Differential Phosphoprotein Labeling (DIPPL), a Method for Comparing Live Cell Phosphoproteomes Using Simultaneous Analysis of 33P- and 32P-Labeled Proteins*

We developed a differential method to reveal kinase-specific phosphorylation events in live cells. In this method, cells in which the specified kinase is inactive are labeled with 32Pi, whereas cells in which the kinase is active are labeled with 33Pi. The two cell extracts are then mixed, and proteins are separated on a single two-dimensional gel. The dried gel is exposed twice. The first exposure reveals both 32P- and 33P-labeled proteins; the kinase-specific spots are revealed because of 33P labeling. The second exposure is conducted with two acetate sheets intervening between the gel and the detection plate. This maneuver screens out the less energetic 33P-labeled proteins while allowing the more energetic 32P-labeled proteins to be detected, thus leaving only those spots that were phosphorylated independently of the specified kinase. We demonstrate the utility of this method for detecting kinase substrates in rare tissue by focusing on extracellular signal-regulated kinase-specific phosphorylation of stathmin/OP18 in primary rat sympathetic neurons.

Phosphorylation is a major reversible protein modification. It regulates a myriad of protein functions including enzyme activity, protein-protein interaction, cellular localization, and protein degradation. It is estimated that there are about 100,000 potential phosphorylation sites in the human proteome of which fewer than 2000 are currently known (1). In addition to identification of phosphorylation sites, there is a need for quantitation of phosphorylation events especially with regard to understanding the regulation of signal transduction. Although novel and sophisticated methods have been developed to enrich for phosphoproteins or phosphopeptides prior to analysis by mass spectrometry, this approach requires large amounts of protein and so is not always feasible when using rare tissue. Moreover from the standpoint of systems biology, it may be useful to acquire an image of protein phosphorylation of the entire proteome prior to homing in on specific proteins.
Labeling cells with phosphorus-32 ( 32 P) has long been used as a means for identifying phosphoproteins. Principle ␤ emission energy for 32 P is 1.709 MeV, making it highly sensitive, and its rapid uptake into cellular ATP makes it very versatile. Although it is possible to run simultaneous gels to compare changes in phosphoprotein profiles between differentially treated samples, it would be an advantage to be able to discriminate changes in phosphorylation between two samples on a single 2D 1 gel just as DIGE is used to highlight changes in protein expression between two samples while eliminating the variability of protein separation patterns (2). We previously noted that it is possible to prelabel cellular proteins metabolically with [ 35 S]methionine (principle ␤ emission, 0.167 MeV) and then label the same cells with 32 P to detect which of these proteins are phosphorylated; as 35 S emission has lower energy it is possible to screen out the lower energy using a simple device such as an acetate sheet while still permitting 32 P radiation to be detected (3). 2 The maximum ␤ emission energy for phosphorus-33 ( 33 P) is 0.249 MeV, which is about 6.8 times lower than that of 32 P. Hence we reasoned that it might be possible to use a similar configuration by mixing differentially treated samples, one labeled with 33 P and the other labeled with 33 P, and running them on a single 2D gel. This would maximize yield (especially when dealing with small samples such as rare tissue, e.g. primary neurons) while eliminating ambiguity in spot detection due to differences between the patterns of two 2D gels. Here we demonstrate the utility of this method by focusing on phosphostathmin as our test protein and primary rat superior cervical ganglion (SCG) neurons as our cell type. We chose this system because it is a good example of a rare tissue type; each ganglion yields only about 10,000 highly purified neurons.
Stathmin is a 19-kDa protein that integrates via its phosphorylation several different signaling pathways (4 -8). At least 16 molecular forms of stathmin have been identified that migrate as two unphosphorylated forms (␣ and ␤; the ␣ form is usually more abundant) and seven increasingly phosphorylated spots each with molecular mass of 19 -23 kDa and pI of 6.2 to Ͼ5.6. Strikingly there are typically three sizes or "tiers" of migration of these phosphoproteins on SDS-PAGE that have been extensively identified at the molecular level. Tier 1, the fastest migrating tier, contains three phosphorylated "spots," ␣1, ␣2, and ␣3, that migrate at about 19 kDa with pI values of Ͼ6, 5.8, and 5.6, respectively, due to addition of one, two, or three phosphates (labeled P1, P2, and P3). Tier 2, "set 16," is comprised of ␣ 1 1 and ␣ 2 1 that migrate at about 21 kDa with pI values of about 5.8 (P2) and 5.6 (P3). Tier 3, "set 17," is comprised of ␣ 1 2 and ␣ 2 2 that migrate at about 23 kDa with pI values of 5.6 (P3) and Ͼ5.5 (P4). The ␤ form is modified similarly except that it is slightly more acidic. In vitro, PKA and Cdc2 phosphorylate stathmin on two mutually exclusive serines each, giving rise to the complete pattern described above (4). In PC12 cells, phosphorylation of set 16 and set 17 is induced by NGF among which serine 25 (the site phosphorylated by Cdc2 in vitro) is also heavily induced by mitogen-activated protein kinase/ERK (9). Serine 38 is mildly phosphorylated as well. We have shown previously that NGF maintains prolonged ERK activity in SCG neurons (10,11) without activating PKA (12). Moreover SCG neurons do not express Cdc2 (13). Knowing the pattern of migration of stathmin and using the cyclic AMP analogue 8-(4-chlorophenylthio)-cAMP to activate PKA in SCG neurons (12), we tested whether we can use dual labeling with 33 P and 32 P and the MEK inhibitor U0126 to distinguish the sites in stathmin/OP18 specifically phosphorylated by NGF-stimulated ERK. We showed that the differential phosphoprotein labeling (DIPPL) procedure is indeed an excellent means of identifying kinasespecific phosphorylation in small numbers of cells.

EXPERIMENTAL PROCEDURES
Materials-32 P (catalog number PBS 13) and 33 P (catalog number BF 1003) were from Amersham Biosciences. 32 P was an aqueous (acid-free) product, but 33 P was in a dilute HCl solution (Ͻ0.1 M, pH 2-3) and had to be adjusted to pH 7.5 with 100 mM NaOH before use. U0126 was from Promega UK (Southampton, UK), cytosine arabinoside (araC) and 8-(4-chlorophenylthio)-cAMP (CPTcAMP) were from Sigma, and phosphate-free RPM I1640 medium was from ICN Biomedicals (now MP Biomedicals UK, London, UK).
Preparation and Culture of Neurons-Single cell suspensions of rat SCG neurons were prepared from 1-day-old Wistar rat pups as described previously (10,14). Neurons were purified to 97-99% by preplating for 30 min twice on collagen in L15-CO 2 medium containing 5% fetal bovine serum. The non-adhering cells were collected by centrifugation and cultured on a poly-L-lysine-and laminin-coated substrate.
Radiolabeling, Inhibition of MEK, and NGF Stimulation-For 2D gel work, SCG neurons (about 200,000/dish) were plated for 1 h in phosphate-free RPMI 1640 medium containing 3% dialyzed rat serum (to remove any inorganic phosphate) and 0.5 mM CPTcAMP (to enable neuronal attachment to the substrate in the absence of any MEK/ERK stimulation (10,12,15)). Neurons were then labeled in the same medium for 3 h with 32 P (ϳ200 Ci/dish) or 33 P (ϳ500 Ci/dish) in the presence of 1 mM araC after which the MEK inhibitor U0126 (10 M final concentration) or an equivalent amount of DMSO (0.5%) (control) was added. After a 30-min incubation with the MEK inhibitor, 100 ng/ml NGF (added from a 200 g/ml concentrate to retain steady state labeling) was added to both experimental and control dishes for an additional 3 h. For exploratory work, SCG neurons were labeled with 0.5 mCi of 32 P or 33 P for 3.5 h but lysed immediately in one-dimensional Laemmli sample buffer (16).
Sample Preparation-Neurons were carefully washed three to four times in L15-CO 2 medium (containing phosphate) without serum, scraped off in 0.5 ml of medium containing 0.01% BSA, and left on ice for 5 min after which neurons were pelleted by spinning for 3 min at 4000 rpm in a microcentrifuge. The pellet containing the neurons was washed with ice-cold PBS or medium without any additions and lysed in 320 l of IPG buffer containing 7 M urea, 2 M thiourea, 4% CHAPS (or alternatively 2% ASB-14), 1.2% Pharmalytes pH 3-10, 20 mM DTT, 10 mM Tris-HCl, pH 8, and bromphenol blue. The pellet was vortexed rigorously a few times for 30 -60 s each time until complete solubilization was achieved and centrifuged at maximum speed in a microcentrifuge to remove any particulate material. Approximately 1% of the volume was subjected to TCA precipitation and scintillation counting to estimate the total amount of 33 P and 32 P incorporation and thus adjust the respective total label content between experimental and control samples before mixing both samples (see text for additional comments). TCA precipitation was conducted in 25% icecold TCA (1 ml) followed by decantation onto a Whatman GF/C glass fiber filter and four washes with ice-cold 5% TCA as described previously (3).
Gel Electrophoresis-For the first dimension of the 2D separation, each sample was absorbed into an inverted immobilized pH gradient gel (Immobiline DryStrip, pH 3-10 nonlinear, 18 cm, Amersham Biosciences) during an overnight incubation at room temperature in a reswelling tray to allow uptake of the proteins. Isoelectric focusing was performed with a mineral oil overlay in a Multiphor II flat bed electrophoresis unit (Amersham Biosciences) set at 2 mA, 5 watts and ramped to 100 V for 1 h, 300 V for 1 h, 500 V for 1 h, 3500 V for 3.5 h, and finally 3500 V for 12 h. The strips were then equilibrated for 10 -15 min with gentle shaking in a buffer containing 50 mM Tris-HCl, pH 6.8, 2% SDS, 6 M urea, 30% glycerol, and 20 mM DTT and placed on top of the second dimension gel. Proteins were separated by SDS-PAGE on a 20 ϫ 20-cm gel containing 11% acrylamide using an in-house built apparatus (3). Protein standards (2D Bio-Rad markers) were run along with the samples to ensure equivalent patterns of protein separation between gels. For exploratory work, proteins were separated by SDS-PAGE on an 8 ϫ 10-cm minigel, stained, dried, and imaged as above. 14 C-Rainbow markers (catalog number CFA756) were from Amersham Biosciences.
Image Analysis-Gels were stained (and fixed) in Coomassie Blue solution, destained, dried, and exposed to a phosphorimaging screen (Eastman Kodak Co.). Screens were scanned at 88-m resolution using an Amersham Biosciences PhosphorImager 425 as described previously (3). Data was stored as a 16-bit tiff. In some cases, acetate sheets (PPCI-SC/LE OHP photocopier film, Lloyd Paton Ltd., Manchester, UK) were placed between the screen and the dried gel as described below. The Coomassie-stained gel image was obtained by scanning with a Hewlett Packard Scanjet 5470C flat bed scanner. Raw images were imported into NIH Image 1.62 to quantify intensity of bands/spots.

RESULTS AND DISCUSSION
We first examined whether it is feasible to use acetate sheets to screen out 33 P signals without affecting 32 P detection. Extracts from 33 P-or 32 P-labeled neurons (each labeled with 0.5 mCi of respective radionuclide) were separated either on two separate lanes of a one-dimensional gel (20 l of each extract) or mixed and separated as a single sample (e.g. 40 l of the mixture loaded). Fig. 1A, left, shows results obtained after exposure of the dried gel to a phosphorimaging screen for 20 h, while on the right is the same gel re-exposed for 30 h with two acetate sheets interposed between the gel and the screen. The intensity of the bands marked i, ii, and iii is given  14 C-Labeled molecular weight markers were run in lane 1. Gels were dried and exposed to a phosphorimaging screen once without (left, 20 h) and then with (right, 30 h) two acetate sheets intervening between the gel and the screen. Note the loss of 33 P labeling in lanes 1 and 3 in the acetate-blocked exposure but nearly full retention of 32 P labeling. Intensities are quantified in Table I. i, ii, iii denote bands whose intensity is quantified in Table I. B, decreasing amounts but equal volumes of each radioisotope (starting intensity, ϳ5 nCi; bottom line) were spotted onto Whatman polyethyleneimine-impregnated p81 paper. After drying, the paper was imaged for 24 h without (left) or with (right) two acetate sheets placed between the paper and the screen; the difference in intensities at this time was about 1:3 ( 32 P: 33 P). Hence 3 times more 33 P isotope compared with 32 P isotope needs to be used for in vivo labeling. Note again that there is only slight attenuation of 32 P intensity with acetate sheets in place. Intensities are quantified in Table I. dil, dilution.
in Table I. It can be seen that 33 P labeling detected during the first exposure (lane 1) was essentially completely eliminated when the gel was re-exposed using two acetate sheets (lane 3). However, near full retention of 32 P signal was obtained in the single 32 P-labeled sample (lanes 2 and 5) or in the mixed sample (lanes 3 and 6) without or with acetate. Moreover there was no interference between the 33 P and 32 P when imaged together as shown by the reconstitution of the combined values measured in the mixed sample (lane 3) when the value measured in the 33 P-labeled sample (lane 1) was added to that of 32 P captured under two acetate sheets (lane 5). With one acetate screen present, 33 P radiation was still marginally detectable, whereas with three screens, the 32 P bands had become more diffuse. Notably the 14 C-markers were also screened out using the acetate sheets.
As the cells in Fig. 1A were labeled with equivalent amounts of radioactivity, the amount of 33 P detected was relatively low compared with that of 32 P. We therefore performed a quick test to ensure detection of equal amounts of radiation for both isotopes. For this purpose equal volumes of 33 P and 32 P were spotted onto polyethyleneimine-impregnated P81 paper at increasing dilutions. Fig. 1B shows that although theoretically the energy emitted from 33 P is 6.8-fold less than that of 32 P, practically we found that the intensity of 33 P as detected by the phosphorimaging screen equalled that of 32 P when the sample of 32 P was diluted between 3-4-fold relative to that of 33 P. Again detection of 33 P (Fig. 1B, right) was completely eliminated by two acetate sheets, whereas that of 32 P was hardly attenuated (see quantitation in Table I). Of course, the relative ratios of the two isotopes would have to be determined for each experiment as the half-life of 32 P is 14 days, whereas that of 33 P is 25.4 days. Using this "spot detection test," the assay is quick and simple.
We next examined whether this method could be used to detect kinase-specific labeling of proteins in SCG neurons. To demonstrate first that we can detect the various phosphoforms of stathmin in the neurons, SCG neurons were 32 Plabeled in the presence of CPTcAMP and NGF (as well as araC; see below), and proteins were separated by 2D electrophoresis. Fig. 2A shows the overall pattern of 32 P labeling achieved (left) together with the Coomassie Blue-stained image of the gel (right). The various tiers of stathmin and the number of phosphates incorporated (P1-P4) are indicated in the rectangle using the notation described by Beretta et al. (4). The most prevalent forms on the Coomassie image are ␣0 (or N1, the nonphosphorylated form of stathmin) and ␣1, the first tier singly phosphorylated form of the protein. On the 32 Plabeled gel, the N1 form is naturally absent, whereas all the other forms reported previously after in vitro phosphorylation with PKA and Cdc2 are present, namely the first tier proteins (␣1, ␣2, and ␣3) aligning with P1, P2, and P3; the second tier spots in set 17 (␣ 1 1 and ␣ 2 1 ) aligning with P2 and P3; and third tier spots in set 16 (␣ 1 2 and ␣ 2 2 ) aligning with P3 and P4, the most acidic phosphorylated spot. Minor ␤ forms were also sometimes noted, but these did not appear in all gels. These spots are annotated in greater detail in Fig. 2, C and D, as explained below.
The underlying purpose of our study was to detect ERK targets that protect against araC-induced apoptosis. NGF maintains a sustained activation of ERK (10), which partially antagonizes the proapoptotic signal induced by araC (a p53dependent type of apoptosis (17)). Protection by ERK is thus eliminated when ERK phosphorylation and its activity are suppressed by the MEK inhibitors PD98059 (75 M) or U0126 (10 M) (11,(17)(18)(19). Stathmin in this context was used to demonstrate that ERK inhibition by the drug had occurred. It is important to note that to detect ERK-dependent labeled phosphoproteins using simultaneous analysis of 33 P and 32 P,  Fig. 1 Raw phosphorimages were imported into NIH Image, and the intensity of the three bands indicated in Fig. 1A (i, ii, and iii) was quantified on a scale of 0 -255. Note the lack of effect of the acetate sheets on the intensity of the 32 P-labeled proteins (compare lane 5 with 2), the loss of 33 P values in the presence of acetate sheets (compare lane 4 with 1 and lane 6 with 3), and the linearity of phosphorimaging, indicated as the restoration of the values measured for lane 3 by adding the values measured in lanes 1 and 6 minus the values in lane 4 (column 7). A similar effect was found when the intensity of the spots shown in Fig. 1B  the inhibitor must be used in conjunction with 32 P as the image of the 33 P-labeled proteins that were phosphorylated by ERK will be eliminated by the acetate sheets, leaving to appear only those ( 32 P-labeled) spots that were phosphorylated independently of ERK. Accordingly one set of SCG neurons was labeled with 33 P (about 0.5 mCi) in the absence of U0126 to label the entire phosphoproteome cohort, whereas the other set was labeled with 32 P (about 0.17 mCi) in the presence of U0126. In this case, a 3-fold higher amount of 33 P radionuclide was added compared with 32 P to equalize loading. Equalizing the signal of the two isotopes by varying the amount of radioactivity of 32 P and 33 P used to label the cells FIG. 2. Evidence for efficacy of the DIPPL method. A, low power overview of a typical 2D gel on which a 32 P-labeled sample of SCG neurons has been separated; on the left is the image collected from the phosphorimaging screen, and on the right is the Coomassie Blue-stained scanned digital image. The rectangle encloses the region where the various multiply phosphorylated stathmin forms are located, the labeling indicating the three tiers identified in the text. B, stathmin phosphorylation. Serines 16 and 63 are predicted to be phosphorylated due to CPTcAMP activation of PKA, whereas serine 25 is predicted to be the major residue phosphorylated by ERK. Serine 38 is also phosphorylated by ERK, but it is not clear whether this is mediated by ERK in the neurons. C and D, schematics of the phosphostathmin forms traced from the adjacent image identified according to Beretta et al. (4). C shows the ␣ forms with the arrows showing the conversion pattern expected in the presence of U0126. D traces the pattern of the minor ␤ forms observed with the respective expected conversion pattern. E and F, comparison of the stathmin pattern obtained when two samples are independently labeled with 32 P and proteins are separated on two independent 2D gels. E, neurons were labeled in the presence of CPTcAMP and araC for 3 h, and then DMSO (U0126 solvent) was added for 0.5 h after which NGF was added for an additional 3 h. F, neurons were labeled in the presence of CPTcAMP and araC for 3 h, and then U0126 was added for 0.5 h after which NGF was added for 3 h. G and H, comparison of the stathmin pattern obtained when two samples are independently labeled as in E and F except that 33 P was used to label cells in the absence of U0126 and 32 P was used to label cells in the presence of U0126. Samples were mixed and run on a single 2D gel. G, no acetate sheets (both radioisotopes imaged). H, re-exposure with two intervening acetate sheets. Note the similarity between E and G on the one hand and F and H on the other hand. Intensities are quantified in Table II. obviates the need to vary the amount of protein loaded per sample, which would evidently distort the results. This is especially important when low numbers of cells are involved as there is very little to spare. Initially TCA precipitation of a small sample from each labeling was used to confirmed that equal amounts of radioactivity were incorporated per sample. However, the spot test could be conducted once, and the amount of each radionuclide was adjusted thereafter according to their half-lives. Samples were mixed and run on a single 2D gel. We also analyzed two independent samples, both labeled with 32 P, under the same regime to validate our approach.
The schematic in Fig. 2B shows the four possible phosphorylation sites in stathmin, two expected to be induced by CPTcAMP and two expected to be induced by NGF via ERK activation. The spots whose intensities are predicted to alter most are those labeled at serine 25 (9). The schematic in Fig.  2C (tracing the pattern in E and G of the ␣ forms of stathmin) indicates the seven major spots detected, while the arrows indicate the conversion pattern expected if serine 25 phosphorylation is attenuated in the presence of U0126; generally, tier 2 and 3 spots should disappear (the same pattern is predicted irrespective of whether serine 38 phosphorylation is eliminated or not (4)). Fig. 2D indicates the location of some of the ␤ form spots detected and their expected conversion pattern; indeed, as expected, these were minor spots.
As predicted, in both the single 32 P-labeled sample treated with U0126 (Fig. 2F) and the mixed 33 P/ 32 P sample exposed with two acetate sheets interposed (Fig. 2H), phosphorylation of the two major tier 2 and 3 spots (sets 16 and 17) was eliminated by the presence of U0126 (see Table II for quantification). The intensity of the two slowly migrating minor ␤ spots was also reduced, consistent with these being similarly regulated (4). Thus, comparing across the two types of labeling, it is clear that the acetate sheets successfully masked the 33 P-labeled proteins (Fig. 2, compare H and G, and Table II). The 32 P-labeled protein pattern is slightly more diffuse with the acetate sheets present than the images collected from the equivalently treated 32 P labeled sample because the gel was imaged sequentially, and there was some decay of 32 P during the time required to collect the 33 P image. Increasing and adjusting the amount of 33 P and 32 P the cells incorporate, thereby shortening the time required for exposure, will com-pensate for this as would imaging first with the acetate sheets present.
Altogether we propose that our method (which we have named DIPPL) is an excellent way to differentiate between, and sensitively compare, two rare samples of phosphoproteins while reducing the time and tedium of having to run two 2D gels for each two-way comparison. Moreover with regard to the problem of identification of ERK substrates that inhibit p53-induced apoptosis, we newly demonstrated that phosphorus-labeled stathmin can be used to monitor and verify that ERK-dependent phosphorylation, and its successful inhibition by U0126, occurred in SCG neurons.  Fig. 2 Raw phosphorimages of the regions described in Fig. 2, E-H, were imported into NIH Image, and the intensity of the seven ␣ phosphoforms of stathmin was quantified. The ratio of the values collected in the presence or absence of U0126 for each spot was calculated and normalized to spot ␣1. Note the similarity between the values using the two different protocols and the marked reduction in spot intensity for spots ␣11, ␣12, ␣21, and ␣22.