Conceived and designed the experiments: NB MS DAL WMK. Performed the experiments: NB MS. Analyzed the data: NB MS. Wrote the paper: NB.
The authors have declared that no competing interests exist.
Oncolytic adenoviruses, such as ONYX-015, have been tested in clinical trials for currently untreatable tumors, but have yet to demonstrate adequate therapeutic efficacy. The extent to which viruses infect targeted cells determines the efficacy of this approach but many tumors down-regulate the Coxsackievirus and Adenovirus Receptor (CAR), rendering them less susceptible to infection. Disrupting MAPK pathway signaling by pharmacological inhibition of MEK up-regulates CAR expression, offering possible enhanced adenovirus infection. MEK inhibition, however, interferes with adenovirus replication due to resulting G1-phase cell cycle arrest. Therefore, enhanced efficacy will depend on treatment protocols that productively balance these competing effects. Predictive understanding of how to attain and enhance therapeutic efficacy of combinatorial treatment is difficult since the effects of MEK inhibitors, in conjunction with adenovirus/cell interactions, are complex nonlinear dynamic processes. We investigated combinatorial treatment strategies using a mathematical model that predicts the impact of MEK inhibition on tumor cell proliferation, ONYX-015 infection, and oncolysis. Specifically, we fit a nonlinear differential equation system to dedicated experimental data and analyzed the resulting simulations for favorable treatment strategies. Simulations predicted enhanced combinatorial therapy when both treatments were applied simultaneously; we successfully validated these predictions in an ensuing explicit test study. Further analysis revealed that a CAR-independent mechanism may be responsible for amplified virus production and cell death. We conclude that integrated computational and experimental analysis of combinatorial therapy provides a useful means to identify treatment/infection protocols that yield clinically significant oncolysis. Enhanced oncolytic therapy has the potential to dramatically improve non-surgical cancer treatment, especially in locally advanced or metastatic cases where treatment options remain limited.
Novel cancer treatment strategies are urgently needed since currently available non-surgical methods for most solid malignancies have limited impact on survival rates. We used conditionally replicating adenoviruses as cancer-fighting agents since they are designed to target and lyse cells with specific aberrations, leaving healthy cells undamaged. Highly malignant cells, however, down-regulate the adenovirus receptor, impairing infection and subsequent cell death. We demonstrated that disruption of the MEK pathway (which is frequently activated in cancer) up-regulated this receptor, resulting in enhanced adenovirus entry. Although receptor expression was restored, disruption of signaling interfered with adenovirus replication due to cell cycle arrest, presenting an opposing trade-off. We developed a dynamical systems model to characterize the response of cancer cells to oncolytic adenovirus infection and drug treatment, providing a means to enhance therapeutic efficacy of combination treatment strategies. Our simulations predicted improved therapeutic efficacy when drug treatment and infection occurred simultaneously. We successfully validated predictions and found that a CAR-independent mechanism may be responsible for regulating adenovirus production and cell death. This work demonstrates the utility of modeling for accurate prediction and optimization of combinatorial treatment strategies, serving as a paradigm for improved design of anti-cancer combination therapies.
Therapeutic options for most patients with locally advanced or metastatic cancer are limited. Surgery is often not an option for these patients because the cancer has diffusely spread, and currently available non-surgical treatments for most solid malignancies have insufficient impact on survival rates. Therefore, novel treatment strategies that incorporate the molecular composition of individual tumors are urgently needed. Conditionally replicating oncolytic adenoviruses are designed to target and lyse cells with specific aberrations, showing promise as a new non-surgical treatment strategy
ONYX-015 is an oncolytic adenovirus that lacks the E1B-55K gene product required for p53 degradation and therefore was predicted to selectively replicate in tumor cells with inactive p53 pathways
In previously published work, we explored the possibility of pharmacologically up-regulating CAR in colon cancer cell lines through inhibition of signal transduction pathways involved in its repression. We were able to demonstrate that inhibition of MEK, as well as TGFβ, up-regulates CAR expression
In order to generate sufficient experimental data quantifying the mechanistic behavior critical to predicting nonlinear dynamics, we systematically assessed CAR expression, cell proliferation, infection, cell viability, and viral replication in the presence and absence of MEK inhibitors (namely, CI1040). In agreement with our previously published work
HCT116 cells were treated with MEK inhibitor CI1040, DMSO, or alone. (
We sought to build a model that captures the key phenotypic behavior of tumor cells responding to combinatorial therapy. We fit an ordinary differential equation (ODE) model to measurements of proliferation, infection, and relative cell viability, characterizing how an
System states are depicted in black bold capital font (
System states (shown in black bold capital font) represent the nonlinear dynamic behavior of (i) uninfected cancer cell density,
Corresponding parameter values govern the rate at which state variables proliferate, arrest in (release from) the G1 cell cycle phase as a result of MEK-inhibitor treatment (removal), infect, and lyse (
Parameter estimation involves changing the model's parameter values until the difference between the model output (i.e., simulation) and experimental data is minimized as defined by the sum of squares error, SSE, weighted by the measurement error associated with each data point. The weight corresponds to the inverse standard deviation of replicate measurements. If the standard deviation is relatively small, our confidence in the measurement is high, so we penalize the simulated error with greater magnitude. Weighted SSE was employed when fitting proliferations kinetics since multiple data replicates were available; standard SSE was employed when fitting infection and viability kinetics (
We expect delays throughout the infection cycle: endocytosis, viral replication, and lysis require a sequence of non-instantaneous sub-cellular events
Model parameters were estimated in sequence to improve the biological relevance of fitted values while emulating experimental conditions. Most resulting parameters were defined as functions of the system's inherent control inputs, where the duration of treatment is set to 2-days:
We made several assumptions to simplify model development and supporting experiments. First, we neglected state transitions between cell cycle phases and characterized only the switch between proliferating cells and cells arrested in G1-phase as a result of treatment with MEK-inhibitor. Regarding enhanced infection and cell death, we presumed that MEK-inhibition caused a switch-like sensitivity to infection instead of a linear progression. Therefore, if 100% of cells were arrested in G1-phase as a result of MEK-inhibition, prolonged treatment would not further enhance infection. We neglected dose and duration of treatment as system control variables and limited simulated MOI between 0.1 and 10 to avoid error due to extrapolation. We assumed cancer cell populations were spatially uniform such that experimental measurements reflected a deterministic (rather than spatial or stochastic) mean behavior. Finally, we did not explicitly characterize virus titer. Instead, the presumed effects of virus dynamics were consolidated into estimated parameters.
We interpolated intermediate parameter values and used the model to predict the extent of cell death as a function of the time of CI1040 treatment initiation, the time of ONYX-015 infection, and the MOI. We employed an exhaustive search algorithm to simulate the effect of various treatment and infection protocols. This algorithm systematically evaluated every possible sequence combination of drug treatment and infection conditions (within a defined interval), with the exception of media change,
Simulated percent cell death (CD) is evaluated on day 8 as a function of the timing of MEK-inhibitor treatment initiation, timing of infection, and multiplicity of infection (MOI). Each Cartesian coordinate reflects an independent simulation or treatment/infection protocol. The timing of ONYX-015 infection is varied on the x-axis; the timing of CI1040 treatment initiation is varied on the y-axis. CI1040 removal by media change occurs 2 days post treatment irrespective of the timing of infection. MOI is held constant in each subplot. Percent cell death is defined as the complement of cell viability. Treatment and infection protocols that yield over 50% cell death are shown. Greater cell death is reflected by larger data points and an increasingly red color (see color bar). Empty data points depict protocols that fail to kill at least 50% of the cellular population.
To experimentally validate the predictive capabilities of the model, we simulated (
HCT116 cells were treated with CI1040 or DMSO, and infected with ONYX-015 at MOI = 0.5 or MOI = 7. The cell viability outcome of three different treatment protocols is compared in each plot:
To further investigate conditions that give rise to increased therapeutic efficacy, we correlated simulated cell death profiles with (i) cell confluency at the time of treatment, (ii) the proportion of cells in CI1040 mediated G1 cell cycle arrest at the time of infection, and (iii) cell confluency at the time of infection. Little correlation between cell death and cell confluency at the time of drug treatment was found by Pearson correlation analysis (R = .2;
HCT116 cells were seeded at 2e4 cells/well (low density) or 1e5 cells/well (high density) in 96-well plates. Low and high density cells were treated with DMSO or CI1040, and infected at an MOI of (
Model simulations and experimental validation confirm that simultaneous treatment with MEK-inhibitor and infection is most advantageous, suggesting that alternate (CAR-independent) regulatory mechanisms may be responsible for enhanced oncolysis. Given the disproportionate increase in virus replication relative to infection (
HCT116 cells were density arrested and released from synchronization. (
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
We performed time course measurements that confirmed previously observed CI1040-mediated CAR up-regulation and G1 cell cycle arrest
The accuracy between simulated time courses and validation measurements (
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
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 colon cancer cell line, HCT116, was kindly provided by Dr. B. Vogelstein (Johns Hopkins Cancer Center, Baltimore, MD). HCT116 cells were cultured in McCoy's 5A medium (UCSF Cell Culture Facility, San Francisco, CA) supplemented with 10% fetal bovine serum (Valley Biomedical Products, Winchester, VA).
Viruses included a wild-type adenovirus, WtD; an E1B-55K-deficient adenovirus mutant, ONYX-015; an E1A-deficient adenovirus, Delta-24
For inhibition of RAF-MEK-ERK signaling, the MEK inhibitor CI1040 (Pfizer, Ann Arbor, MI) was used at a final concentration of 5 µM. As a control, cells were treated with DMSO (0.1%).
For CAR staining, cells were treated with CI1040, DMSO, or cell culture medium alone (as stated previously). Over the course of 4 days, the cells were harvested daily using 0.05% trypsin (UCSF, Cell Culture Facility, San Francisco, CA). After media change in PBS (UCSF), cells were incubated for 45 minutes at 4°C with the mouse monoclonal anti-CAR antibody RmcB (1∶50)
For cell proliferation, HCT116 cells were seeded in 6-well plates and immediately treated with CI1040 or DMSO (as stated previously) for 1, 2, or 3 days and harvested 1–7 days after treatment. Cells were counted using a C6 Flow Cytometer (Accuri Cytometer). Treated cells were also analyzed for cell cycle distribution (please refer to
HCT116 cells were seeded in 96-well plates overnight and infected with WtD, ONYX-015, Delta-24, or Delta-24RGD at different MOI. Cell viability was measured by the CellTiter 96 Aqueous One Solution Cell Proliferation Assay (MTS) (Promega, Madison, WI) 1 to 7 days post-infection. Cell viability was expressed as percentage of the uninfected medium control (i.e. MOI = 0). Therefore, any relevant toxic effects are normalized from the relative viability measurements.
HCT116 cells were density-arrested by plating at 5e5 cells/cm2 for 2 days. Cells were released from arrest by re-plating at low density, 1e5 cells/cm2. Cell cycle was analyzed by Propidium Iodide (Sigma) staining as described above. CAR expression was analyzed at 7, 16, and 24 hours after release from arrest using RmcB antibody as described above. Synchronized cells were infected with WtD, ONYX-015, Delta-24, or Delta-24RGD at an MOI of 1 and subsequently measured for viability by adding Propidium iodide (Sigma) to a final concentration of 1 µg/mL just prior to acquisition to exclude dead cells and counting cell numbers using a C6 Flow Cytometer (Accuri Cytometer). Cell viability was expressed as percentage of the uninfected medium control (i.e. MOI = 0).
Viral titers of harvested cells were determined by the Adeno-X Rapid Titer Kit (Clontech) as described by Shiina et al.
Parameters were fit to experimental measurements by minimizing the sum of squares error (SSE) between the simulation and the data using the genetic algorithm function in MATLAB. When multiple data replicates were available, the SSE was weighted by the inverse standard deviation of experimental measurements. A gradient search algorithm, fmincon, was used post-estimation to ensure convergence to a local minimum. Please refer to
Contents: 1) parameter estimation; 2) parameter convergence; 3) model fitness; 4) interpolation methods; 5) model simulations; 6) MATLAB syntax for ordinary differential equation model; 7) MIFlowCyt outline.
(0.89 MB PDF)
NB would like to thank Dr. David C. Clarke and Dr. Nathan C. Tedford for helpful discussions.