Reconstructing the Regulatory Kinase Pathways of Myogenesis from Phosphopeptide Data

EXPERIMENTAL PROCEDURES

Cells and Sample Preparation—

C2C12 myoblasts were cultured in growth medium (10% fetal bovine serum, Dulbecco’s modified Eagle’s medium), and whole cell lysates were collected either 0 or 24 h after a change to differentiation medium (2% horse serum, Dulbecco’s modified Eagle’s medium) as described previously (13).

Phosphoprotein Enrichment—

Phosphoprotein enrichment was performed by affinity column purification of total myoblast proteins using PhosphoProtein Purification kits (Qiagen) as described previously (13).

Analysis of Putative Phosphoproteins by One-dimensional Gel Electrophoresis—

Phosphoprotein-enriched samples (typically 100–150 μg) were washed in water and concentrated using Amicon 10,000 molecular weight cutoff ultrafiltration columns (Millipore). Aliquots of retentate were separated by SDS-PAGE (12.5% acrylamide “mini” format). Gels were stained with Bio-Safe colloidal G-250 Coomassie Blue (Bio-Rad). 23 1-mm gel sections were excised from each lane, and in-gel digest and extraction of proteins was performed.

In-gel Digest—

Gel pieces cut into 1-mm cubes were destained in 100 mm ammonium bicarbonate (NH₄HCO₃) in 30% ACN. Destained gel pieces were washed in dH₂O and shrunk in ACN. Gel pieces were incubated for 1 h at 56 °C in 10 mm DTT, 50 mm NH₄HCO₃; washed in 50 mm NH₄HCO₃; and incubated for 1 h at room temperature in 55 mm iodoacetamide, 50 mm NH₄HCO₃. Gel pieces were shrunk with ACN and reswollen in 50 mm NH₄HCO₃ with Promega modified trypsin (10 ng/μl). Sufficient 50 mm NH₄HCO₃ was added to cover the gel pieces, and sealed tubes were incubated overnight at 37 °C.

Preparation of Samples for IMAC—

Approximately 15 μg of phosphoprotein-enriched protein extract was suspended in 50 mm NH₄HCO₃ containing Promega modified trypsin (10 ng/μl). After overnight incubation at 37 °C digest solutions were concentrated to ∼40 μl and desalted through 20-μl GELoader tips (Eppendorf) packed with 20 μl of OligoR3 resin (Applied Biosystems) as described previously (16). Peptides were washed once by 20 μl of 1% acetic acid and eluted by 20 μl of 50% ACN, 1% acetic acid. Elution solvent was removed by SpeedVac and replaced by 40 μl of the appropriate IMAC binding buffer (see below).

Preparation of Metal-chelating Resin for IMAC—

A 50% slurry of POROS 20MC resin (Applied Biosystems) or Propac-IMAC resin (Dionex) was prepared in dH₂O. Resin was washed twice in water, stripped by 50 mm EDTA, washed in dH₂O, and then washed in 1% acetic acid. Resin was incubated in 100 mm iron chloride, 1% acetic acid for 15 min with agitation. Activated resin was rinsed in 30% ACN, 1% acetic acid and then in 1% acetic acid.

IMAC Enrichment Using GELoader Tips—

30 μl of prepared IMAC resin was added to the peptide sample in binding buffer (100 mm NaCl in 1% acetic acid) and agitated for 15 min before loading into a GELoader tip. Resin was rinsed once by 20 μl of a solution of 100 mm NaCl in 30% acetonitrile and 1% acetic acid and eluted in two steps by 250 mm ammonium phosphate, pH 9. In the first step, 5 μl was pushed into the resin and left for 5 min to neutralize the remaining acetic acid. In the second step, 15 μl of the elution buffer was pushed through the resin. The 20 μl of phosphopeptide-carrying eluate was evaporated by SpeedVac, and the phosphopeptides were suspended in 20 μl of 5% acetonitrile in 0.1% acetic acid.

IMAC Enrichment Using Pierce Phosphopeptide Isolation Kit—

Gallium-chelated columns were used according to the manufacturer’s recommendation with the following modifications. Using 40 μl of a solution of 5% acetic acid as binding buffer, the tryptic peptides were transferred to a Pierce column, left for 15 min at room temperature, and then centrifuged for 1 min at 3000 × g. The Gallium-chelated disc was rinsed twice with 50 μl of 100 mm NaCl in 1% acetic acid, twice with 30% acetonitrile in 1% acetic acid, and once with dH₂O and then eluted by three additions of 20 μl of 250 mm ammonium phosphate, pH 9. Eluates were pooled and evaporated by SpeedVac, and phosphopeptides were suspended in 20 μl of 5% ACN, 0.1% acetic acid before MS analysis.

Q-TOF Analyses—

A fraction of each tryptic digest (typically 5–10 μl) was analyzed by nano-LC-MS/MS using a Q-TOF Ultima hybrid quadrupole time-of-flight mass spectrometer coupled to a CapLC capillary HPLC system (Waters, Milford, MA). Digests were separated on a 75-μm-inner diameter × 150-mm Inertsil ODS 3, 5-μm nano-HPLC column (Dionex/LC Packings, Sunnyvale, CA) using the following gradient conditions: 5–60% ACN, 0.2% formic acid in 30 min, and 60–90% ACN in 5 min. The mass spectrometer was set to automatically acquire MS/MS spectra on doubly, triply, and quadruply charged ions. The MS survey scan range was 400–1600 m/z; the MS/MS range was 50–2000. After the first LC-MS/MS run, an exclusion file was created listing all previously analyzed ions that appeared to represent non-phosphorylated peptides, and a second fraction of each tryptic digest was injected with the mass spectrometer set to ignore the ions selected during the first run.

Data Processing—

Database searching was carried out using Mascot Daemon™ (Matrix Science) against the National Center for Biotechnology Information (NCBI) genome sequence database. In addition to carbamidomethylcysteines and oxidized methionine, Mascot was asked to search for the potential presence of phosphate groups on serine, threonine, or tyrosine residues by looking in the MS/MS fragmentation spectra for the mass of a HPO₃ group on these residues or for ions resulting from a neutral loss of HPO₃ or H₃PO₄ (79.9663 and 97.9769 Da, respectively). All putative phosphopeptide MS/MS spectra identified by Mascot were manually verified for both the presence of a phosphate group and the peptide sequence. In many cases, the MS/MS spectrum allowed precise identification of the modified residue.

Identification of Phosphopeptides in Existing LC-MS/MS Data—

Previously phosphoprotein-enriched samples were separated by 2D-GE, and 190 proteins were identified by LC-MS/MS (13). Many of these proteins showed evidence of differential phosphorylation between the undifferentiated and differentiating states (13). These data was extensively reanalyzed against the latest available NCBInr (non-redundant) database using Mascot server software version 2.0.04 (Matrix Science) allowing up to two missed cleavages and the variable modifications protein N-terminal acetylation, carbamidomethylation, deamidation (NQ), oxidation (Met), pyro-Glu (N-terminal Gln), and phosphorylation (YST). All putative phosphopeptide MS/MS spectra were manually reviewed.

Reconstructing a Pathway Map from Phosphopeptides—

For each experimentally determined phosphorylation site, any kinase(s) known to phosphorylate the site in vivo or in vitro was identified using the bioinformatics tools KinaSource (A. Knebel, Dundee, Scotland, UK) and PhosphoSite (Cell Signaling Technology, Danvers, MA). In a small number of cases where indicated, the relevant kinase was predicted using ScanSite (17). Then KinaSource was queried for other known substrates of each kinase. The process of mapping substrates to kinases was then repeated for all proteins in the model, including the kinases themselves. Database-derived results were verified and extended by examination of the primary research literature. The identified substrate-kinase interactions were then diagrammed using the bioinformatics tool HubView (18).

RESULTS AND DISCUSSION

53 phosphopeptides were identified from C2C12 cell cultures undergoing growth or differentiation (Table I; see Supplements 1 and 2 for additional details). 17 phosphopeptides were identified after IMAC, 17 after one-dimensional gel electrophoresis (Fig. 1), and 30 after 2D-GE and LC-MS/MS. Only nine phosphopeptides were detected by multiple methods suggesting that each method introduces a bias into the selection of peptides. The apparently greater success of the 2D-GE method may indicate a benefit from greater sample fractionation or may simply reflect the larger number of such experiments performed. All phosphopeptides listed in Table I were matched to known or predicted proteins, and in 43 cases additional non-phosphorylated peptides were identified that supported the identification. Some phosphorylations were detected only in either growing or differentiating cells (Table II). Of the 55 identified phosphopeptides, at least 21 contained phosphorylation sites that have been reported previously.

A major challenge in systems biology is to transform large datasets into biologically relevant interpretations. To address this problem, we chose to use a novel bioinformatics approach. We reasoned that because phosphopeptides are the product of protein kinase activity we could construct a proposed kinase-substrate interaction network based upon our data. To perform this task, 19 experimentally detected phosphorylations for which the responsible kinase was known or strongly predicted were used as the “bootstrap” to create an initial network of kinase-substrate interactions (Fig. 2A; see also Supplement 3). Then other known substrates of those kinases, as well as interactions upstream of each kinase, were determined and incorporated into the model (Fig. 2B). For clarity, the abundant and promiscuous kinases PKC and CK2 were excluded from the model, although PKC was included in cases where phosphorylation was attributed to a specific PKC isoform.

The generally accepted view of myogenic signaling is that AKT, p38, MAPKAPK2, and CDK5 activity promote cell cycle exit and myotube formation, whereas ERK1/2 and GSK3 activity are associated with cell cycling and maintenance of the undifferentiated state. Of 114 kinases listed in KinaSource, only 21 appeared in our extrapolated network, and satisfyingly, this group prominently featured AKT, CDK5, GSK3, and MAPKAPK2 kinases (Fig. 2C). Furthermore some kinase-substrate interactions known to be relevant in myogenic signaling such as CDK5-myocyte enhancer factor 2 and AKT-FOXO1A were inferred by the model. Whereas the list of phosphopeptides alone provided little biological context, the extrapolated kinase-substrate network suggested a number of features of interest, such as the implication of DYRK kinases as promyogenic factors.

DYRK—

DYRK interactions were prominent in the model network. Induction of DYRK family kinases is required for muscle differentiation (19) at least in part via DYRK-mediated phosphorylation and regulation of histone deacetylases (20). Here we found that the putative DYRK substrates CRHSP24 and CRMP4 are also phosphorylated in C2C12 (Table I and Fig. 3). Bioinformatics modeling revealed five more DYRK-substrate interactions (p27kip, p21waf, FOXO1A, glycogen synthase, and STAT3), suggesting additional mechanisms by which DYRK may contribute to myogenesis. Two of the phosphorylation sites ascribed to DYRK2 activity are thought to be priming sites for GSK3. In CRMP4 (Dpysl3, Drp3) Ser-522 appears to represent the priming site for a series of GSK3-mediated phosphorylations at serines 518, 514, and 509 (21). In C2C12 cells both site-specific (Table I) and general phosphorylation (13) of CRMP4 was detected only in differentiating cells, whereas Ser-522 is constitutively phosphorylated in brain. Although CRMP4 has been studied almost exclusively in the context of neurons, these data and the observation that the CRMP4 promoter contains a MyoD/myogenin binding site (22) indicate that CRMP proteins likely play an important role in muscle differentiation. eIF2B (eukaryotic translation initiation factor 2B) is also a substrate of both DYRK and GSK3. DYRK-mediated phosphorylation at Ser-539 is thought to permit subsequent phosphorylation of Ser-535 by GSK3 leading to repression of translation (23). Because muscle hypertrophy is associated with GSK3 inhibition and eIF2 derepression, this suggests that promyogenic functions of DYRK could be blocked by GSK3 activity. These observations are consistent with the model that DYRK activity is increased and GSK3 is repressed during myogenesis. In agreement with this principle, several putative GSK3 substrate proteins that were not DYRK substrates, pyruvate dehydrogenase and CRMP2, showed evidence of phosphorylation only in non-differentiated myoblasts, although specific phosphorylation sites were not determined (13).

CRHSP24—

CaRHSP24/CRHSP24 is phosphoprotein of unknown function. DYRK2 can phosphorylate CRHSP24 at serines 31 and 33 in vitro, and Ser-43 (equivalent to Ser-42 in mouse) is phosphorylated in human embryonic kidney 293 cells (24). We detected phosphorylation at serines 31, 33, and 53 in differentiating C2C12 and constitutive phosphorylation at Ser-42 (Table I). Phosphorylation of CRHSP24 at Ser-53, ascribed to AKT activity, was also detected. CRHSP24 may play an important role in myogenesis. IGF-1 treatment, which stimulates muscle hypertrophy, stimulates phosphorylation of the Ser-53 site in human embryonic kidney 293 cells. CRHSP24 is a calcineurin substrate (25). Calcineurin activity is important in normal muscle development, and augmented calcineurin activation has been shown to limit the pathogenesis of muscular dystrophies (26).

PEA-15—

Our previous observation of PEA-15 phosphorylation in C2C12 (13) was confirmed and extended with evidence of a second phosphorylation site (Table I). Ser-116 phosphorylated PEA-15 was repeatedly observed in growing cells but never in differentiating cells. This observation is consistent with the recent report that the Ser-104/Ser-116 dephosphorylated form of PEA-15 promotes cell cycle arrest via sequestration of ERK (27), and indeed, reduced ERK activity is a prerequisite for myoblast differentiation (28).

PKC-δ—

A potential role for PKC-δ was inferred in the model (Fig. 2C), and the potential PKC-δ substrates 14-3-3, HMGA1 (high mobility group AT-hook 1), and HNRNPK were detected in phosphoprotein enrichment experiments (13). Although site-specific evidence of PKC-δ-mediated phosphorylation is lacking, these observations are interesting in light of a recent report that constitutively active PKC-δ promotes whereas kinase-inactive PKC-δ inhibits IGF-1-mediated proliferation of myoblasts (29). PKC-δ, as well as DYRK1a and ERK1/ERK2, can also phosphorylate STAT3, which is reported to promote myoblast proliferation and inhibit MyoD function and muscle differentiation (30–32). Because the aforementioned substrates contain multiple phosphorylation sites, additional studies will be required to determine the precise role of PKC.

Limitations to Application and Interpretation—

Because the experimental data cover only a very small proportion of the phosphoproteome, it is important to consider the validity of extrapolating a model kinase network under these circumstances. The strategy used relies on three assumptions: that substrates have a limited number of kinases, that kinases have a large number of substrates, and that the overall topology of the kinase-substrate network is scale-free. First, if a substrate has many potential kinases then inference of specific kinase activities is not possible. Second, if a kinase has few substrates then the odds that one of those substrates will be found experimentally is extremely low. The last assumption determines whether the model is likely to encompass major features of the true in vivo network or if it reveals only local details. In a non-scale-free network a small data sample will show only local features, much as mapping a few roads in a city would reveal nothing about the overall layout of the traffic flow. However in a scale-free network, any two points on the network are connected by a very small number of interactions, and the overall picture emerges quickly, just as a sampling of commercial airline flights would quickly reveal the existence of traffic hubs at the major airports. Work on kinase-substrate interactions in yeast supports these three assumptions. An exhaustive evaluation of 1325 phosphoproteins and 87 kinases by Ptacek et al. (15) found that 73% of substrates were associated with only one or two kinases but that most kinases had multiple substrates. Analysis of the kinase-substrate subset of the extensive protein-protein interaction data available for yeast confirmed that this network is scale-free (18).

Another consideration was protein abundance. If myogenesis resulted in altered protein phosphorylation in only a subset of very low abundance proteins against a large backdrop of abundant invariant phosphoproteins then no relevant results would have been obtained. However, we found previously that the subset of the phosphoproteome that is sufficiently abundant for silver stain detection on gels does in fact undergo extensive alteration during early myogenesis (13). The use of data obtained from two-dimensional gels was especially valuable in retaining a biologically relevant focus. Comparative analysis of gels prepared from growing versus differentiating myocytes allowed us to focus the MS analysis on proteins that exhibited evidence of altered post-translational modification during myogenesis. All of these factors contributed to the ability to produce a model network that exhibited relevant features of interest. However, a number of caveats remain. First, because proteome coverage is incomplete, lack of detection of phosphorylation cannot be taken as strong evidence of dephosphorylation. Second, the results are biased toward the detection of relatively abundant peptides. Third, kinase-substrate databases are far from complete, although the amount of protein-protein interaction data available is increasing rapidly. We anticipate that increased coverage should not invalidate or contradict the results presented here. Because kinase networks appear to be scale-free (18), increased data coverage would likely increase the number of substrates more rapidly than the number of kinases, limiting any changes in the overall topology of the network. However, this remains to be experimentally demonstrated. Another challenge remaining to be addressed is the role of regulated phosphatases in the network. At this time inclusion of a comprehensive phosphatase dataset is not possible given the limited experimental observations in the literature combined with the absence of robust screening tools for phosphatase activity or dephosphorylation events.

Conclusion—

Experimental phosphopeptide data and bioinformatics databases of kinase-substrate interactions were used to produce a model signaling network for C2C12 myogenesis. From a relatively small number of phosphopeptides, a complex network was readily constructed that inferred the existence of several known features of myogenic signaling. In addition, places for newly described phosphorylations in signaling cascades were predicted. New protein-protein interaction datasets including kinase-substrate datasets are being produced at an increasing rate, and the utility of this approach is certain to increase as more data become available.