Dissection of the insulin signaling pathway via quantitative phosphoproteomics

SILAC Quantification Reveals 40 Insulin-Induced Effectors in Differentiated Brown Adipocytes.

To characterize specific signaling events downstream of the insulin/IGF-1 receptors, we stimulated differentiated brown adipocytes with insulin and quantitatively analyzed the tyrosine phosphoproteome by MS compared with unstimulated control samples. For accurate quantitation of tyrosine phosphorylated (pY) proteins, stable isotope labeling in cell culture (SILAC) was used (8). In each experiment, one cell population was grown in unlabeled medium as a control and a second population was grown in medium substituted with ¹³C₆¹⁵N₄-Arg (Arg-10) and ¹³C₆¹⁵N₂-Lys (Lys-8). Peptides derived from proteins of the two cell populations then occur in doublets, with either 10-Da or 8-Da mass differences. After complete incorporation of the heavy or light SILAC amino acids in brown preadipocytes, cells were allowed to differentiate into brown adipocytes as described in ref. 9 (Fig. 1) and stimulated with 100 nM insulin for 5 min. Tyrosine phosphorylated proteins were immunoprecipitated by anti-pY antibodies from the lysates of the combined cell populations, separated by 1D gel electrophoresis and measured by MS. Quantitative analysis of the peptides identifying each protein resulted in a fold change (“activation”) between control and insulin-stimulated cells. The relative standard deviation (%SD) of all proteins revealed a quantitation precision better than ±20% [supporting information (SI) Fig. 4]. As in the experiments reported in ref. 6, we chose a minimum of a 1.3-fold change as biologically significant. In four independent experiments, we identified 33 proteins precipitated by anti-pY antibodies that had a 1.3-fold or higher change in the peak intensities upon 5 min of insulin stimulation (Table 1 and SI Tables 2–5).

Using this immunoprecipitation protocol, both the tyrosine phosphorylated proteins and molecules that have bound tightly to them can be detected. Of the 33 proteins identified, in most cases, the protein was known to be tyrosine phosphorylated or to associate with such a protein. These included proteins in the insulin pathway with a wide variety of cellular functions, including the insulin receptor substrates IRS-1, IRS-2, APS, and Shc, which served as positive controls. The identification of other effectors from different branches of the pathway, such as MAP kinases, showed that experimental conditions were also suitable to detect downstream factors. For this and subsequent experiments, we term the up-regulated proteins as activated proteins, defined as proteins with a fold change >1.3 after insulin stimulation. Note that this activation could in principle down-regulate insulin signaling. Proteins with a fold change <0.7 were termed “deactivated.” The quantitative MS data were confirmed by standard Western blot analysis (Fig. 2).

Analysis of Temporal Activation Profiles Revealed Different Categories of Insulin-Induced Effectors.

Signaling events are not regulated by a simple “on” and “off” switch at a single time point, but require a highly coordinated kinetic cascade of effectors and events. To capture the temporal dynamics of tyrosine phosphorylated proteins, we used a triple labeling SILAC experiments in which three differentially labeled cell populations were stimulated for different times (see also SI Fig. 5). The temporal proteome is encoded by the SILAC state of each cell population and the relative intensities of peptides from each state are a direct measure of their presence in the pY immunoprecipitate (6). The combination of two three-state experiments allowed us to generate a time profile for 0, 1, 5, 10, and 20 min (Fig. 3). The kinetic analysis revealed distinctive temporal patterns of insulin signaling. Based on the similarity of patterns, we grouped known and unknown candidate effectors into different branches of the pathway. As expected, one of the first tyrosine phosphorylation events upon insulin stimulation occurred on the IR itself. The IR was activated ≈20-fold compared with unstimulated control within the first minute and remained fully activated for the next 5 min. After 20 min, IR phosphorylation/activation levels still remained at 85% of the maximum, reflecting the saturating insulin concentrations in this experiments (Fig. 3a and SI Tables 6 and 7). Insulin can also bind to the IGF-1 receptor and vice versa, albeit with a 10- to100-fold lower affinity (10). In addition, some IGF-1 and insulin receptors exist as heterodimer hybrids allowing cross-phosphorylation of their intracellular kinase domains (11). Reflecting these mechanisms, the IGF-1R showed an early activation profile, similar to the insulin receptor, but with a fivefold lower activation. Therefore, at high insulin concentrations, downstream effectors may potentially reflect activity of either or both receptors (see below for a dissection of these pathways).

To date, at least 11 substrates of the IR and IGF-1R have been identified in different cellular systems (1). Of these, we detected the activation of five substrates in brown adipocytes. These included IRS-1, IRS-2, Gab-1, Shc, APS, and c-Cbl. For most of these substrates, we also sequenced the tyrosine phosphorylated peptides (Table 1, SI Fig. 6, and SI Table 8). For an accurate assignment of the phosphorylation sites in phosphopeptides, we used the posttranslational modification (PTM) score. The PTM score is based on an algorithm that tests and scores all possible combination of serine, threonine, and tyrosine phosphorylations within a potential phosphopeptide (12). The substrate proteins IRS-1 and Shc showed a similar kinetics of activation as the insulin receptor (Fig. 3 a and b), suggesting physical proximity as has recently been found for Shc upon EGFR activation (13). In contrast, IRS-2 showed its highest activation after 1 min of stimulation, followed by rapid deactivation. This transient activation profile is consistent with the results of earlier experiments using L6 muscle cells (14). The rapid abrogation of the IRS-2 signal suggests fast negative feedback mechanisms, such as serine phosphorylation or activation of phosphotyrosine phosphatases. In support of this notion, we identified several IRS-2 peptides with phosphorylated S/T- sites (SI Fig. 6 and SI Table 8).

Tyrosine phosphorylation of substrate/adaptor proteins creates docking sites for several SH2-containing proteins that are central to insulin signaling (15), most importantly the regulatory subunit (p85) of the phosphatidylinositol 3-kinase (PI3K) (16). We detected the α and β isoforms of both the regulatory p85 and catalytic p110 subunit of PI3K (Fig. 3c) with the apparently most abundant species being p85β. PI3K activation leads to the generation of 3-phosphorylated lipids (PIP₃) and their synthesis is counteracted by the SH2-domain-containing lipid phosphatase SHIP2 (17). The appearance of SHIP2 in the activation profile showed slower kinetics, being half-max at ≈3 min and reaching a fivefold maximum after ≈15 min (Fig. 3e).

The other major branch of the insulin signaling pathway is the activation of the Ras-MAP kinase cascade, which plays a role in control of cell growth and gene expression. Initial docking proteins in this cascade are the SH2-domain-containing protein Shc (18) and Gab-1 (19). After Shc activation (Fig. 3b) a strong activation of Erk-1 and Erk-2 was observed (Fig. 3g). Both MAP-kinases reached a maximum activation after 5 min and decreased at the 10- and 20-min time points. The stress induced MAP-kinase p38 showed only weak activation between 10 and 20 min (Fig. 3g).

An acute effect of insulin treatment is the rapid formation of caveolin-enriched lipid raft subdomains within the plasma membrane (20). Upon stimulation, the adapter protein containing a PH and SH2 domain (APS) (21) interacts with Cbl and CAP and relays the insulin signal to the accruing lipid raft structures. Caveolin-1 is known to become tyrosine phosphorylated and acts as a scaffolding protein to organize signaling molecules and receptor tyrosine kinases, such as the insulin receptor (20). Consistent with this finding, we detected Caveolin-1 (pY 14) and Caveolin-2 in the pY immunoprecipitated fraction (Fig. 3d). Both proteins showed interesting biphasic kinetics, with two maxima at 1 and 10 min and an intermittent decrease at the 5-min time point. This biphasic stimulation was supported by Western blot analysis (Fig. 3i). Polymerase I release transcription factor (PTRF) showed a similar activation profile to the caveolins. PTRF has been implicated in caveolar formation and is also known as Caveolin-60 (22). Two proteins not previously implicated in insulin signaling, serum deprivation response protein (Sdr) (23) and the PKCδ binding protein (also known as sdr-related gene product that binds to C-kinase) (24), had a biphasic activation profile, suggesting they might also associate with insulin dependent formation of lipid rafts.

The family of low density lipoprotein receptor-related proteins (LRP) is also known to be recruited to caveolae structures in response to insulin treatment (25). We identified LRP-1 as part of the activation response (Fig. 3d); however, in contrast to the caveolins, LRP-1 showed a slower increase and only one maximum at the 5-min time point, as did LRP-6, another lipoprotein that had not been implicated in insulin signaling (26). LRP-6 has also been reported to be a part of the Wnt/beta-catenin pathway (27) and could be an interesting candidate for cross-talk of the insulin and Wnt pathways.

Another effect of insulin treatment is the rapid reorganization of the actin cytoskeleton mediated mainly by small G proteins (28). Focal adhesion signaling complexes are also involved in this process. Fig. 3h (see also Fig. 2 for Western blot) shows a number of proteins involved in focal adhesion formation, such as the focal adhesion kinase FAK, paxillin, Beta-PIX, vinculin, and the G protein-coupled receptor kinase interactor 1 (GIT1). Somewhat surprisingly, these proteins showed deactivation in response to insulin. In most cases this was transient, reaching a nadir at 5 min. Annexin A2, a calcium-dependent phospholipid-binding protein, displayed a similarly rapid down regulation, followed by recovery and a second downward trend after 10 min. Other effectors, such as the Nck binding protein Spin 90, α-actinin-4, the cytoskeleton-like bicaudal D protein homolog 2, the tight junction ZO-1 protein, and the PDZ-containing protein PDZK11/PISP showed prominent activation and are likely further candidates to connect insulin signaling to cytoskeletal reorganization.

Signal attenuation is critical for the fine-tuning of signaling events. This attenuation can be achieved by serine/threonine phosphorylation or ubiquitination. A number of ubiquitin ligases have been identified as important for the fast ubiquitination and subsequent degradation of receptor tyrosine kinases (29). Both c-Cbl and Nedd4 (30), two ubiquitin ligases, were identified after 5 min, and the Nedd4 activation profile is shown in (Fig. 3f).

PISP/PDZK11-pY7 Peptide Pull-Down Revealed Interaction with Myosins and with the Calcium Transporting ATPase SERCA2A.

Among the proteins not previously described as insulin-induced candidate effectors, we were intrigued by the PDZ-containing protein PISP/PDZK11 because its dynamic profile and fold activation are very similar to IRS-1 (Fig. 3b). This ubiquitously expressed protein is named PISP for plasma membrane Ca²⁺ ATPase (PMCA)-interacting single-PDZ-protein (31). The early activation profile and the high fold change of PISP/PDZK11 suggest a direct function in the early steps of insulin signaling, which might link to the previously observed acute effects of insulin on calcium flux (32) To gain more insight into the function of the activation of the PISP/PDZK11 protein, we synthesized two tyrosine-containing desoxy-biotin linked peptides derived from PISP/PDZK11 (including peptides containing tyrosines pTyr7 and pTyr10 in a phosphorylated and nonphosphorylated version to find phosphorylation-dependent binding partners (33). These phosphorylated and nonphosphorylated peptides were used as bait and incubated with lysates of SILAC light and heavy labeled cells, respectively. Peptides and binding partners were purified by streptavidin-coated beads, and the eluates were combined for in solution digest and MS analysis. We found that the PISP/PDZK11-pY7-containing sequence specifically interacted with myosin regulatory light chain 2, and two calcium binding proteins—SERCA2 and S100 calcium-binding protein A4 (SI Fig. 6 and SI Tables 9 and 10). Because PISP/PDZK11 is a potential calcium ATPase binding partner and we find that it interacts with proteins involved in calcium homeostasis, it is a possible link between insulin and calcium signaling.

Cell Culture and SILAC.

Brown preadipocyte cell lines were established from wild type and IRS-1 or IRS-2 deficient mice as described in ref. 9. After a 16-h starvation period, the cells were treated with insulin (1 μg/ml; Sigma). After hormone incubation for 1–20 min at 37°C, the media were removed, and the cells were immediately lysed with ice-cold modified RIPA buffer. For SILAC experiments, the DMEM was custom-made and was deficient for l-Arg, and l-Lys (Invitrogen). l-argine-6, l-arginine-10, l-lysine-4, and l-lysine-8 were from Sigma–Aldrich.

Antibodies and Immunoprecipitation.

Antibodies against the following proteins were used: immobilized anti-phosphotyrosine antibody (4G10; Upstate) and anti-pY antibody (P-Tyr-100; Cell Signaling); for Western blot analysis, insulin receptor β, IRS-1, p85α (Santa Cruz), IRS-2, p42/p44 MAP kinase, phospho-p42/p44 MAP kinase, phospho p38 (Cell Signaling), Shc, Grb-2, SHP2, Paxillin, Nedd4, c-CBL (BD Transduction ), FAK, Caveolin-1 (Upstate), hnRNP Q, GAP-1, APS (Abcam), and BICD2 (kindly provided by C. Hoogenraad, Erasmus University, Rotterdam, The Netherlands). Immunoprecipitation was performed as described in ref. 6.

Mass Spectrometric Analysis.

A detailed description of the mass spectrometric measurements and peptide pull downs can be found in SI Methods. Briefly, after SDS/PAGE and peptide extraction (49), samples were processed for liquid chromatography-mass spectrometry (GeLC-MS). Reverse phase nano-LC was performed on an Agilent 1100 nanoflow LC system (Agilent Technologies). The LC system was coupled to a 7-Tesla LTQ-FT instrument (Thermo Electron) equipped with a nanoeletrospray source (Proxeon). The following search parameters were used in all MASCOT searches: fully tryptic peptides, maximum of two missed trypsin cleavages, cysteine carbamidomethylation, methionine oxidation, pSTY, N-term protein acetyl, and SILAC labels Lys-D4, Lys-8, Arg-6, and Arg-10. The maximum initial mass tolerance for MS scans was 10 ppm and 0.5 Da for MS/MS and MS3 scans. Only proteins identified and quantified with at least two peptides with cores >20. MSQuant were considered for validation.