We report the development and use of a synthetic, discrete event, discrete space model that functions as an epithelio-mimetic device (EMD). It is intended to facilitate the study of intestinal transport of drug-like compounds. We represent passive paracellular and transcellular transport, carrier-mediated transport and active efflux using stand-alone components. Systematic verification of the EMD over a wide physiologically realistic range is essential before we can use it to address questions regarding the details of the interacting mechanisms that are believed to influence absorption. We report details of key verification experiments. We demonstrate that this device can generate behaviors similar to those observed in the in vitro Caco-2 transwell system. To do that we used a series of hypothetical drugs and we simulated behaviors for two clinically used drugs, alfentanil and digoxin. The results support the feasibility and practicability of the EMD as a tool to expand the experimental options for better understanding the biological processes involved in intestinal transport and absorption of compounds of interest. (c) 2005 Elsevier Ireland Ltd. All rights reserved.

Rolling, activation, and adhesion on endothelial surfaces are necessary steps for the proper recruitment of leukocytes from circulating blood to sites of inflammation. Once at the target site, leukocytes help destroy pathogens and decompose damaged tissue. Inflammatory mechanisms, when uncontrolled, are also associated with diseases such as asthma, rheumatoid arthritis, and atherosclerosis. Improved therapeutics can be developed with a better understanding of the molecular-level events that mediate this process.

This dissertation reports the development of a synthetic, in silico model for use as an experimental system for testing the plausibility of mechanistic hypotheses of how molecular components may interact to cause leukocyte behaviors during rolling, activation, and adhesion to endothelial surfaces. Object-oriented software components were designed, instantiated, verified, plugged together, and then operated in ways that can map concretely to mechanisms and processes believed responsible for leukocyte rolling, activation, and adhesion. The result is an in silico analogue of the wet-lab experimental systems to study leukocyte rolling and adhesion; the experimentally measured phenotypic attributes of the analogue can be compared and contrasted to those of leukocytes from the referent systems.

We first demonstrate that the in silico devices can generate behaviors similar to those observed in vitro at the cell-level and population-level, by comparing simulation results with data from three different in vitro experimental conditions. The validated model was then used to test the hypothesized mechanism of LFA-1 integrin clustering in adhesion. The in silico device showed that clustering was not necessary to achieve adhesion as long as densities of LFA-1 and its ligand ICAM-1 were above a critical level. Importantly, at low densities LFA-1 clustering enabled improved efficiency: adhesion exhibited measurable, cell level positive cooperativity.

Our results provide convincing evidence for the feasibility of using synthetic in silico systems to improve our understanding of the molecular- and cellular-level events mediating leukocyte rolling, activation, and adhesion during inflammation. This work represents early, but important advances for future in silico research into plausible causal links between molecular-level events and the variety of systems-level behaviors that distinguish normal leukocyte adhesion from disease-associated adhesion.

The study of epithelial morphogenesis is fundamental to increasing our understanding of organ function and disease. Great progress has been made through study of culture systems such as EMT6/Ro mouse mammary tumor spheroids and Madin-Darby canine kidney (MDCK) cells, but many aspects of even simple morphogenesis remain unclear. For example, are specific cell actions tightly coupled to the characteristics cell environment or are they more often cell state dependent? Our objective was to discover plausible representations of the operating principles realized during characteristic growth of EMT6/Ro mouse mammary tumor spheroids and MDCK cells in culture. To reach that objective we engineered and iteratively falsified agent-oriented analogues of EMT6 spheroid growth (Chapter 2) and MDCK cystogenesis (Chapters 3 and 4). An approach to understanding how individual cell behaviors contribute to system behaviors is to discover a set of principles that enable abstract agents to exhibit closely analogous behaviors using only information available in an agent's immediate environment. EMT6 spheroids and MDCK cysts display consistent and predictable growth characteristics, implying that individual cell behaviors are tightly controlled and regulated. We listed key attributes of EMT6 spheroid growth and MDCK cystogenesis, which became our behavioral targets.

To understand the in vitro systems we created analogues made up of quasi-autonomous software agents and an abstract environment in which they could operate. The EMT6 analogue was designed so that upon execution it could mimic EMT6 cells forming spheroids in culture. Each agent used an identical set of axiomatic operating principles. In sequence, we used the list of targeted attributes to falsify and revise these axioms, until the analogue exhibited behaviors and attributes that were within prespecified ranges of those targeted, thereby achieving a level of validation. For the analogue of MDCK cystogenesis we tested our hypotheses through in vitro experimentation and quantitative validation. We observed novel growth patterns, including a cell behavior shift that began around day five of growth.

We posit that the validated analogues' operating principles are reasonable representations of those utilized by EMT6/Ro and MDCK cells during morphogenesis. Simulations provide an observable theory for cystogenesis based on hypothesized, cell-level operating principles.

Hepatocytes interact with new compounds: metabolism is a frequent consequence. Hepatocyte micromechanisms have evolved to adjust in real time to interact with any never-before-seen compound. To demonstrate understanding of those mechanisms, next

generation, pharmacologically useful hepatocyte models must strive to do the same: their micromechanisms recognize a compound as not previously seen and respond based on prior experience. How might that be accomplished when uncertainty is large and detailed data are chronically limited? My first goal was to answer that question: engineer a new class of hepatocyte models that draw on their own past experiences and compound physicochemical properties (molecular weight, pKa, logP, etc.) to respond uniquely to a new compound: the events that emerge during model execution provide a useful prediction of that compound's metabolic clearance. I used the new synthetic (combining elements to form a whole) method of modeling and simulation. The method involves building extant--actually existing and observable--working biomimetic micromechanisms. The micromechanisms within my successful in silico hepatocytes are comprised of autonomous, interactive objects and agents that map to six key components: extracellular media, cells, transporters, metabolic enzymes, cytosolic binding factors, and compounds. In silico clearance emerges from their interactions.

Within livers, hepatocytes are spatially organized into functional units called lobules. Those hepatocytes exhibit location dependent, possibly cooperative, properties, including gene expression and metabolic clearance. These properties frequently adapt to changes

in compound type and exposure. Drug-induced hepatocyte damage can also be location dependent. Such spatially heterogeneous phenomena are referred to as hepatic zonation. Next generation, pharmacologically useful, liver models must be capable of exhibiting

similar phenomena under analogous conditions. My second goal was to extend my new class of models to the level of hepatic lobules. I organized autonomous agents, which map to small units of lobular function, into structures called Zonally Responsive Lobular Analogues (ZoRLA). The micromechanisms within each that enabled achieving a degree of validation were given two tasks: protect hypothetical external "tissues" by eliminating simulated toxins and minimize resource consumption. The zonation patterns

that emerged are striking similarities to reported patterns. ZoRLA are designed to become components of future, virtual organisms.

Models are frequently and ubiquitously used in all pharmacokinetic investigations. The familiar inductive equation-based pharmacokinetic models formulate hypotheses about data; they alone cannot provide mechanistic insights. We need models that have extant, working mechanisms that generate emergent properties analogous to how phenomena emerge during wet-laboratory experiments.

In this dissertation, I report a new class of synthetic, agent-based, discrete events models and simulations, with the objective to provide mechanistic insights. Validated, biomimetic software components are plugged together to form in silico analogues of the referent experimental systems. Each synthetic analogue is a mechanistic hypothesis: execution produces an observable phenomenon.

The recirculating in silico livers (RISLs) are in silico analogues of isolated perfused rat livers during an experiment in which digoxin is administered, alone or in combination with either an uptake or efflux inhibitor. A RISL that comprised four time-variant mechanisms and new enzyme and transporter components achieved the most stringent similarity measure: simulated digoxin and metabolite perfusate levels were experimentally indistinguishable from the referent data. The mechanisms simulated unanticipated loss of hepatic viability during the original experiments: erosion of hepatic accessibility and of enzyme and transporter activities.

In silico experimental Caco-2 (cell monolayer) cultures (ISECC) are analogues of the confluent, asymmetric cell monolayer used in vectorial transport studies. To seek an explanation for the observed paradoxical saquinavir transport data, I followed an iterative refinement protocol that enabled discovery of plausible, new mechanistic details. The ISECC surviving the most stringent similarity challenge produced transport data statistically indistinguishable from referent observations. It required heterogeneous intracellular spaces; a biased distribution of metabolizing enzymes; and restrictions on intracellular drug movement.

Experimenting on synthetic analogues, such as RISLs and ISECCs, provides a formerly unavailable means of discovering and testing new mechanistic hypotheses. It is a powerful expansion of the scientific method: an independent, scientific means to challenge, explore, and improve any inductive mechanism. Validated, biomimetic analogues are concrete instances of hypothetical yet plausible mechanisms, and would replace vague, unverified concepts. The collection of mechanisms, rules, assemblies, and interactions of components can be subjected to testing and falsification, and in the absence of other competing theories, stands as the current best mechanistic hypothesis for the phenomena.

Biomedical science can be characterized as the pursuit of deeper, more useful mechanistic insight into biological phenomena to facilitate advancing health. In silico modeling and simulation can accelerate this process when models are designed for particular use cases aimed at challenging explanatory, mechanistic hypotheses. Modeling and simulation is ubiquitous in biology. Yet, a mental gap exists between computational versus wet-lab models in terms of their perceived scientific usefulness. Bridging this gap requires focusing on computational models as scientific tools that can explore and challenge explanatory, mechanistic hypotheses.

I begin by exploring the roles of computational, mechanistic models in science and their relationship to traditional wet-lab science. I analyze biological modeling and simulation from a philosophy of science perspective, culminating in a requirements list for computational models that are to be scientifically useful. To fulfill these requirements, I present the synthetic analog approach, a developing simulation methodology for simulating complex, multi-scale biological systems, and discuss agent-based modeling as an appropriate modeling formalism. I then extend this approach to include the development and execution of virtual biomedical experiments, which are simulations of wet-lab or clinical biomedical experiments. Executing virtual biomedical experiments allows one to mimic all relevant aspects of the wet-lab scientific process—from hypothesis formation to data analysis, and key concepts in between. I envision virtual experimentation not as a supplement to traditional wet-lab experimentation, but rather as an essential part of the scientific method itself.

I demonstrate the above methodology and virtual experiment vision by developing software analogs that quantitatively mimic wet-lab experiments related to drug metabolism, drug-induced toxicity, and immune system interactions with the liver. I use these models to explore several mechanistic hypotheses, reaching several validation milestones and falsifying several mechanistic explanations along the way. I highlight the importance of modularity in achieving the above vision, and develop a modularization approach to facilitate model reuse and repurposing. Finally, I discuss important technical issues in agent-based modeling and offer practical solutions. Taken together, this dissertation aims to demonstrate and lay the preliminary groundwork for engineering scientifically useful virtual biomedical experiments.