Identifying therapy targets & evolutionary potentials in ovarian cancer

For those of us attending the 7th annual Integrated Mathematical Oncology workshop (IMO7) at the Moffitt Cancer Center in Tampa, this week was a gruelling yet exciting set of four near-all-nighters. Participants were grouped into five teams and were tasked with coming up with a new model to elucidate a facet of a particular type of cancer. With $50k on the line and enthusiasm for creating evolutionary models, Team Orange (the wonderful team I had the privilege of being a part of) set out to understand something new about ovarian cancer. In this post, I will outline my perspective on the initial model we came up with over the past week.

Ovarian cancer

Comparatively, very little is known about ovarian cancer. The poor prognosis of those diagnosed with ovarian carcinomas has not seen significant improvement over the last 30 years, and this has been attributed to little knowledge about the history of this highly-heterogeneous cancer Labidi-Galy et al., (2017).  In fact, it was only this year that new evidence showed ovarian carcinomas originating in the fallopian tubes, not the ovaries themselves. While the place of origin is of paramount importance, so are the sites of distant metastasis. Similarly, little is known about what leads to the success of some ovarian carcinomas in different metastatic sites, but we understand that reduction of distant metastasis is certainly desirable for improved outcome and increased time to recurrence.

A rather unfortunate facet of ovarian carcinoma is that both the cancer and its environment are extremely heterogenous. Among ourselves (Team Orange), we conjectured that the intra and inter-tumor heterogeneity may be caused by the ‘lush’ environment. That is, because the environment is so accommodating of cell needs,  there is little selection pressure which leads to a much higher rate of neutral evolution. For us modelers, this meant that there was not a well-defined starting point; no strong signals from genomic analysis, no distinguished key-proteins or proteases in the evolution of the cancer — though a review of many factors can be found in Davidson et al. (2014).

Luckily, the clinician on our team, Dr. Robert Wenham, recommended that we use clinical data to come up with a mechanistic model for tumor evolution that would allow us to identify key players in the mechanisms of tumor-stroma interactions that could lead to distant metastasis. The hope here is that with a little data and modeling magic, we can start to identify the combination of factors in this lush environment that leads to the creation of extraordinary cells which take up residence in distant metastatic sites.

Given the workshop’s focus on stroma — a broad term referring to all non-cancer factors in the oncological ecology — we attempted to build an evolutionary model of ovarian cancer that would help us exhume greater knowledge of the molecular factors, evolutionary players, and therapy responses from existing and forthcoming data.

Clinical data

How we model this ‘lush’ environment was largely shaped by the data that we had in hand and the data that is currently being collected. Thanks to current and previous trials taking place at the Moffitt Cancer Center, we had access to mass spectrometry imaging of tumor biopsies and began to analyse those to identify clusters of stromal factors that may define different ‘niches’. We define a niche as any spatially related area expressing similar levels of pre-chosen biomarkers:

Team Orange image analysis illustrates pipeline for niche identification from mass spec imaging. Completed by Chandler Gatenbee. (a) Cells marked based on nucleus and spatially separated via Voronoi Tesselation (b) Image of collagen expression in the cells (c) The automatically identified niche for the collagen expression.

Identifying these niches and their spatial composition by molecular markers will allow us to have an understanding of the spatial distributions and ecologies our model should produce. Analysing the outcome and time to recurrence of patients with similar ecological niches will allow us to further validate our model and use its simulated outcome as predictive of prognosis and time to recurrence based on future patient biopsies.

Coarse grained abstraction of the major players in evolution of ovarian cancer. (Right loop) – Cancer cells express VEGF which promotes formation of blood vessels, which increases the resources a cancer cell receives. (Left loop) Predation of cancer cells by T-Cells is inhibited by expression of PD-L1. (Center loop) – Changing of strategy to quiescense. Participation in these loops may lead to increased performance in metastatic sites.

Designing the model

Though I have discussed how molecular markers are indicative of different niches and may therefore provide clues about the probability of distant metastasis, I have not yet defined what markers we would like to look at.  In fact, this is something the team struggled with quite heavily, but in the end, we came up with the abstraction at right of tumor-stroma interactions in ovarian cancer.

The thought here is that modeling the promotion or inhibition of three related yet competing factors — vascular, immune, and quiescence — would allow us to characterize the main players in the ovarian cancer ecology. Attacking any one of these three systems may lead to the promotion of selective pressures which may in turn lead to reduction of the cancer’s heterogeneity. Of course, if this is the case, we would expect the geno/phenotype which was selected for to play a large roll in the success or failure of metastasis.

A highly hypoxic-resistant (loosely-termed quiescent) cells form a tumor surrounded by blood vessels and T-cells within the model designed by team orange. PDE for resources is in pink and PDE for chemo is in red.

The above system was represented as a hybrid cellular automata where the cancer cells are characterized by their levels of expression of PD-L1, VEGF, and ability to survive hypoxia (loosely-termed quiescence). Other cells in the space are representative of vasculature (red cells) and T-Cells (green cells). Vasculature emanates resources that the cells need (i.e. oxygen) and if a cell is unable to receive this resource then it dies after a certain amount of time this time is extended if the cell expressed the gene responsible for quiescence. Similarly, those cells expressing PD-L1 are able to survive encounters with T-Cells, and cells which express VEGF are able to more easily recruit new vasculature to survive. We modeled the emanation of resources from vasculature with a partially differentiable equation (PDE).

We believe that this relatively simple model which encapsulates the primary interactions of three main phenotypes will allow us to test different combinations of observed niches. More importantly, we can observe how these niches respond to different therapies and how those therapies introduce a selective pressure which may affect metastasis. In order to do so, we introduce another PDE representing the chemotherapy. Cells which encounter one of the six blasts of chemotherapy are likely to die. In our model, we observe a higher number of quiescent and PD-L1 expressing cells in a post chemotherapy ecology. Initial models of metastatic sites have been attempted, but were unable to be refined in the context of a four-day workshop.

Gif of the model in action and plots of the observed behavior. Though this is only a preliminary model, we see the rise of hypoxic-resistant cells and PD-L1 expressing cells post chemo therapy.

Right now, we are able to access data which is part of several on-going Moffitt projects, but have not refined the variables of our model, nor were we able to pull together realistic models of metastatic sites. In the future, we hope to create a more analytic representation of the dynamics, and we hope to begin diving deeper into specific niches using the data collected from ongoing clinical trials.

References and Acknowledgements

Davidson, B., Trope, C., & Reich, R. (2014). The role of the tumor stroma in ovarian cancer. Frontiers in Oncology, 4.

Labidi-Galy, S.I., Papp, E., Hallberg, D., Niknafs, N., Adleff, V., Noe, M., … & Hruban, C. A. (2017). High grade serous ovarian carcinomas originate in the fallopian tube. Nature Communications, 8(1): 1093.

Thanks again to all of the members of the Orange Team who worked really long hours to make this model. It was truly a unique experience to hear and collaborate with such intelligent people. IMO7 Orange team was comprised of:

Robert Wenham, Ryan Schenk, Maximilian Strobl, Linggih Saputro, Jill Gallaher,  Martijn Koppens, Chandler Gatenbee, Salvador Cruz García, Vikram Adhikarla, Andrew Shockey, Pantea Pooladvand, Eszter Lakatos, and Mehdi Damaghi.

About Matthew Wicker
I am currently majoring in computer science at the University of Georgia with a minor in applied mathematics. In my free time I enjoy investigating open problem and discussing complex software engineering challenges. My research interests include computation biology, verification, graph theory, and I'm trying to get into game theory more.

3 Responses to Identifying therapy targets & evolutionary potentials in ovarian cancer

  1. Pingback: Cataloging a sparse year of blogging: IMO workshop and preprints | Theory, Evolution, and Games Group

  2. Pingback: IMO Workshop 7: Stroma – Ryan O. Schenck

  3. Pingback: Blogging community of computational and mathematical oncologists | Theory, Evolution, and Games Group

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: