ERK and Wnt signaling pathways implicated in the EMT process contain multiple CFLs that form a highly interconnected network. However, little is known about the functional role of these feedback loops. In this study, we found that the positive feedback in which activated ERK counteracts the inhibition of PKCδ by GSK3β has a significant role in inducing a large state change of E-cadherin in response to EGF stimulation while the positive feedback mediated by RKIP decreases the state transition interval in the state change of E-cadherin ( and ). CFLs composed of the RKIP-mediated PFLs and the negative feedback loop where ERK phosphorylates and inhibits the SOS/Grb2 complex increase EC50 in response to the EGF stimulation ( and ). On the other hand, it turns out that the negative feedback loop where the β-catenin/TCF complex induces Axin dose not have any significant role in the regulation of E-cadherin expression.
In a previous study (18
) we found that the positive feedback loop formed by the phosphorylation of RKIP by ERK induces a switch-like behavior of ERK and MEK activities. In this study, we have further revealed that such switch-like behaviors are cooperatively caused by CPFLs through phosphorylation of RKIP by ERK and transcriptional repression of RKIP by Snail (). If one of these feedback loops is blocked, the switch-like behavior becomes weaker or even disappears.
From extensive in silico
simulation of the dose-response characteristics to oncogenic stimulation and the Karnaugh-map (K-map) analysis method (), we found that particular feedback combinations had a significant effect on E-cadherin regulation. Some of them were commonly observed for gradual increase of EGF with a fixed level of Wnt, while none was observed for a gradual increase of Wnt with a fixed level of EGF (Table S1 of Supplementary Data C
). This suggests that the dose-response of E-cadherin expression to EGF stimulation is mostly not affected by the crosstalk with Wnt, but that Wnt stimulation is significantly affected by the crosstalk with EGF. This result might raise the interesting possibility that the ERK and Wnt signaling pathways play different roles in inducing EMT, which can also be partially supported by other recent experimental results. For instance, both the ERK and Wnt signaling pathways are activated by the growth factors secreted by malignant tumor cells with epithelial phenotype, and they contribute to the EMT process possibly at an initial phase (6
). However, in transformed mesenchymal cells, the ERK signaling pathway seems to be strongly activated compared to the Wnt signaling pathway since the mesenchymal cells secrete various growth factors that strongly stimulate ERK, such as fibroblast growth factors (FGF), EGF, and HGF (37
). Thus, it seems that both the ERK and Wnt pathways participate in the initial phase of the EMT process, while the ERK pathway takes the major role in the later phase when the epithelial cells are transformed to the mesenchymal cells.
Our simulation results show that RKIP controls the EMT process through regulation of E-cadherin (). A number of recent studies showed a statistically significant inverse relationship between RKIP expression and metastasis and overall survival both in animal models and in human cancer patients (32
). For instance, Fu et al
) showed that highly metastatic C4-2B prostate cancer cells lack RKIP expression, and that enforced expression of RKIP decreased cell invasiveness in in vitro
assays and suppressed the development of lung metastasis when implanted into mouse prostate without affecting the growth of the primary tumor. In human cancer patients it was shown that the retention of RKIP expression in colorectal tumors correlates with a low risk of metastatic relapse and improved survival (32
). Similar results were obtained by Minoo et al. (36
) who used colon cancer tissue microarrays to show that the loss of cytoplasmic RKIP is associated with distant metastasis, vascular invasion, and worse survival. Thus, while an important role for RKIP in metastasis suppression especially in colon and prostate cancer is now emerging, the underlying mechanisms how RKIP expression is altered in tumors and how RKIP counteracts the development of metastasis are less well understood. The regulation of RKIP expression seems to involve at least two mechanisms. One is the silencing of the rkip
gene promoter by hypermethylation (38
), the other is the repression of the rkip
gene promoter by Snail (19
). The molecular function of RKIP as inhibitor of the ERK pathway that inhibits MEK phosphorylation by Raf is well characterized (16
). Recent data show that ERK can inactivate RKIP as part of a feedback loop and thereby confer non-linear dynamics on ERK activation (18
). However, so far this feedback loop only has been analyzed in isolation. The same is true for most other feedback loops described here.
Our analysis of the combinatorial effects of such feedback loops revealed several new findings. One is that RKIP is an important feedback loop that exerts major control over both STI and EC50, and that RKIP levels are crucial for these effects (). Thus, the CPFLs where ERK phosphorylates RKIP, and Snail transcriptionally represses the expression of RKIP cause a switch-like behavior of E-cadherin expression (), where the RKIP expression determines the EMT progression by antagonizing the suppression of E-cadherin. Another result is that subtle changes in the combination of feedback links may substantially change the output of the network. This is interesting in light of the observation that β-catenin is found in the nucleus of the mesenchymal-like cells at the invasive front of colorectal tumors, while it is cytosolic in the central epithelial areas of the same tumor (41
). These changes were attributed to slightly different microenvironments. Our results suggest that combinatorial changes in the network topology could be the molecular substrate for such dramatic influences of the microenvironment. This flexible topology afforded by different feedback link combinations also would enable the facile transition of cells between epithelial and mesenchymal morphologies as observed in tumors. In summary, our analysis shows that feedback loops formed by combination of feedback links do not only have roles in shaping dynamic response kinetics of networks, but also in the flexible specification and diversification of responses.