Introduction

Protein palmitoylation (or, more correctly, protein S-acylation) is the thioesterification of fatty acyl moieties, typically the 16-carbon palmitoyl moiety, to selected protein cysteines. Similar to prenylation and myristoylation, and often in combination with these two lipidations, palmitoylation may serve to tether proteins to membrane cytosolic surfaces (for reviews, see refs. 1, 2, 3). Many signaling proteins, including key players in cancer and synaptic signaling, are palmitoylated. Notably, many G proteins rely on palmitoylation for proper membrane-localized function, including H- and N-Ras, some Rho proteins, as well as the α subunits of most heterotrimeric G proteins. In distinction to prenylation and myristoylation, palmitoylation frequently also is found as a transmembrane (TM) protein modification. For the TM proteins, embedded in the membrane by its hydrophobic, bilayer-spanning TM domains, the addition of palmitoyl tethers seems superfluous. For these TM palmitoyl-proteins (PPs), a palmitoylation role in targeting to raft-like membrane domains often is invoked; the saturated palmitoyl moiety should increase affinity for the cholesterol- and sphingolipid-rich, liquid-ordered membrane domain. Another unique feature of protein palmitoylation that excites interest is its reversibility. The regulated addition and removal of palmitoyl tethers provides an attractive mechanism for controlling membrane and/or raft association. Such control of targeting by reverse palmitoylation, however, has yet been demonstrated only for a small handful of PPs.

The protocol detailed below, first published as part of a global analysis of palmitoylation in the yeast Saccharomyces cerevisiae4,5, purifies and identifies the subset of proteins that are palmitoylated from highly complex protein extracts. This PP purification is a proteomic extrapolation of the acyl-biotinyl exchange (ABE) chemistry of Drisdel and Green, an in vitro method that substitutes biotinyl moieties for the thioester-linked palmitoyl modifications6. The ABE-generated biotinylated proteins can then be affinity-purified using streptavidin–agarose and identified by proteomic mass spectrometry (MS). ABE comprises a sequence of three chemical steps: (i) an exhaustive blockade of free thiols with N-ethylmaleimide (NEM); (ii) hydroxylamine treatment to release thioester-linked palmitoyl moieties, restoring the modified cysteine to thiols, (iii) which are then biotinylated using a thiol-reactive biotinylation reagent (Fig. 1). As the biotinylation reagent, we have opted to use biotin–HPDP, which has several advantages over the biotin–BMCC originally used by Drisdel and Green6. Biotin–HPDP, which disulfide bonds to the thiols, is more thiol-specific; furthermore, it facilitates an easy release of proteins bound to the streptavidin–agarose affinity matrix, through β-mercaptoethanol-mediated cleavage of the biotin–Cys linkage.

Figure 1: Schematic of the proteomic acyl-biotinyl exchange methodology.
figure 1

Schematic of the proteomic acyl-biotinyl exchange methodology.

Full size image

An important control, included in all our analyses, is the processing of an equal portion of the initial protein extract through a parallel ABE protocol that omits the hydroxylamine thioester-cleavage step. In the absence of hydroxylamine, palmitoyl modifications should not be removed and PPs should not be biotinylated or purified. However, owing either to inappropriate biotinylation or non-specific streptavidin–agarose binding, some proteins do end up purified into the minus-hydroxylamine (−HA) control sample (Fig. 2). These −HA sample proteins also are non-specifically purified into the experimental plus-hydroxylamine (+HA) sample and thus represent a source of false-positive identifications. In addition to contaminant proteins, there also is a clear set of +HA sample proteins that show the HA-dependent purification indicative of likely PP status (Fig. 2). The challenge is to distinguish true PPs from the non-specifically purified contaminant proteins; in the protocol below, this is accomplished using a quantitative MS analysis that compares each protein's representation in parallel +HA and −HA samples.

Figure 2: SDS–polyacrylamide gel electrophoresis (SDS-PAGE) of purified parallel plus- and minus-hydroxylamine (HA) yeast samples just before proteomic multi-dimensional protein identification technology (MuDPIT) analysis.
figure 2

A protein extract derived from total yeast membranes was split in half and purified through the +HA and −HA acyl-biotinyl exchange (ABE) protocols in parallel. After the final streptavidin–agarose purification, a comparison was made of 1% of each of the two samples by SDS-PAGE and silver-staining. Presumptive contaminant proteins, showing HA-independent purification, are marked to the right by hash marks, while the candidate PPs, which show HA-dependent purification, are indicated with arrows. Reprinted with permission from ref. 4.

Full size image

Multi-dimensional protein identification technology analysis

HA samples (+ and −) are analyzed by multi-dimensional protein identification technology (MuDPIT), a high-throughput, tandem MS–based proteomic technology developed for the analysis of highly complex protein samples7,8. MuDPIT is distinguished by its orthogonal 2D peptide separation, which occurs before, but in-line with, the tandem MS. Following exhaustive proteolysis of the protein samples, the peptides are fractionated from one another both by charge and by hydrophobicity, continuously eluting into the tandem MS for sequencing over an 8–12 h timeframe. A single typical MuDPIT run of a complex sample may sequence tens of thousands of peptides that link to hundreds or thousands of database proteins. As MuDPIT is a highly complex technology that utilizes expensive MS instrumentation, it is assumed that the researcher embarking on the protocol below will first establish a collaboration with an experienced scientist with ample experience in running MuDPIT.

MuDPIT, in addition to providing comprehensive sample coverage, also affords a crude but facile quantitative analysis. Unlike other quantitative proteomic approaches, which typically require differential labeling of samples with heavy and light isotopes and complex downstream data analysis9, MuDPIT allows a crude quantitation which is based upon a parameter that is part of the standard MuDPIT dataset: the spectral count number. The spectral count number is the number of sequenced peptides that link to each identified protein. Within a single MuDPIT run, abundant sample proteins often are independently re-identified multiple times, both through identification of the protein's different component peptides and through iterative re-identification of the same peptide eluting in multiple fractions owing to peak broadening within the initial chromatographic separations. The spectral count number, which includes these redundant peptide identifications, has been demonstrated to be a useful metric for comparing a protein's relative abundance among samples10.

Quantitation based on spectral count has proven critical both for identifying PPs and for analyzing the effects of treatments that perturb palmitoylation4. Distinguishing PPs from the contaminant protein background is based on comparing spectral counts for each protein from parallel +HA and −HA samples. Non-specifically purified contaminant proteins show roughly equivalent abundances in +HA and −HA samples, whereas the PPs show exclusive, or substantially higher, +HA sample abundances. Figure 3, a spectral count analysis of the proteins identified from +HA and −HA ABE samples from the yeast S. cerevisiae, illustrates the utility of the spectral count parameter. All 1,558 yeast proteins identified from MuDPIT analyses of four parallel +HA and −HA samples are plotted. Each protein (each dot) is plotted as averaged +HA sample spectral counts on the x-axis versus averaged −HA sample spectral counts on the y-axis. Of the 1,558 total yeast proteins identified, the vast majority are not palmitoylated, but rather contaminant proteins, showing significant representations in both +HA and −HA samples and trending, therefore, toward the x, y-diagonal (Fig. 3). These contaminant proteins tend to be proteins of known yeast cell high abundance11. Many of the PPs (red dots), a group that includes both previously known PPs and PPs newly identified by this analysis, are detected only from the +HA samples and thus map onto the x-axis (Fig. 3). Other PPs map near, but not directly on top of, the x-axis (Fig. 3), reflecting a +HA sample bias that is not fully exclusive. This second class of PPs, which are also detected at low levels from −HA samples, would be overlooked by a non-quantitative analysis that simply subtracts the list of −HA proteins from the list of +HA proteins. In addition to identifying 12 of the 15 known yeast PPs, our analysis also identified and confirmed palmitoylation for an additional 35 proteins4. The identified PPs encompassed all the known types of PPs—proteins that tether to membranes solely through palmitoylation, proteins that are palmitoylated in addition to being also either N-terminally myristoylated or C-terminally prenylated; as well as many TM PPs.

Figure 3: An x, y-scatter plot depiction of yeast proteins identified by multi-dimensional protein identification technology (MuDPIT) of plus- and minus-hydroxylamine (HA) samples.
figure 3

Each of the 1,558 proteins identified by MuDPIT analysis in our yeast proteomic analysis is plotted as averaged, normalized +HA sample spectral counts (x-coordinate) against averaged, normalized −HA sample spectral counts (y-coordinate)4. The palmitoyl proteins (PPs), including both the 15 PPs that were known to be palmitoylated at the outset of this analysis and the 35 PPs newly identified and confirmed by this analysis, are shown in red. The right-hand panel shows an expanded portion of the graph at the left. Reprinted with permission from ref. 4.

Full size image

Our spectral count–based approach, in addition to allowing PPs to be distinguished from co-purifying contaminants, may be used to assess the global changes in palmitoylation that can be induced by drugs or mutation. The power of such an approach is illustrated by our recent mapping of the yeast PPs to their cognate palmitoylation enzymes4. The palmitoyl proteome of a wild-type yeast strain was compared to the proteomes of mutant strains deleted for the genes encoding the seven members of the newly identified DHHC (Asp-His-His-Cys) protein acyl transferase (PAT) family12,13,14. PP substrates of a particular PAT are expected to be lost from the palmitoyl proteome of strains deficient for that PAT; thus, substrates are highlighted by their absence from the DHHC deletion strain palmitoyl proteome (Fig. 4). This analysis finds not only PPs that fully drop out but also PPs with partial under-representations, indicative of a partial requirement for the deleted PAT. Analysis of strains multiply deleted for the different DHHC genes allows the overlapping specificity relationships among the different DHHC PATs to be discerned. Many of the enzyme–substrate relationships uncovered in this comparative proteomic analysis have indeed been confirmed in targeted testings of the palmitoylation of individual PPs in different DHHC mutant backgrounds4. In principle, this approach should allow global changes in palmitoylation to be monitored in response to a wide variety of perturbants and stimuli—e.g., hormonal signals, drugs and changes in growth conditions. Indeed, the mammalian DHHC PAT family is an important set of future drug targets with possible utility in the treatment of cancer and other diseases. Recently, a first generation of compounds with inhibitory effects on palmitoylation has been developed15. Comparing the profiles from drug-treated cells with those of mutants deficient in the individual PATs should provide a means of mapping drugs to their target PATs.

Figure 4: Colorimetric depiction of palmitoylation changes seen in strains deficient for the different yeast DHHC protein acyl transferases (PATs).
figure 4

Plus-hydroxylamine (HA) samples purified from 11 different wild-type and mutant yeast strains were analyzed by multi-dimensional protein identification technology (MuDPIT) with the 30 most prominent palmitoyl proteins (PPs) (listed at the bottom) being compared for relative abundance with the spectral count metric. Wild-type:mutant spectral count ratios for each protein were converted to color, with proteins showing 20-fold or greater mutant sample under-representations shown in red, and proteins with intermediate under-representations depicted by intermediate red shadings (for details on the colorimetric conversion, see ref. 4). Relevant strain genotypes are indicated on the left: for each tested strain, the seven different yeast DHHC PAT-encoding genes are indicated as being wild-type (+), deleted (Δ) or replaced by a conditional GAL1-driven depletion allele strain with the gene encoding the indicated DHHC PAT fully deleted (Δ) or, in a few cases, depletion of the indicated PAT was induced via glucose-mediated repression of a GAL1-driven PAT allele (downward arrow; glucose depletion periods are indicated at right). Note that for many PPs, palmitoylation is significantly blocked only in strains concomitantly mutated for multiple DHHC PATs, presumably indicating substantial specificity overlaps among the seven different DHHC PATs. Reprinted with permission from ref. 4. DHHC.

Full size image

The protocol that follows should be applicable to tissue from any source organism with the available sequence data to allow MS-based protein identification. As indicated above, this protocol has been used in yeast both to identify many new PPs and to map substrate PPs with their cognate modifying DHHC PAT4. Recently, we have applied the same approach to the analysis of various mammalian palmitoyl proteomes—from mouse or rat whole brain, and from primary cultures of embryonic rat brain neurons. Similar to our yeast analyses4, our initial mammalian ABE purifications also readily pull out a mix of both known and new PPs; indeed, our analysis of rat embryonic neurons identifies 24 known PPs from among the 100 top-scoring proteins (R. Kang, J.W., A.O.B., J. Yates, N.G.D. and A. El-Husseini, unpublished results). For both the yeast and mammalian analyses, starting samples are typically scaled to approximately 10 mg of starting total protein, although successful analyses have also been performed on samples that start with as little as 2 mg of total protein. The initial steps of the protocol describe the preparation of starting protein extracts both from a yeast cell culture and from mouse whole brain. For other source tissues, initial homogenization may have to be varied appropriately. However, protocol steps subsequent to these initial homogenization steps should be identical to those detailed below.

The protocol is divided into the following sub-sections: (i) preparation of the starting protein extracts (the preparation of lysates from both yeast cultures and mouse brain as well as an optional crude membrane purification step to enrich for PPs is described); (ii) ABE; (iii) affinity-purification of biotinylated proteins; (iv) proteolysis and preparation of samples for MuDPIT analysis; (v) MuDPIT analysis (it is assumed that the MuDPIT analysis will be done collaboratively with a proteomics facility well versed in the procedure; consequently, the MuDPIT portion of the protocol is presented in overview fashion, with the particular parameters that may be unique to our analyses of palmitoylation indicated. For an excellent step-by-step MuDPIT protocol detailing both the set-up and running of MuDPIT, the reader is referred to ref. 16); (vi) spectral count–based quantitation.

Materials

Reagents

  • Pepstatin (Sigma)

  • Antipain (Sigma)

  • Chymostatin (Sigma)

  • Leupeptin (Sigma)

  • Triton X-100 10% solution (Anatrace Anapoe, cat. no. X-100)

  • NEM (Pierce, cat. no. PI 23030)

  • Hydroxylamine (Sigma)

  • HPDP–biotin (Pierce, cat. no. PI 21341)

  • Streptavidin–agarose (Pierce, cat. no. PI 20349)

  • β-Mercaptoethanol (Fisher, cat. no. BP176)

  • Tris(2-carboxyethyl)phosphine (TCEP; Sigma)

  • Iodoacetamide (Sigma)

  • Endoproteinase Lys-C (Roche)

  • Glucose

  • Peptone

  • Yeast extract

  • Trypsin (Roche)

  • Reversed-phase resins—Aqua C18 (3-μm beads, 100-Å pores) and Aqua C18 (5-μm beads, 300-Å pores) (Phenomenex)

  • Strong cation exchange resin—Luna SCX (5-μm beads, 100-Å pores; Phenomenex)

  • Acetonitrile (HPLC grade)

  • Ammonium acetate

  • Formic acid

  • 1 M Tris/Cl, pH 7.4

  • 1 M Tris/Cl, pH 8.5

  • 0.5 M EDTA, pH 8.0

  • 1 M NEM in ethanol

    Critical

    Prepare fresh for every experiment; store on ice.

  • 1 mg ml−1 pepstatin in methanol

    Critical

    Store at −20 °C.

  • 10 mg ml−1 leupeptin in water

    Critical

    Store at −20 °C.

  • 10 mg ml−1 antipain in DMSO

    Critical

    Store at −20 °C.

  • 10 mg ml−1 chymostatin in ethanol

    Critical

    Store at −20 °C.

  • 0.1 M phenylmethanesulfonyl fluoride(PMSF; Sigma, cat. no. P7626) in ethanol

    Critical

    Store at 4 °C.

  • 1 M hydroxylamine, pH 7.4

    Critical

    Prepare fresh for each experiment; store on ice.

  • 50 mM HPDP–biotin in DMSO

    Critical

    Store at −20 °C; solution may become somewhat cloudy after −20 °C storage.

  • 4 mM HPDP–biotin in N,N-dimethyl formamide

    Critical

    Dilute from 50 mM stock just before use; keep on ice until use.

  • 100 mM TCEP

    Critical

    Store at −20 °C.

  • 1 M CaCl2

  • 0.5 M iodoacetamide

    Critical

    Prepare fresh for every experiment; store on ice.

  • MuDPIT buffer A (5% acetonitrile, 0.1% formic acid)

  • MuDPIT buffer B (80% acetonitrile, 0.1% formic acid)

  • MuDPIT buffer C (500 mM ammonium acetate, 5% acetonitrile, 0.1% formic acid)

Equipment

  • Coffee grinder (Krups, cat. no. F2037051)

  • Sorvall (Dupont) RC-5B centrifuge with GSA and HS-4 rotors

  • 50-ml disposable centrifuge tubes (Sarstedt, cat. no. 62.547.205)

  • 12-ml screw-cap centrifuge tube (Sarstedt, cat. no. 60.540)

  • 1.5-ml screw-cap centrifuge tube (Sarstedt, cat. no. 72.692)

  • IKA tissue homogenizer (T25 basic; Ultra Turrax, cat. no. T25BS1) with S25N-8G blade attachment

  • Sonicator (Sonic Dismembrator Model 500; Fisher Scientific) with micro-probe (Branson model 1020)

  • Ultra-centrifuge and fixed angle rotor (optional)—e.g., Beckman L8-M ultra-centrifuge with Type 80Ti rotor (Beckman)

  • Tube rotator (Thermolyne Labquake, cat. no. 400110)

  • Micro-centrifuge

  • Tube rocker (Thermolyne Vari-Mix, cat. no. M48725)

  • LTQ ion-trap mass spectrometer (Thermo-Finnigan)

  • HPLC (Agilent)

  • Fast computer system for analysis of tandem mass spectra and for database searching

  • MuDPIT microcapillary chromatography set-up (see EQUIPMENT SETUP)

  • Microfilter assembly (UpChurch Scientific, Oak Harbour, WA) (see EQUIPMENT SETUP)

Reagent setup

  • YPD (1% yeast extract, 2% peptone, 2% glucose) For 1 l, dissolve 10 g yeast extract, 20 g peptone in 960 ml water. Autoclave to sterilize, then add 40 ml of sterile 50% glucose.

  • Lysis buffer (LB; 150 mM NaCl, 50 mM Tris, 5 mM EDTA, pH 7.4) For 100 ml, combine 3 ml 5 M NaCl, 5 ml Tris/Cl pH 7.4, 1 ml 0.5 M EDTA and 91 ml water. Supplement with necessary components (e.g., Triton X-100, NEM, PI) to the concentrations indicated in the protocol.

  • 100 × protease inhibitors (PIs) Combine 25 μg ml−1 each of pepstatin, leupeptin, antipain and chymostatin. For 1 ml, combine 250 μl 1 mg ml−1 pepstatin, 25 μl 10 mg ml−1 leupeptin, 25 μl 10 mg ml−1 antipain, 25 μl 10 mg ml−1 chymostatin and 675 μl ethanol.

  • 4% SDS buffer (4SB; 4% SDS, 50 mM Tris, 5 mM EDTA, pH 7.4) For 10 ml, combine 4 ml 10% SDS, 0.5 ml 1 M Tris/Cl pH 7.4, 0.1 ml 0.5 M EDTA and 5.4 ml water. Supplement with necessary components (e.g., NEM) to the concentrations indicated in the protocol.

  • +HA buffer (0.7 M hydroxylamine, 1 mM HPDP–biotin, 0.2% Triton X-100, 1 mM PMSF, 1 × PI pH 7.4) For 10 ml, combine 2.5 ml 4 mM HPDP–biotin, 0.2 ml 10% Triton X-100, 0.1 ml 100 × PI, 0.1 ml 0.1 M PMSF, 7 ml 1 M hydroxylamine pH 7.4, and 0.1 ml water.

  • −HA buffer (50 mM Tris, 1 mM HPDP–biotin, 0.2% Triton X-100, 1 mM PMSF, 1 × PI, pH7.4) For 10 ml, combine 2.5 ml 4 mM HPDP–biotin, 0.5 ml Tris/Cl pH 7.4, 0.2 ml 10% Triton X-100, 0.1 ml 100 × PI, 0.1 ml 0.1 M PMSF and 6.6 ml water.

  • Low HPDP–biotin buffer (150 mM NaCl, 50 mM Tris, 5 mM EDTA, 0.2 mM HPDP–biotin, 0.2% Triton X-100, 1 mM PMSF, 1 × PI, pH 7.4) For 10 ml, combine 0.5 ml 4 mM HPDP–biotin, 0.3 ml 5 M NaCl, 0.5 ml Tris/Cl pH 7.4, 0.2 ml 10% Triton X-100, 0.1 ml 0.5 M EDTA, 0.1 ml 100 × PI, 0.1 ml 0.1 M PMSF and 8.2 ml water.

  • 2% SDS buffer (2SB; 2% SDS, 50 mM Tris, 5 mM EDTA, pH 7.4) For 10 ml, combine 2 ml 10% SDS, 0.5 ml 1 M Tris/Cl pH 7.4, 0.1 ml 0.5 M EDTA and 7.4 ml water.

  • Proteolysis buffer (8 M urea, 0.1 M Tris, pH 8.5) For 100 ml, dissolve 6 g urea in 10 ml 1 M Tris/Cl pH 8.5, with water added to volume.

Equipment setup

  • MuDPIT microcapillary chromatography set-up (See ref. 16 for additional details). Desalting column is fused silica capillary column (250-μm i.d., 365-μm o.d.; Agilent) filled with 4 cm of Aqua C18 RP resin (5-μm beads, 300-Å pores; Phenomenex, Torrance, CA).

  • Microfilter assembly Analytical microcapillary column (100-μm i.d., with laser-pulled 5-mm orifice) packed with 3 cm of Luna SCX resin (5-μm beads, 100-Å pores; Phenomenex) above 9 cm Aqua C18 RP resin (3-μm beads, 300-Å pores; Phenomenex).

  • Software Xcalibur (Thermo-Finnigan), PARC, SEQUEST (Thermo-Finnigan), Excel (Microsoft), DTASelect v1.9. Note: Although we have primarily relied on DTASelect v1.9 for our analyses, the newer version 2.0 utilizes an improved method for evaluating data quality that compares the number of database matches to the number of matches obtained from a decoy database in which protein sequences are reversed. By monitoring the frequency of matches to the decoy database, correlation coefficients and ΔCn cut-off value are automatically adjusted ad hoc to generate a dataset with the desired false-positive rate (D. Cociorva and J. Yates, personal communication).

Procedure

Preparation of starting protein extracts

  1. 1

    Tissue homogenization and cell lysis. Conditions are described for preparation of lysates from either yeast cultures (option A) or freshly resected mouse brain (option B). For other source materials, homogenization conditions may need to be altered.

    1. A

      Yeast lysates

      1. i

        Collect by centrifugation (4,000g, 5 min, 4 °C in Sorvall GSA rotor) a 1-l late log-phase culture (A600 = 1.0; 1.5 × 107 ml−1) of YPD-grown yeast.

      2. ii

        Re-suspend cell pellets in 40 ml ice-cold LB with 10 mM NEM, 2 × PI, 2 mM PMSF. Transfer to a 50-ml disposable centrifuge tube. Re-pellet cells (5,000g, 5 min, 4 °C).

      3. iii

        Aspirate buffer and freeze cell pellet on dry ice.

        Pause point

        Frozen cell pellet may be stored for several days at −80 °C.

      4. iv

        Cell lysis. Break open cell pellet–containing centrifuge tubes with one sharp hammer blow. Rapidly transfer the frozen cell pellet to the Krups coffee grinder, together with chunks of dry ice estimated to correspond to one-third of the cell pellet volume. Grind for a total of approximately 30 s (as grinding reaches completion, tone of grinder transitions to higher-pitch whine; continue for an additional 10 s). Scoop cell lysate (viscous, partially frozen slurry) into cold 50-ml centrifuge tube. Incubate, unlidded, at 0 °C, to allow residual dry ice to sublimate.

      5. v

        Add 10 ml of cold LB with 10 mM NEM, 1 × PI, 1 mM PMSF.

    2. B

      Mouse brain homogenate

      1. i

        Flash-freeze freshly resected mouse brain in liquid nitrogen.

        Pause point

        Brain may be stored for several days at −80 °C.

      2. ii

        Transfer frozen brain to ice-cold 12-ml centrifuge tube. Add 1 ml of ice-cold LB with 10 mM NEM, 2 × PI, and 2 mM PMSF. Homogenize for 2 min at 24,000 r.p.m. using an IKA tissue homogenizer with S25N-8G blade attachment.

      3. iii

        To further homogenize, sonicate, using micro-probe, with ten duty cycles of 1 s ON and 2 s OFF.

  2. 2

    (Optional) Collect total membranes by high-speed centrifugation (200,000g, 30 min, 4 °C) in fixed-angle rotor. This membrane purification is intended to enrich for PPs.

  3. 3

    (Optional) Re-suspend membrane pellet in 3 ml LB buffer with 10 mM NEM, 1 × PI and 1 mM PMSF.

  4. 4

    Detergent solubilization. Add Triton X-100 to 1.7%. Incubate with end-over-end rotation at 4 °C for 1 h.

  5. 5

    Remove particulates and unbroken cells with low-speed centrifugation (250g, 4 °C, 5 min).

  6. 6

    Chloroform–methanol (CM) precipitate sample (see Box 1).

  7. 7

    To each protein pellet, add 300 μl 4SB with 10 mM NEM. Incubate for 10 min at 37 °C with occasional agitation of tube to dissolve pellet.

    Critical Step

    The protein denaturation accompanying this step facilitates access of NEM to Cys that are buried within the folded protein interior.

    Troubleshooting

  8. 8

    To each tube, add 900 μl of LB with 1 mM NEM, 1 × PI, 1 mM PMSF, and 0.2% Triton X-100. Transfer to 1.5-ml screw-cap centrifuge tubes. Incubate overnight at 4 °C with gentle rocking.

    Pause point

    Overnight incubation with NEM.

Acyl-biotin exchange reactions

  1. 9

    Remove NEM from samples by three sequential CM precipitations. Transfer samples to 12-ml screw-cap tubes and CM precipitate as described in Box 1. After the first two CM precipitations, add 300 μl 4SB and dissolve protein pellet by incubating at 37 °C for 10 min with occasional vortex mixing. Then, dilute with 900 μl LB containing 0.2% Triton X-100.

    Critical Step

    Residual NEM can greatly reduce biotinylation, irreversibly modifying thiols as they become exposed from palmitoylated Cys by hydroxylamine and thus blocking subsequent reaction with thiol-specific biotinylation reagent.

    Pause point

    After first or second CM precipitation, sample may be stored in 4S buffer overnight at −20 °C. The next day, thaw at room temperature (RT; 23–25 °C), and dilute with LB.

  2. 10

    After the third and final CM precipitation, dissolve protein pellet in 250 μl 4SB, 37 °C, 10 min. At this point the sample is divided into two equal portions (+HA sample and −HA sample). To ensure that the two samples are equalized, the entire protein sample is pooled into a single tube and then distributed at 240 μl per 1.5-ml screw-cap centrifuge tube. +HA samples are diluted fivefold with the addition of 960 μl of the hydroxylamine-containing +HA buffer; for −HA samples, 960 μl of −HA buffer is used. Incubate at RT for 1 h with end-over-end rotation.

  3. 11

    Transfer samples to 12-ml screw-cap tubes and CM precipitate (Box 1).

  4. 12

    Dissolve each resulting protein pellet in 240 μl 4SB. Dilute with addition of 960 μl low-HPDP–biotin buffer. Incubate at RT for 1 h with end-over-end rotation.

  5. 13

    To remove unreacted HPDP–biotin before the streptavidin–agarose affinity purification, subject samples to three sequential CM precipitations (Box 1). After the first and second precipitations, dissolve and dilute precipitated proteins as described for Step 10. After the third CM precipitation, dissolve each pellet in 120 μl 2SB at 37 °C, 10 min.

    Caution

    Unremoved HPDP–biotin competes with biotinylated protein for streptavidin–agarose binding.

    Pause point

    After the first or second CM precipitation, sample is typically stored in 4SB overnight at −20 °C. The next day, thaw at RT, before LB addition.

Affinity purification of biotinylated proteins

  1. 14

    Dilute SDS to 0.1% before adding samples to streptavidin–agarose. For this, first pool like samples (e.g., +HA samples with +HA samples), then dilute 20-fold with addition of LB containing 0.2% Triton X-100, 1 × PI and 1 mM PMSF. Aliquot 1 ml per 1.5-ml screw-cap centrifuge tube and incubate at RT for 30 min with end-over-end rotation.

  2. 15

    Centrifuge 15,000g (13,000 r.p.m. in the micro-centrifuge) for 1 min to remove particulates, transferring supernatant to new tubes containing 15 μl streptavidin–agarose, pre-equilibrated with LB containing 0.1% SDS and 0.2% Triton X-100. Incubate at RT for 90 min with end-over-end rotation.

    Critical Step

    Particulates removed by this pre-spin otherwise would pellet along with the streptavidin–agarose through subsequent wash steps and would nonspecifically contaminate the final purified sample.

  3. 16

    Remove unbound proteins by four sequential 1-ml washes with LB containing 0.1% SDS and 0.2% Triton X-100.

  4. 17

    Release bound proteins from the affinity resin through reduction of the protein–biotin disulfide linkages. For this elution, resuspend the resin in each tube in 150 μl LB containing 0.1% SDS, 0.2% Triton X-100 and 1% β-mercaptoethanol. Incubate for 15 min at 37 °C with occasional gentle mixing to resuspend settling resin.

  5. 18

    Pool like eluants together into 1.5-ml screw-cap centrifuge tubes, then concentrate by trichloroacetic acid (TCA) precipitation: add TCA to 10% (add one-tenth volume of 100% TCA), incubate 20 min on ice then collect precipitates by 15,000g, 10 min, 4 °C centrifugation. Dissolve the pellet in 30 μl 2SB. Then dilute to 150 μl with LB.

    Pause point

    At this point, samples may be stored at −80 °C for up to 1 week.

  6. 19

    Compare 1% of the final +HA and −HA samples by SDS–polyacrylamide gel electrophoresis (PAGE) and silver-staining (e.g., Fig. 2).

    Troubleshooting

  7. 20

    Immediately before sending to MS collaborators, clean up the samples further using two sequential CM precipitations to remove residual detergents. For this, the CM precipitation (described above) is scaled down: with 150-μl samples in 1.5-ml screw-cap centrifuge tubes, added methanol, chloroform and water volumes are scaled down proportionately.

  8. 21

    After the first CM precipitation, dissolve the pellet in 15 μl 2SB, then dilute to 150 μl with LB. Ship the final 'wet' protein pellet overnight to MS collaborators on dry ice (i.e., to avoid dislodging the pellet from the tube bottom during transport, the sample is not desiccated).

    Pause point

    At this point, samples may be stored at −80 °C for up to 1 month.

Sample proteolysis

  1. 22

    Dissolve protein pellets containing 1–20 μg total protein in 50 μl proteolysis buffer.

  2. 23

    Add TCEP to 5 mM and incubate at RT for 30 min to reduce disulfide bonds.

  3. 24

    Add iodoacetamide to 10 mM and incubate at RT for an additional 30 min to alkylate free thiols.

  4. 25

    Add 0.15 μg endoproteinase Lys-C (approximate substrate to enzyme mass:mass ratio of 100:1) and digest proteins for 4 h at 37 °C.

  5. 26

    Dilute sample fourfold with addition of 100 mM Tris/Cl pH 8.5; add CaCl2 to 2 mM, then 0.5 μg trypsin. Incubate overnight (12–16 h) at 37 °C.

    Pause point

    Samples may be stored at this point for up to 1 month frozen at −80 °C.

  6. 27

    Just before MS/MS analysis, formic acid is added to 5%, and insoluble particulates are removed by centrifugation (16,000g, 15 min).

MuDPIT chromatography and collection of tandem mass spectra

  1. 28

    Refer to Box 2 for information about MuDPIT. Load peptide samples onto desalting column. Desalt with 5% acetonitrile, 0.1% formic acid.

  2. 29

    With peptides still retained, couple the desalting column via the microfilter assembly in-line to the analytical column, which is dually packed with SCX resin over RP resin.

  3. 30

    Use the six-step chromatographic program (Box 3) to move the peptides from the desalting column onto and through the analytical column with electrospray-mediated elution, and ultimately into the tandem MS. Buffer changes are effected by the HPLC, driven from the tandem MS using Excalibur software.

Data analysis: protein identifications

  1. 31

    MS/MS spectra are analyzed on an Intel Xenon 80-processor cluster running under the Linux operating system. Use PARC17 to analyze charge state and filter data quality of MS/MS spectra.

  2. 32

    Search relevant protein database, supplemented with common contaminants (e.g., keratins) using SEQUEST18.

  3. 33

    Filter SEQUEST results using DTASelect v1.919 with the following parameters:

    Table 1 Table 2
    Full size table

Spectral count–based quantitation

  1. 34

    Extract spectral counts from DTASelect MS/MS datafiles. Open DTASelect datafile (in .txt format) using Excel. Data fields sort into the rows and columns of the worksheet.

  2. 35

    Collect DTASelect summary lines. Heading the data for each identified protein within the DTASelect datafile is a summary line reporting various information, including the protein's unique database ID, its database annotation, its predicted molecular weight and pI, as well as a summary of the MuDPIT data associated with the identification (including the spectral count). DTASelect summary lines are marked at the left with the identifier “U,” which should sort to the leftmost column (column A) of the Excel worksheet. “Sort” worksheet on the basis of column A in “descending” order. Summary lines should now be collected near the top of the worksheet (Fig. 5a). Delete all rows that do not contain the “U” identifier, conserving just the summary line for each identified protein.

    Figure 5: Example spectral count analysis.
    figure 5

    Analysis is shown for an arbitrary set of five proteins of similar mid-abundance range identified from published analysis of yeast acyl-biotinyl exchange (ABE) samples4. (a) DTASelect summary lines. After the DTASelect .html file [from multi-dimensional protein identification technology (MuDPIT) of a yeast plus-hydroxylamine (HA) sample] is pasted in its entirety into an Excel spreadsheet, the “Sort” command is used to collect summary lines that are uniquely identified by the “U” in the leftmost column. All other rows are deleted. (b) Extraction of spectral count data. Delete all columns except for the ones containing database IDs and spectral counts. (c) “Consolidate” spectral count data from MuDPIT runs under comparison. (d) Normalize data. Each spectral counts is divided by the total spectral count for the sample (strategy 1 of Box 4; 17,494 for the +HA run and 11,293 for the −HA run) then multiplied by the arbitrary multiplier 15,000. (e) Average normalized spectral count data from like experiments. The data from experiment 1 are assembled (using the “Consolidate” command) with data, manipulated as above, from experiment 2. Normalized spectral counts from the two experiments are averaged. (f) +HA:−HA ratios identify candidate PPs. Before dividing averaged and normalized +HA spectral counts by the averaged and normalized −HA spectral counts, averaged and normalized −HA spectral counts of zero are arbitrarily replaced by a value of 0.5 (to avoid division by zero). The highlighted proteins encoded by YDR104W and YOR106W (corresponding to Pin2 and Vam3, respectively) are identified as candidate PPs.

    Full size image
  3. 36

    Delete all columns containing extraneous information (Fig. 5b).

  4. 37

    Use the Excel “consolidate” function to combine spectral count datasets from the MuDPIT runs under comparison onto a single spreadsheet (Fig. 5c).

  5. 38

    Normalization of spectral count data (Fig. 5d and Box 4). Unavoidable variation introduced at the level of the ABE purification and/or the running of the MuDPIT 2D chromatography results in spectral count yields that may vary by as much as twofold to threefold for identical proteins in equivalent samples. Despite these differences in overall spectral count yields, the spectral count ranking of individual identified proteins, particularly for the more abundant sample proteins identified by high spectral count numbers, typically is very well reproduced. Two normalization strategies have been employed to accommodate sample-to-sample spectral count yield variation (see Box 4).

  6. 39

    Average like data from iterative MuDPIT analyses. Spectral count numbers are statistics, and thus confidence in conclusions derived from spectral count data is increased through repeat analysis. Our identification of yeast PPs was based on the analysis of four paired +HA and −HA samples4. Generate a new Excel spreadsheet column that averages the normalized spectral counts for equivalent samples—e.g., a column reporting averaged spectral count numbers from +HA samples and a column reporting averaged −HA spectral count numbers (Fig. 5e).

  7. 40

    Compare the abundance level of each protein within the different samples through ratiometric comparisons of averaged and normalized spectral count data (Fig. 5f). Candidate PPs are identified, for instance, by comparing each protein's +HA sample representation with its −HA sample representation (PPs show higher +HA sample abundance).

  8. 41

    Generate a new spreadsheet column that divides normalized and averaged counts for each identified protein from the +HA samples by the normalized, averaged counts from the −HA samples. Before ratio calculations, to avoid division by 0, replace zeros present in the divisor column with some small arbitrary number (choose a number somewhat smaller than the smallest value present in divisor column). Such replacement is easily automated with the Excel “IF” function. Candidate PPs are identified by their high +HA:−HA spectral count ratios.

  9. 42

    Graphically compare spectral count data using either x, y-scatter plots (see Box 5) or colorimetric depictions of the ratios (see Box 6).

Troubleshooting

Troubleshooting advice can be found in Table 1.

Table 1 Troubleshooting table.
Full size table

Timing

From start to finish, approximately 2 weeks are required.

Day 1—Steps 1–8, extract preparation, culiminating in overnight NEM blockade: 4–6 h

Day 2—Steps 9–13, acyl-biotin exchange: 8 h

Day 3—Steps 14–18, streptavidin–agarose affinity purification: 8 h

Day 4—Steps 19–21, SDS-PAGE analysis of purified samples: 8 h

Day 5—Steps 22–27, sample proteolysis: 16 h

Days 6 and 7—Steps 28–30, column preparation, loading, and overnight MuDPIT analysis: 18 h

Days 8–10—Steps 31–33, computer analysis of tandem MS: 48–72 h (depending on computing power)

Day 11—Steps 34–42, spectral count analysis: 4 h

Anticipated results

PPs should be substantially over-represented in +HA versus −HA samples and thus should be highlighted by the ratiometric comparison of +HA and −HA sample spectral counts for each identified protein. For abundant sample proteins (i.e., those identified by high spectral count numbers), the bona fide PPs should be cleanly resolved from the background of co-purifying contaminant proteins. Contaminant proteins tend to be proteins of known high abundance, e.g., cytoskeletal proteins, chaperones, ribosomal proteins and glycolytic enzymes, that generally are purified in hydroxylamine-independent fashion and thus show plus-to-minus spectral count ratios near 1. At lower spectral counts, distinguishing PPs from contaminants becomes a little more problematic; at these low numbers, some non-palmitoylated contaminant proteins show skewed plus-to-minus spectral count ratios owing purely to chance.

Spectral count statistical variation

Spectral counts correlate with, but do not directly measure, abundance10. For proteins identified by low spectral count numbers, substantial run-to-run spectral count variation may be seen in parallel MuDPIT runs of identical samples, a reflection, presumably, of variations in both the peptidyl chromatography and peptide sampling by the tandem MS. Ultimately, spectral count is a statistic, and confidence in a result is reinforced by repeat analyses and also, perhaps, by statistical tests of significance. For our analyses to date, we have tended to derive conclusions mainly from the set of proteins that are identified by high spectral count numbers. For instance, our mapping of PP substrates with DHHC PATs (Fig. 5), which relied on comparing spectral count scores from +HA samples derived from DHHC PAT mutant yeast strains with scores from wild-type yeast +HA samples, focused on just the 30 top-scoring yeast PPs. Spectral count reductions of tenfold or more specific to the DHHC mutant strain samples easily highlighted the links between substrate PPs and DHHC PATs. Confidence in detected palmitoylation changes in the two- to fivefold range may require repeat analyses coupled with statistical tests of significance.

False positives

Two classes of false-positive proteins typically are seen. One class, 'statistical false-positives', result from the spectral count statistical variation discussed above, i.e., proteins that show high plus-to-minus spectral count ratios simply owing to chance under-detection from the −HA samples. Such false positives become increasingly dominant as one proceeds further down the list to proteins identified by low spectral count numbers. Repetition of the MuDPIT analysis reduces this statistical noise. A second class of false positives comprises proteins that are not palmitoylated, but that are strongly and specifically detected by ABE none the less. As ABE detects PPs through detection of the palmitoyl–cysteinyl thioester linkage, it is not surprising that some proteins that utilize thioesters for chemistries other than palmitoylation also are purified by ABE. Two examples of this class, prominently detected by our yeast analysis, are Pdx1 and Lat1, which are both subunits of the mitochondrial pyruvate dehydrogenase complex. In the decarboxylation of pyruvate, these two subunits transiently accept acetyl moieties in thioester linkage to lipoic acid prosthetic groups. Although strongly detected by ABE, Pdx1 is not palmitoylated and it is not labeled in experiments that assess palmitoylation through metabolic [3H]palmitic acid incorporation4. Other thioester-utilizing false positives detected in our yeast proteomic analysis include Gcv3, which uses the lipoic acid prosthetic group for glycine decarboxylation, the E2 ubiquitin conjugase Ubc1, which transiently accepts ubiquitin moieties in thioester linkage, and the acyl-carrier protein Acp1, which carries growing fatty acyl chains in thioester linkage to a phosphopantetheinyl prosthetic group. An orthologous set of thioester-utilizing false positives also are prominently detected in mammalian analyses (R. Kang, J.W., J. Yates, N.G.D. and A. El-Husseini, unpublished results). These thioester-utilizing false positives are always seen; indeed, their detection provides some measure of the efficacy of the ABE reactions and purifications. Furthermore, as their ABE labeling is palmitoylation independent, these proteins should remain unchanged in analyses aimed at the global palmitoylation effects of perturbants, e.g., mutation or drugs, and may prove useful as a standard for normalizing sample-to-sample spectral counts (Box 4). Particularly useful in this regard are two strongly detected lipoic acid–utilizing subunits of the mitochondrial pyruvate dehydrogenase complex, namely, the dihydrolipoamide S-acetytransferase and the lipoyl-containing component X (in yeast, Lat 1 and Pdx1, respectively).

Candidate testing

To confirm palmitoylation and to eliminate false positives, the newly identified PP candidates should be independently tested for palmitoylation. Statistical false positives can be eliminated in small-scale experiments that use ABE chemistry to detect palmitoylation. Such a scaled-down ABE protocol involves a scaled-down ABE processing of extracts and immunoprecipitation of an epitope-tagged version of the protein under consideration, with a final western blot detection using anti-biotin antibodies4. However, the alternative approach, the direct testing of the candidate protein incorporation of a label from [3H]palmitic acid via classic metabolic labeling4, may be preferred: it has the added advantage of eliminating the thioester-utilizing class of false positives.