PMC:4331679 / 23677-30795 JSONTXT

{"project":"2_test","denotations":[{"id":"25707537-17057752-14870731","span":{"begin":361,"end":363},"obj":"17057752"},{"id":"25707537-22579283-14870732","span":{"begin":1084,"end":1085},"obj":"22579283"},{"id":"25707537-15286780-14870733","span":{"begin":3627,"end":3629},"obj":"15286780"},{"id":"25707537-20221256-14870734","span":{"begin":3953,"end":3954},"obj":"20221256"},{"id":"25707537-22579283-14870735","span":{"begin":4075,"end":4076},"obj":"22579283"},{"id":"25707537-22056502-14870736","span":{"begin":4077,"end":4079},"obj":"22056502"},{"id":"25707537-22705015-14870737","span":{"begin":4257,"end":4259},"obj":"22705015"},{"id":"25707537-9545235-14870738","span":{"begin":4512,"end":4514},"obj":"9545235"},{"id":"25707537-11801527-14870739","span":{"begin":4697,"end":4699},"obj":"11801527"},{"id":"25707537-20221256-14870740","span":{"begin":5772,"end":5773},"obj":"20221256"},{"id":"25707537-22579283-14870741","span":{"begin":5774,"end":5775},"obj":"22579283"},{"id":"25707537-11350724-14870742","span":{"begin":5776,"end":5778},"obj":"11350724"},{"id":"25707537-17616683-14870743","span":{"begin":5779,"end":5781},"obj":"17616683"}],"text":"Results\n\nPLSR for simulation data from N1\nWe first used Partial Least Squares Regression (PLSR) to test if the simulation data generated using our basic network N1 were reliable for performing the prediction of network rewiring. PLSR is used for mapping independent variable with dependent variables (e.g. mapping signalling molecules to phenotypic responses) [30]. In a PLSR plot, if a predictor variable is closely associated with the predicted variable, then such a predictor variable influences the predicted variable relatively more strongly than the predictor variable who is far away in the PLSR plot. For reaction parameters, we ensured that the selected parameter values enabled us to associate the important predictor variables like Caspases, XIAP, SMAC etc. closely with the apoptosis. The Partial least Square Regression (PLSR) plot for the simulation data from the basic network from N1, is given in Figure 2, which PLSR plot shows the close association between apoptosis and several apoptotic proteins. This result is consistent with the results reported by Lee et al. [5].\nFigure 2 PLSR plot for simulation data for apoptosis from network N1.\n\nInferring the tumorigenic network N2\nFrom network N1, we will infer the rewired network N2 corresponding to cancer cell lines by fitting simulated data to the cancer specific dataset. To this end, we adopt Genetic Algorithm for multiple generations to search for the most likely network rewiring events and collected 50 rewired networks in each generations along with corresponding scores. The Genetic Algorithm performed rewiring on the basic apoptotic network of Figure 1. The comparison for simulated data from network N1 and Yaffe's data for control treatments using Genetic Algorithm, before and after rewiring event in network N1, is shown in Figure 3 and 4.\nFigure 3 Simulation results for Genetic Algorithm: fitting experimental data with simulation data. (A) Simulation result for basic apoptotic network, (B) Simulation result for DMSO treatment. (C) Comparing N2 based simulation data with ERL-DOX treatments data by correlation of apoptosis rates over time points. (D) Simulation result for ERL-DOX treatments.\nFigure 4 Due to inherent randomness in Genetic Algorithm, the edges added and deleted in N1 (while inferring N2) were different among different run of simulations. Some of the changes in the network were found to be more frequent compared to other. So, just few runs of simulation couldn't be relied upon to find the changes in the N1 network causing cancer. Therefore, to find consistent (or conserved re-wirings) changes to the network, we performed the simulation for 150 times for inferring N2 from N1 and another 150 rounds of simulation for inferring N3 from N2. The changes which occurred in the network with the highest frequency were selected to construct N2 (and subsequently N3). The most frequently removed edges in the N1 network while inferring N2 were SMAC-XIAP, Casp8-Casp3 and p53-PUMA. For 150 rounds of simulations, each of these deletions occurred with a frequency of 146, 143 and 141 respectively. The network edges found to be inserted in the N1 by the Genetic Algorithm with the highest frequencies were XIAP-Casp8 and TNFR-Stat3, with frequencies of 149 and 146.\nWe verified the various rewiring events using the existing literature and confirmed the validity for the above mentioned network changes. p53 is a tumour suppressor and controls the regulation of PUMA. PUMA induces apoptotic signals inside the cells. Less activity of PUMA leads to apoptosis deficiency which in turn leads to the increased risk of cancer [31]. We also realised that SMAC down-regulates XIAP and XIAP down-regulates Casp3, a pro-apoptotic protein. So if SMAC is working properly, there will be more cell death leading to less tumorigenic behaviour by cells. But if SMAC is not able to down-regulates XIAP, there will be less apoptosis and so more chances of cancer [2]. We also came across literature evidence highlighting the critical role of Casp8 for affecting breast cancer directly [5,32]. STAT3 interacts with various molecules involved in programmed cell death, regulating their functions including MOMP formation, which leads to the release of the cytochrome c [33]. XIAP inhibits the processing of Casp8. In fact, proteolytic processing of several caspases, including Casp9, is not allowed in the presences of XIAP. The lack of processing of Casp8 and Casp9 allows them to interact with Casp3 and promote apoptosis [34]. TNFR activates the Stat3 signalling. The stimulation of TNFR leads the phosphorylation of the Stat3 and Stat5b and the phosphorylated molecules are trans-located into the nucleus [35].\n\nInferring the drug sensitive network N3\nAfter inferring N2 from N1, our aim was to infer the drug sensitive network (N3), by fitting modified topology of N2 to the real data of drug treatment. Before predicting rewiring of N2, we plotted the N2 based simulation data with the real dataset of signalling and cell fate of cancer cells after treatment with drugs. The plot is shown in Figure 5. We again used Genetic Algorithm to search for the network rewirings that allow better fitting of simulation to the real data. For each run of simulation, we compared the simulation data with the drug sensitive experimental apoptosis data for ERL-DOX treatments. The best network plot, uncovered using Genetic Algorithm, is shown in Figure 6. The edges of N2 that were most frequently removed were BIM-Casp9 and EGFR-Casp8, deleted 141 and 137 times respectively. The network edges inserted with highest frequencies by the Genetic Algorithm were Casp8-RIP and Casp8-MOMP, inserted 144 and 135 times respectively. These rewiring events have been mentioned in related literature [2,5,17,36].\nFigure 5\nFigure 6\n\nSensitivity analysis of the model\nTo analyse the behaviour of our model with respect to the fluctuations in model parameters, we verified our model by performing sensitivity analysis for the rate constants. We initially tested apoptosis level for each of the molecules by reducing the parameter by 30%. Then we increased the parameter by 30%. Among all the parameters, \"Casp3-Casp8\" and \"Casp3-Apop\", were found to be very sensitive, affecting the apoptosis level to the maximum. We found that the level of apoptosis for most of the parameters changed only slightly, within +/-5%, except for the most sensitive parameters, \"Casp3-Casp8\" and \"Casp3-Apop\", which affected the level of apoptosis by +/-20%.\nTo test the robustness of our model, we also performed the perturbation test. We changed the initial concentration of each of the molecules and identified the most sensitive parameters of the model. The apoptosis level was predicted for each of the molecule whose initial concentration was reduced by 30%. Then, for each molecule we increased the initial concentration by 30%. We found that the level of apoptosis for most of the initial concentrations was stable, except for the concentration of the molecules with parameters of \"Casp3-Casp8\" and \"Casp3-Apop\". This suggests an overall robustness of our model."}