Dynamical effects of epigenetic silencing of 14-3-3σ expression

Abstract

The development and progression of malignant tumours are often due to deregulated cell cycle control involving a plethora of different molecules. Among these, tumour suppressor proteins like p53 play a crucial role. p53 induces 14-3-3σ, a multifunctional protein kinase inhibitor, centrally involved in cell cycle control and DNA damage repair after genotoxic stress. Recently, it has been shown that 14-3-3σ is epigenetically silenced in a variety of tumours, which might contribute to tumour development and progression via impaired cell cycle control. In addition, p53, its inhibitor MDM2 and 14-3-3σ form a signalling module in which 14-3-3σ positively regulates the activity of p53 through feedback regulation. Here we present a mathematical model integrating the effects of 14-3-3σ gene silencing, the dynamics of 14-3-3σ induction and compartmentalisation by genotoxic stress and the role of interacting molecules p53 and MDM2. In vitro experiments with different melanoma cell lines were performed and our mathematical model was subjected to computer simulations to analyse different scenarios of activation depending on genemethylation status and DNA damage levels. Our analysis indicates that 14-3-3σ expression is silenced by high genemethylation, but also that strong stimulation is necessary to induce 14-3-3σ expression in cases of intermediate levels of genemethylation. More intriguingly, the model suggests that epigenetic silencing of 14-3-3σ affects p53 dynamics in a synergistic way, such that the accumulative effect of partial downregulation of p53 expression and reduction of its nuclear fraction could affect drastically the activity of p53 as a transcription factor.

---

Introduction

Malignant melanoma is a highly aggressive tumour with increasing incidence.¹ Moreover, it is associated with poor prognosis in the metastatic stage. The molecular events underlying initial tumour development and further tumour progression are still poorly understood. Activating mutations were found in BRAF and NRAS oncogenes in a significant portion of primary melanomas and metastases.^2,3 Activating mutations were also found in members of the Akt signalling pathway, namely in phosphatidylinositol 3-kinase CA (PI3KCA) and Akt kinase, and inactivating mutations were found in the tumour suppressor phosphatase and tensin homolog (PTEN).³ However, these findings alone may not fully explain melanoma development and progression.⁴

Earlier reports implicated that classical tumour suppressor molecules such as p16 (CDKN2A) and retinoblastoma protein might play an additional role in malignant melanoma development.⁵ Indeed, familial melanoma patients show mutations in the CDKN2Agene in almost half of the cases. In line with this, mice genetically engineered for CDKN2A inactivation and overexpression of activated Ras develop primary melanomas and distant metastases.⁵ However, only a small percentage of non-familial melanoma patients show inactivating mutations in the CDKN2A tumour suppressor gene. Therefore, it is tempting to speculate that further tumour suppressor molecules might be involved in melanoma suppression, and loss of function of these might give rise to primary tumours as well as metastases.

In a recent large-scale gene expression analysis of primary melanomas and melanoma metastases, we identified a series of differentially expressed genes, many of which are involved in cell cycle regulation, cell adhesion and cell migration.⁶ The most significantly downregulated molecule in this study was 14-3-3σ, a member of the family of 14-3-3 proteins which comprises seven small molecules termed 14-3-3β, γ, ε, η, σ, τ, and ζ. These molecules are important regulators of intracellular signalling and cell cycle and may act as tumour suppressors.^7,8 14-3-3 proteins bind to phosphoserine and phosphothreonine residues in target proteins with the consensus motif RSXpS/TXP. The list of interacting proteins comprises protein kinase C (PKC), RAF1, small G proteins, BAD, and CDC25C.⁷ As mentioned, 14-3-3 proteins, and in particular 14-3-3σ, interact with the cell cycle machinery at different checkpoints in order to allow DNA damage repair or to establish a permanent arrest of cells that have severe damage with consecutive induction of cellular senescence or even apoptosis. By this means, 14-3-3 proteins may be involved in tumour suppression.^8,9 Earlier reports have shown that 14-3-3σ induces cell cycle arrest mainly in G2-M.⁹

Using tandem affinity purification technology, 117 different proteins were identified as putative interaction partners of 14-3-3σ.¹⁰ Many of the 92 proteins with known biological functions were involved in mitogenic signalling such as APC, A-RAF, B-RAF, and c-RAF, casein kinase II, PI3K-C2β, or cell cycle regulation such as AJUBA, WEE1, and c-TAK1. Overall, these findings were suggestive of an important role for 14-3-3σ in cell cycle control. Several lines of evidence indicate that 14-3-3σ is implicated in the pathogenesis of many malignant tumours.⁷ Firstly, 14-3-3σ is a downstream target of tumour suppressor p53, which is inactivated in almost half of all cancers.¹¹ Secondly, it is involved in a positive feedback loop for p53 expression via stabilization and inhibition of ubiquitination.^12,13 Thirdly, overexpression of 14-3-3σ leads to reduced tumourigenicity of oncogene expressing NIH 3T3 cells.^12,13 Fourthly, absence of 14-3-3σ is associated with increased genomic instability resulting in the so-called mitotic catastrophe.¹⁴ Finally, the 14-3-3σ gene is silenced in many cancers viagenemethylation, e.g. in breast cancer,¹⁵ hepatocellular carcinoma,¹⁶ ovarian¹⁷ and prostate cancer.¹⁸ Moreover, our recent results suggest that epigenetic silencing of 14-3-3σ might contribute to tumour progression in malignant melanoma via loss of cell cycle control, impaired cellular senescence programming and support of migratory capacity.¹⁹ Together, 14-3-3σ appears to play an important role in the development and progression of a variety of different tumours including malignant melanoma. To further substantiate these findings and create a full picture of the role of 14-3-3σ in malignant melanoma growth regulation, a systems biology approach was used in our work.

Little is know about the temporal and spatial regulation of 14-3-3σ and its downstream effects in tumour cells. Moreover, no mathematical model has been developed to describe its dynamics up to now. However, the p53 pathway has been investigated through mathematical modelling by several groups in the last decade, due to its essential role as tumour suppressor and master controller of the cellular response to DNA damage. Lev Bar-Or and co-authors²⁰ developed a simplified mathematical model accounting for p53 activation in response to DNA damage and MDM2 inhibition and suggested that under certain circumstances, oscillations in p53 and MDM2 protein levels can emerge in response to a stress signal. Their investigations about the oscillatory response of the p53 pathway were continued in a series of papers accounting for the modelling of the p53 response in individual cells²¹ and the variability of the oscillatory behaviour of the pathway in living human cells.²² In parallel, Ma and co-workers²³ presented a mathematical model accounting for the digital nature of the oscillations in p53, Ramalingam and co-workers²⁴ used high throughput data to characterise a mathematical model of MDM2-mediated p53 inhibition and Puszyński et al.²⁵ investigated the effect that stochasticity in p53 regulation has in the high heterogeneity of cell responses. In general we could say that the modelling efforts undertaken to characterise the p53 pathway focus on the analysis of its oscillatory behaviour mediated through inhibition by MDM2 under certain experimental conditions, but very little has been discussed about the effects that the modulation of p53 has on downstream molecules and pathways.

It has been well established in recent years that 14-3-3σ regulates the degradation of MDM2, a well known inhibitor of p53.^12,13 Moreover, p53 mediates the expression of both 14-3-3σ and MDM2 proteins under conditions of DNA damage. Taking all these facts together, the three proteins form a signalling module in which 14-3-3σ positively regulates the activity of p53 through feedback regulation, which is investigated via mathematical modelling in the present work. The aim of the current paper is the investigation of the feedback-loop regulated p53–MDM2–14-3-3σ signalling module through the integration of experimental data and mathematical modelling. Moreover, the effects of genemethylation on this pathway¹⁹ were also investigated with our systems biology approach. To this end, we derived, characterised and analysed a mathematical model that accounts for p53-mediated synthesis and activation of 14-3-3σ and MDM2, compartmentalisation of p53, MDM2 and 14-3-3σ, as well as the dynamics induced by modulation of 14-3-3σ genemethylation.

Materials and methods

Cell lines

Three different human melanoma cell lines , Sk-Mel-19, SK-Mel-29 and SK-Mel-103 were used in the present study. Cell lines were kindly provided by M. Soengas, Department of Dermatology, University of Michigan, Ann Arbor, MI, USA.²⁶ Cells were maintained in DMEM medium supplemented with 10% fetal calf serum and 100 μg ml⁻¹ penicillin–streptomycin.

Methylation-specific PCR of 14-3-3σ

To test the methylation levels of the 14-3-3σ gene in the above mentioned melanoma cells, genomic DNA was extracted from melanoma cell lines using DNeasy™ Tissue Kit (Qiagen, Hilden, Germany). SK-Mel-19 cells were treated additionally with demethylating agent 5-aza-2′-deoxycytidine (5-Aza-CdR). 1 μg of DNA was treated with sodium bisulfite using CpGenome™ Fast DNA Modification Kit (Chemicon, Hofheim, Germany). Subsequently, methylation-specific PCR (MSP) was performed with a specific primer set for methylated or unmethylated DNA. Primers specific for methylated DNA had the sequence: 5′-TGGTAGTTTTTATGAAAGGCGTC-3′ (sense) and 5′-CCTCTAACCGCCCACCACG-3′ (antisense), primers specific for unmethylated DNA had the sequence: 5′-ATGGTAGTTTTTATGAAAGGTGTT-3′ (sense) and 5′-CCCTCTAACCACCCACCACA-3′ (antisense). The PCR conditions were as follows: one cycle of 95 °C for 5 min; 30 cycles of 95 °C for 45 s, 56 °C for 30 s and 72 °C for 30 s; and one final cycle of 72 °C for 10 min. PCR products were separated on 1% agarosegels and stained with ethidium bromide. For quantification of PCR products the volumes of the different bands of the gel were analyzed using Progenesis PG200 software (Nonlinear Dynamics, Newcastle, UK).

Immunoblotting and protein quantification

Melanoma cell lines were lysed on ice for 30 min using radioimmunoprecipitation (RIPA) buffer. 40 μg of total protein was denatured in electrophoresis sample buffer for 5 min at 95 °C and subjected to SDS-polyacrylamide gel electrophoresis (PAGE). Gels were electroblotted onto nitrocellulose membranes (Highbond ECL®, Amersham, Braunschweig, Germany) and subjected to immunodetection. The following primary antibodies were used for immunodetection: anti-14-3-3σ mouse monoclonal antibody (ab14123, abcam/BIOZOL, Eching, Germany), anti-p53 mouse monoclonal antibody (cat. no. 554293, BD Biosciences, Heidelberg, Germany), anti-β-tubulin rabbit polyclonal antibody (sc-9104, Santa Cruz Biotechnologies , Heidelberg, Germany) and anti-histone H3 methylated on K9 (ab61231, abcam/BIOZOL). Signal detection was performed by appropriate secondary IRDye 680 labelled goat anti-mouse or IRDye 800CW labelled goat anti-rabbit antibodies (LI-COR Biosciences, Bad Homburg, Germany). The Odyssey infrared imaging system was used for signal visualization (LI-COR Biosciences). For staining of 14-3-3σ in fixed cell preparations a fluorescent secondary antibody was used. Odyssey infrared imaging system was used to quantify the immunoblots. Expression of β-tubulin was used to scale and normalise the data.²⁷

Results

Mathematical modelling

For our investigation we derived and used a mathematical model based on ordinary differential equations, which describes spatio-temporal changes in protein concentrations and other biological products such as RNAs with kinetic equations. The model is depicted in Fig. 1. The proposed model is not only an extension of previous models accounting for the dynamics of the p53–MDM2 core module, but also considers the following additional dynamical features: (a) p53-mediated synthesis and activation of 14-3-3σ; (b) compartmentalisation of p53, MDM2 and 14-3-3σ into the cytosol and the nucleus; and (c) dynamical features associated with the gene methylation of 14-3-3σ.

Fig. 1 Scheme of the model. Dashed green arrows represent activation, while dashed red lines ending with a bar represent inhibition. Solid lines represent synthesis (when starting with a circle and curved line) or degradation (when finishing in the symbol ∅). Double dashed arrows represent nuclear shuttling. Finally, the “clock” symbol accounts for a potential time-delay associated to gene expression processes.

For p53, we considered state variables accounting for cytosolic p53 (p53_c) and nuclear (activated) p53 (p53_n) fraction of the protein. The processes considered in the model are: (i) enhancement in p53 expression due to stress signals provoking DNA damage (DD); (ii) cytosolic basal degradation of p53; (iii) nuclear shuttling (import and export) of p53; and (iv) enhanced degradation of nuclear p53 promoted by nuclear MDM2.

In the case of MDM2, we considered the cytosolic (MDM2_c) and nuclear (MDM2_n) MDM2 fraction of the protein, and also the expression levels of messenger RNA accounting for MDM2 expression (RNA_m2). The model has rate equations accounting for: (i) p53-mediated synthesis of MDM2 messenger RNA; (ii) linear-like degradation of RNA_m2; (iii) RNA_m2-mediated MDM2 expression; (iv) cytosolic basal degradation of MDM2; (v) nuclear shuttling (import and export) of MDM2; and (vi) enhanced degradation of nuclear MDM2 promoted by nuclear 14-3-3σ.

In the case of 14-3-3σ, the cytosolic (σ_c) and nuclear (σ_n) fractions were considered, as well as expression levels of messenger RNA for 14-3-3σ (RNA_sg). The model describes the following mechanisms: (i) p53-mediated synthesis of 14-3-3σ messenger RNA; (ii) linear-like degradation of RNA_sg; (iii) RNA_sg-mediated 14-3-3σ expression; (iv) cytosolic basal degradation of 14-3-3σ; and (v) nuclear shuttling (import and export) of 14-3-3σ. We included an additional rate in the synthesis of 14-3-3σ messenger RNA accounting for the effect of changes in gene methylation, represented by the tuneable parameter γ_m (current percentage of methylation for the gene, with γ_o original basal level of methylation in normal cells). The model is composed of the following equations:

p53 dynamics

MDM2 dynamics


14-3-3σ dynamics


Equations accounting for the p53-mediated synthesis of mRNA for MDM2 and 14-3-3σ were taken from the generic ODE model for RNA synthesis of Bartlett and Davis.²⁸ This model was modified to include the promoting effect of nuclear p53 in the mRNA synthesis rate, as well as the repressive effect of genemethylation in 14-3-3σ mRNA synthesis. In our equations we considered non-linear features in the p53-mediated induction of MDM2 and 14-3-3σ and in the activation of p53 and MDM2 suggested by Ramalingan and co-authors,²⁴ allowing non-integer kinetic orders (g_ind, g_am2, g_a53).²⁹ Nevertheless, in our initial model we assumed that all of them were equal to one and discuss later the effect of other values on the stability of the system. We also considered non-linear kinetics for the term describing the effect of genemethylation (g_met = 2, low cooperativity). In addition, we considered in our equations a time-delay accounting for the elapsed time between transcription factor-mediated gene activation and effective messenger RNA synthesis.³⁰ We assumed such a time-delay would be much smaller than the time scale of our experiment (and therefore have negligible effects), but afterwards we discussed the potential effects of time-delay in the stability of the system.

Values for parameters were estimated using data extracted from Yang et al. (2003)¹² and Urano et al. (2002).³¹ More precisely, time series accounting for the half-life of MDM2 and p53¹² and 14-3-3σ³¹ were used, together with processed and quantified immunoblots to account for dynamics of the system under different conditions of expression for the three proteins and information about their subcellular distribution.¹² ImageJ was used for image processing and data quantification.³² In addition, data concerning the nucleocytoplasmic shuttling of 14-3-3σ were used to characterise 14-3-3σ nuclear import/export rates.³³ For parameter estimation we used optimisation algorithms and manual tuning functions contained in the SBtoolbox.³⁴ The estimated values are included in Table 1 (Appendix A2 in the ESI† ). In Appendix A1 we show the comparison between the data and the model predictions.† Finally for parameters accounting for the p53-mediated synthesis of mRNA for MDM2 and 14-3-3σ we took the values in the generic ODE model for RNA synthesis in Bartlett and Davis,²⁸ but normalised them in a way that mRNA levels are equal to one in the steady-state of the system for DNA damage equal to one (DD = 1).

Model predictions for melanoma cell lines

In order to verify the predictive abilities of our model concerning the dynamics in melanoma cell lines , we performed additional experiments and compared those data with the predictions of our model. We first analysed the dynamics of stress-modulated activation of the system. Towards this end, melanoma cells were exposed to a constant dose of the cytostatic agent adryamicin (3.2 μg ml⁻¹). Samples were taken at indicated time points and the expression levels of the proteins involved in the system were estimated after immunoblotting (Fig. 2, top), and quantified following the procedure described in the Materials and methods section. β-tubulin was used as normaliser and staining for histone H3 methylated on K9 (anti-K9M-H3) was used as a surrogate marker for DNA damage. In the simulations we assumed a basal level of DNA damage around 20% of the peak stimulation and used data for K9M-H3 to represent dynamics of the input signal (Fig. 2, middle). In order to allow a comparison, data from the simulations were normalised such that total levels of proteins were equal to one at the maximum level for the experimental data (8 h). Finally, we compared the experimental data with the prediction of our model (Fig. 2, bottom). In general terms, there is a good agreement between the experimental data and the predictions of our model, especially for the fast, initial stress-driven activation of the system. The match between model predictions and data for the slow signal termination is however not totally satisfactory. We notice that expression of histone H3 methylated on K9 reflecting DNA damage still remains at 50% of the maximum value by the end of the experiment, while proteins are nearly at the initial expression value. This indicates that either p53 and 14-3-3σ expression levels are overestimated at the final data points or more complicated dynamics are necessary to get perfect fitting. This study is however beyond the scope of the present work.

Fig. 2 Model predictions for dynamical stimulation of the system. Top: immunoblot accounting for the experiment; bottom left: estimated data on K9M-H3 used as input signal; bottom right: comparison between data (stars) and model predictions (lines) for the total amount of p53 (red) and 14-3-3σ (blue).

In order to test the compartmentalisation of the proteins described in the model, we investigated the subcellular localisation of 14-3-3σ (Fig. 3). Cells were incubated and later exposed to a constant dose of the cytostatic agent adryamicin (3.2 μg ml⁻¹) in an experiment identical to one described in Fig. 2. 14-3-3σ was stained and cell imaging was performed at the indicated time points by fluorescence microscopy (Fig. 3, top). In our simulations (Fig. 3, bottom left), we took the same input signal used in Fig. 2, computed the evolution over time of the cytosolic (red) and nuclear (blue) fractions of 14-3-3σ and normalised the results such that the total levels of 14-3-3σ are one at the maximum level for the experimental data (8 h). As we can see, the model is in good agreement with the experimental observations. At basal levels of stimulation (0 h) most of the existing 14-3-3σ is accumulated in the cytosol. Upon stimulation, expression levels of 14-3-3σ increases to a maximum between 8 and 16 h and then slowly decreases to 24 h. In all the cases live imaging reveals that most of the protein stays in the cytosol up to 8 h after stimulation, which is in accordance with the predictions of the model, but shuttles to the nucleus after 8 h. This point was confirmed when we estimated the nuclear and cytosolic fraction of 14-3-3σ using immunoblots (Fig. 3, bottom right). In any case, the model and the data suggest that a significant portion of the protein remains cytosolic during the whole experiment. We notice that the results of our analysis are in line with the investigation by van Hemert et al. (2004),³³ where it was shown that 14-3-3σ remains essentially cytosolic for all the human cells lines analysed.

Fig. 3 Subcellular localisation of 14-3-3σ. Top: live imaging of 14-3-3σ dynamics in cells (green); bottom left: model predictions for the dynamics of cytosolic (red) and nuclear (blue) fractions of 14-3-3σ; bottom right: immunoblots for the nuclear and cytosolic fractions of 14-3-3σ for stimulation of SK-Mel-103 cells with 3.2 μg ml⁻¹ adriamycin.

We further compared the response predicted by the model with results from experiments where increasing amounts of stimulus were applied to the system (Fig. 4). Cells were incubated and exposed in parallel experiments to different amounts of adriamycin. Samples were taken after 24 h of stimulation and the protein expression levels were estimated by immunoblots (Fig. 4, top) and quantified following the procedure described in the Materials and methods section (Fig. 4, bottom left). β-tubulin was used for normalisation. In order to allow a comparison, data from the simulation were normalised such that total levels of proteins were one at the maximum level simulated (3.2 μg ml⁻¹). Finally, we compared the experimental data (Fig. 4, bottom left) with the prediction of our model (Fig. 4, bottom right). As we can see, the predictions of the model concerning responsiveness are in good agreement with the experimental data. The model slightly underestimated the expression of 14-3-3σ for low doses of adriamycin, while in the case of p53 the overestimation is due to problems in the experimental estimation of p53 at 24 h for the highest drug dose (3.2 μg ml⁻¹). In dose-response experiments performed with increasing amounts of a similar cytotoxic drug we have verified that the expression of p53 at 24 h increases always with the amount of the cytotoxic drug (data not shown), suggesting that the behaviour of p53 at 24 h for the highest dose of adryamicin (3.2 μg ml⁻¹) is an outlier rather than a systemic property.

Fig. 4 Dose-dependent response of the system measured 24 h after stimulation with different amounts of adryamicin. Top: immunoblot; bottom: comparison between model predictions (left) and normalised experimental data (right).

The comparison between model simulations and experiments reveals that the proposed model is able to predict the behaviour of the system under the experimental conditions assayed. Moreover, the data available suggests that the predictions of the model concerning the subcellular distribution of the proteins are also in accordance with the real behaviour.

We also investigated the effects of time-delay on the p53-mediated expression of 14-3-3σ and MDM2. Firstly, we analysed with simulations the stability of the system steady-states for different levels of sustained DNA damage and time-delay; the computed steady-states of the system are stable and do not show sustained oscillations (data not shown). Afterwards, we computed a series of simulations for transient DNA damage stimulation of the system with different values of time-delay (ESI,† Fig. A3.1). The transient dynamics of the system were not significantly affected for physiologically feasible values of time-delay (τ smaller than 2 h), while for longer, abnormal time-delays the transient dynamics were affected in the intensity of the signal peak, although our simulations do not show evidence for the emergence of transient oscillations in the system. Our analysis indicates that, in the current parameterisation of the model and for a wide interval of values in DNA damage and time-delay, the stability of the system is not altered by the addition of a time-delay in the p53-mediated expression of 14-3-3σ and MDM2.

Investigating the dynamical effects of genemethylation

We were interested in investigating how different levels of genemethylation of 14-3-3σ affect the behaviour of the system. The methylation status of the 14-3-3σgene in the different melanoma cell lines was tested by methylation-specific PCR and is shown in the ESI, Fig. A4.1.† In addition, SK-Mel-19 melanoma cells were treated with demethylating agent 5-Aza-CdR. We performed simulations where different levels of 14-3-3σgene methylation were simulated by tuning the parameter γ_m in the model. The dynamics of the different realisations of the system were simulated under the same basal conditions (Fig. 5, right panel). In order to allow a comparison, the data obtained were normalised such that the total levels of proteins were one at the maximum level simulated. The model simulations suggest that increasing methylation levels provoke the expected decrease in the expression of 14-3-3σ but there also exists a feedback mechanism modulating downregulation of p53, due to the enhanced expression of MDM2. Thus, our investigation suggests that 14-3-3σgene methylation negatively affects 14-3-3σ and p53 and positively affects MDM2.

Fig. 5 Expression patterns for the proteins involved in the system under different levels of genemethylation of 14-3-3σ. Melanoma cell lines with different levels of 14-3-3σgene methylation (SK-Mel-19 cells treated with 5-Aza-CdR, 30% methylation; untreated SK-Mel-19 cells, 50% methylation; SK-Mel-29 cells, 65% methylation; SK-Mel-103 cells, 85% methylation). Protein expression of 14-3-3σ and p53 was analysed by immunoblot. Left: immunoblot; Right: model predictions.

In order to verify experimentally the hypothesized feedback mechanisms, cell lines with different methylation levels were cultured in parallel experiments. Cell lines were either left untreated or treated with 5-Aza-CdR (SK-Mel-19) to enhance 14-3-3σ expression, and protein expression levels for 14-3-3σ and p53 were measured by immunoblots (Fig. 5, left panel). We compared the experimental data (Fig. 5, left panel) with the prediction of our model (Fig. 5, right panel) where basal expression was simulated as a constant small value for the input signal of the system. In general terms, the agreement between the model predictions and the experimental data is good. For strong methylation, basal levels of 14-3-3σ are reduced to almost complete removal, while p53 expression is significantly affected as well. Additional experiments would be necessary to verify the effect of methylation on MDM2 expression.

We furthermore used the model to simulate the expression and compartmentalisation of 14-3-3σ and the other proteins involved in the system under different scenarios of DNA damage and gene methylation (Fig. 6). We simulated the response of the system under different levels of DNA damage in a normalised scale ranging from low (DD = 0.1) to severe damage (DD = 10) and different 14-3-3σmethylation levels, from (unfeasible) no methylation (γ_m = 0) to complete gene methylation, γ_m = 100. We computed for each protein in our model the total level of expression, measured as the sum of the cytosolic and nuclear fractions (Fig. 6, left), and the percentage of nuclear fraction over the total level of expression in each simulated scenario (Fig. 6, right). The model suggests that p53 activity is clearly affected by methylation of the 14-3-3σgene as a consequence of the feedback loop, but remains (weakly) expressed even for intense 14-3-3σmethylation (Fig. 6, top left). Moreover, an increase in methylation shifts the threshold of DNA damage necessary to get significant p53 expression (Fig. 6, top left). In addition, the model predicts that changes in the methylation of the 14-3-3σgene affect the expression of MDM2 as well, getting high MDM2 expression levels even for moderate DNA damage levels when the gene methylation is very high (Fig. 6, middle left). On the other hand, with just moderate levels of methylation (0–50%), the model predicts a large reduction in MDM2 expression levels and a shift in the threshold of significant amounts of MDM2 towards high levels of DNA damage (Fig. 6, middle left). Our simulations predict that 14-3-3σ expression will be virtually silenced even for high levels of DNA damage when gene methylation is very high, but also that strong stimulation will be necessary to get significant levels of 14-3-3σ for intermediate levels of methylation (Fig. 6, bottom left).

Fig. 6 Protein expression patterns when different levels of DNA damage and 14-3-3σgene methylation are considered. Every point in the grid represents the expression of the considered protein for a given DNA damage level (ranging from low to severe damage, DD = (0.1, 10)) and percentage of methylation (ranging from zero methylation to complete gene methylation, γ_m = (0, 100)). Every protein expression level is represented using the code encoded in the colour bar at the right side of the figure.

Regarding the effect of DNA damage and methylation in the subcellular distribution of the proteins, the model predictions suggest that p53 localisation could be drastically affected for high levels of 14-3-3σmethylation, reducing importantly the nuclear fraction of p53 (Fig. 6, top right). This suggests a synergistic effect of 14-3-3σmethylation in the regulation of p53: high levels of methylation provoke an intermediate reduction in p53 expression, but also a significant change in the value of its nuclear fraction; the combination (cross-correlation) of both effects suggests that high levels of 14-3-3σmethylation could reduce drastically the activity of p53 as transcription factor. In the case of MDM2, 14-3-3σgenemethylation enhances the value of the nuclear fraction, in a way the model predicts that the protein will virtually localise in the nucleus for any level of DNA damage for high or complete gene methylation (Fig. 6, middle right). This suggests that 14-3-3σmethylation enhances MDM2 activity through nuclear localisation. We notice that our results concerning subcellular localisation of the p53 and MDM2 are in accordance with the results of Yang et al. (2003),¹² who found drastic changes in the nuclear–cytosolic distribution of both proteins under experimental ectopic modulation of 14-3-3σ expression.

Finally, the simulations based on our model indicate that the distribution of 14-3-3σ between cellular compartments is not affected by methylation under different regimes of stimulation (Fig. 6, bottom right). The changes in the subcellular distribution of p53 and MDM2 induced by reduced 14-3-3σ expression are a consequence of the feedback-loop structure of the system. We notice that our results do not indicate that reduced gene expression of 14-3-3σ affects its own subcellular distribution, but the subcellular distribution of the other proteins that are feedback-loop regulated by 14-3-3σ.

Discussion and conclusions

The p53 pathway plays an important role in the maintenance of genomic integrity viacell cycle arrest and consecutive DNA repair or induction of apoptosis in cases of severe DNA damage.¹¹ Recent data suggest that 14-3-3σ, which is induced by p53 viagene regulatory mechanisms, is involved in this process.⁸ In line with this, loss-of-function of 14-3-3σ via epigenetic silencing is associated with deregulated cell growth and tumour development in a series of malignant tumours.^15–18

Here we developed a mathematical model for the description of the p53–MDM2–14-3-3σ pathway in melanoma cells under genotoxic stress executed by the chemotherapeutic agent adriamycin. We further addressed the question of whether epigenetic silencing of 14-3-3σ affects the dynamics of the feedback loop signalling system integrated by p53, MDM2 and 14-3-3σ. To this end, we derived and analysed a mathematical model describing the DNA damage-mediated activation of the system, p53-mediated synthesis of 14-3-3σ and MDM2, and also the subcellular distribution of the three proteins. The effect of gene methylation on the effective synthesis of 14-3-3σ was also included in our model.

We verified that the proposed model is able to predict the dynamics of the system for melanoma cell lines by comparing experimental data and computational simulations for different experimental conditions including: (a) dynamics of stress-modulated activation of the system, (b) stress-induced dynamics of subcellular localisation for the protein 14-3-3σ; and (c) a drug-dose response curve. The comparison between model simulations and experiments reveals that the model proposed is able to predict the qualitative behaviour of the system under the experimental conditions assayed. This was particularly true for the initial stress-driven activation of the system. However, the model predictions and experimental data of the slow signal termination of DNA damage (indicated by K9M-H3 expression) was not totally satisfactory. This might be due to irreversible induction of DNA damage or cellular senescence—K9M-H3 expression is also a surrogate marker for cellular senescence—in melanoma cells after exposure to adriamycin. Indeed, adriamycin has been described as an inducer of cellular senescence in prostate and breast cancer cells.^35,36 We observed cytoplasmic-nuclear import of 14-3-3σ after 8 h of stimulation, but our analysis indicates that an important fraction of the available protein remains cytosolic. In line with these observations, van Hemert et al. (2004) have shown that 14-3-3σ remains essentially cytosolic for all the human cells lines analysed.³³

In addition, we investigated the effects of changes in 14-3-3σgene methylation levels in the responsiveness of the system. Towards this end, we simulated the basal expression levels of p53 and 14-3-3σ for different levels of gene methylation by tuning the parameter γ_m in the model and compared the results with experimental data generated in similar conditions. The predictions of our model match with the experimental data generated and indicate that under high methylation, the expression levels of not only 14-3-3σ but also p53 are significantly affected. We further used the model to simulate different scenarios of activation of the system with increasing levels of DNA damage and gene methylation. Interestingly enough, our simulations suggest that 14-3-3σ expression will be virtually silenced when the gene methylation is very high (>80%), but also that strong DNA damage signalling will be necessary to get 14-3-3σ expression for intermediate levels of gene methylation. Furthermore, the model suggests that p53 activity will be feedback loop affected by 14-3-3σgene methylation in two different ways: (a) p53 expression is partially downregulated; and (b) subcellular localisation is modified for high levels of 14-3-3σmethylation, reducing importantly the nuclear fraction of p53. This suggests an intriguing synergistic effect of 14-3-3σmethylation in the regulation of p53: the convolution of intermediate downregulation in p53 expression and the reduction of its nuclear fraction under epigenetic silencing of 14-3-3σmethylation could reduce drastically (non-linearly) the net activity of p53 as transcription factor.

Although the existence of feedback loops, non-linearity and time-delay indicate that our model is structurally able to generate sustained oscillations,³⁰ we have not addressed this question in this paper. The rationale for this choice is that in our investigation a genotoxic agent has been used to induce DNA damage, while in every previous work investigating the emergence of oscillations in p53 signalling, the experimental data concerned stimulation with ionizing radiation.^20–22,24 Our data did not show oscillatory behaviour, which suggests that an adequate experimental set up is necessary to induce and analyse oscillations in p53.

Our model does not include a complete description of all the processes involved in the regulation of the pathway and further investigations are required to address the following questions. Silencing methylation occurs exclusively under oncogenic conditions,^37,38 for which the p14/p19ARF loop is expressed supporting the action of p53 and feedback regulating MDM2. Moreover, some authors suggest that the p16 tumour suppressor protein is also inactivated when 14-3-3σ appears methylated,³⁹ suggesting cooperativity in the control of the DNA damage response. Furthermore, 14-3-3σ forms complexes with others regulators like MDMX, that also affect MDM2 dynamics, suggesting an additional way of regulation in the system.

Acknowledgements

The initial idea for the investigation was suggested by M.K. and O.W. The model was derived, computed and analysed by J.V. Experiments were designed and performed by M.K., J.S., Y.R. and S.I. All the authors drafted the manuscript. The authors thank the comments and suggestions of Daniel Guebel and Xin Lai.

This work was supported by the German Federal Ministry of Education and Research (BMBF) as part of the project CALSYS under the FORSYS initiative, contract 0315264 (http://www.sbi.uni-rostock.de/calsys).

References

1. A. J. Miller and M. C. Mihm Jr, Melanoma, N. Engl. J. Med., 2006, 355, 51–65.
2. J. A. Curtin, J. Fridlyand, T. Kageshita, H. N. Patel, K. J. Busam, H. Kutzner, K. H. Cho, S. Aiba, E. B. Brocker, P. E. LeBoit, D. Pinkel and B. C. Bastian, Distinct sets of genetic alterations in melanoma, N. Engl. J. Med., 2005, 353, 2135–47.
3. R. K. Thomas, A. C. Baker, R. M. Debiasi, W. Winckler, T. Laframboise, W. M. Lin, M. Wang, W. Feng, T. Zander, L. MacConaill, J. C. Lee, R. Nicoletti, C. Hatton, M. Goyette and L. Girard, et al., High-throughput oncogene mutation profiling in human cancer, Nat. Genet., 2007, 39, 347–51.
4. V. Gray-Schopfer, C. Wellbrock and R. Marais, Melanoma biology and new targeted therapy, Nature, 2007, 445, 851–57.
5. L. Chin, The genetics of malignant melanoma: lessons from mouse and man, Nat. Rev. Cancer, 2003, 3, 559–70.
6. J. Jaeger, D. Koczan, H. J. Thiesen, S. M. Ibrahim, G. Gross, R. Spang and M. Kunz, Gene expression signatures for tumour progression, tumour subtype, and tumour thickness in laser-microdissected melanoma tissues, Clin. Cancer Res., 2007, 13, 806–15.
7. H. Hermeking, The 14-3-3 cancer connection, Nat. Rev. Cancer, 2003, 3, 931–43.
8. H. Hermeking and A. Benzinger, 14-3-3 proteins in cell cycle regulation, Semin. Cancer Biol., 2006, 16, 183–92.
9. H. Hermeking, C. Lengauer, K. Polyak, T. C. He, L. Zhang, S. Thiagalingam, K. W. Kinzler and B. Vogelstein, 14-3-3σ is a p53-regulated inhibitor of G2/M progression, Mol. Cell, 1997, 1, 3–11.
10. A. Benzinger, N. Muster, H. B. Koch, J. R. Yates, 3rd and H. Hermeking, Targeted proteomic analysis of 14-3-3σ, a p53 effector commonly silenced in cancer, Mol. Cell. Proteomics, 2005, 4, 785–95.
11. A. J. Levine, p53, the cellular gatekeeper for growth and division, Cell, 1997, 88, 323–31.
12. H. Y. Yang, Y. Y. Wen, C. H. Chen, G. Lozano and M. H. Lee, 14-3-3σ positively regulates p53 and suppresses tumour growth, Mol. Cell. Biol., 2003, 23, 7096–7107.
13. H. Y. Yang, Y. Y. Wen, Y. I. Lin, L. Pham, C. H. Su, H. Yang, J. Chen and M. H. Lee, Roles for negative cell regulator 14-3-3σ in control of MDM2 activities, Oncogene, 2007, 26(52), 7355–62.
14. T. A. Chan, H. Hermeking, C. Lengauer, K. W. Kinzler and B. Vogelstein, 14-3-3σ is required to prevent mitotic catastrophe after DNA damage, Nature, 1999, 401, 616–20.
15. A. T. Ferguson, E. Evron, C. B. Umbricht, T. K. Pandita, T. A. Chan, H. Hermeking, J. R. Marks, A. R. Lambers, P. A. Futreal, M. R. Stampfer and S. Sukumar, High frequency of hypermethylation at the 14-3-3σ locus leads to gene silencing in breast cancer, Proc. Natl. Acad. Sci. U. S. A., 2000, 97, 6049–54.
16. N. Iwata, H. Yamamoto, S. Sasaki, F. Itoh, H. Suzuki, T. Kikuchi, H. Kaneto, S. Iku, I. Ozeki, Y. Karino, T. Satoh, J. Toyota, M. Satoh, T. Endo and K. Imai, Frequent hypermethylation of CpG islands and loss of expression of the 14-3-3σ gene in human hepatocellular carcinoma, Oncogene, 2000, 19(46), 5298–5302.
17. P. Mhawech, A. Benz, C. Cerato, V. Greloz, M. Assaly, J. C. Desmond, H. P. Koeffler, D. Lodygin, H. Hermeking, F. Herrmann and J. Schwaller, Downregulation of 14-3-3σ in ovary, prostate and endometrial carcinomas is associated with CpG island methylation, Mod. Pathol., 2005, 18(3), 340–48.
18. D. Lodygin, J. Diebold and H. Hermeking, Prostate cancer is characterized by epigenetic silencing of 14-3-3σ expression, Oncogene, 2004, 23(56), 9034–41.
19. J. Schultz, S. M. Ibrahim, J. Vera and M. Kunz, 14-3-3σ gene silencing during melanoma progression and its role in cell cycle control and cellular senescence, Mol. Cancer, 2009, 8, 53.
20. R. Lev Bar-Or, R. Maya, L. A. Segel, U. Alon, A. J. Levine and M. Oren, Generation of oscillations by the p53-Mdm2 feedback loop: a theoretical and experimental study, Proc. Natl. Acad. Sci. U. S. A., 2000, 97(21), 11250–55.
21. G. Lahav, N. Rosenfeld, A. Sigal, N. Geva-Zatorsky, A. J. Levine, M. B. Elowitz and U. Alon, Dynamics of the p53-Mdm2 feedback loop in individual cells, Nat. Genet., 2004, 36(2), 147–50.
22. N. Geva-Zatorsky, N. Rosenfeld, S. Itzkovitz, R. Milo, A. Sigal, E. Dekel, T. Yarnitzky, Y. Liron, P. Polak, G. Lahav and U. Alon, Oscillations and variability in the p53 system, Mol. Syst. Biol., 2006, 2, 2006.0033.
23. L. Ma, J. Wagner, J. J. Rice, W. Hu, A. J. Levine and G. A. Stolovitzky, A plausible model for the digital response of p53 to DNA damage, Proc. Natl. Acad. Sci. U. S. A., 2005, 102(40), 14266–71.
24. S. Ramalingam, P. Honkanen, L. Young, T. Shimura, J. Austin, P. S. Steeg and S. Nishizuka, Quantitative assessment of the p53-Mdm2 feedback loop using protein lysate microarrays, Cancer Res., 2007, 67(13), 6247–52.
25. K. Puszyński, B. Hat and T. Lipniacki, Oscillations and bistability in the stochastic model of p53 regulation, J. Theor. Biol., 2008, 254(2), 452–65.
26. M. S. Soengas, P. Capodieci, D. Polsky, J. Mora, M. Esteller, X. Opitz-Araya, R. McCombie, J. G. Herman, W. L. Gerald, Y. A. Lazebnik, C. Cordón-Cardó and S. W. Lowe, Inactivation of the apoptosis effector Apaf-1 in malignant melanoma, Nature, 2001, 409, 207–11.
27. M. Schilling, T. Maiwald, S. Bohl, M. Kollmann, C. Kreutz, J. Timmer and U. Klingmüller, Computational processing and error reduction strategies for standardized quantitative data in biological networks, FEBS J., 2005, 272, 6400–11.
28. D. W. Bartlett and M. E. Davis, Insights into the kinetics of siRNA-mediated gene silencing from live-cell and live-animal bioluminescent imaging, Nucleic Acids Res., 2006, 34(1), 322–33.
29. J. Vera, E. Balsa-Canto, P. Wellstead, J. R. Banga and O. Wolkenhauer, Power-Law Models of Signal Transduction Pathways, Cell. Signalling, 2007, 19, 1531–41.
30. S. Nikolov, J. Vera, V. Kotev, O. Wolkenhauer and V. Petrov, Dynamic properties of a delayed protein cross talk model, BioSystems, 2008, 91(1), 51–68.
31. T. Urano, T. Saito, T. Tsukui, M. Fujita, T. Hosoi, M. Muramatsu, Y. Ouchi and S. Inoue, Efp targets 14-3-3σ for proteolysis and promotes breast tumour growth, Nature, 2002, 417(6891), 871–75.
32. M. D. Abramoff, P. J. Magelhaes and S. J. Ram, Image Processing with ImageJ, Biophotonics International, 2004, 11(7), 36–42.
33. M. J. van Hemert, M. Niemantsverdriet, T. Schmidt, C. Backendorf and H. P. Spaink, Isoform-specific differences in rapid nucleocytoplasmic shuttling cause distinct subcellular distributions of 14-3-3σ and 14-3-3ζ, J. Cell Sci., 2004, 117(8), 1411–20.
34. H. Schmidt and M. Jirstrand, Systems Biology Toolbox for MATLAB: a computational platform for research in systems biology, Bioinformatics, 2006, 22(4), 514–15.
35. L. W. Elmore, C. W. Rehder, X. Di, P. A. McChesney, C. K. Jackson-Cook, D. A. Gewirtz and S. E. Holt, Adriamycin-induced senescence in breast tumour cells involves functional p53 and telomere dysfunction, J. Biol. Chem., 2002, 277, 35509–15.
36. S. R. Schwarze, V. X. Fu, J. A. Desotelle, M. L. Kenowski and D. F. Jarrard, The identification of senescence-specific genes during the induction of senescence in prostate cancer cells, Neoplasia, 2005, 7, 816–23.
37. S. Liang, Y. Xu, G. Shen, X. Zhao, J. Zhou, X. Li, F. Gong, B. Ling, L. Fang, C. Huang and Y. Wei, Gene expression and methylation status of 14-3-3σ in human renal carcinoma tissues, IUBMB Life, 2008, 60, 534–40.
38. U. K. Bhawal, K. Tsukinoki, T. Sasahira, F. Sato, Y. Mori, N. Muto, M. Sugiyama and H. Kuniyasu, Methylation and intratumoural heterogeneity of 14-3-3σ in oral cancer, Oncol. Rep., 2007, 18(4), 817–24.
39. M. Gasco, A. K. Bell, V. Heath, A. Sullivan, P. Smith, L. Hiller, I. Yulug, G. Numico, M. Merlano, P. J. Farrell, M. Tavassoli, B. Gusterson and T. Crook, Epigenetic inactivation of 14-3-3σ in oral carcinoma association with p16(INK4a) silencing and human papillomavirus negativity, Cancer Res., 2002, 62, 2072–76.

---

Footnotes

† Electronic supplementary information (ESI) available: A comparison between data used for parameter estimation and model predictions; a list of model parameter values; simulations of the effect of time-delay on the dynamics of the system; the methylation status of 14-3-3σ in the different cell lines used. See DOI: 10.1039/b907863k

‡ Equal contributors.

---