Biochemical Journal

Research article

RSK phosphorylates SOS1 creating 14-3-3-docking sites and negatively regulating MAPK activation

Madhurima Saha , Audrey Carriere , Mujeeburahiman Cheerathodi , Xiaocui Zhang , Geneviève Lavoie , John Rush , Philippe P. Roux , Bryan A. Ballif


The extent and duration of MAPK (mitogen-activated protein kinase) signalling govern a diversity of normal and aberrant cellular outcomes. Genetic and pharmacological disruption of the MAPK-activated kinase RSK (ribosomal S6 kinase) leads to elevated MAPK activity indicative of a RSK-dependent negative feedback loop. Using biochemical, pharmacological and quantitative MS approaches we show that RSK phosphorylates the Ras activator SOS1 (Son of Sevenless homologue 1) in cultured cells on two C-terminal residues, Ser1134 and Ser1161. Furthermore, we find that RSK-dependent SOS1 phosphorylation creates 14-3-3-binding sites. We show that mutating Ser1134 and Ser1161 disrupts 14-3-3 binding and modestly increases and extends MAPK activation. Together these data suggest that one mechanism whereby RSK negatively regulates MAPK activation is via site-specific SOS1 phosphorylation.

  • 14-3-3
  • negative feedback
  • phosphorylation
  • ribosomal S6 kinase (RSK)
  • signal transduction
  • Son of Sevenless (SOS)


Cells respond to an array of biological and environmental stimuli which trigger intracellular signalling pathways governing a diversity of cellular states. Differences in the magnitude and duration of signalling can lead to very different biological outcomes. This has been studied by examining cellular commitment to proliferation as a function of strength and duration of Ras–MAPK (mitogen-activated protein kinase) signalling in mammalian cells [1]. Intriguingly, although some cells proliferate due to increased Ras–MAPK signalling, others have varied responses including differentiation, senescence, survival and death [25]. Thus genetic and biochemical context can dramatically alter the cellular interpretation of a given signal.

Furthermore, although activating mutations of Ras–MAPK signalling can be found in greater than 30% of all human tumours, and greater than 60% in specific tumour types such as in pancreatic cancers [58], hyperactive Ras–MAPK signalling is responsible for several human developmental disorders which display defects which cannot be defined simply or exclusively as over-proliferation disorders. These so-called Rasopathies include NF-1 (neurofibromatosis 1), Noonan syndrome, LEOPARD (multiple lentigines, electrocardiographic conduction abnormalities, ocular hypertelorism, pulmonary stenoisis, abnormal genitalia, retardation of growth and sensorineural deafness) syndrome, Costello syndrome and CFC (cardio-facio-cutaneous) syndrome [810]. Together, these disorders display a striking variety of cellular phenotypes and clinical manifestations not only underscoring the importance of developing therapeutic approaches to control Ras–MAPK signalling, but also obviating the need for evolved cellular mechanisms of developmental and homoeostatic pathway control.

A number of molecular mechanisms have been uncovered that negatively regulate Ras–MAPK signalling in vivo. These include dephosphorylation of activating phosphorylation events, inhibitory binding of 14-3-3 to pathway components Raf (an MAPK kinase kinase) [11,12] and KSR (kinase suppressor of Ras) [13,14], and negative feedback loops. The known negative feedback loops are dependent on the activation of the canonical MAPKs ERK (extracellular signal-regulated kinase) 1/2 which desensitize Raf-1 by direct phosphorylation [11], which phosphorylate the RasGEF (Ras guanine-nucleotide-exchange factor) SOS (Son of Sevenless) 1 leading to SOS1 dissociation from Grb2 (growth factor receptor-bound protein 2, the adaptor protein which links SOS1 to tyrosine phosphorylated growth factor receptors and scaffolds [15]), and which transcriptionally regulate dual-specificity phosphatases [16]. In recent years, an additional negative feedback mechanism of the Ras–MAPK pathway was experimentally observed when pharmaceutical and genetic disruption of the MAPK-activated 90 kDa RSK (ribosomal S6 kinase) family led to increased or sustained ERK1/2 activity [1719]. On the basis of the evidence that RSK could phosphorylate SOS1 in vitro [20], it has been assumed that RSK mediates negative feedback in Ras–MAPK signalling via modulation of SOS1. However, the demonstration of RSK phosphorylation of SOS1 in cells and its putative negative regulatory effect on MAPK activation is still lacking.

In the present study we establish that RSK phosphorylates SOS1 in vivo at Ser1134 and Ser1161. Furthermore, we show that RSK-dependent phosphorylation of these residues creates 14-3-3-binding sites on SOS1. Finally, expression of an unphosphorylatable form of SOS1 modestly increases ERK1/2 activation and duration in response to EGF (epidermal growth factor) stimulation. Together these data indicate that RSK phosphorylation of SOS1 is one mechanism whereby RSK provides negative feedback control in Ras–MAPK signalling. However, these data also argue for additional mechanisms of RSK-dependent negative feedback to MAPK activation that remain to be identified.


Plasmids and site-directed mutagenesis

The following plasmid constructs were as described previously: pGEX-2TK-14-3-3β (human), pGEX-4T-14-3-3ϵ (rat) wild-type and K49E [21] (from Professor Michael Yaffe, MIT, Cambridge, MA, U.S.A.), pCGN-SOS1-HA [15] (from Professor Dafna Bar-Sagi, NYU Langone Medical Center, New York, NY, U.S.A.), pCMV6-Myristoylated-Rsk1 (avian) [22], pKH3-HA-RSK2 (mouse) [23] and FLAG-MEK1-DD [24]. For the expression of GST (glutathione transferase)–14-3-3ϵ variants in mammalian cells we PCR amplified 14-3-3ϵ wild-type and K49E in pGEX-4T inserting a 5′ BamHI site and a 3′ NotI site, which was then ligated in-frame with GST in pEBG (Professor Bruce Mayer, University of Conneticut Health Center, Farmington, CT, U.S.A.). The generation of the S1134A and S1161A mutants was done using the QuikChange strategy (Stratagene) and mutants were verified by DNA sequencing at the Vermont Cancer Center's DNA Analysis Facility (Burlington, VT, U.S.A.).

GST–14-3-3ϵ fusion proteins

Overnight cultures (50 ml) of BL21 Escherichia coli transformed with pGEX-2TK-14-3-3β, pGEX-4T-14-3-3ϵ wild-type or K49E were diluted into 500 ml and cultured for 2 h followed by induction with 1mM IPTG (isopropyl β-D-thiogalactopyranoside) for 6 h. Cells were pelleted and resuspended in 10 ml of PBS, 0.1 M EDTA, 5 mg/ml pepstatin A, 10 mg/ml leupeptin and 1 mM PMSF. The suspensions were sonicated on ice using a probe sonicator eight times for 30 s with 30 s delays between blasts. Triton X-100 (1 ml) was then added and the sonicates were centrifuged at 13000 g for 30 min. The supernatants were incubated with 300 μl of a washed 50% slurry of glutathione–agarose (G Biosciences). The beads were washed four times with the bacterial lysis buffer with 1% Triton X-100 and then three times with mammalian cell lysis buffer (see below). All of the manipulations were performed at 4°C.

Cell culture, transfections, pull-down assays, immunoprecipitation, immunoblotting, densitometry, antibodies and RNAi (RNA interference)

E1A-transformed HEK (human embryonic kidney)-293, COS7, NIH 3T3 and MEF (mouse embryonic fibroblast) cells immortalized using the ‘3T3’ protocol were maintained in DMEM (Dulbecco's modified Eagle's medium; Mediatech) supplemented with 10% fetal bovine serum (Hyclone), 50 units/ml penicillin and 50 μg/ml streptomycin. For the stimulations the cells were starved of serum for 16–18 h prior to the treatments. Pharmacological inhibitors and stimulants were from the following sources (with final concentrations indicated): BI-D1870 (10 μM; Biomol), SL0101 (50 μM; Toronto Research Chemicals), UO126 (20 μM; Biomol), PD184352 (10 μM; Calbiochem), PMA (25 ng/ml; Biomol and Calbiochem) and EGF (25 ng/ml; Invitrogen and Cell Signaling Technology). Transfections were done by calcium phosphate precipitation. For the pull downs, cells were lysed in 25 mM Tris (pH 7.4), 137 mM NaCl, 1% Igepal, 10% glycerol, 25 mM NaF, 10 mM Na4P2O7, 1 mM Na3VO4, 5 mg/ml pepstatin A, 10 mg/ml leupeptin and 1 mM PMSF. Clarified lysates (containing ~500 μg of protein) were incubated with 30 μl of a 30% slurry (~5 μg) of GST–14-3-3ϵ wild-type or K49E. After incubating for 2 h to overnight the beads were washed four times with lysis buffer prior to SDS/PAGE (10% gel) and immunoblotting. For anti-HA (haemagglutinin) immunoprecipitations cells were lysed in 0.5% Nonidet P40 (or Igepal), 0.1% Brij-35, 0.1% sodium deoxycholate, 1 mM EDTA, 5 mM EGTA, 7 mM K2HPO4, 3 mM KH2PO4, 10 mM MgCl2, 50 mM 2-glycerophosphate, 1 mM Na3VO4, 5 μg/ml pepstatin A, 10 μg/ml leupeptin and 1 mM PMSF. Immunoblotting was performed using a submersible transfer apparatus and nitrocellulose membranes. Blocking was performed in 5% milk/TBST (0.05% Tween 20, 8 mM Tris base, 25 mM Tris/HCl and 154 mM NaCl). Primary antibodies were incubated with the membranes in 1.5% BSA in TBST and washes were done with TBST. Secondary antibodies conjugated to horseradish peroxidase were from Chemicon/Millipore and visualization was done using enhanced chemiluminescence and exposure to X-ray film. Densitometry analysis was conducted using identical areas for each lane of a given blot using inverted histograms in Adobe Photoshop CS2. After subtracting the background, phospho-ERK1/2 levels were normalized to the levels of total ERK1/2 and HA–SOS1 levels. Commercial primary antibodies were from the following sources: anti-RSK1 used in Figures 3(E) and 4(C) (Santa Cruz Biotechnology), anti-RSK1 used in Figures 1 and 2(A) (Cell Signaling Technology), antiRSK2 used in Figure 1(B) (Cell Signaling Technology), anti-(Ser380 phospho-Rsk) (Cell Signaling Technology), anti-GST (Upstate Biotechnology/Millipore), anti-HA (Covance), anti-(Thr202/Tyr204 phospho-ERK1/2) (Cell Signaling Technology), anti-SOS1 (Upstate Biotechnology/Millipore) and anti-phospho-Akt substrate (anti-RXXpS, Cell Signaling Technology). Anti-(avian Rsk1) and anti-ERK1/2 were gifts from Professor John Blenis (Harvard University Medical School, Boston, MA, U.S.A.) and were described previously [25]. siRNA (small interfering RNA) against RSK1 (SI02223067), RSK2 (SI00288190) and the negative control siRNA (1027280) were obtained from QIAGEN. A total of 1–2 μg (50 nmol) of siRNA was transfected per 6-cm dish using calcium phosphate precipitation as described previously [26].

Figure 1 Identification of one or more potential RSK phosphorylation sites in SOS1 using an anti-(RXXpS motif) antibody

(A) PMA and EGF induce phosphorylation of an RXXS motif in SOS1. HEK-293 cells transfected with HA–SOS1 were starved of serum and left untreated or stimulated with PMA or EGF for the indicated times. At 20 min prior to stimulation the indicated cultures were treated with U0126 (U0), PD184352 (PD) or BI-D1870 (BI). HA–SOS1 was immunoprecipitated from whole cell extracts and immune complexes and whole cell extracts were subjected to immunoblotting (IP) with the indicated antibodies. (B) RSK1 and RSK2 contribute to MEK-dependent phosphorylation of SOS1. Cells were treated as in (A) except with the co-transfection of RNAi and an activated MEK1 allele (MEK-DD) as indicated.

Figure 2 Ser1134 and Ser1161 are the PMA- and EGF-induced SOS1 phosphorylation sites that conform to the minimal RSK consensus motif

(A) HEK-293 cells were treated as described in Figure 1 and whole cell extracts as well as 10% of each immune complex was subjected to immunoblotting (IP) with the indicated antibodies. (B) A total of 90% of the immune complexes shown in (A) were subjected to SDS/PAGE and Coomassie Blue staining. (C) The gel bands from (B) were subjected to in-gel tryptic digestion and the extracted peptides were mixed with stable isotope-containing reference peptides. Peptides were subjected to technical replicate LC–MS/MS analyses. Representative example of raw data showing the isotopic envelopes from the MS1 full scans of the stable isotope-containing and native peptides inclusive of Ser1134, phosphor-Ser1134 and Phe648–Arg660 from starved, PMA-stimulated and EGF-stimulated cells are shown. Monoisotopic peak height for the native (open stars) and labelled (closed stars) peptides are indicated. Broken lines connect monoisotopic peaks of peptide pairs. Increases and decreases in abundance are indicated by arrows. (D) Averages and standard deviations of technical replicates for fold changes relative to unstimulated are indicated. MS/MS spectra for labelled and native peptides are shown in Supplementary Figure S1 (at (E) Ser1134 and then Ser1161 are the major RXXS phosphorylation sites in SOS1 following stimulation with PMA and EGF. HEK-293 cells transfected with wild-type HA–SOS1 or the indicated HA–SOS1 mutants were starved of serum and stimulated with the indicated factors. Cell extracts were subjected to anti-HA immunoprecipitation and immunoblotting with the indicated antibodies.

Quantitative MS and data analysis

Immunoprecipitated HA–SOS1 from transfected cells starved of serum and then untreated or stimulated with either EGF or PMA was subjected to SDS/PAGE (10% gel) and staining with Coomassie Blue. Bands were digested in-gel with trypsin, and tryptic peptides were extracted and prepared as described previously [27]. Dried peptides were resuspended in 8.5 μl of 2.5% acetonitrile and 2.5% formic acid containing 100 fmol/μl of each stable isotope-containing AQUA (absolute quantification) [28] peptide standard prior to LC-MS/MS (liquid chromatography tandem MS) analysis in a linear ion trap–orbitrap hybrid mass spectrometer (Thermo Electron) set up and run as described previously [29]. A total of two 4 μl technical replicates were analysed and quantified using monoisotopic peak heights as described previously [30,31]. For quantification of fold changes, the abundance of each peptide was first determined by dividing the ratio of the labelled (heavy) peptide to the native (light) peptide and then normalizing this ratio by dividing it by the heavy/light ratio of the reference peptide Phe648–Arg660 found in the same sample. The fold changes were then determined by dividing the ratios obtained from stimulated samples by the ratios obtained in the unstimulated states. Technical replicates were averaged and the standard deviations were calculated. Supplementary Figure S1 (at shows representative MS/MS spectra for each heavy reference peptide and the corresponding identified native peptides. Stable isotope-containing heavy AQUA peptides were synthesized by Cell Signaling Technology.


The Ras–MAPK pathway promotes SOS1 phosphorylation at basophilic sequences via RSK

RSK, like many of its basophilic kinase relatives, shows a preferred minimal substrate target motif, RXXS [3234], where X is any amino acid and S is the serine targeted for phosphorylation. Supplementary Figure S2 (at lists a selection of 18 characterized RSK substrates [33,34] and the amino acid sequences surrounding the sites of phosphorylation. Also shown are two strong potential RSK target sites (Ser1134 and Ser1161) in the C-terminus of SOS1 (not found in SOS2) which exhibit high conservation among vertebrate SOS1 orthologues. Given that SOS1 has putative RSK phosphorylation sites we asked if EGF or the phorbol ester PMA, both known in vivo activators of RSK [34,35], would induce phosphorylation of SOS1 in a minimal RSK consensus motif. We made use of a phospho-specific motif antibody that recognizes the phosphorylated minimal RSK consensus motif (RXXpS; where pS is phosphoserine and X is any amino acid) [26,35], and we observed increased SOS1 phosphorylation following both EGF and PMA stimulation. This phosphorylation was blocked when cells were pre-treated with inhibitors of MEK (MAPK/ERK kinase) activation (UO126 and PD184352) or RSK (BI-D1870) (Figure 1A). Furthermore, overexpression of a constitutively active allele of MEK1 (MEK-DD) [24] induced SOS1 phosphorylation as recognized by the anti-(RXXpS motif) antibody (Figure 1B). This effect was eliminated by RNAi-dependent silencing of RSK1 and RSK2, with RSK2 silencing having the more profound effect (Figure 1B). Together these data provided strong evidence for the presence of at least one in vivo RSK phosphorylation site on SOS1.

RSK phosphorylates SOS1 at Ser1134 and Ser1161

Earlier work described RSK phosphorylation of SOS1 in vitro using purified RSK, [γ-32P]ATP and a GST–C-terminal (Ser1132–Ser1343) SOS1 [20]. Although not definitive, phosphopeptide mapping best placed a single phosphorylation site at Ser1161 [20]. Of note, Arg1131 was not present in the GST–C-terminal SOS1 fusion and would therefore have truncated the RSK target motif surrounding Ser1134. To definitively monitor PMA- and EGF-induced changes in SOS1 Ser1134 and Ser1161 phosphorylation, we took a quantitative MS approach using stable isotope-containing ‘heavy’ reference peptides [28]. The advantage of this quantitative MS approach is that a given native peptide can be quantified directly relative to a known amount of the internal reference peptide based on the fact that the ionization efficiency and chromatographic behaviour of the two peptides are identical. Furthermore, the CID (collision-induced dissociation) tandem mass spectra can provide unambiguous identification and quantification of site-specific phosphorylation (see below and Supplementary Figure S1). To monitor site-specific changes in Ser1134 and Ser1161 phosphorylation we immunoprecipitated HA–SOS1 from cells stimulated with either EGF or PMA. Of each immunoprecipitation 10% was subjected to Western blotting using the anti-(RXXpS motif) antibody and an increase in phosphorylation of the putative RSK target motif was confirmed (Figure 2A). The rest of the immune complex was subjected to SDS/PAGE and staining with Coomassie Blue (Figure 2B). The bands were digested in-gel with trypsin and extracted peptides were resuspended in a solution containing a fixed amount of each stable isotope-containing peptide standard prior to LC–MS/MS analysis in a linear ion trap–orbitrap hybrid mass spectrometer. The heavy reference peptides included peptides harbouring Ser1134 and Ser1161 in their phosphorylated and unphosphorylated states. Given that ERK1/2 phosphorylate the proline-directed Ser1167 in the same tryptic peptide as Ser1161, we monitored the doubly phosphorylated phospho-Ser1161- and phospho-Ser1167-containing peptide. To normalize SOS1 protein levels across samples, an additional heavy peptide (Phe648–Arg660) was monitored that was not predicted to become modified following stimulation. Tandem mass spectra for all native and reference peptides are provided in Supplementary Figure S1. Exemplifying the raw data, Figure 2(C) displays the MS1 spectra for the heavy and native unphosphorylated and phosphorylated Ser1134-containing peptides as well as peptide Phe648–Arg660 of SOS1 immunopurified from starved, PMA-stimulated and EGF-stimulated conditions. An increase in the relative abundance of the phospho-Ser1134-containing peptide was observed following PMA and EGF stimulation with an expected corresponding decrease in the unphosphorylated Ser1134-containing peptide. After technical replicates were analysed the average fold changes were plotted (Figure 2D). The quantitative MS data are in good agreement with the anti-RXXpS data shown in Figure 2(A) and definitively show PMA- and EGF-induced phosphorylation of Ser1134, Ser1161 and Ser1167. To determine if Ser1134 and Ser1161 were the major SOS1 sites of anti-RXXpS immunoreactivity, we mutated these residues singly or in combination to generate non-phosphorylatable alanine mutants. These HA-tagged constructs were expressed in cells as well as wild-type HA–SOS1. The cells were starved of serum and either left untreated or stimulated with PMA or EGF. Anti-HA immune complexes and immunoblotting with the anti-RXXpS antibody showed that Ser1134 is the major RXXpS site on SOS1 followed by Ser1164 in agreement with the MS data. The double mutant showed no appreciable stimulation-induced phosphorylation (Figure 2E).

Phosphorylation of SOS1 at Ser1134 and Ser1161 creates 14-3-3-binding sites

The minimal target motif (RXXS) as visualized by a frequency plot [36] of RSK substrates overlaps with the preferred mode 1 binding motif (RSXpSXP; Figure 3A) of the 14-3-3 family of phosphoprotein regulators [21,37]. As SOS1 Ser1134 and Ser1161 each show conserved aspects of the canonical mode 1 14-3-3-binding motif (Figure 3A), and given that 14-3-3 has been shown to negatively regulate Ras–MAPK signalling by binding to both Raf and KSR, we asked if SOS1 might also bind to 14-3-3 in a phospho-dependent manner. We expressed HA–SOS1in HEK-293 cells and lysed the cells while they were growing in complete medium. Extracts were subjected to a GST–14-3-3 pull-down assay using either wild-type 14-3-3ϵ or a K49E mutant that shows dramatically reduced binding to phosphorylated substrates [38]. Consistent with a phospho-dependent interaction, HA–SOS1 bound to wild-type 14-3-3ϵ, but showed little to no binding to the K49E mutant (Figure 3B). Similar results were observed using 14-3-3β (Figure 3C) and when using transfected HA–SOS1 with transfected GST–14-3-3 (Figure 3D). We also observed PMA-induced binding of endogenous SOS1 with GST–14-3-3 in a pull-down assay from three immortalized cell lines (Figure 3E).

Figure 3 SOS1 binds to 14-3-3 in a stimulus and likely phospho-dependent manner

(A) Comparison of Weblogo frequency plots of the RSK target motif (upper panel), the top 500 Scansite [49] predicted 14-3-3-binding motifs in proteins from the SwissProt database, and the evolutionary conservation around SOS1 Ser1134 and Ser1161 (lower two panels) from sequences presented in Supplementary Figure S2 (at (B) SOS1 binds to wild-type (wt), but not K49E, 14-3-3ϵ in pull-down assays. HEK-293 cells were mock-transfected or transfected with HA–SOS1. Whole cell extracts were subjected to immunoblotting as indicated. Equal portions of cell extract containing HA–SOS1 were subjected to pull-down assays using bacterially produced GST–14-3-3ϵ wild-type or K49E as indicated. The pull downs were subjected to immunoblotting with anti-HA antibodies. Amounts of each GST–14-3-3ϵ fusion protein are shown by Ponceau staining of the membrane prior to immunoblotting. *Wild-type or mutant GST–14-3-3ϵ as indicated. (C) PMA-stimulated SOS1 interacts with GST–14-3-3β and ϵ. The pull downs and immunoblots were conducted with the indicated fusion proteins and antibodies as described in (B). (D) SOS1 binds to co-transfected GST–14-3-3ϵ wild-type, but not GST–14-3-3ϵ K49E. HEK-293 cells growing in complete media were mock-transfected or transfected with HA–SOS1 with or without the indicated expression plasmids for GST–14-3-3ϵ variants. Cell extracts (lower panel) and glutathione agarose pull-down assays were subjected to immunoblotting with the indicated antibodies (upper panel). (E) Endogenous SOS1 binds to GST–14-3-3ϵ following stimulation with PMA. HEK-293, COS7 and NIH 3T3 cells were starved of serum for 16 h following stimulation with PMA for the indicated times. Cell extracts (lower panel) and GST–14-3-3ϵ pull-down assays (upper panel) were subjected to immunoblotting with the indicated antibodies. The amounts of GST–14-3-3ϵ fusion protein are shown by Ponceau staining of the membrane prior to immunoblotting.

RSK regulates 14-3-3 binding to SOS1 in response to Ras–MAPK activation

To determine if RSK is the kinase responsible for the interaction of 14-3-3 with SOS1, we examined the interaction from three angles. First, we asked whether the PMA-induced binding of SOS1 to 14-3-3 would be blocked by pharmacological inhibition of MEK–MAPK signalling and found that it was the case (Figure 4A). Secondly, we expressed a constitutively active allele of RSK1 [22] and found that 14-3-3 binding to SOS1 was dramatically induced in serum-starved cells (Figure 4B). Finally, we knocked down RSK expression using RNAi and found that this prevented the PMA- and EGF-induced interaction of SOS1 with 14-3-3 (Figure 4C). We next examined the functional consequences of a SOS1 allele that could not be phosphorylated at the RSK phosphorylation sites. The SOS1 S1134A/S1161A double mutant displayed no detectable binding to 14-3-3 and whereas the two sites each contributed to the binding to 14-3-3, phosphorylation at Ser1134 was most important (Figure 4D). Together, these data demonstrate that RSK regulates 14-3-3 binding to SOS1 by promoting phosphorylation of Ser1134 and Ser1161.

Figure 4 RSK phosphorylates SOS1 inducing 14-3-3 binding

(A) Pharmacological disruption of MEK1/2 activation blocks PMA-induced SOS1 binding to 14-3-3ϵ. HEK-293 cells mock-transfected or transfected with HA–SOS1 were starved for 16 h. Where indicated, cells were pre-treated with U0126 for 30 min prior to stimulation with PMA for 20 min. Cell extracts and GST–14-3-3ϵ pull-down assays were subjected to immunoblotting with the indicated antibodies. The amounts of GST–14-3-3ϵ fusion protein are shown by Ponceau staining of the membrane prior to immunoblotting. (B) Constitutively active, myristoylated (Myr)-RSK1 induces SOS1 binding to 14-3-3ϵ in the absence of stimulation. Cells were treated as in (A) except for the co-transfection of Myr-RSK1 (avian) as indicated. Av, an antibody specific to avian RSK1. Starved cells were lysed and extracts were subjected to pull-down assays and immunoblotting as in (A). (C) RSK is required for PMA and EGF-induced binding of 14-3-3ϵ to SOS1. Cells were treated as in (A) except for the co-transfection of RNAi for RSK1/2 or a control RNAi (Cont. RNAi) where indicated. Starved cells were then treated with either PMA (20 min) or EGF (10 min) prior to pull downs and immunoblotting. (D) Phosphorylation at Ser1134 and Ser1161 mediate the binding of SOS1 to 14-3-3ϵ. HEK-293 cells were transfected with the indicated HA-SOS1 wild-type (wt) or serine-to-alanine mutant constructs. Cell extracts were subjected to GST–14-3-3ϵ pull-down assays and immunoblotting as in (A).

SOS1 phosphorylation by RSK inhibits EGF-induced MAPK activation

We next examined a time course of EGF-dependent ERK1/2 activation in cells overexpressing either wild-type SOS1 or the SOS1 Ser1134Ala/Ser1161Ala double mutant. Consistent with phosphorylation of these sites contributing to negative regulation of Ras–MAPK signalling, we observed an increase in the magnitude and duration of EGF-dependent ERK1/2 activation (Figures 5A and 5B). These data support the notion that RSK is a negative regulator of Ras–MAPK activity, and is consistent with previous literature. We found that pharmacological disruption of RSK using two different inhibitors (BI-D1870 and SL0101) resulted in increased ERK1/2 phosphorylation in response to EGF stimulation in HEK-293 cells (Figure 1A) and MEFs (Figure 5C). The effect of the RSK inhibitors appeared stronger compared with the effect of overexpressing the double phosphorylation site SOS1 mutant, but this could be due to several factors including contributions of endogenous SOS1 in the overexpression experiments, compensatory effects during the transfection time compared with acute effects during drug treatments or that SOS1 is not the only target of RSK-dependent negative feedback on Ras–MAPK signalling. Together, our results demonstrate that RSK phosphorylates SOS1 on residues that create 14-3-3-binding sites and which play a role in the negative regulation of the Ras–MAPK pathway.

Figure 5 RSK phosphorylation of SOS1 negatively regulates ERK1/2 phosphorylation

(A) SOS1 S1134A/S1161A increases the magnitude and duration of EGF-stimulated ERK1/2 phosphorylation. The indicated HA–SOS1 wild-type (Wt) and mutant constructs were transfected into HEK-293 cells. Cells were starved and stimulated with EGF for the indicated times prior to lysis. Cell extracts were subjected to SDS/PAGE and immunoblotting with the indicated antibodies. Both a short exposure (SE) and a long exposure (LE) are provided for phospho-ERK1/2 specific blots. (B) The left-hand panel shows relative phospho-ERK1/2 quantified by densitometry (as described in the Experimental section) from three experiments similar to the one shown in (A). The right-hand panel shows the mean fold increase in ERK1/2 phosphorylation of the three experiments comparing wild-type (HA–SOS1) and S1134A/S1161A HA–SOS1. Results are means±S.E.M. One-tailed unequal variance Student's t test P values are indicated for a comparison of the means of the wild-type and mutant (S2A) phospho-ERK1/2 levels. *significance at a 95% confidence interval. (C) As shown for HEK-293 cells in Figure 1(A), inhibition of RSK reveals the presence of a negative feedback loop. MEFs were starved of serum and then stimulated with EGF or pre-treated with RSK inhibitors BI-D1870 or SL-0101for 30 min prior to stimulation with EGF, and immunoblotting of extracts as indicated.


The importance of understanding the various molecular mechanisms negatively regulating Ras–MAPK signalling is profound given the prevalence of its hyperactivation in human tumours as well as in rare, but devastating, developmental disorders. Notwithstanding the strong pharmacological and genetic evidence for a negative feedback role for RSK in Ras–MAPK signalling, little progress towards elucidating the relevant molecular mechanisms has been made. In the present study we identify RSK phosphorylation sites on SOS1 that negatively regulate EGF-dependent ERK1/2 activation, potentially as they create 14-3-3-binding sites that could interfere with molecular interactions of Ras pathway components. 14-3-3 has already been shown to interfere with other components of Ras–MAPK signalling, including Akt-phosphorylated Raf [11,12,39], and the MAPK signalling cassette scaffold, KSR phosphorylated by CTAK1 [Cdc25C (cell division cycle 25C)-associated kinase 1] [13]. We propose a two-stage negative feedback loop towards SOS1 in Ras–MAPK signalling with the first stage at the level of ERK1/2 and the second stage at the level of RSK (Figure 6). ERK1/2 have long been known to phosphorylate SOS1 at four critical residues reducing the binding of SOS1 to GRB2 [15], the effect of which may be to diffuse signalling away from the receptor complex even if SOS1's membrane-binding domains retain SOS1 near lipid products at the membrane. In stage two of the negative feedback loop RSK phosphorylation of SOS1 induces 14-3-3 binding which may prevent the interaction of SOS1 with Ras by directly blocking the interaction or by inducing conformational changes that reduce the interaction of SOS1 with Ras, or with the membrane itself (Figure 6). Evidence suggesting that the negative effect of RSK is functioning upstream of Raf can be found in Figure 1(A), where the RSK-specific inhibitor BI-D1870 leads to a dramatic up-regulation of ERK1/2 phosphorylation following EGF stimulation, but this is not observed following PMA stimulation. This is consistent with the mechanism of action of PMA at the level of Raf [40].

Figure 6 Model of a two-stage phospho-dependent Ras–MAPK feedback loop to negatively regulate SOS1

Activation of Ras–MAPK signalling by EGF engaged with EGFR induces autophosphorylation of EGFR and the recruitment of Grb2 bound to SOS1. SOS1, which also interacts with lipids at the plasma membrane, activates Ras which can then interact with Raf leading to sequential activation of MEK and ERK. KSR localizes the three kinases as a cassette at the plasma membrane. 14-3-3 proteins prevent membrane localization of both KSR and Raf unless dephosphorylated by PP2A (protein phosphatase 2A) and PP1 (protein phosphatase 1) respectively. ERK activation leads to the activation of RSK. In stage one of the negative feedback loop on SOS1, ERK phosphorylates several sites in the C-terminus of SOS1 which reduces its binding to Grb2. This may lead to SOS1 diffusing away from the proximity of the activated signalling complex. In stage two RSK phosphorylates SOS1 at Ser1134 and Ser1161 leading to 14-3-3 binding which might reduce SOS1 catalytic activity or prevent its interaction with Ras. Alternatively, 14-3-3 might reduce the capacity for SOS1 to bind to the plasma membrane and thereby attenuate Ras–MAPK signalling. C-TAK1, Cdc25C (cell division cycle 25c)-associated kinase 1

A large proportion of human cancers are characterized by hyperactivation of the MAPK pathway, whereby deregulated ERK1/2 helps drive unrestricted cell growth and proliferation. RSK was confirmed to be constitutively active in several cancers [41], such as melanoma [42], suggesting that it regulates substrates that contribute to tumorigenesis. On the basis of the findings of the present study and those of others [17,18], RSK negatively regulates MAPK activation in response to growth factors. Thus, in addition to its growth-related functions, RSK may also limit MAPK signalling in cells with perturbed Ras regulation, such as in NF-1 [43]. The biological impact of RSK activation is therefore the result of its dual roles downstream of the MAPK pathway, and it is conceivable that hyperactivation of RSK may in some cases be detrimental to MAPK-dependent cell proliferation. Very few studies have reported increased tumour-derived RSK expression. Rather, accumulating evidence supports the idea that at least some RSK isoforms are down-regulated in certain cancers (in particular RSK3 and RSK4 [44,45]). Although our data indicate a requirement for RSK1 and RSK2 (with RSK2 playing a more important role) in Ser1134 and Ser1161 phosphorylation of SOS1 in response to growth factor stimulation, further investigation will be necessary to determine the extent to which RSK3 or RSK4 participate in the regulation of SOS1 in other cellular contexts. However, significant disruption of negative feedback is observed in muscle and neuronal cells of RSK2-deficient animals [18,19] suggesting that RSK2 plays a primary role in the negative feedback mechanism.

Previously, canonical mammalian Ras–MAPK signalling has been characterized as a system with significant robustness to change, given its graded, rather than switch-like, signal amplification [4648]. The robustness of the system is dependent on intact negative feedback loops and is characteristic of negative feedback amplification systems in engineering that facilitate the buffering or smoothening of a given output even when inputs fluctuate [16,46,47]. Although ERK phosphorylation of Raf was observed by others to provide strong negative feedback in serum- and EGF-dependent signalling [16,47], genetic disruption of RSK2 alone leads to significant loss of negative feedback following insulin stimulation in skeletal muscle and glutamate signalling in neurons [18,19].

To understand the topology of any signalling system, it is important to delineate the strength and multiplicity of negative feedback mechanisms acting within it. This is particularly true if imposition of targeted pharmacological control is desired as is the case with hyperactive Ras–MAPK signalling. For example, if RSK inhibitors were to be used as anti-cancer therapy then they would likely be more effective when tumours are driven by activating mutations in Raf since RSK-dependent negative feedback appears to act upstream of Raf, at least in part, on SOS1. Thus the RSK-dependent negative feedback mechanisms, in part described in the present paper, highlight the need for critical evaluation of both the positive and negative contributions initiated by RSK isoforms in Ras–MAPK signalling in both normal and aberrant biology.


Madhurima Saha and Audrey Carriere conducted the majority of the experiments with additional experiments performed by Mujeeburahiman Cheerathodi, Xiaocui Zhang, Geneviève Lavoie, Philippe Roux and Bryan Ballif. John Rush helped to design and provided the stable isotope-containing reference peptides. Bryan Ballif wrote the paper with the assistance of Philippe Roux. Bryan Ballif and Philippe Roux were the principals directing and designing experiments and interpreting the data.


This work was supported by the Vermont Genetics Network through the NIH (National Institutes of Health) INBRE (IDeA Network of Biomedical Research Excellence) program of the NIMGS (National Institute of General Medical Sciences) [grant number P20 RR16462 (to B.A.B. and M.S.)], the Canadian Cancer Society Research Institute [grant number 700878 (to P.P.R.)] and the Cancer Research Society [grant number DF121153 (to P.P.R.)]. P.P.R. holds a Canada Research Chair in Signal Transduction and Proteomics. A.C. is recipient of a fellowship from the Cole Foundation and X.Z. was awarded a doctoral studentship from the China Scholarship Council (CSC). The University of Vermont neuroscience molecular core provided film developing equipment and is funded by the NIH COBRE (Centers of Biomedical Research Excellence) program [grant number P20 RR016435].


We acknowledge Professor Michael Yaffe, Professor Dafna Bar-Sagi, Professor Bruce Mayer and Professor John Blenis for reagents; Jason Reynolds and Jeffrey Knott (Cell Signaling Technology, Danvers, MA, U.S.A.) for AQUA peptide preparation; and the Vermont Cancer Center DNA Analysis Facility for DNA sequencing.

Abbreviations: AQUA, absolute quantification; EGF, epidermal growth factor; ERK, extracellular signal-regulated kinase; Grb2, growth factor receptor-bound protein 2; GST, glutathione transferase; HA, haemagglutinin; HEK, human embryonic kidney; KSR, kinase suppressor of Ras; LC, liquid chromatography; MAPK, mitogen-activated protein kinase; MEF, mouse embryonic fibroblast; MEK, MAPK/ERK kinase; MS/MS, tandem MS; NF-1, neurofibromatosis1; RNAi, RNA interference; RSK, ribosomal S6 kinase; siRNA, small interfering RNA; SOS, Son of Sevenless; TBST, 0.05% Tween 20, 8 mM Tris base, 25 mM Tris/HCl and 154 mM NaCl


View Abstract