Signaling through the mitogen activated protein kinase (MAPK) network affects virtually every major aspect of cell physiology and is among the most intensely studied cell signaling pathways in biology. Multiple MAPK regulatory features related to signal strength, signal duration, compartmental localization, coordination through scaffolds/inhibitors, and dynamic behaviors have been characterized.¹⁻⁵ Four MAPK branches have been defined and are referenced according to their terminal kinase (p38, extracellular signal regulated kinase (ERK), JNK, and ERK5).⁶ MAPK branches are frequently associated with opposing cellular programs, such as p38/JNK regulating differentiation, stress-responsive signaling/apoptosis, and ERK/ERK5 regulating mitogenic signaling.^7,8 Exceptions have been observed, such as a role for p38 in proliferation^9,10 or ERK in cell death,¹¹ making it difficult to rely on measurements of kinase activities for predicting biological response.

Our group has defined oscillatory behavior in the ERK pathway,¹² and the function of this dynamic behavior in cells and tissues is currently unknown. ERK oscillations are observed in primary cell culture and become deregulated with prolonged passage.¹³ The mechanisms by which ERK oscillations are deregulated in vitro remain unclear; however, these observations suggest that the oscillating phenotype is under-represented in cell culture models used for toxicological investigations. ERK oscillations regulate a tumor suppressor activity that is conserved in vitro/in vivo,¹⁴ suggesting that oscillatory behavior can significantly impact biological outcomes. Therefore, it is prudent to consider this regulatory feature in toxicological investigations, and there is a need to bridge the molecular dynamics observed in vitro with complex biology in tissues to improve the interpretation of experimental models used for hazard assessment and risk prediction. ERK oscillations are measured at the single cell level in vitro, which is not experimentally tractable in animal models. In contrast, genes that are uniquely regulated by ERK oscillations could provide a bridge that can be examined in vivo by standard molecular/histological methods.

To date, the molecular features regulating canonical ERK signaling also regulate oscillatory behavior.^5,12 Stress-responsive signaling is known to antagonize the ERK pathway,¹⁵ but this cross-talk has not yet been demonstrated to regulate oscillatory behavior and may be of toxicological interest. As a first step toward advancing our understanding of ERK dynamics in complex tissues, we have defined gene expression patterns regulated by ERK oscillations and provide an initial assessment of regulatory nodes that may mediate oscillation-specific gene expression that are also targets of stress-responsive signaling.

Three independent cell strains have been subcloned from an hTERT immortalized normal human keratinocyte cell line. Strain #1 exhibits transient ERK activation induced by TGFα, without detectable oscillations (Figure 1). Strain #2 exhibits persistent ERK oscillations that are dependent on stimulation with exogenous TGFα. Strain #3 exhibits spontaneous ERK oscillations in the absence of exogenous TGFα. An EGFR neutralizing antibody (mAb 225)¹⁸ fully inhibits ERK oscillations in strains #2 and #3 (Figure 1), indicating dependence on the EGFR in both cases. The time period for ERK oscillations is slightly prolonged in strain #3, as compared with strain #2, as might be expected for an autocrine process requiring the shedding of the ligand. However, we cannot distinguish whether strain #3 is regulating ERK oscillations via an EGFR autocrine loop or a mutant EGFR based on results using mAb225 alone. Collectively, all cell strains share the same genetic background and EGFR-dependent ERK activation, which is an ideal scenario for direct comparison of groups in a microarray experiment.

DNA microarray was used to define transcriptomics of the human keratinocyte cell lines described in Figure 1. Conditions for high and low ERK activity were defined for each cell strain as described in Materials and Methods, and gene expression patterns for the low ERK activity condition were subtracted from the high ERK activity condition to enrich for ERK-dependent gene expression patterns. We then defined the genes that were common or unique to each cell strain from this enriched data set. As illustrated in the Venn diagram (Figure 2A), we identified 45 genes that are common to the two oscillation proficient strains (#2 and #3) and are absent in the oscillation deficient strain (#1). The Metacore annotated database was used to predict the transcription factor networks encompassed within the 45 genes common to ERK oscillations (Figure 2B). Network reconstruction suggests that ERK oscillations are coupled to transcription factor networks through the MED1 coactivator and CD44.

Studies were conducted to determine whether an index of MED1 activation showed differential regulation under oscillating vs nonoscillating conditions as validation. ERK phosphorylates MED1 on T¹⁴⁵⁷, which increases MED1 stability and activity,¹⁹ and we examined MED1 T¹⁴⁵⁷ phosphorylation status (P-MED1) by Western blot analysis in cell strains #1–3. P-MED1 levels were below detection in strains #1 and #2 under serum-starved conditions, and treatment with EGF for 10 min resulted in a dramatic increase in P-MED1 levels (Figure 3A). At 8 h post-EGF stimulation, cell strain #2 (ligand-induced oscillations) displayed higher P-MED1 levels compared with that of cell strain #1 (nonoscillating). Cell strain #3 (spontaneous oscillations) displayed constitutive P-MED1 levels that were reduced by treatment with mAb 225 or ERK inhibitor (PD98059) for 60 min (Figure 3B). Thus, oscillatory behavior in the ERK pathway is associated with increased P-MED1 status at an established ERK-responsive site.

We subsequently considered the interplay between growth- and stress-responsive signaling on the ERK-MED1 axis. Stress-responsive p38 signaling can antagonize the ERK pathway,¹⁵ and high oxygen tension in vitro can induce tonic stress.²⁰ Thus, we asked whether p38-dependent antagonism of ERK oscillations was apparent under standard cell culture conditions. TGFα-stimulated keratinocytes were treated with DMSO or 1 μM SB203580 (p38 inhibitor), and ERK oscillation characteristics were quantified. Treatment with SB203580 resulted in a marked increase in TGFα-dependent ERK oscillation amplitudes but did not alter the time period, rise time, or decay time (Figure 4A, white bar; rise/decay time not shown). Additional evidence that was consistent with p38-dependent negative feedback included a small increase in the number of TGFα-stimulated cells displaying ERK oscillations (45% of population for TGFα-stimulated cells → 53% of population in cells treated with TGFα + SB203580). Treatment of keratinocytes with vehicle (DMSO) or a JNK inhibitor (100 nM SP600125)²¹ did not alter ERK oscillation characteristics in the presence or absence of TGFα (data not shown). We then confirmed that SB203580 increased ERK oscillation amplitudes at the single cell level. ERK oscillations were monitored in TGFα-treated human keratinocytes for approximately 1 h prior to the addition of 1 μM SB203580. ERK oscillation amplitudes were clearly increased at the single cell level following SB203580 treatment, as compared with the pretreatment time period (Figure 4B), confirming the effects of the p38 inhibitor on oscillation amplitudes. In mathematical terms, ERK oscillations are classified into two categories, termed clean and noisy oscillations.²² At present, it is unclear whether any biological difference exists between clean and noisy oscillators, and Figure 4B illustrates that both clean (bottom trace) and noisy (top trace) oscillators show a comparable response to the p38 inhibitor.

The increase in ERK oscillation amplitude induced by the p38 inhibitor suggests increased ERK activation under this condition since this index requires a quantitative increase in nuclear ERK-GFP protein levels. We next examined nuclear P-MED1 levels to determine whether the ERK-MED1 axis was regulated in a similar manner. Human keratinocytes were treated with SB203580 for 5–10 min, and nuclear P-ERK and P-MED1 levels were defined by epifluorescence microscopy following in situ staining with the appropriate antibodies. SB203580 treatment resulted in a significant increase in nuclear P-ERK levels (Figure 5A square), which preceded increased nuclear P-MED1 levels (Figure 5A, circle). P-MED1 and MED1 protein levels in nuclear extracts were increased following treatment of human keratinocytes with SB203580 for 20–60 min, while actin levels served as the loading control (Figure 5B). The increase in P-MED1 protein level induced by the p38 inhibitor is suppressed by an ERK inhibitor (PD98059) but not by a JNK inhibitor (SP600125) (Figure 5C). Thus, p38 inhibition enhances the ERK-dependent phosphorylation and stabilization of MED1.

Bromate is a probable human carcinogen produced as a disinfection byproduct from the ozonation of water containing bromide. The stress response to bromate has been carefully characterized and includes activation of p38.^23,24 ERK oscillations in human keratinocytes and JB6 cells treated with 1 mM bromate were inhibited following bromate treatment (Figure 6). Western blot analysis demonstrated that 1 mM bromate increased phospho-p38 (P-p38) and decreased P-MED1 protein levels at 60 min in both model systems. Basal P-p38 levels were clearly detectable in human keratinocytes, consistent with results illustrated in Figure 4 showing that p38 inhibitor alone increases ERK oscillation amplitudes in human keratinocytes. We have observed that JB6 cells are highly sensitive to hydrogen peroxide-mediated toxicity as compared with other in vitro models we have used (Weber et al., unpublished observation). Bromate treatment induced the accumulation of ERK-GFP in the nucleus immediately following the cessation of oscillations to a greater extent in JB6 cells as compared with human keratinocytes. Although the mechanism for nuclear accumulation is beyond our present scope, bromate induces toxicity via the generation of an oxidative stress.²³ Oxidative stress inhibits nuclear export regulated by the CRM1 export protein,²⁵ which is relevant because CRM1 regulates the export of ERK from the nucleus.¹² Thus, it is plausible that oxidative stress may modulate multiple mechanisms (e.g., p38 and CRM1) capable of inhibiting ERK oscillations, and the antioxidant status of the cell system may be an additional variable for consideration. Movies illustrating ERK oscillations in human keratinocytes and JB6 cells before and after the addition of bromate can be found in Supporting Information (S1 and S2).

We attempted to extend our investigation to normal rat kidney cells (NRK) and primary salivary gland epithelial cells but met technical limitations that constrained our interpretation of p38-dependent cross-talk. Specifically, we used our established retroviral transduction system¹² for ERK-GFP expression in both model systems. NRK cells showed efficient transduction with ERK-GFP; however, NRK cells rapidly acquired a malignant phenotype (within 2–3 passages) characterized by a stellate morphology and loose adherence to tissue culture plates (data not shown). Safety protocols dictate that cells cannot be removed from the virus laboratory until after 3 passages, and the NRK ERK-GFP cells we were able to examine under this condition did not display ERK oscillations in response to EGF treatment (data not shown). We have shown that JB6 cells transformed to a malignant phenotype by phorbol ester treatment lose the ability to regulate oscillatory behavior;¹² therefore, the lack of oscillations in NRK cells are difficult to interpret because the transformed phenotype may inhibit this dynamic behavior. Primary salivary gland epithelial cells were also examined for comparison with immortalized cell line responses; however, the primary cells were difficult to transduce, and we were unable to generate a stable model system for analysis.

Although the status of ERK oscillations remains unclear in NRK and salivary epithelial model systems, we believe that the antagonistic cross-talk between p38 and the ERK pathways is relevant to both canonical and dynamic signaling features and used these models to further investigate the ERK-MED1 axis. P-MED1 levels are detected in NRK cells and inhibited by 10 μM PD98059 (Figure 7A), indicating that ERK-dependent regulation of MED1 was conserved in this model. Treatment of NRK cells with 0.06–0.3 mM bromate for 1 h reduced P-MED1 levels (Figure 7B), consistent with the activation of p38 by bromate in NRK cells as previously established.^14,22 P-MED1 levels remained depressed in bromate-treated cells at 24 h (data not shown). NRK cells were then treated with 0.06 mM bromate with or without pretreatment with a p38 inhibitor (10 μM SB202190). Bromate treatment reduced P-MED1 levels, and this effect was inhibited by SB202190 (Figure 7C and D).

Salivary gland epithelial cells afforded a comparison of the ERK-MED1 axis in a primary cell system. Salivary epithelial cells grow as epithelial islands and show localization of the tight junction protein ZO-1 at points of cell–cell contact (Figure 8A). Treatment of salivary epithelial cells with 1 mM bromate for 60 min reduced P-MED1 levels, and this effect was inhibited by the p38 inhibitor SB203580 (Figure 8B and C).

ERK oscillations represent a new regulatory mode in the ERK pathway^5,12 whose biological function remains unclear. The results of the present study advance our understanding of ERK oscillations through the identification of unique gene expression patterns that can be used to infer possible biological function. The sensitivity of MED1 phosphorylation status to stress-responsive signaling raises the possibility that MED1 may represent a transcriptional node for signal integration that is relevant to both canonical and dynamic ERK signaling.

ERK phosphorylates MED1 on T¹⁰³²/T¹⁴⁵⁷, which imparts two distinct functions: (1) it stimulates MED1 transcriptional activity, and (2) it increases the stability and half-life of MED1.¹⁹ Therefore, the activation of MED1 by ERK is established, and what remains to be determined is how oscillatory behavior uniquely impacts transcriptional regulation through the MED1 node. MED1 regulates the activity of a battery of transcription factors;²⁶⁻³⁰ therefore, sustained MED1 activation by ERK oscillations could impact a broad transcriptional program. The stabilization of MED1 protein levels by ERK oscillations might result in the development of a MED1 gradient (Figure 3A), and future studies will determine the relationship between MED1 protein and ERK activity levels. Molecular gradients associated with pathways exhibiting oscillatory behavior have been documented,^31,32 and morphogens demonstrate the importance of gradients as a regulatory mode in biology.^33,34 Theoretically, an increase in MED1 stoichiometry is expected to permit interaction of MED1 with low abundance transcription factors following saturation of MED1 binding sites on high abundance transcription factors. Other mechanisms are possible, and additional studies focused on the MED1 node are warranted.

p38 antagonizes canonical ERK signaling,¹⁵ and it was prudent to demonstrate that this antagonism extended to ERK oscillations and the ERK-MED1 axis (Figures 6–8). Prior work by Wiley’s group²² has demonstrated an important balance between positive and negative feedback loops in regulating oscillatory behavior. In fact, they demonstrated that treatment of cells with a phosphatase inhibitor alone in the absence of growth factor was sufficient to induce ERK oscillations. p38-dependent antagonism of ERK oscillations illustrates another example of negative feedback that is relevant to toxicological investigations in light of the broad association of p38 with stress-responsive signaling. Because ERK oscillations regulate a unique subset of genes, these observations raise the possibility that antagonistic signaling by p38 might manifest differently in cells displaying transient versus dynamic ERK signaling behaviors.

Bromate was chosen as a test agent because it is a probable human carcinogen that activates p38 via the generation of oxidative stress.^23,24 We have previously shown that reactive oxygen species and ionizing radiation inhibit ERK oscillations;¹³ therefore, the inhibition of ERK oscillations by bromate is consistent with our previous observations. Prior studies have demonstrated that a bolus dose of hydrogen peroxide inhibits ERK oscillations and induces ERK accumulation in the nucleus, while a low dose of ionizing radiation inhibits ERK oscillations without nuclear ERK accumulation.¹³ Bromate at a high dose (1 mM) inhibits ERK oscillations in association with both p38 activation and nuclear accumulation of ERK in JB6 cells (Figure 6). Thus, it appears that multiple mechanisms can contribute to the inhibition of ERK oscillations and that the p38 stress response may be activated at lower doses of free radicals, relative to other possible mechanisms, such as the inhibition of CRM1-dependent nuclear export. This may provide a useful gauge for interpreting the level of oxygen free radicals produced by toxicants in vivo, based on a comparison of p38 activation, the nuclear accumulation of ERK/CRM1, and target oscillation-specific gene expression patterns for oscillation positive cell types.

In the context of carcinogenesis, bFGF-dependent ERK oscillations regulate tumor suppressor activity in JB6 cells that is conserved in mouse skin carcinogenesis.¹⁴ We have defined gene expression patterns that are unique to bFGF in JB6 cells and not associated with the fully transformed JB6 phenotype (RT101 cells).¹⁴ We have analyzed this data set and identified tumor suppressors unique to bFGF-dependent ERK oscillations (lysyl oxidase,³⁵ ≈7-fold ↑; aspartoacylase 2,³⁶ ≈9-fold ↑; insulin-like growth factor binding protein 5,³⁷ ≈9-fold ↑; and scavenger receptor class A, member 5,³⁸ ≈7-fold ↑). In human keratinocytes (Figure 2B), ERK oscillations were uniquely associated with increased tumor suppressor mRNA expression, including PDCD6,³⁹ CD44,⁴⁰ WNT7A,⁴¹ HOPX,⁴² WNT5A,⁴³ and others. Thus, the regulation of multiple tumor suppressor genes was a common function of ERK oscillations in murine and human model systems. Antagonism of ERK oscillations might decrease tumor suppressor activity, which has obvious implications for increasing cancer risk.

In considering the ERK-MED1 axis, MED1 is known to form a complex with tumor suppressor genes such as BRCA1.⁴⁴ Alternatively, MED1 is a transcriptional coactivator for several nuclear hormone receptors, including members of the vitamin D and retinoic acid families.⁴⁵ Our experimental model systems are derived from skin, and both the vitamin D receptor⁴⁶ and retinoid receptors⁴⁷⁻⁴⁹ are implicated as tumor suppressors in skin. Therefore, increasing MED1 transcriptional coactivation is expected to increase tumor suppressor activities mediated by vitamin D and retinoic acid pathways. In the murine model system where bFGF-dependent ERK oscillations directly inhibit the cell transformation response,¹⁴ retinoid X receptor alpha mRNA expression is increased (2-fold), which could further sensitize retinoid-mediated tumor suppression. In human keratinocytes, retinoid receptor beta (RXRB) mRNA showed decreased expression with ERK oscillations, and a RXRB-causal network in human cancer has been hypothesized by independent investigators.⁵⁰ Therefore, retinoid signaling is a common theme in microarray data sets defining oscillation-specific gene expression.

In summary, we have demonstrated that oscillatory behavior in the ERK pathway couples to the regulation of a unique subset of genes, including many tumor suppressor genes. Network reconstruction suggests that ERK oscillations are decoded into a transcriptional response, at least in part, through MED1 whose activation is antagonized by p38. Our results indicate that oscillatory behavior in the ERK pathway can alter homeostatic gene regulation patterns and that the cellular response to perturbation may manifest differently in oscillating vs nonoscillating cells.