Enhancing apoptosis in TRAIL-resistant cancer cells using fundamental response rules

Abstract

The tumor necrosis factor related apoptosis-inducing ligand (TRAIL) induces apoptosis in malignant cells, while leaving other cells mostly unharmed. However, several carcinomas remain resistant to TRAIL. To investigate the resistance mechanisms in TRAIL-stimulated human fibrosarcoma (HT1080) cells, we developed a computational model to analyze the temporal activation profiles of cell survival (IκB, JNK, p38) and apoptotic (caspase-8 and -3) molecules in wildtype and several (FADD, RIP1, TRAF2 and caspase-8) knock-down conditions. Based on perturbation-response approach utilizing the law of information (signaling flux) conservation, we derived response rules for population-level average cell response. From this approach, i) a FADD-independent pathway to activate p38 and JNK, ii) a crosstalk between RIP1 and p38, and iii) a crosstalk between p62 and JNK are predicted. Notably, subsequent simulations suggest that targeting a novel molecule at p62/sequestosome-1 junction will optimize apoptosis through signaling flux redistribution. This study offers a valuable prospective to sensitive TRAIL-based therapy.

Figure 1. TRAIL signaling pathway and experimental activation profiles of signaling molecules.

(A) Schematic topology of TRAIL signaling pathway. See maintext for details. (B) Experimental activation profiles of p38, IκB, JNK, caspase-8 and -3 in wildtype, RIP1 KD*, FADD KD*, caspase-8 KD*, and TRAF2 KD in arbitrary units (a.u.) at t = 0, 10, 30, 60** and 120 min after TRAIL stimulation of HT1080 cells. The original source was obtained from Figure 3A of ref. and was processed through imageJ (see Methods). *data is unavailable for caspase-8 and -3, ** available only for caspase-8 and -3. Note: interpolated dotted lines between experimental data points are inserted as a guide, they might not represent the actual temporal dynamics.

Figure 2. Computational modeling framework for TRAIL signaling.

Parameters of the initial model based on the original topology (1) are determined by overall fitting of experimental data using a genetic algorithm (GA, see Methods) for all molecules (i = 1,2,..n) and at all conditions (k = 0,1,…m), here n = 5 and m = 4. (2). If the overall error E = max(ε_ι,κ) between experimental and simulation profiles is higher than the set tolerance (3) (see Eq. 5) in Methods), the model is not acceptable. As the next step, the n molecules' activation profiles are ranked from the one showing highest (i = 1) to the lowest (i = n) error (4) for individual molecule's (5) best fitting in wildtype (6) and m other experimental conditions (6′). If the simulation of the i^th molecule fit reasonably in the k^th condition (individual error ε_i,k ≤ 0.15) (7), we check the next condition (k+1), else we modify the current topology according to response rules (8) (see Figure 3) and restart the procedure from wildtype condition again (6). If all m conditions fit for the i^th molecule (9–10) without any changes applied to the topology, we proceed to the next molecule (i+1) (11). If any change is necessary to the topology, the parameters have to be refitted for all molecules from the first molecule (i = 1). The whole procedure is repeated until the resultant model fit all experimental profiles of the n molecules within the error tolerance (12).

Figure 3. Response rules to modify signaling topologies when the first molecule is perturbed.

(A) Analyzing time to activation: Rule 1, Time delay and Rule 2, Rapid bypass, (B) Analyzing peak activation levels: Rule 3, Missing flux, Rule 4, Signaling Flux Redistribution (SFR), Rule 5, Lack of SFR and Rule 6, Dominant and Recessive flux, (C) Analyzing activation patterns: Rule 7, Reversible flux, Rule 8, Superposing flux, Rule 9, Continuous flux, and Rule 10, Oscillations. See maintext for descriptions. Note that rules 1–6 are developed from first-order response and the law of signaling flux conservation in pulse perturbation. Rules 7–10 are introduced to interpret any non-linear response or those that do not obey the law of conservation. These rules are not exhaustive and can possibly be further categorized if detailed experimental data for each molecule is available. The rules serve as guide to modify the overt topology highlighting the key missing features only.

Figure 4. Simulation of initial TRAIL signaling model.

(A) Static topology of the TRAIL signaling pathway used in developing our computational model. Note that we lump the similar effects of DR4/5 as TRAILR1/2, and ignore the response of DcR1/R2/OPG. Also, note that we include molecular conditions such as receptor clustering as additional first-order terms. (B) Comparison of simulations (solid lines) with experimental data (dotted lines) in wildtype, RIP1 KD, FADD KD, caspase-8 KD* and TRAF2 KD in arbitrary units (a.u.). The error between simulations and experimental data for the i^th molecule in the k^th condition is calculated based on the area between experimental and simulation curves (see Eq. 5 in Methods). *caspase-8 KD also refers to pro-caspase-8 KD.

Figure 5. Revealing novel features of TRAIL signaling using modeling strategy and response rules.

Model simulations compared with experiments. For p38, (A) M0, the initial model, (B) M1 with the addition of a rapid bypass, and (C) M2 with the addition of a missing link between RIP1 and p38 pathway. For JNK (D) M2, (E) M3 with intermediates to introduce delay in activation, (F) M4 with a missing link for the activation of JNK in FADD and caspase-8 KDs, and (G) M5 a missing link between p62 and JNK pathway to show enhancement through SFR in TRAF2 KD.

Figure 6. Simulations of the proposed TRAIL signaling topology.

(A) Comparison of M5 simulations (solid lines) with experimental data (black points) in wildtype, RIP1 KD, FADD KD, caspase-8 KD and TRAF2 KD. (B) Static topology of the proposed model for TRAIL signaling pathway. Modifications are indicated by blue arrows.

Figure 7. Identifying key target for sensitizing TRAIL resistance.

(A) Simulation profiles of p38, JNK, IκB, caspase-8 and -3 in Y and Z KDs. (B) Cell survival metric (CSM) for all KDs. (C) Survival ratio, SR, (experimental versus evaluated, from t = 0 to 120 min) in all conditions. Evaluated data is obtained using experimental data of RIP1 and FADD KDs (see Methods). (D) Wildtype HT1080 and HT29 (control) cells shows 60% and 95% survival, respectively, for 1000 ng/mL of TRAIL stimulation.