust. The system typically provides uninterrupted analysis of hundreds of fractions, often spanning several months of analysis time, without a major failure. In addition, we have observed that vigorous regeneration of column resins (alternating high salt/high organic injections, for example 1 M KCl with 10% acetic acid, 900 mM ammonium formate with 10% acetic acid, 90% acetonitrile) and use of PEEK coated fused silica for delivering high pH solvents has improved overall column lifetime and system up-time. DISCUSSION Results from recent large scale proteomics studies suggest that experimental protocols for phosphopeptide enrichment and analysis have stabilized within individual labs. In fact, related work from our lab (8) and other groups (93,94) demonstrated that enrichment of phosphopeptides from complex biological matrices can now be performed in an automated fashion. Despite this progress we noted that the yield of phosphopeptides as a function of input material had remained relatively constant across studies from numerous labs (ours included), spanning a wide range of biological sample types and quantities (Table I). Although data that describe an absolute number of phosphopeptides in mammalian systems are not currently available, previous studies have attempted to address relative levels of phosphorylation, and suggest that (1) 30 -50% of the proteome is phosphorylated at any given time (95), and (2) a surprising number of proteins are phosphorylated at a stoichiometry of greater than 90% (96). Based on these observations, we asked whether improvements in fractionation could provide a viable avenue for significant increases in the efficiency of phosphopeptide identification. Toward this end we first explored the performance of our recently described (32) implementation of RP-RP fractionation (28) for phosphopeptide analysis. Interestingly, we observed similar behavior for fractionation of phosphopeptides via RP-RP as has been reported for SCX-RP. That is, binding efficiency to the first dimension RP column (ammonium formate, pH ϭ 10) was inversely related to the degree of peptide phosphorylation in particular, as well as overall peptide acidic content. We sought to improve phosphopeptide retention under alkaline conditions through the use of aliphatic-functionalized quaternary amines as ion pairing agents, and found that tetrapropylammonium ion completely abrogated breakthrough of phosphopeptides; moreover, we observed that peptide elution order from the first dimension column was well correlated with both the degree of phosphorylation and relative acidic content (D and E) of phosphopeptides. Despite the successful use of ion pairing chromatography in the analysis of a wide range of molecule classes (50,54,97), we observed significant fraction-to-fraction overlap of phosphopeptides in RP-RP fractionation augmented with tetrapropylammonium formate, suggesting that further optimization will be required to utilize these compounds for the analysis of complex biological lysates.
Based on the above results we next placed an anion exchange (SAX) column between the first and (now) third dimension RP columns to ensure efficient capture of phosphopeptides (e.g. those phosphopeptides that were otherwise un-retained on the first dimension column). We found that discrete salt steps facilitated controlled elution of phosphopeptides from the SAX column, providing the basis for an integrated, three-dimensional fractionation platform. Using K562 as a model mammalian system we observed reproducible and orthogonal separation and identification of phosphopeptides across a range of fractionation depths, with efficiencies more than an order of magnitude greater than those previously reported, in some cases exceeding 450 phosphopeptide IDs per g of cell lysate consumed (Table II). These studies revealed that RP-SAX-RP provides a unique combination of figures of merit as compared with previously reported multidimensional fractionation systems (10, 15-22, 29, 31, 45, 46), including: (1) orthogonal separation mechanisms in each dimension; (2) high separation peak capacity; (3) efficient retention of singly-and multiply phosphorylated peptides; (4) compatibility with automated, online LC-MS analysis. The results reported in Fig. 3 and Table II may represent a lower bound in performance given the improvements in associated proteomics technologies such as specialized algorithms (40,98,99) that improve phosphopeptide sequence and site assignment as compared with standard database search algorithms, along with mass spectrometers having higher acquisition rates (100 -102) or equipped with complimentary dissociation techniques (103)(104)(105)(106). Although the online, capillary format of our RP-SAX-RP platform provides for automated LC-MS/MS acquisition, it is important to note the trade-off in total sample capacity as compared with systems that utilize one or more stages of offline separation. Although these schemes provide for fractionation of large sample quantities, recent reports suggest that sample handling, lyophilization, and nonspecific adsorption of peptides to tube surfaces can reduce yield significantly. For example, Dowell and colleagues reported that online SCX-RP detected twice the number of peptides as compared with an equivalent offline SCX-RP format. Not surprisingly these deleterious effects scale inversely with total input, limiting the performance for offline systems for sample quantities below Ϸ50 -100 g (107, 108). In addition, it remains to be seen whether efficiencies similar to those demonstrated herein will translate to a wider range of sample types and quantities. Ideally, improved fractionation can directly offset ionization suppression and stochastic undersampling of precursors during MS/MS, both of which compress observed dynamic range. However, the relatively narrow range of efficiencies observed across multiple experiments and cell types listed in Table I could already reflect improved dynamic range from higher performance mass spectrometers or other technological developments.
Given that the distribution of phosphoprotein abundances in vivo is not known, the relative impact of improved fractionation, along with other analytical figures of merit in LC-MS/MS (109), will need to be empirically tested across multiple experimental scenarios. Finally, although we have chosen the specific combination of RP and SAX for the first and second dimensions, respectively, it will be interesting to see whether the use of an orthogonal fourth dimension or an alternative separation phase in the current three-dimensional configuration, such as HILIC (19,45) SCX (10), or ERLIC (30, 31) will provide still further improvements in performance.
Having established several analytical figures of merit for our RP-SAX-RP platform, we next analyzed phosphopeptides in the context of an in vitro model of AML to decipher divergent signaling downstream of FLT3-ITD and -D835Y, two constitutively active kinases commonly associated with leukemogenesis. We found that RP-SAX-RP provided sufficient peak capacity to support accurate iTRAQ-based quantification of discrete signaling events associated with each kinase, and in toto these data recapitulated results from murine bone marrow transplant models and human clinical phenotypes. As a more challenging test case for RP-SAX-RP we analyzed phosphopeptides derived from primary CD8 ϩ T lymphocytes that were activated in vivo, and subsequently isolated from a single mouse. We identified an average of 2782 unique phosphopeptide sequences across biological triplicates, starting with 1-3 ϫ 10 6 highly enriched T cells (Ϸ5-10 g of total protein lysate) for each analysis. We observed reproducibility in phosphopeptide identifications comparable to that obtained when analyzing technical replicates of K562 lysate, suggesting that our RP-SAX-RP platform is very robust with respect to variation across biological replicates. These results highlight the analytical capabilities of RP-SAX-RP, and are particularly encouraging in the context of two recent studies, one of which identified 281 phosphorylation sites from 2 mg of biological lysate derived from partially sorted, primary lymphocytes (110), and a second which reported a total of 6248 phosphopeptides across 46 LC-MS/MS analyses of whole mouse spleens (Ϸ2 mg total protein per analysis, or 92 mg total input) (111) in a model designed to study immune response to anthrax exposure. In both studies, the samples were composed of heterogeneous cell populations, each of which will contribute a distinct signaling response to the final ensemble of phosphopeptides measured. The first case included a diverse repertoire of lymphocytes normally found in blood, including: subtypes of B cells, CD4 ϩ T cells, CD8 ϩ T cells, NK cells, and NK-T cells. Interestingly in the latter study, the authors noted an inability to analyze phosphorylation in mouse primary T cells, and hence used phosphoprofiling of whole spleens as a surrogate for host immune response to anthrax exposure. However, analysis of whole spleen encompasses all the aforementioned B, T, NK, and NK-T lymphocytes, in addition to myeloid (dendritic, macrophage, and neutrophil), erythroid, as well as cells of endothelial and stro-mal origin. In contrast, our RP-SAX-RP platform facilitates analysis of highly enriched, antigen specific CD8 ϩ T cells responding to an ongoing bacterial infection. Accordingly, our data are largely free of confounding effects from contaminating cells and should therefore provide direct insights into the pathways regulating CD8 ϩ T-cell responses in clinically relevant contexts (e.g. infection, vaccine development, etc.). Collectively our data and results demonstrate that RP-SAX-RP provides for reproducible, high efficiency phosphopeptide fractionation and will be particularly valuable for applications in which availability of biological materials is limited.