Abstract:

Molecular interactions and catalytic transformations are the fundamental building-blocks out of which essential cellular processes are composed. Evolution through random mutation and natural selection is mainly responsible for the formation of modular structures and networks that enable organisms to express certain phenotypic behaviors. However, we have limited knowledge on whether and under what circumstances randomly evolving networks of such components can capture the structure of real world environments. From a systems and evolutionary biology perspective, elucidating the formation, evolution and dynamical characteristics of these evolved networks would be considered a seminal advancement towards understanding molecular life, while experiments regarding the effect of environmental parameters, such as the mutation rate, on the structure of these networks may provide valuable answers to fundamental biological questions.

In an effort to bring more insight to the evolution of biological systems, I created a simulation framework, where organisms compete and evolve in structured environments. Through appropriate abstractions that balance biochemical realism and computational feasibility, I attempted to make direct connections between the evolved networks and their biological counterparts. Under a variety of selections, the evolved organisms captured an internal representation of the dynamical multi-dimensional structure of their environments and were able to precisely predict the future abundance of resources. Surprisingly, this primordial cognitive capacity emerged rapidly and reproducibly in randomly evolving networks of small size. The observations that are presented here strongly argue that the formation of cognitive internal representations may be widespread in nature, with far-reaching implications for understanding and controlling microbial behavior. Homeostasis, the canonical framework for understanding microbial physiology, may be, in fact, inadequate in describing the most complex of microbial behaviors.

Additionally, my work shows that the notion of module or sub-network functionality itself is only accurate when considering the environmental context in which the organism has evolved. In a modern and contextual adaptation of Dobzhansky's words, that "nothing in biology makes sense except in light of evolution" [1], our ability to fathom an organism's behavior and its underlying mechanisms crucially depends on an intimate understanding of the correlation-structure of the environment in which the organism has evolved. Along these lines, I here discuss my findings regarding basic evolution questions such as the effect of mutation rate to network topology and network size. I will provide evidence over the cost and advantages of the emergence of mutational robustness [2] and will relate it to other evolutionary parameters like the evolvability and adaptation to fluctuating environments.

Aside from studying properties of evolved biological networks, the framework that is presented here is an ideal platform to answer basic epistemological questions that are difficult to address due to the limitations of biological measurements. Since the blueprint, fossil record and complete environmental information of naturally evolved organisms are available in this setting, any network inference technique can be rigorously evaluated. As an example I analyze two approaches that aim to discover functional modules and interactions by analyzing epistatic and expression data. Finally, in the context of synthetic biology this approach can be used by genetic engineers to evolve systems that have the desired computational, dynamic, or steady state capacity and characteristics (robustness, fault tolerance, genetic buffering) and then use off-the-shelf molecular analogs to construct them. The usefulness of a tool that has the ability to supply optimally tuned networks that can compete in silico with man-made proposed genetic designs under controlled conditions is apparent considering the difficulty of the in vitro synthetic circuit implementation[3][4].