Stimulus-induced uncoupling of extracellular signal-regulated kinase phosphorylation from nuclear localization is dependent on docking domain interactions

A wide variety of extracellular signals activate the Raf–MEK–ERK pathway. In this cascade, Raf isoforms are recruited to the cell membrane for activation, which phosphorylate and activate MEK [mitogen-activated protein kinase kinase (MAPKK), also known as ERK kinase). Activated MEK in turn phosphorylates ERK1 (MAPK3) and ERK2 (MAPK1) (referred to as ERK herein) on Thr and Tyr residues of a TEY activation loop (Caunt et al., 2006a; Raman et al., 2007). ERK is chiefly cytoplasmic in resting cells, but activation causes its accumulation in the nucleus (Chen et al., 1992; Lenormand et al., 1993; Lidke et al., 2010). This is crucial for ERK to access transcriptional targets, and correct localization is essential for the integrity of cell fate decisions, such as proliferation and differentiation (Brunet et al., 1999; Robinson et al., 1998). ERK does not contain intrinsic nuclear localization or export signals and relies on dynamic association with a wide repertoire of proteins for appropriate subcellular targeting (von Kriegsheim et al., 2009). The distribution of ERK between the nucleus and the cytoplasm is dependent upon rates of nuclear entry and exit (Costa et al., 2006; Lidke et al., 2010). ERK can enter and exit the nucleus by means of an energy-independent process that is facilitated by direct interaction with nuclear pore proteins, and TEY-phosphorylated ERK can also be imported by a second process requiring both energy and cytosolic factors (Ranganathan et al., 2006; Yazicioglu et al., 2007; Whitehurst et al., 2002). Cytoplasmic and nuclear binding proteins can also influence the distribution of ERK by affecting the fraction of ERK available for movement to or from the nucleus. MEK is a major cytoplasmic scaffold of ERK, and MEK-directed ERK phosphorylation of the TEY motif causes its liberation from MEK, facilitating its interaction with other proteins, including those of the nuclear pore complex. In this way, MEK activation characteristically increases the nuclear localisation of ERK, an effect that is dependent upon the TEY motif (Adachi et al., 1999; Chuderland et al., 2008b; Fukuda et al., 1997; Lenormand et al., 1998; Wolf et al., 2001) but not on ERK catalytic activity (Adachi et al., 1999; Gonzalez et al., 1993; Khokhlatchev et al., 1998; Yazicioglu et al., 2007). Other proteins that that can cause cytoplasmic retention of ERK include PEA-15, dual specificity phosphatase 6 (DUSP6), β-arrestin and Sef (IL17RD) (Brunet et al., 1999; Caunt et al., 2006b; Formstecher et al., 2001; Karlsson et al., 2004; Luttrell et al., 2001; Tohgo et al., 2002; Torii et al., 2004). By contrast, expression of nuclear DUSP2, DUSP4 and DUSP5 can mediate dephosphorylation and nuclear anchoring of ERK in sustained phases of signaling (Caunt et al., 2008a; Caunt et al., 2008b; Mandl et al., 2005; Volmat et al., 2001). Thus, many proteins influence ERK compartmentalization, and correct interpretation of these often-conflicting signals is essential during dynamic changes in localization.

The specificity of the association of ERK with its binding proteins is dictated by docking domains. The ERK common docking (CD) motif is distal to the catalytic site and binds to Lys- and Arg-rich sequences, known as D (docking)-domains (Tanoue et al., 2001). Another key docking site in ERK protein partners is termed the DEF (docking site for ERK, F/Y-X-F/Y-P) motif, which binds to a DEF-binding pocket (DBP) adjacent to the catalytic site of ERK (Jacobs et al., 1999; Lee et al., 2004). ERK protein partners can contain either or both D- and DEF-domains, allowing intricate regulation of ERK recognition (Dimitri et al., 2005; Jacobs et al., 1999; Shin et al., 2010). The MEK–ERK interaction is multivalent, involving both the CD motif and the DBP of ERK (Robinson et al., 2002). Phosphorylation and release of ERK from MEK therefore exposes these docking sites during stimulus-mediated responses. Recent studies have shown that D319N and Y261A mutations can interfere with ERK binding to partner D- or DEF-domain-containing proteins, respectively, in intact cells without affecting phosphorylation or overall catalytic activation by MEK (Caunt et al., 2008a; Caunt et al., 2008b; Dimitri et al., 2005; Shin et al., 2010). These mutations can therefore be used to identify ERK responses that are dependent on D- or DEF-domain interactions.

Uncoupling of ERK phosphorylation and nuclear localization responses occurs in a number of models, but the underlying mechanisms remain largely unknown. Most previous work exploring the relationships between ERK phosphorylation and compartmentalization in attached cells has involved biochemical assays with cell homogenates or microscopy with relatively small cell numbers. Here, we have taken a novel approach, using automated fluorescence microscopy and image analysis to monitor whole cell levels of ppERK1 and ppERK2 (hereafter referred to as ppERK1/2) as well as ERK1 and ERK2 (hereafter referred to as ERK1/2) and the nuclear:cytoplasmic (N:C) ERK1/2 ratio in large numbers of single attached cells. This enables response characteristics to be defined from frequency distributions and in cell subpopulations selected (i.e. ‘binned’) according to the ppERK1/2 level. Using population-averaged data, we found that protein kinase C (PKC) activation using phorbol 12,13-dibutyrate (PDBu) and epidermal growth factor (EGF) caused the expected dose- and time-dependent increases in ppERK1/2 and the N:C ratio of ERK1/2. However, with data binned according to ppERK1/2 levels, both activators still caused a pronounced increase in ERK1/2 N:C ratio. This reveals a clear component of the endogenous ERK nuclear localization response that cannot be attributed to TEY phosphorylation of ERK. Exploring the mechanisms (with chemical inhibitors and a system in which endogenous ERK is replaced with wild-type or mutated ERK2–GFP) revealed that this component of the response was dependent on MEK activity and TEY phosphorylation of ERK. It was also dependent upon D-domain-dependent binding but not on DEF-domain-dependent binding, protein neosynthesis or ERK catalytic activity. Thus, we have found that MEK-directed TEY phosphorylation of ERK is necessary but not sufficient for the full PDBu- and EGF-stimulated nuclear localization response. There is an additional TEY-phosphorylation-unattributable component of the response that is dependent upon D-domain-mediated binding of ERK to, as yet, unidentified partners.

We examined spatial and temporal aspects of ERK regulation by immunofluorescence microscopy. Total ERK1/2 and ppERK1/2 were stained simultaneously and quantified using a high-content image platform for automated image acquisition and analysis. Stimulation of conventional and atypical PKC isoforms with PDBu caused dose and time-dependent increases in whole-cell levels of dual-phosphorylated ERK (ppERK1/2), providing a measure for ERK activation. It also caused dose- and time-dependent increases in the nuclear:cytoplasmic (N:C) ERK1/2 ratio, providing a measure for nuclear localization. The kinetics of these two responses differed (Fig. 1). The effect of PDBu on ppERK1/2 levels was maximal at 5 minutes and then decreased to a plateau level (10–30% of maximal) that was maintained from 60 minutes to 6 hours. This is consistent with earlier data (Caunt et al., 2008a; Caunt et al., 2008b) and with response kinetics obtained by western blotting for ppERK1/2 in whole-cell lysates (Fig. 1). Western blotting revealed sustained MEK phosphorylation from 5–240 minutes, and so the reduction in ppERK1/2 levels occurred in spite of ongoing MEK activation over this period. The ERK nuclear localization response also increased to a peak at 5 minutes, then reduced briefly, before increasing again to a maximum at 2 hours (1 μM PDBu) or 6 hours (0.1 μM PDBu). This uncoupling of the relatively transient effect of PDBu on ppERK1/2 from its more sustained effect on ERK1/2 N:C ratios has been attributed previously to nuclear DUSP proteins that can both dephosphorylate and anchor ERK in the nucleus (Caunt et al., 2008a; Caunt et al., 2008b; Mandl et al., 2005; Volmat et al., 2001). Consistent with this, we found that PDBu also caused a pronounced increase in whole-cell levels of DUSP1 (the prototypic nuclear DUSP) at 1–4 hours after stimulation (Fig. 1).

The data described above were derived from population averages (thousands of imaged or lysed cells), but the microscopy methods used also provide data on each individual cell. This is particularly important for ERK signaling because, at the single-cell level, stimuli cause digital ‘all-or-nothing’ responses in some models and graded responses in others (Ferrell and Machleder, 1998; Lin et al., 2009; Mackeigan et al., 2005; Whitehurst et al., 2004). Moreover, a recent study found that graded ERK phosphorylation causes digital increases in immediate-early gene expression, raising the question of whether ERK translocation to the nucleus is digital or graded (Mackeigan et al., 2005). To assess this, we generated frequency-distribution histograms for whole-cell ppERK1/2 levels and ERK1/2 N:C ratios in control and PDBu-stimulated cells (Fig. 2A). As expected, stimulation for 5 minutes with 1 μM PDBu shifted the frequency–response plots rightward, yielding near-normal distribution curves for both measures, with ppERK1/2 and ERK1/2 N:C mode values increased approximately eight- and two-fold above those of controls, respectively. The histograms obtained with 0.1 μM PDBu were also distributed near normally, with intermediate mode values (six- and 1.6-fold above control for ppERK1/2 and ERK1/2 N:C, respectively). This is in sharp contrast to the bimodal distribution seen with digital responses (Lin et al., 2009), revealing that ERK1/2 phosphorylation and nuclear localization are both graded responses in HeLa cells. Similar data were seen in MCF7 cells (data not shown).

A further feature of these plots is that whole-cell ppERK1/2 values were broadly spread (99% of cells within a 30-fold range of ppERK1/2 levels), as compared with the much tighter N:C ERK1/2 range (99% of cells within three-fold N:C ERK1/2 range) (Fig. 2A). This suggests uncoupling of the responses, with relatively constant nuclear localization over a broad range of ppERK1/2 levels. To test for this, we sorted the cells into bins according to ppERK1/2 level [each spanning 80 arbitrary fluorescence units (AFU)] and, for each bin, plotted the mean ppERK1/2 value against the mean ERK1/2 N:C ratio in the same cells. In control cells, this revealed a positive correlation between the ppERK1/2 and N:C ERK1/2 measures (Fig. 2B). By contrast, when cells were treated for 2 hours with 1 μM PDBu, an entirely different relationship was observed, with the N:C ERK1/2 ratio decreasing with increased ppERK1/2 levels but remaining greater than that of control cells in all ppERK1/2 bins (Fig. 2B). This uncoupling of the two responses most probably reflects the ability of PDBu to increase expression of nuclear DUSP proteins that dephosphorylate and scaffold ERK in the nucleus (Caunt et al., 2008a; Caunt et al., 2008b; Mandl et al., 2005; Volmat et al., 2001). Surprisingly, however, 5 minutes stimulation with PDBu also caused a substantial increase in ERK1/2 N:C ratio over a wide range of ppERK1/2 bins in spite of the fact that this is insufficient time for increased DUSP expression (Fig. 2B). Similar data were obtained in MCF7 cells and in primary mouse embryonal fibroblast cultures. In all three cell types, data binning revealed a pronounced uncoupling of the ppERK1/2 response from the ERK1/2 N:C response after 2 hours of stimulation with 1 μM PDBu, and a less-pronounced but statistically significant uncoupling at 5 minutes (Fig. 2 and supplementary material Fig. S2).

The data outlined above demonstrate that the effects of PDBu on ERK nuclear distribution cannot solely be attributed to TEY phosphorylation because PDBu causes pronounced increases in ERK1/2 N:C in the absence of any measurable increase in ppERK1/2 (i.e. using bins with matched and indistinguishable ppERK1/2 levels). They also reveal that analysis of the subset of cells within a relatively narrow ppERK1/2 range provides a simple means of quantifying this component in isolation. We next stimulated cells for 5 minutes with varied concentrations of PDBu and compared population-average responses with those in cell subsets matched for ppERK1/2 levels. With population-averaged data, PDBu caused the expected concentration-dependent increases in ppERK1/2 and N:C ERK1/2 ratio with comparable potency for both effects (log EC₅₀ values −6.9±0.2 and −7.2±0.2, respectively; Fig. 3). When similar plots were generated only for cells within a narrow range of ppERK1/2 levels (240–280 AFU), PDBu caused no measurable increase in ppERK1/2 but did cause a concentration-dependent increase in N:C ERK1/2 ratio. These data reveal that PDBu-mediated nuclear localization of ERK1/2 is proportional to stimulus, even under conditions where it is not proportional to ppERK1/2 level. Thus, we show for the first time the extent to which mechanisms other than phosphorylation-dependent release from MEK are employed to achieve the full ERK nuclear localization response elicited by PKC activation.

We also determined the time-course of PDBu action by stimulating cells with 1 μM PDBu for 5 minutes to 6 hours and generated single-cell frequency-distribution plots (for ppERK1/2 and ERK1/2 N:C) and found, as expected, that the mode values for ppERK1/2 and ERK1/2 N:C paralleled the population-averaged responses (supplementary material Fig. S1 and Fig. 4A). We also binned the data according to ppERK1/2 level and plotted this against ERK1/2 N:C (as above), and this revealed a clear uncoupling of the responses at all time-points (supplementary material Fig. S1). This can be visualized readily by plotting the time-course of the PDBu effect using only cells within a single relatively low ppERK1/2 bin (240–320 AFU). As shown (Fig. 4), PDBu caused a pronounced increase in N:C ERK1/2 ratio in this subpopulation in spite of the fact that it caused no measurable increase in ppERK1/2. The time-courses of effects on ERK1/2 distribution were very similar in the binned and population-averaged responses, although the transient increase in ppERK1/2 levels was clearly seen in the population-averaged data (compare Fig. 4A,B).

We next used the same approach to explore stimulus specificity. Population-averaged data revealed that epidermal growth factor (EGF) causes a transient increase in ppERK1/2 and also increases the N:C ERK1/2 ratio. As in earlier work (Caunt et al., 2008a; Caunt et al., 2008b; Whitehurst et al., 2004), the magnitude and kinetics of the EGF effect on ppERK1/2 levels were similar to those seen with PDBu, but EGF was less effective than PDBu at driving ERK1/2 nuclear localization (Fig. 5A). The tyrosine phosphatase inhibitor sodium orthovanadate (vanadate) caused a slow and near-linear increase in ppERK1/2 (without any increase in N:C ERK1/2 ratio), demonstrating involvement of tyrosine phosphatases in basal ppERK1/2 turnover (Fig. 5A). We also binned the single-cell data according to ppERK1/2 level in order to compare effects of these stimuli under conditions with comparable average ppERK1/2 levels (5 minutes stimulation for PDBu and EGF, and 6 hours for vanadate). This revealed the expected positive correlation between ppERK1/2 levels and N:C ERK1/2 ratio in control cells and clear uncoupling of the responses, with PDBu increasing N:C ERK1/2 as compared with control cells matched for ppERK1/2. Similar uncoupling was seen in EGF-stimulated cells, but not in response to vanadate, where only a modest elevation in ERK1/2 N:C occurred, even at very high ppERK1/2 levels (Fig. 5B). These data echo earlier results showing that ERK nuclear accumulation is dictated by stimulus type (Whitehurst et al., 2004) and support the idea that stimulus-induced and basal trafficking of ERK operate through distinct mechanisms (Casar et al., 2007).

We next used a range of pharmacological inhibitors to explore the possible mechanisms underlying ERK activation and nuclear localization. Using maximally effective inhibitor concentrations established in control assays (data not shown), we treated HeLa cells for 10 minutes before acute (5 minute) stimulation with PDBu. Population-averaged data revealed that the effects of PDBu on ppERK1/2 and N:C ERK1/2 ratio were dependent upon PKC and MEK (inhibited by Ro31-8425 and PD184352) and independent of Src (uninfluenced by SU6656). The responses were also unaltered by vanadate or cycloheximide (CHX), demonstrating that tyrosine phosphatases and protein neosynthesis do not influence these acute responses [although they do alter responses to long-term PDBu stimulation in this model; unpublished data and the findings of Caunt and colleagues (Caunt et al., 2008a; Caunt et al., 2008b)]. Focusing on cell subsets with a narrow range of ppERK1/2 levels (240–320 AFU) again revealed a dose-dependent effect of PDBu on ERK1/2 N:C ratio (Fig. 6B). This nuclear localization response was completely blocked by Ro31-8425 and PD184352, demonstrating dependence on PKC and MEK, respectively. However, it was uninfluenced by SU6656, vanadate and CHX, demonstrating independence from Src, tyrosine phosphatase activity and protein neosynthesis. These effects were seen under conditions matched for low ppERK1/2 levels and in which ppERK1/2 was not measurably altered by the stimulus or the inhibitors (Fig. 6B, lower-right panel). Thus, the component of the acute PDBu-induced translocation response that cannot be solely attributed to TEY phosphorylation of ERK1/2 is nevertheless dependent on MEK activation. This might explain why this component is seen with stimuli that activate MEK (PDBu and EGF) but not with a treatment that increases ppERK1/2 by phosphatase inhibition (vanadate) (Fig. 5B).

To explore further the relationships between ERK phosphorylation and localization, we used a previously characterised system in which siRNAs are used to reduce expression of endogenous ERK, and recombinant adenoviruses (Ad) are used to introduce GFP-tagged wild-type (WT) or mutated ERK2 reporters. As we show (Fig. 7A), the siRNAs reduced endogenous ERK levels by >95%. This was paralleled by reduced PDBu-stimulated ERK phosphorylation (as measured by cell imaging or by western blotting) and Egr-1–Luc activity (early growth response-1 promoter coupled to a luciferase reporter, used as a transcriptional readout for ERK activation). Addition of Ad ERK2–GFP restored ERK2 expression levels, as well as PDBu-stimulated ERK2 phosphorylation and Egr-1–Luc responses to wild-type levels [Fig. 7A, data not shown and the data of Caunt and colleagues (Caunt et al., 2008a; Caunt et al., 2008b)]. A similar approach was used to add-back mutated ERK2–GFP constructs (after siRNA knockdown), and western blotting revealed comparable expression levels for WT ERK2–GFP and for reporters containing K52R (catalytically inactive), T183–Y185A (non-TEY phosphorylatable), Y261A (deficient in binding DEF domains) and D319N (deficient in binding D-domains) variants of ERK2–GFP. None of the mutations inhibited MEK expression or PDBu-stimulated MEK phosphorylation (Fig. 7A).

Imaging assays (population-averaged data) were then used to define the effects of these mutations on responses to PDBu. As we show (Fig. 7B), stimulation for 5 minutes with PDBu caused a dose-dependent increase in ERK2–GFP phosphorylation (whole-cell ppERK2) and nuclear localization (ERK2–GFP N:C), and stimulation for 6 hours caused a corresponding increase in Egr-1–Luc activity. All three responses were reduced markedly in cells expressing the non-TEY-phosphorylatable T183–Y185A ERK2–GFP mutant, confirming the effectiveness of the mutation as well as the dependence of nuclear localization and nuclear signaling on TEY phosphorylation. PDBu also dose dependently increased phosphorylation and nuclear localization of the K52R mutant, but the Egr-1 response was greatly reduced, demonstrating that activation of the transcription reporter requires ERK catalytic activity, whereas the phosphorylation and nuclear localization responses do not (Gonzalez et al., 1993). Interestingly, the basal ppERK2 levels and ERK2–GFP N:C ratio were increased with this mutant (Fig. 7B), which might well reflect a lack of negative feedback that is dependent upon ERK catalytic activity (Catalanotti et al., 2009; Dougherty et al., 2005). The Y261A mutation, which inhibits DEF-domain-dependent binding, did not affect ERK2–GFP phosphorylation in response to PDBu but did increase N:C ERK2–GFP ratios in basal and stimulated cells, and actually reduced PDBu-stimulated Egr-1–Luc activity (Fig. 7B). The latter effect is known to reflect dependence on DEF-directed phosphorylation and activation of immediate-early gene transcription factors (Dimitri et al., 2005; Murphy et al., 2004; Murphy et al., 2002). By contrast, inhibition of ERK2 binding to D-domain partners (D319N mutation) did not affect ERK2–GFP phosphorylation but substantially reduced nuclear localization and doubled Egr-1–Luc responses (Fig. 6B). Together, these findings show that each of these biochemical motifs is necessary for correct regulation of the ERK1/2 localization and downstream response and confirm that effects of catalysis and docking interactions can be functionally separated in vivo using this system.

We next asked whether the uncoupling of ERK phosphorylation from nuclear localization seen with endogenous ERK would also be seen with the knockdown/add-back model. To do so, cells expressing endogenous ERK were stimulated for 5 minutes with 0, 0.1 or 1 μM PDBu before staining for ppERK1/2 and ERK1/2. Cells transduced with construct encoding ERK2–GFP were treated similarly (except that the GFP fluorescence was used instead of total ERK1/2 staining), and the single-cell data were binned according to ppERK1/2 level. In control cells expressing endogenous ERK, the expected positive correlation occurred between ppERK1/2 and ERK1/2 N:C, as did the marked uncoupling of responses (i.e. the PDBu-induced increase in ERK1/2 N:C in cell subsets with matched ppERK1/2 level). Very similar data were obtained in cells transduced with ERK2–GFP (Fig. 8).

The previous data show that the GFP tag does not interfere with the component of the nuclear localization response that cannot be attributed to TEY phosphorylation and provided the opportunity to explore mechanisms using the ERK2–GFP mutants. Accordingly, we compared the effect of the ERK2–GFP mutants on stimulus-dependent localization within matched ppERK2 levels. This analysis clearly cannot be performed with the T183–Y185A mutant (which lacks the TEY phosphorylation epitope). As expected, there was clear uncoupling of nuclear localization from phosphorylation in cells expressing WT ERK2–GFP and stimulated for 5 minutes with 0.1 or 1 μM PDBu. Very similar data were obtained in cells expressing the K52R and Y261A mutants (Fig. 9A). By contrast, the uncoupling of phosphorylation from nuclear localization was much less pronounced in cells expressing D319N ERK2–GFP (Fig. 9A). This distinction was also evident in dose–response studies, where data are plotted for the subset of cells within a single narrow ppERK2 bin (160–240 AFU). This revealed a clear dose-dependent increase in N:C ERK2–GFP under conditions where there was no measurable PDBu-stimulated increase in ppERK2 (Fig. 9B). The nuclear localization response seen with the binned data was greatly reduced in cells expressing D319N ERK2–GFP. Consistent with the population-averaged data (Fig. 7), we also found that the K52R and Y261A mutants increased ERK2–GFP N:C ratios in the absence of stimulus, and this was additive with the PDBu effects. Together, these data demonstrate that there is an additional component of the nuclear localization response that cannot be attributed to TEY phosphorylation, does not require catalysis or DEF-domain-dependent binding but is largely prevented by the D319N mutation and therefore requires D-domain-dependent binding.

The compartmentalization of ERK dictates substrate access and, in turn, controls cell fate decisions. ERK contains neither nuclear localization nor export sequences, and so its compartmentalization is determined largely by binding to scaffolds, anchors and substrates (Caunt et al., 2006a). In unstimulated cells, ERK is bound mainly to MEK in the cytoplasm (Chuderland et al., 2008b; Fukuda et al., 1997). When MEK catalyses the TEY phosphorylation of ERK, this not only activates the enzyme but also causes its dissociation from MEK and exposes binding motifs enabling its interaction with other binding partners, including components of the nuclear transport pore (Adachi et al., 1999; Lee et al., 2004; Wolf et al., 2001; Yazicioglu et al., 2007). The importance of MEK for cytoplasmic retention of ERK in unstimulated cells is illustrated by the characteristic redistribution of ERK from the cytoplasm to the nucleus occurring when MEK is activated (Adachi et al., 1999; Chen et al., 1992; Lenormand et al., 1993). TEY phosphorylation of ERK might be insufficient for maximal stimulation of nuclear ERK localization as additional phosphorylation sites (Ser244, Ser246 and Thr188) are also implicated in the control of nuclear localization (Chuderland et al., 2008a; Lorenz et al., 2009). Moreover, ERK can be largely restricted to the cytoplasm by elevation of the cytoplasmic Ca²⁺ concentration, which can inhibit its transport into the nucleus (Chuderland et al., 2008b) and by engaging signaling pathways in which β-arrestin binding retains ERK in the cytoplasm (Caunt et al., 2006b; Luttrell et al., 2001; Tohgo et al., 2002). Conversely, many activators of ERK also increase expression of nuclear-inducible DUSP proteins that can dephosphorylate and anchor ERK, thereby increasing the proportion of ERK in the nucleus (Caunt et al., 2008a; Caunt et al., 2008b; Mandl et al., 2005; Volmat et al., 2001).

The data above illustrate the potential for uncoupling of ERK phosphorylation from nuclear localization, but most of the supporting evidence derives from studies of large cell populations (i.e. homogenates of cultured cells) or imaging of small numbers of cells. Here, we have developed a novel approach based on automated fluorescence microscopy to determine the relationship between ERK phosphorylation and compartmentalization in large numbers of individual cells. A useful feature of this approach is that it enables the cells to be sorted (binned) into subpopulations according to their ppERK1/2 level. Plotting ppERK1/2 level (bin mean) against nuclear:cytoplasmic (N:C) ERK1/2 ratio then provides a simple measure of the relationship between phosphorylation and nuclear localization and also enables comparison of control and stimulated cells under conditions of matched ppERK1/2 level. This analysis revealed the expected positive correlation between ppERK1/2 and ERK1/2 N:C ratios in control cells (Fig. 2B). We suspected that stimulation would simply extend this relationship, driving more cells into the higher ppERK1/2 bins and causing a corresponding increase in N:C ERK1/2. This is precisely what was seen when tyrosine phosphatases were inhibited with vanadate, but the relationship between ppERK1/2 and N:C ERK1/2 was entirely different in cells treated with PDBu for 2 hours (Fig. 2B). Under this condition, increasing ppERK1/2 did not increase the N:C ERK1/2 ratio, and importantly the N:C ERK1/2 ratio was greater in PDBu-stimulated cells than in control cells in all ppERK1/2 bins. Thus, exposure to PDBu for 2 hours causes two distinct effects: first, it increases TEY phosphorylation of ERK1/2, as evidenced by the increase ppERK1/2 seen in analysis of the whole-cell population (Fig. 1) and by the increased mode value seen in frequency distribution plots (supplementary material Fig. S1); second, it causes an increase in N:C ERK1/2 ratio that is not attributable to the increased TEY phosphorylation because it occurs under conditions matched for ppERK1/2 levels (Fig. 2B). Similar uncoupling (i.e. stimulus-induced increases in N:C ERK1/2 ratio with cells binned for indistinguishable ppERK1/2 levels) was seen at early time-points (5 minutes) and in three different cell types (HeLa, MCF7 and mouse embryonal fibroblasts), although, in all cases, it was less pronounced at 5 minutes (Fig. 2 and supplementary material Fig. S2).

To our knowledge, this is the first study in which ppERK levels and ERK compartmentalization have been related in large numbers of individual adherent cells (data herein derive from measures of ppERK and N:C ERK ratio in several million individual cells). However, a key earlier study used flow cytometry to explore the issue of graded versus switch-like signalling in the ERK pathway (MacKeigan et al., 2005). This revealed that the PDGF-stimulated ppERK1/2 response in Swiss 3T3 fibroblasts was graded, whereas c-Fos induction was an all-or-nothing response at the individual-cell level (MacKeigan et al., 2005). As c-Fos induction was used as a readout for nuclear ERK activity, it was suggested that graded ERK phosphorylation stimulated all-or-nothing nuclear localization. This provides another potential explanation for uncoupling of ppERK1/2 and N:C ERK1/2 responses – if maximal nuclear localization occurred at low ppERK levels, further increases in ppERK1/2 would have no effect on nuclear localization. However, this appears not to be the case in our model as we see graded responses to PDBu for both ERK phosphorylation and nuclear localization (Fig. 2). Moreover, any such relationship would be expected to yield a step-wise increase in binned data plots of ppERK1/2 against N:C ERK1/2, and no such increase was seen in control cells or in the presence of any of the stimuli used (Fig. 5). Interestingly, we observed graded N:C ERK1/2 and graded nuclear ppERK1/2 responses in HeLa and MCF7 cells (Fig. 2, and data not shown) suggesting that, where ERK drives switch-like downstream responses (e.g. cell cycle entry and apoptosis), the transduction from graded to switch-like behaviour occurs distal to TEY phosphorylation and nuclear localization of ERK.

A key finding of this study is that the relationship between ppERK1/2 level and N:C ERK1/2 is entirely different between unstimulated and PDBu-stimulated cells (Fig. 2); namely, that PDBu has an effect on nuclear ERK1/2 localization beyond that simply attributable to TEY phosphorylation of ERK (Figs 3 and 4). As we have imaged fixed cells, it is not possible to know whether PKC activation does so by altering ERK trafficking to or from the nucleus or through effects on anchoring in the cytoplasm or nucleus. However, the uncoupling at later time-points is most probably due to the PKC-mediated increase in expression of nuclear-inducible DUSP proteins (Caunt et al., 2008a; Caunt et al., 2008b). Nevertheless, this clearly cannot be the case for the similar uncoupling seen in cells stimulated for only 5 minutes with PDBu (Fig. 2C, Figs 3, 4 and supplementary material Fig. S2). In confirmation of this, the presence of protein synthesis or tyrosine phosphatase inhibitors failed to have any effect on PDBu-mediated ERK redistribution in HeLa cells at these early time-points but did affect signaling in sustained phases [Fig. 5 (Caunt et al., 2008a; Caunt et al., 2008b) and data not shown]. Thus, we have found a previously unidentified component of ERK signaling downstream of PKC – a TEY-phosphorylation-unattributable nuclear localization response that does not reflect PKC-induced expression of nuclear anchors. Indeed, our observation of an acute TEY-phosphorylation-unattributable nuclear localization response is not readily explained by any of the above-described mechansisms for uncoupling of TEY phosphorylation from nuclear localization (Chuderland et al., 2008a; Chuderland et al., 2008b; Lorenz et al., 2009; Luttrell et al., 2001; Tohgo et al., 2002; Caunt et al., 2008a; Caunt et al., 2008b; Mandl et al., 2005; Volmat et al., 2001).

This additional component of PKC-mediated ERK nuclear localization was also seen using a knockdown/add-back model in which siRNAs are used to reduce expression of endogenous ERK and recombinant adenovirus is used to express ERK2–GFP reporters (Figs 8 and 9). This enabled the use of pharmacological inhibitors to probe mechanisms in cells with endogenous ERK and mutated ERK2–GFP reporters in the knockdown/add-back model. This revealed that the uncoupling of ERK phosphorylation from nuclear localization was prevented either by inhibition of MEK (PD184352) or by expression of a non-TEY-phosphorylatable (T183–Y185A) mutant. Thus, the phosphorylation-unattributable component of the response cannot be described as phosphorylation independent. Instead, we show that MEK-catalysed TEY phosphorylation of ERK is necessary, but not sufficient, for the full ERK nuclear localization response.

Comparison of cells treated with PDBu, EGF and vanadate revealed that the ERK nuclear localization response is governed by stimulus rather than phosphorylation levels. PDBu and EGF both caused phosphorylation-unattributable nuclear localization of ERK (Fig. 4). The EGF response was not affected by a PKC inhibitor (Ro 31-8425), indicating that this component of the ERK nuclear localization response can be induced by more than just PKC activation (data not shown). By contrast, vanadate was unable to cause this component of the nuclear localization response, despite causing robust increases in ERK phosphorylation (Fig. 4). These data reveal that simply increasing the levels of phosphorylated ERK does not necessarily provoke this additional component of the nuclear localization response. Accordingly, they add to the growing body of evidence showing that compartmentalization of ERK is stimulus specific (von Kriegsheim et al., 2009; Whitehurst et al., 2004).

In contrast to mutation of the TEY motif, we found that inhibition of catalysis (using a K52R mutation) or inhibition of DEF-domain binding (using a Y261A mutation) increased nuclear localization of ERK2–GFP in basal and stimulated cells. This was true in both average cell populations and in matched-phosphorylation bins (Fig. 7B and Fig. 9). This could be because K52R ERK2–GFP is not retained in the cytosol, as suggested by studies showing that the K52R mutant inhibits ERK traffic to the ruffling membrane but not the nucleus (Gonzalez et al., 1993; Yazicioglu et al., 2007). Previous studies have also implicated DEF-domain interactions in basal nuclear traffic of ERK (Casar et al., 2007), but exactly why K52R or Y261A mutants distinguish basal from stimulus-regulated ERK2–GFP localization patterns is currently unclear. More importantly in the context of this study however, we were able to establish that neither mutation significantly affected the phosphorylation-unattributable component of the ERK nuclear localization response (Fig. 9).

Intriguingly, we found that reduction of D-domain binding (using a D319N mutation) not only inhibited 5 minutes PDBu- or EGF-induced nuclear accumulation of ERK2–GFP (Fig. 7B, and data not shown) but also removed the phosphorylation-unattributable component of the nuclear localization response (Fig. 9). Thus, PKC and EGFR both activate pathways causing regulation of an unidentified D-domain-containing protein (or proteins) that is/are required for the full ERK nuclear localization response. In spite of this, the D319N mutation actually increased PDBu-stimulated Egr-1–luc activity, demonstrating the overriding importance of D-domain-containing phosphatases in terminating sustained ERK signaling (Caunt et al., 2008a; Caunt et al., 2008b; Mandl et al., 2005; Volmat et al., 2001). Clearly, more-specific methods will be needed to explore this additional component of the localization response without perturbation of ERK binding to phosphatases, and we are currently focusing on identifying specific D-domain-containing ERK binding partners in this model.

Taken together, we show that phosphorylation of ERK by MEK on the TEY motif is necessary, but not sufficient, for the full nuclear localization response to occur. Instead, there is an additional TEY-phosphorylation-unattributable component of the ERK translocation response that is provoked by PKC or EGFR activation and does not reflect induced expression of nuclear anchors. We also show that this component of the response is not dependent on ERK catalytic activity or DEF-domain association but is reliant on binding to stimulus-regulated D-domain-containing proteins. These studies illustrate how spatially distinct ERK domains control functionally separate aspects of stimulus-regulated subcellular targeting and will ultimately inform identification of the candidate binding partners responsible for these different modes of regulation.

Viral shuttle vectors were constructed initially by subcloning a KpnI–NotI digest of ERK2–GFP in pEGFP-N1 (a gift from L. Luttrell, Medical University of South Carolina, Charleston, SC) into a corresponding digest of pacAd5CMV K-N pA (donated by B. Davidson, University of Iowa, Iowa City, IA). K52R, T183–Y185A, Y261A and D319N mutations were introduced using a QuikChange PCR-based mutagenesis kit (Stratagene, Amsterdam, The Netherlands) and the following primers: 5′-CAA AGT TCG AGT TGC TAT CAG GAA AAT CAG TCC TTT TGA GC-3′, 5′-TGA TCA TAC AGG GTT CTT GGC AGA GGC TGT AGC CAC GCG TTG GTA C-3′, 5′-AAT TTA AAA GCT AGA AAC GCT TTG CTT TCT CTC CCG CAC-3′ and 5′-GCA GTA TTA TGA CCC AAG TAA TGA GCC CAT TGC TGA AGC-3′, along with antisense primers according to the manufacturer's recommendations and using the pacAd5CMV ERK2–GFP vector as the template. Adenovirus (Ad) Egr-1–luciferase and Ad CMV β-galactosidase reporter shuttle vectors were constructed as described previously (Caunt et al., 2008a; Armstrong et al., 2010). Viruses were generated from shuttle vectors as described previously (Anderson et al., 2000; Caunt et al., 2006b). Briefly, 4.5 μg of the shuttle vectors were digested alongside 1.5 μg of pacAd5 9.2-100 sub360 backbone vector (kindly donated by B. Davidson, University of Iowa, Iowa City, IA) with PacI. Cut shuttle and backbone vectors were then mixed and transfected into low-passage HEK293 cells using Superfect (Qiagen, Crawley, UK). Cells were left to allow recombination between shuttle and backbone vectors. Verification of recombination was performed by restriction digest and sequence analysis, and Ad vectors were grown to high titre and purified according to standard protocols.

HeLa and MCF7 cells (from ECACC) were cultured in 10% FCS-supplemented Dulbecco's modified Eagle's medium (DMEM) without sodium pyruvate. Mouse embryonal fibroblasts (a gift from S. Keyse, University of Dundee, UK) were cultured in 10% FCS-supplemented DMEM with sodium pyruvate. For 96-well plate experiments, cells were harvested by trypsinization and seeded at 3–5×10³ cells per well. Where necessary, cells were transfected using RNAiMAX reagent (Invitrogen, Paisley, UK) and the manufacturer's reverse transfection protocol. Cells were transfected with two siRNA duplexes (Qiagen) each for ERK1: 5′-CGU CUA AUA UAU AAA UAU AdTdT-3′, 5′-UAU AUU UAU AUA UUA GAC GdGdG-3′ and 5′-CCC UGA CCC GUC UAA UAU AdTdT-3′, 5′-UAU AUU AGA CGG GUC AGG GdAdG-3′ and for ERK2: 5′-CAC UUG UCA AGA AGC GUU AdTdT-3′, 5′-UAA CGC UUC UUG ACA AGU GdTdT-3′ and 5′-CAU GGU AGU CAC UAA CAU AdTdT-3′, 5′-UAU GUU AGU GAC UAC CAU GdAdT-3′, which have been validated in recent publications (Caunt et al., 2008a; Caunt et al., 2008b; Dimitri et al., 2005). A mixture of all four ERK1–ERK2 duplexes or control siRNA against GFP (Ambion, Warrington, UK) was used in experiments at 2.5 nM total concentrations. Sixteen hours after siRNA transfection, cells were transduced with 0 or 1.5×10⁶ plaque-forming units (pfu)/ml Ad WT, K52R, T183–Y185A, Y261A or D319N ERK2–GFP vector in DMEM with 10% FCS. The Ad-containing medium was removed after 4–6 hours and replaced with fresh DMEM supplemented with 0.1% FCS. The cells were then maintained for 16–24 hours in culture before stimulation with GnRH, EGF (Calbiochem, San Diego, CA, USA), PDBu (phorbol 12, 13-dibutyrate, Sigma) or sodium orthovanadate (Sigma). In some experiments, cells were treated with 10 μM PD184352 (Enzo Life Sciences, Exeter, UK), 200 nM Ro31-8425 (Roche, Welyn Garden City, UK) 30 μM cycloheximide (Sigma) or 1 mM sodium orthovanadate (Sigma) for 10 minutes before stimulation. Expression levels of GFP-tagged fusions were compared using western blotting techniques (see Fig. 6A) as well as comparison of mean cell fluorescence in microscopy assays.

HeLa cells were plated in six-well plates at 2.5×10⁵ cells per well. Where required, cells were transfected with 2.5 nM ERK1/2 siRNAs before Ad transduction, as described above. Following the treatments noted in the figure legends, cells were lysed, as described, before western blotting. Total and ppERK1/2 were detected using rabbit anti-ERK1/2 monoclonal (clone 137F5, 1:1000; Cell Signaling Technology, Hitchin, UK) and mouse anti-ppERK1/2 monoclonal antibody (clone MAPK-YT, 1:2000, Sigma), respectively. Total MEK1 and MEK2 (hereafter referred to as MEK1/2) was detected using rabbit anti-MEK1/2 monoclonal (clone 47E6, 1:1000, Cell Signaling Technology) and phospho Ser217 and Ser221 MEK1/2 was detected using rabbit anti-ppMEK1/2 monoclonal (clone 41G9, 1:1000, Cell Signaling Technology), respectively. DUSP1 was detected using rabbit anti-DUSP1 polyclonal (M-18, 1:500, Santa Cruz Biotechnology, CA). Antibodies were visualized using horseradish-peroxidase-linked secondary antibody and enhanced chemiluminescence kit (GE Healthcare, Amersham, UK).

Cells were plated (and transfected with siRNA, and Ad vectors as necessary) as described above on Costar plain black-wall 96-well plates (Corning, Arlington, UK) before the treatments as noted in the figure legends. Cells were fixed in 4% paraformaldehyde in PBS and permeabilized in −20°C methanol. After blocking in 5% normal goat-serum in PBS, cells were probed with mouse anti-ppERK1/2 monoclonal antibody (clone MAPK-YT, 1:200, Sigma) and rabbit anti-ERK1/2 monoclonal (clone 137F5, 1:100, Cell Signaling Technology) in PBS. Alexa-488-conjugated goat anti-mouse and Alexa-546-conjugated goat anti-rabbit secondary antibodies (1:200, Invitrogen) and DAPI in PBS (600 nM) were used to visualize ppERK1/2, ERK1/2 antibodies and stained nuclei, respectively, in imaging of endogenous kinases. For imaging cells transduced with Ad ERK2–GFP, cells were counterstained with rabbit anti-ERK1/2 monoclonal (1:100) and Alexa-546-conjugated goat anti-rabbit secondary (1:200) for assessing expression levels and setting filter levels (to ensure only ERK2–GFP-expressing cells within physiological ranges were included in analysis in subsequent assays). For imaging ppERK2 in cells expressing ERK2–GFP, cells were counterstained with mouse anti ppERK1/2 monoclonal (1:200) and Alexa-546-conjugated goat anti-mouse secondary (1:200), as described above. Image acquisition in each well was performed on an IN Cell Analyzer 1000 (GE Healthcare) microscope, using a ×10 objective and 360 nm (DAPI), 475 nm (Alexa 488 and GFP) and 535 nm (Alexa 546) excitation filters and monitored through 460 nm, 535 nm and 620 nm emission filters, respectively, with a 61002 trichroic mirror (GE Healthcare). Analysis of endogenous ERK1/2 and ppERK1/2 was performed using the Multi-target Analysis algorithm and ‘region growing’ cell identification module in the IN Cell Analyzer Workstation (GE Healthcare) using DAPI and ERK1/2-stained images to define nuclear and cytoplasmic regions, respectively, which were used as a mask for detection of changes in ppERK1/2 staining. ERK2–GFP- and ERK1/2- or ppERK1/2-stained cells were similarly analysed using the Multi-target Analysis algorithm, but using the more sensitive ‘multiscale top hat’ cell identification module (IN Cell Analyzer Workstation, GE Healthcare). In both ppERK2 and GFP readouts, cells expressing sub- or super-physiological levels of ERK2–GFP were excluded from analysis (based on comparisons of frequency histograms of ERK1/2 staining intensity in cells transfected with control siRNA, ERK1/2 siRNAs and with ERK1/2 siRNAs plus Ad ERK2–GFP). Mitotic and apoptotic nuclei were excluded from analysis, as were aberrantly identified cells using a ‘decision tree’ filter based on multiple readouts of intensity and shape defined in nuclear and cellular regions. Imaging data are reported as ppERK1/2 intensity (mean fluorescence intensity per cell) or as a ratio of nuclear to cytoplasmic stain intensity (N:C ratio). Sorting of cells into linked subpopulations according to ppERK1/2 staining intensity was performed using a macro in Microsoft Excel (kindly designed by N. Rich, University of Bristol, UK).

Cells were transfected with siRNA, transduced with Ad vectors and plated, as described above, on Costar plain black-wall 96-well plates (Corning) but including Ad Egr-1–luciferase and Ad CMV β-galactosidase reporter vectors. Following treatment with PDBu, cells were washed in ice-cold PBS, lysed and assessed for luciferase activity by chemical luminescence following the addition of the luciferin substrate (Promega, Southampton, UK). β-galactosidase activity was used to correct luciferase activity for transfection efficiency, as measured following the addition of chlorophenol red-β-d-galactopyranoside substrate (Roche).

We thank Stephen Keyse for his kind gift of mouse embryo fibroblasts, Beverly Davidson for the donation of viral shuttle vectors and Neil Rich for his help in data analysis. This work was funded by Wellcome Trust grants 084588 and 078407. Deposited in PMC for release after 6 months.