A growing number of studies make use of mathematical modeling techniques to better analyze and predict increasingly complex, dynamic data. While several groups have employed computational approaches to optimize oncolytic virotherapy 
, only two other groups have investigated the combinatorial dynamics that govern MEK-inhibitor mediated oncolytic adenovirus therapy 
. Here, we report our findings on improved treatment strategies for oncolytic adenovirus therapy, being the first to fully integrate modeling and experiment in the same study.
We performed time course measurements that confirmed previously observed CI1040-mediated CAR up-regulation and G1 cell cycle arrest 
. Based on these findings, we postulated that treating cells with CI1040 prior to infection, followed by its removal at the time of ONYX-015 infection, would (i) maximize virus uptake due to increased up-regulation of CAR, and (ii) maximize cell death (and consequently viral replication) due to the release of cells from G1-phase arrest. To explore this hypothesis, we developed an ODE model that characterized the proliferation, infection, and relative cell viability of a population of cancer cells subjected to MEK inhibition and ONYX-015 infection. We simulated combinations of different timings of MEK-inhibitor treatment initiation, timings of infection, and multiplicities of infection to ascertain their combinatorial effect on oncolysis. Surprisingly, our simulations suggested that, at low MOI, the greatest efficacy of virus-mediated cell death results from MEK-inhibition that coincides with the time of infection. This scenario is particularly relevant from a clinical perspective, since exposure to low MOIs is a likely limiting factor of treatment efficacy in vivo
, particularly following systemic virus administration. At higher MOI, our model predicts maximal cell killing when inhibitor treatment occurs at the time of, or immediately after, infection. We confirmed our predictions experimentally, showing that sensitizing cells via MEK-inhibition prior
to infection was less effective than treatment protocols that maintained CI1040 treatment during and following ONYX-015 infection. Consistent with our findings, simulations from an independent partial differential equation free boundary problem model presented by Tao and Guo 
suggested that greater tumor treatment is achieved when oncolytic adenovirus infection and MEK inhibitor treatment occur at the same time. Experimental validation was not carried out in that study.
The accuracy between simulated time courses and validation measurements () were striking given that validation conditions were beyond the scope of training data. More specifically, we predicted cell dynamics associated with MOIs 0.5 and 7; neither condition was considered in model development. Data reflecting MOIs 0.1, 1, 2, 5, and 10 were used for parameter estimation and subsequently interpolated to predict intermediate values. Furthermore, model fitting was based on time course measurements quantifying the effect of pre-treatment with MEK inhibitors (or DMSO) followed by infection, yet we are able to extrapolate cell viability dynamics for mechanistically unique protocols. Specifically, the simultaneous and post-treatment protocols involve experimental procedures that were unaccounted for in model development. Simulations also assumed that the MOI of the system remained unaffected upon removal of MEK-inhibitor by media change. This theory is accurate when treatment occurs prior to infection (as was the case in our training data). When treatment occurs at the time of or after infection, it is reasonable to imagine removal of MEK-inhibitor affecting virus titer, and hence the MOI. Despite these differences, model simulations accurately predict and extrapolate the nonlinear cellular response to combinatorial treatment strategies.
Further investigations of simulated predictions identified critical virus-host mechanisms responsible for enhanced combinatorial therapy. In particular, we explored how cell cycle phase affected oncolysis and virus production. Shepard and Ornelles 
demonstrated that ONYX-015 replicates more effectively in HeLa cells when infection occurs during S-phase rather than G1-phase. Later, Zheng et al. 
found that adenovirus E1B55K is required to enhance cyclin E expression; the failure to induce cyclin E expression due to E1B55K mutation in ONYX-015 prevents viral DNA from undergoing efficient replication in HeLa (and other non-permissive) cells when infected during G0-phase. In contrast, cyclin E induction is less dependent on the function encoded in the E1B55K of HCT116 cells whether the cells are in S- or G0-phase. This finding is consistent in other cancer cells that are permissive for replication of ONYX-015. As a result, we expanded our analysis to two additional oncolytic adenoviruses: Delta-24, which carries a deletion in the E1A region; and Delta-24RGD, which has an RGD-4C peptide motif inserted into the adenoviral fiber 
. The latter virus is able to anchor directly to integrins, providing CAR-independent mechanisms for infection. We found greater cell killing and virus production with Delta-24RGD when cells were infected during G1 cell cycle arrest. This result verifies the existence of a regulatory pathway that governs virus production in a CAR-independent manner. The mechanism underlying virus replication in G1-arrested cells remains unclear and warrants further investigation.
Model development is an ongoing process that needs to be tightly coupled with experiments in order to maximize mechanistic relevance and reflect the nonlinear complex dynamics critical to understanding and predicting biological function. However, it is important to note that our current model does not fully encompass the physiological complexities of malignant tumors in humans. It is clear that factors influencing drug distribution and elimination play a major role in this context. For example, the extent of vascular leakiness observed in tumors will impact viral extravasation 
. The immune responses directed against oncolytic viruses or tumor cells will also impact viral anti-tumor effects 
. Interestingly, recent in vivo
experiments demonstrate that the efficacy and specificity of virus replication in tumors modulate the immune response, highlighting yet another layer of complexity 
. Our aim is to develop multi-scale models that account for the greater dimensionality of oncolytic virus replication. Nevertheless, our current study demonstrates that dynamical mathematical models of oncolytic virus replication, tightly coupled with experimental studies, have the potential to optimize central aspects of this therapeutic approach.