to BioTechniques free email alert service to receive content updates.
Mighty Modelers: The Art of Virtual Cell Culture

07/20/2016
Kelly Rae Chi

When translational biology problems are complicated, math comes to the rescue. Read more...


Image of a DCIS simulation.

Credit: Paul Macklin

By now, the idea that the Human Genome Project provided a “parts list” is a familiar notion. We’re awash in data and parts lists that extend well beyond genes. Computational modeling—in particular, incorporating multiple scales of biological and clinical data into new computer simulation—provides one means of wading through the mess of biology to come up with testable hypotheses. In the past decade, more biologists and clinicians have become enthusiastic about modeling. And it’s become a two-way street: Modelers now want to swim in biological data.

James Glazier: CompuCell3D

A decade ago, a biologist might hand data to a modeler who was content to run with it, creating theoretical explanations that may or may not map back to clinical situations. Not today, and certainly not for James Glazier, director of the Biocomplexity Institute at Indiana University Bloomington (IU). He and his colleagues have spent more than 15 years developing and applying an open-source modeling platform called CompuCell3D (1), which makes it easier for biologists to build their own models using molecular, cellular, and tissue-level data.

In a recent study, Glazier and his collaborators used CompuCell3D to study the etiology of polycystic kidney disease (2). Glazier’s collaborator, Robert Bacallao noticed a different complement of cell adhesion molecules lining the cells of diseased kidneys compared with normal kidneys. Complicating matters, the adhesion molecules played two major roles; they helped cells stick together and signaled for other cells to grow into an empty space. It was unclear which role might be involved in disease.

The simulations took a few weeks to develop, but helped reveal that both mechanisms were at play and that defects in each one led to noticeable differences in the shapes and behaviors of cysts. Bacallao confirmed these predictions in patients’ kidney cells.

Using other programs, it might take 40,000 lines of code to simulate the kidney, but with CompuCell3D, it takes about 100 lines. Glazier will continue to refine the kidney model and, because it is open-source, it can serve as a template for subsequent models. According to Bacallao, doing the necessary experiments in mice to reach the same conclusions would have taken 10–20 years.

Glazier is now forming a team of modelers within IU’s new Intelligent Systems Engineering department to work with clinicians. “When you solve these problems often enough—the eye, kidney, liver are some recent examples—there’s a workflow, he said. “We believe we can actually standardize this whole process.” His group is also working to make CompuCell3D more user-friendly.

Kristin Swanson: Patient-Specific Models

How can you determine how well an individual cancer patient is doing on a particular treatment? Should you switch to a different therapy? Clinicians compare a patient’s response to that of an average patient, a value obtained from clinical trial data. The problem is, we don’t know where on that curve any one patient exists, said Kristin Swanson, co-director of the Precision Neurotherapeutics Lab at the Mayo Clinic.

As a mathematical biology graduate student in the late 90s, Swanson served on the University of Washington Medical Center’s Tumor Boards, where oncologists of different subspecialities discuss individual cases and treatment decisions; she become immersed in these clinical decisions. During her service, she learned of glioblastoma, a deadly brain cancer with a median survival time of 15 months.

Partly motivated by the recent death of her father, who succumbed to a brain metastasis of lung cancer, Swanson decided to focus on glioblastoma. She worked with her mentor to incorporate serial imaging data into a “virtual control”—an understanding of what was going to happen to a person’s tumor if they received no therapy or the standard therapy. Published in a series of studies, the virtual controls gave clinicians a better baseline with which to gauge response to a change in treatment (3).

When exploring new therapies, “we need to know early on whether we’re onto something,” Swanson said. “We don’t have those tools. That’s what our methods help with.”

More recently, her group applied the virtual control approach to a 4-patient cohort receiving a new experimental therapy for glioblastoma and saw an early signal of response that would have gone undetected using standard tools.

In a study of nearly 250 patients with glioblastoma, Swanson’s team sorted patients into categories based on how diffuse their tumor margins were using modeling methods she pioneered during her graduate work. They found that patients with the most diffuse margins were least amenable to surgery, normally a first-line treatment (4).

The cancer field has not spent enough time matching patients to the right treatment, Swanson said. For many years, she was the lone mathematician on the Tumor Boards. But as the world of mathematical oncology is growing, more quantitative scientists will integrate into clinical decision-making—they have to, she said.

Paul Macklin: Ductal Carcinoma in Situ

As a new postdoc at the University of Texas Health Science Center in 2007, Paul Macklin was eager to apply modeling efforts closer to the clinic. He and his mentor met with a breast pathologist named Mary Edgerton at the MD Anderson Cancer Center in Houston, who like others in the field, was trying to understand ductal carcinoma in situ (DCIS), a worrisome precursor to breast cancer that appears on mammograms as tiny, calcified specks. DCIS is contained within the milk ducts of the breast, but these cells can unpredictably invade the surrounding breast tissue.

Now in his own lab at the University of Southern California, Macklin and his collaborators focus on modeling the natural course of DCIS by first building two-dimensional models of individual DCIS cells in a duct. The team makes do with what they have, which is often a single biopsy that provides only a snapshot. This gives them a sense of the overall balance between cell proliferation and cell death, as well as cell sizes. To predict how DCIS progresses, they mined the literature to incorporate the length of cell cycles and a cell’s approximate lifespan.

Their simulations found that DCIS grows surprisingly slowly and steadily, which mirrored clinical observations and gave Macklin the confidence to keep going (5,6). He and his collaborators went on to build a more flexible and realistic model of the duct wall, which suggested that DCIS cells penetrate the duct wall not by chewing at it gradually, but by charging through in less than an hour (7).

“Something happens that either changes the cell’s phenotype or changes communication in the system in some way that flips a switch, and the cell decides to plow through and move to the next steps,” Macklin said.

That model and pathology slides—which showed different cell types presumably in communication—“has really been motivating us for the past few years," said Macklin. "We can’t just model the tumor cells; we now have to model the whole system.”

This year, they released new code that models 10–15 diffusing signals in the microenvironment of DCIS (8) and plan to publish more aspects of their system this year.

Although Macklin’s models are becoming more complex, he is surprised by how well the earlier models matched with clinical data. “To me, that has been eye-opening,” he said. “Perfect is the enemy of good. You should take the simple model you have and run with it.”

References

1. Swat MH, Thomas GL, Belmonte JM, Shirinifard A, Hmeljak D, Glazier JA. Multi-scale modeling of tissues using CompuCell3D. Methods Cell Biol. 2012;110:325-66.

2. Belmonte JM, Clendenon SG, Oliveira GM, Swat MH, Greene EV, Jeyaraman S, Glazier JA, Bacallao RL. Virtual-Tissue Computer Simulations Define the Roles of Cell Adhesion and Proliferation in the Onset of Kidney Cystic Disease. Mol Biol Cell. 2016 May 18. [Epub ahead of print]

3. Adair JE, Johnston SK, Mrugala MM, Beard BC, Guyman LA, Baldock AL, Bridge CA, Hawkins-Daarud A, Gori JL, Born DE, Gonzalez-Cuyar LF, Silbergeld DL, Rockne RC, Storer BE, Rockhill JK, Swanson KR, Kiem HP. Gene therapy enhances chemotherapy tolerance and efficacy in glioblastoma patients. J Clin Invest. 2014 Sep;124(9):4082-92.

4. Baldock AL, Ahn S, Rockne R, Johnston S, Neal M, Corwin D, Clark-Swanson K, Sterin G, Trister AD, Malone H, Ebiana V, Sonabend AM, Mrugala M, Rockhill JK, Silbergeld DL, Lai A, Cloughesy T, McKhann GM 2nd, Bruce JN, Rostomily RC, Canoll P, Swanson KR. Patient-specific metrics of invasiveness reveal significant prognostic benefit of resection in a predictable subset of gliomas. PLoS One. 2014 Oct 28;9(10):e99057.

5. Edgerton ME, Chuang YL, Macklin P, Yang W, Bearer EL, Cristini V. A novel, patient-specific mathematical pathology approach for assessment of surgical volume: application to ductal carcinoma in situ of the breast. Anal Cell Pathol (Amst). 2011;34(5):247-63. doi: 10.3233/ACP-2011-0019.

6. Macklin P, Edgerton ME, Thompson AM, Cristini V. Patient-calibrated agent-based modelling of ductal carcinoma in situ (DCIS): from microscopic measurements to macroscopic predictions of clinical progression. J Theor Biol. 2012 May 21;301:122-40.

7. D'Antonio G, Macklin P, Preziosi L. An agent-based model for elasto-plastic mechanical interactions between cells, basement membrane and extracellular matrix. Math Biosci Eng. 2013 Feb;10(1):75-101.

8. Ghaffarizadeh A, Friedman SH, Macklin P. BioFVM: an efficient, parallelized diffusive transport solver for 3-D biological simulations. Bioinformatics. 2016 Apr 15;32(8):1256-8. doi: 10.1093/bioinformatics/btv730. Epub 2015 Dec 12.