PMC

Combinatorial drug viability responses of BTK inhibitor Ibrutinib in combination with additional inhibitors targeting BCR network intermediates were predicted and compared with experimental data.

(A) Modes of interaction of all pairwise inhibitions under Bliss Independence model. β<1, β=1, β>1 correspond to synergism, additive and antagonism respectively. (B) In vitro validation of predicted synergistic and antagonistic drug combination in ABC DLBCL cell line TMD8, HBL1 and OCI-LY10. R406, Dasatinib and MI-2 are inhibitors against SYK, LYN and MALT1 respectively.

The central BCR signaling network constructed from literature. Antigen binding induces BCR aggregation and subsequent phosphorylation, which further triggers a complex signaling cascade initiated by phosphorylated LYN and SYK. The BTK-PLCγ2-PKCβ pathway activates downstream NFκB and ERK through divergent paths, while membrane recruitment of PI3K leads to AKT activation. Pathway crosstalks and feedback regulations are highly abundant in the network.

Training and prediction of single drug viability response in ABC DLBCL cell line TMD8. (A) Tumor growth model parameterization using single drug viability response of inhibitors targeting NFκB, AKT and MEK. Gray dashed lines correspond to simulation results of model without SYK to LYN negative feedback, while brown dashed lines correspond to simulation results of model without PI3K-NFκB crosstalk. (B) Single drug viability response of inhibitor targeting various kinases against BCR signaling network.

Simulation of normal BCR signaling and estimation of kinetic parameters. (A) Simulated trajectory of ten parameter sets in comparison with published phosphorylation time course data. (B) Comparison between literature-retrieved parameter values with simulation-estimated parameter ranges. Box-plot indicates the simulation-estimated parameter ranges while black dots represent literature-retrieved parameter values. (C) Empirical cumulative distribution of the number of parameters that would fall within simulation-estimated parameter ranges by chance derived by shuffling the literature-retrieved parameters 10,000 times.

Outline of the approach taken in the present study. The central BCR signaling network was constructed based on validated protein-protein interactions from experimental literature. Parameters of molecular reaction kinetics were estimated from phosphorylation time course data and protein concentrations were retrieved from MOPED protein expression database. A phenotypic tumor growth model was trained on cell viability assays of inhibitor treatments to link signaling response to viability outcome. In the end, simulation of the signaling model in combination with the tumor growth model was conducted to optimize treatment strategy. The model’s prediction was compared to published drug response data and new prediction-driven hypotheses were tested independently in vitro.

Computational optimization of treatment strategy against chronic active BCR signaling (A) Viability response surface of three drug target combinations. Two horizontal axis corresponds to virtual dosage of two different inhibitors while the vertical axis indicates cell viability normalized to untreated condition. For each drug target combination, 10 virtual dosages (between 0% to 99% inhibition evenly spaced in log10 space) of each single drug are tested. (B) Barplot of simulated viability response of all possible dual inhibition on 11 kinases in the BCR signaling network that are currently targetable. Here viability responses are calculated area under the combinatorial viability response surface (as shown in Fig 6A) as an overall indicator of drug combination potency. Binary codes on the bottom indicate the treatments applied (black represents targeted inhibition).