(B-55) Population dynamics model of ovarian cancer resistance for adaptive therapy strategies

Ovarian cancer is the second most common gynecological cancer accounting for 13,270 deaths in the United States in 2023. The difficulty in ovarian cancer diagnosis is due to its unnoticeable symptoms resulting in late detection and a 5-year survival rate of 50.8%. The standard of care for the disease involves neoadjuvant and adjuvant chemotherapy and surgery. However, around 80% of the patients develop resistance to platinum-based chemotherapies leading to cancer recurrence. Based on eco-evolutionary concepts, this treatment modality aims to kill treatment-sensitive cells allowing resistant clones to dominate in the tumor population. To combat the development of resistance, clinicians can leverage mathematical models of tumor population dynamics to determine alternative treatment schedules and doses known as adaptive therapy. This therapeutic regime allows clinicians to control the overall size of the tumor and reduce treatment toxicities. With this application in mind, we have utilized pure population ecology concepts to analyze the growth dynamics of two ovarian cancer cell lines (A2780 and Tyk-nu) and their cisplatin-resistant counter parts (A2780cis and Tyk-nu cp.r) undergoing different continuous treatment conditions and estimated their growth rate under exponential and logistic model assumptions. The growth rates were then used in a Lotka-Volterra model to understand the interactions of the sensitive and resistant populations.

Sensitive and resistant cell lines were cultured according to the manufacturer and were counted daily using two different techniques. Cells were lifted using Typsin-EDTA and counted with either the Countess 3 FL Automated Cell Counter or abstracted from images of live/dead labeled cells using CellProfiler. To study cell growth dynamics under cisplatin treatment, the cells were exposed to no treatment, 6µM cisplatin and 10 µM cisplatin media, which was changed every other day. We tested two mathematical models to estimate growth for all the cell lines under each of the three treatment conditions using procedures established in ecology. For the first model, exponential growth rates were calculated following Dennis et al. (1991). For the second model, the population growth rates were estimated assuming logistic growth which is characterized by a linear decline in growth rate with increasing population size using Williams et al. (2002). Growth curves were computed and compared by R² values that were calculated to evaluate accuracy of model estimates.

Assuming that cancer cells are able to access unlimited resources from their host and that only sensitive cells can convert to the resistant phenotype, the exponential growth rates found here were used to simulate a Lotka-Volterra system at different initial conditions. To identify cellular dominance, a population residual was then calculated by subtracting the resistant cell counts from the sensitive and taking the average. A positive population residual indicates higher average sensitive counts and thus an optimal tumor composition for adaptive therapy.

Under treatment naïve conditions, A2780 had a higher exponential growth rate compared to A2780cis whereas the opposite trend was shown in Tyk-nu and Tyk-nu cp.r populations (Fig1A). Under cisplatin treatment, sensitive populations presented a 1.5- to 2-fold decrease in growth rate, while resistant populations presented a maximum of 1.6-fold decrease in growth rate. Taken together, resistant cells had higher growth rates compared to their sensitive cell counterpart when treated. However, low R² values (< 0.6) were observed for most cells under the different treatments indicating that the exponential growth model is not optimal to estimate these growth patterns (Fig 1B-C).

Logistic growth rates of untreated cell are shown in Fig1D and are generally higher than exponential growth rates. Under cisplatin treatment, sensitive populations presented a 1.4- to 2-fold decrease in growth rate, whereas resistant populations presented a similar growth rate under all treatments. For these fits, most R² were >0.9 (Fig1E-F), therefore the logistic model explains cells growth dynamics more accurately.

To simulate clinically relevant interactions, the exponential growth rates were used in our Lotka-Volterra framework assuming unlimited resources provided by host. Simulations under treatment show higher sensitive cell counts throughout a 10-day period when the initial cell concentrations are at a 90/10 ratio (Fig1G-H). This resulted in a positive population residual suggesting that at a 90/10 sensitive/resistant composition, the tumor had enough sensitive cells to still dominate and control the resistant population at a 10 µM cisplatin dose. For initial conditions of 50/50 and 10/90 cells, the population residuals were negative, indicating that the resistant population could outcompete the sensitive population.

In conclusion, we showed the population dynamics of two ovarian cancer cell lines and their resistant counterparts under different treatments and fitted their growth to population ecology models. Here, the logistic model was the best fit for the data collected in 2-D culture. The Lotka-Volterra framework simulated interactions between sensitive and resistant populations to determine potential treatment strategies. While population ecology concepts were able to provide insights of cellular growth, we will further develop these methods by optimizing parameter estimation of the growth rates and cellular interaction terms.