SE, DH, and IRC conceived and designed the experiments, contributed reagents/materials/analysis tools, and wrote the paper. SE performed the experiments and analyzed the data.
The authors have declared that no competing interests exist.
Experiments have generated a plethora of data about the genes, molecules, and cells involved in thymocyte development. Here, we use a computer-driven simulation that uses data about thymocyte development to generate an integrated dynamic representation—a novel technology we have termed
Biological systems are the embodiment of complexity that defies intuitive understanding. Biologists have accumulated masses of data about the molecules, cells, and discrete interactions that compose living systems, but the list of facts alone cannot explain how such systems work dynamically. We have developed a hybrid, computational approach to the simulation of complex systems called
The mammalian thymus receives stem cells from the bone marrow. These cells—thymocytes—go through a series of anatomical subcompartments in a process termed
Extensive research in disparate disciplines has uncovered a mass of data regarding thymocyte development. Subfields of thymus research include genes, gene expression and differentiation; molecules (integrins, chemokines, cytokines, receptors, antigens, and other ligands); cells (stem cells, thymocytes, epithelial cells, dendritic cells, and macrophages); cell behavior (adhesion, migration, and anatomic localization); cell states (differentiation states, cell cycle, proliferation, and apoptosis); and physiology (antigen expression, positive and negative selection, lineage choice, and antigen-receptor repertoires). The technologies used to study thymopoeisis include genetics, transgenes and gene knockouts, protein chemistry, microscopy and immunohistochemistry, in vitro cell cultures and interactions, in vivo phenotypes, cell and organ transfers, immunizations, and more. A systematic integration of these data into an accurate and comprehensive representation is much needed. We address this need using reactive animation (RA) to reveal multiscalar emergent properties and to guide experimentation in thymocyte development.
RA is a computational approach to simulating complex dynamic systems. The technology of RA has been described elsewhere [
(A) Different colors stand for different stages during thymocyte selection. The legend box shows the different developmental stages and their corresponding colors.
(B) Different icons beneath the figure stand for different tools for the control and the visualization of the simulation. The tools include the ability to zoom into parts of the visualization, show and hide the two tiers of the simulation, highlight interacting cells, display or hide chemokine gradients, display statistical information through real-time communication with Matlab, visualize apoptotic levels in real-time, initiate and stop a backtrack of a cell's migration, and more.
(A,B) In the lower right hand side, the elapsed time is displayed, together with a utility that enables the control of time progression.
RA differs fundamentally from other approaches directed to network modeling [
Data at the level of single cells and their microenvironment culled from hundreds of papers were coded to the simulation. Anatomic localization is critical to thymus development; thymocytes at different developmental stages migrate to specific thymic compartments [
We generate the visual trace by utilizing the built-in tracing tool. The trace is color-coded: the current time is highlighted in red and the beginning of the trace is enriched with blue. Intermediate times are marked as mixtures of these two colors on the trace line.
The effects on thymus fine anatomy of chemokine receptors CXCR4 [
(A) The left panel shows a figure borrowed from [
(B) A wild-type thymus (labeled
In contrast to the CXCR4 knockout, deleting CCR9 had no major effect experimentally on intrathymic T cell development [
The RA simulation results, presented in
(A) Shows the distribution of wild-type cells (color coded as in
(B) Shows the altered distribution of the knockout cells (colored gray). The normal cells and the altered cells are colored differently to be compatible with
RA also made it possible to observe the dynamics in silico of a competitive experiment, in which equal numbers of CCR9−/− and wild-type cells are seeded into the thymus:
The three panels at the bottom, with context lines leading to the time-graph, show the ratios between the two cell populations developing over time. An initial peak of maturing wild-type cells is followed by a decrease and an eventual asymptotic ratio, as the buildup of random pressure of CCR9−/− cells eventually generates homeostasis. An asymptotic value of four wild-type thymocytes to every CCR9−/− thymocyte is reached. See text for further discussion.
Thymocytes need to traverse developmental niches; thus, when the number of thymocytes exceeds the space available for antigen presentation sites on epithelial cells, the thymocytes pile up and those waiting their turn for stimulation may undergo apoptosis from the lack of interaction [
We visualize the different levels of apoptosis by using different colors to show the relative numbers of apoptosing cells across the thymic grid. Red zones correspond to higher levels of apoptosis.
(A) Shows the normal distribution of apoptosis primarily to the SCZ; some apoptosis is seen in the outer cortex (OC) and some in the medulla (M); the inner cortex (IC) and CMJ show relatively fewer apoptotic cells.
(B) Shows the influence of removing competition between thymocytes for developmental niches. The lack of competition moves the bulk of apoptosis from the SCZ to the CMJ and the M zones. In the wild-type thymus, most of the cells die in their DP stages, in the cortex, and in the SCZ. This is in agreement with experimental results, where only 10% of cells survive to the SP stage (see review; [
RA in silico experimentation suggests that competition also selects for differential speeds of trafficking in response to chemokine gradients.
The faster cells are more likely to survive thymic selection by contacting the process of an epithelial cell; but slower cells also can have an advantage (see text).
Another prediction emerging from cell competition relates to lineage commitment. A developing thymocyte must choose whether to become an SP CD4 T cell (helper) or an SP CD8 T cell (cytotoxic). The decision-making process is obscure because mature SP CD4 and CD8 T cells evolve from precursors that are DP for both CD4 and CD8, yet CD4 cells predominate at a 2:1 ratio. Current theories of lineage commitment deal with the molecular details of the choice. The two most significant themes in the theories distinguish between an “instruction” approach and a “stochastic” approach [
However, the emergence of competition between thymocytes for interaction space provides a novel solution to the CD4:CD8 2:1 paradox. If the dissociation rates of CD8 cells from epithelial cells are lower than those of CD4 cells, then the CD8 cells will remain longer at their epithelial-cell interaction stations (peptide-MHC I sites). As long as a CD8 thymocyte lingers at a peptide-MHC 1 niche, this niche is unavailable for other, competing CD8 thymocytes. CD8 thymocytes, we propose, do not compete with the CD4 thymocytes, because CD4 thymocytes compete among themselves for stimulation by interacting with peptide–MHC II stations on epithelial cells. We tested the outcome on lineage frequency of simulating different dissociation rates for interactions between epithelial cells and CD4 and CD8 thymocytes. The results are shown in
Measuring the ratio of CD4 mature cells to CD8 mature cells, in silico, as a function of the dissociation rate, we find that to achieve the experimentally measured 2:1 ratio, the dissociation rate of CD8 cells should be anywhere between 0.38 to 0.45 that of CD4 thymocytes. The expanded insert zooms in on this zone, which produces the experimentally observed ratio of two CD4 cells for every CD8 cell.
RA analysis of thymocyte development sheds new light on the dynamic relationship between molecules and cells in generating the structure and function of the thymus organ. First, we can see that the existing body of data, however discrete and piecemeal, can be integrated by RA simulation into a representation of the functional anatomy of the thymus seen in histologic sections. What we know about cells and molecules can indeed account for what we see; the macroscale organ emerges from the microscale mass of data in hand. In this regard, RA can be said to validate the database. Note, however, that classical histologic sections are two-dimensional slices of a three-dimensional organ frozen in time; RA simulation adds the dimension of time—dynamics—and so shows us the formative power of the dynamic flux of cells, molecules, and interactions that give rise to the higher-scale organ. In another project involving a different organ, we are currently extending RA simulations to accommodate the third dimension in space; hopefully, the added complexity of the representation will enhance our understanding of the biology.
Second, RA simulation offers novel explanations for the observed outcomes of experimental intervention. In our case, for example, RA simulation suggested that the lack of phenotype observed in mice with CCR9 knocked out (CCR9−/−) might be explained by dynamic compensation through population pressure. RA simulation also explains the competitive growth advantage enjoyed by wild-type cells over CCR9−/− cells. Indeed, overexpression of CCR9 on thymocytes leads to an in vivo phenotype that can be explained by RA as an untimely attraction of the thymocytes by cortical epithelial cells. RA simulation also suggests that the absence of thymic output resulting from CXCR4 inhibition can be attributed to the nonmigratory behavior of cells entering the thymus.
Third, the visualization of cell dynamics through RA provides a view of emergent physiology. Although the thymus is packed full of cells, the existence of competition among thymocytes for space and stimulation has not been a subject for experimentation or even discussion; competition is simply not seen in static histologic sections. Since competition was not recorded in the database, we did not explicitly program competition into our model. Nevertheless, cell competition emerged before our very eyes as we witnessed, via RA, the animated struggle between individual thymocytes for productive interactions with thymic epithelial cells. In silico manipulation of various parameters suggested that thymocyte competition might function as an important factor in three emergent properties of T cell maturation: the functional anatomy of the thymus, the selection of thymocytes with a range of migratory velocities, and the relative preponderance of SP CD4 T cells. Obviously, these suggestions require experimental validation
The RA simulation was written in C++ using the Rhapsody tool, and so RA code was generated by Rhapsody's code-generation engine, initiated by the language of Statecharts. To this automatically generated code, manually encoded objects and function were added. RA is the bridge made between the running simulation and the animation. Communication is made over a TCP/IP connection between a server implementing the Statecharts simulation and built-in animation functions in Flash. We used Matlab to analyze populations and population-level behavior. See [
(74 KB PDF)
(57.8 MB AVI)
(10.9 MB MOV)
(6.3 MB MOV)
(3.3 MB AVI)
The authors thank Dr. Daniel C. Douek, Vaccine Research Center/NIH for his helpful discussions and intellectual input.
cortico–medulary junction
double-negative
double-positive
major histocompatibility complex
reactive animation
subcortical zone
single-positive