Introduction

Branching morphogenesis represents a complex biological process, involving many participants interacting over time and space. The dynamic nature of the process makes construction of multiple static models of the system only marginally useful. It is the purpose of this project to draw on multiple computational approaches to construct a simulation model of branching morphogenesis that can be used to more realistically represent hypotheses about how branching morphogenesis occurs in-vivo. The ultimate goal is to design a system that is sufficiently detailed that it can be used to make predictions about the behavior of the biological system.

The Biology

Tissue morphogenesis is the process whereby a group of cells develops into an organized, multicellular structure. It has been demonstrated that the process involves not only communication between the cells themselves, but between the cells and noncellular environment, called the extracellular matrix (ECM). The extracellular matrix is composed of structural proteins (principally collagens and laminins), as well as bioactive molecules. It is now known that the interactions of cells with each other and with the ECM has a significant effect on the final structure and function of the developing tissue. Branching morphogenesis is a particular form of tissue morphogenesis; it is a process whereby ducts are formed in a developing tissue. A purpose of branching morphogenesis is to develop topologically distinct compartments with sufficient surface area to aid in transport processes within the organ. It occurs in mammary glands, liver, kidney, lung, and the vasculature, among other organs. Branching morphogenesis involves two principal processes: branch elongation, and new branch formation. Modifications of the extracellular matrix have been shown to influence branching morphogenesis in-vitro, and introduction of specific factors into a living organism can alter various observable features of branching morphogenesis and the behaviors of the resulting structures. In this project we focus on the mammary gland, where the process of branching morphogenesis is used in the development of the ducts that are used for milk production. This process has received a significant amount of attention because a number of the genes that have been shown to be important in branching morphogenesis in mammary tissue have also been shown to be important in tumor growth. Genes that have been shown to be important to both processes include those for the matrix metalloproteinases (MMPs) and the steroid hormones. The ductal epithelial cells express some of these factors. Others are expressed by cells found in the extracellular matrix, or are components of the ECM. The nature of the factors suggests that a few key processes are important in the development of the mammary gland. One is invasion of the mammary fat pad by the terminal end buds that give rise to new ducts. Another is the coordinated apoptosis of epithelial cells post lactation. The complexity of these processes makes it difficult to construct a sequence of "static" models. It is the goal of this project to use and explore new simulation methods as an approach to challenging existing hypotheses and formulating new hypotheses about these complex processes; in this case, the processes involved in mammary gland morphogenesis. It is also expected that by developing simulation models, further information can be obtained about the relevant importance of the various processes and factors involved. The usefulness of using our simulation models for testing new hypotheses will also be explored.

The principal cellular participants in branching morphogenesis include the epithelial cells, the myoepithelial cells, and cells found in the ECM, the scaffold within which cells reside. These cells have been shown to communicate with each other, either directly through secretion of growth factors or through cell-cell contacts, or indirectly through secretion of factors that modify the ECM. Particular sequences of communication have been hypothesized to give rise to branching morphogenesis. For example, the process of branching morphogenesis in-vivo is initiated by the release of ovarian hormones, which in turn results in the release of matrix metalloproteinases, and they then degrade a particular component of the ECM called the basement membrane (BM). Degradation of the BM is a necessary but not sufficient feature of branching morphogenesis. Other factors, such as TGF-beta, are involved in, for example, the formation of new branch sites. A diagram including some of the key players is presented in figure 1.

The Simulation

In order to simulate branching morphogenesis, an agent-oriented simulation is being constructed to represent key features of the biological system. Agent-oriented systems have been used to simulate various complex processes such as the A decomposition of the biology components that are important to the process of branching morphogenesis suggest that an agent-oriented representation of the will likely be the most fruitful approach for simulating morphogenesis

There are additional features that may be essential to included in simulations. One is the vasculature which provides blood supply to the tissue. Another is the presence of immune cells within the tissue. In order to start as simply as possible, some features will need to be excluded. Examples include features where their effects are simply to provide a consistent environment within which cells are to function, such as by providing nutrients or fighting an infection. Such features are often taken for granted in static models of branching morphogenesis. While they may also be excluded from our early simulations, these simplifying assumptions will be made explicit during the design of the simulation. Later, during iterative refinements, we aim to shrink the listed of required assumptions. Developing a simulation model of branching morphogenesis requires a detailed understanding of the biology. In order to address this issue, a database is being constructed as a repository for relevant biological information. The database includes tables of the proteins that are involved in branching morphogenesis; interactions that have been identified between proteins; types of ECM and their components; the cell types which have been identified in developing mammary tissue; and the proteins which have been identified as being expressed in these cell types. The database also stores references, which are used to annotate entries in the various tables. While there exist large databases such as protein-protein interaction databases, and cell signaling databases, these databases are often not up-to-date and rarely provide the amount of detail that is actually available in the literature. In addition to capturing biological data, our database will also be used to store functions that may be useful in describing the behavior of components within the simulation. It is these functional relationships which are less easily characterized through traditional experimentation but may be demonstrated as critical to understanding the behavior of the system being studied.

Testing of Existing Hypotheses Through Comparison of Simulation and Experiment

Typical experiments in mammary gland morphogenesis include the generation of knockout mice, introduction of particular factors into the tissue through implantation, or growth of cells using in-vitro systems. Results from such studies have identified a substantial number of factors capable of influencing branching morphogenesis in mammary tissue, either by increasing or decreasing the observed number of branches or duct length. While this information is readily tested against a simulation, it is difficult to compare qualitative descriptions such as "increased branching" or "increased cell division rate" with more detailed descriptions that can be readily determined from a simulation, such as number of branch nodes, the distribution of branch lengths, etc. There exists a substantial amount of image data that could be used to obtain such information; also, the existence of in-vitro systems of branching morphogenesis can be used as sources of such information against which a simulation can be validated. This requires the generation of a scoring function that can be applied to both a simulation and to in vitro experimental results. Our initial scoring function will be implemented as a linear combination of differences in features that are deemed important to the simulation. This will include information such as the ratios of cell types, growth rates, and number of branch nodes.

Formation of Hypotheses

In addition to designing simulations that are capable of testing current hypotheses about branching morphogenesis, we also require simulations that can also be used to test new hypotheses about how morphogenesis occurs. These new hypotheses could be identified automatically by altering simulation parameters such as ECM component levels or cell growth properties in an attempt to minimize our scoring function. Or they could be manipulated intentionally, in order to explore the effects of changing particular system features on the outcome of an experiment. Assuming the simulation is "suitably descriptive", it should be possible to use the simulation model to explore the effects of changes in components of the simulation on branching morphogenesis. In a later phase of the project agents representing normal epithelial cells can be replaced by agents representing cancerous cells in order to study the effects of changing operational rules.

Conclusion

It is the goal of this project to design simulation models that are sufficiently flexible to be incorporated into the scientific process to assist researchers in designing better, more informative experiments. So doing requires that we pay attention to user interface issues, and construct our models so that their components in a simulation can be easily added or removed. It is our operating hypothesis that the use of simulations as representations of hypotheses is necessitated by the complexity of morphogenesis. If a simulation can be designed that exhibits reproducible behavior that resemble in-vitro or in-vivo biological systems, it suggests that such a simulation may be an adequate representation of the in-vitro system given the expected criteria. There is a danger of "enscription error", where the simulation is intentionally or unintentionally "built" to represent the biological system. We will strive to avoid enscription errors. Longer term, this project is expected to contribute to the development of more formal methods for directing such refinements.