Project Summary
Biomimetic Evolutionary Analysis:
The Origin of Vertebrae via Computational and Robotic Simulations of Fish
PI: John Long,
Department of Biology, Vassar College.
coPI: Thomas Koob,
Skeletal Biology, Shriners Hospital for Children, Tampa.
coPI: Chun Wai
Liew, Department of Computer Science, Lafayette College.
coPI: Robert
Root, Department of Mathematics, Lafayette College.
Research funded by the National Science Foundation (award DBI-0442269
from the CRUI program).
In spite of their name, the first vertebrates evolved without vertebrae, which are the bones in the backs of most living species. Vertebrae have evolved independently at least five different times, in five different groups of ancient fishes, from a non-bony skeleton. Biologists have sought for years to understand these critical events in vertebrate evolution. Because the vertebral column plays an important mechanical role in the body flexures that living fish use to swim, we sought to test the idea that vertebrae evolved in response to natural selection for enhanced swimming performance and navigation. The trouble with reconstructing the evolution of vertebrae is that since events occurred nearly 400 million years ago we are unable to directly observe and test natural selection in action. To understand how selection may have operated, we build, swim, compete, and reproduce artificial fish.
These artificial fish mimic the shape and internal anatomy of living fish; one kind is created as a computer simulation and the other is created as a robot. We then vary the structure of the vertebral columns of the artificial fish using the same genetic principles that create variable offspring in real fish. Each generation of variable artificial fish is then tested, using the principles of how evolution works in biological systems, and each individual receives a fitness score based on its relative performance in swimming and navigation trials. Individuals who perform better than others, those with higher relative fitness, have more of their genes passed on to the next generation of artificial fish. When repeated, this process of selection and reproduction has the potential to change the structure of the vertebrae, creating artificial evolution that allows us to examine, step by step, how vertebrae may have originally evolved.
In our computerized and robotic fish, we predict that vertebrae will evolve repeatedly in response to selection for increased swimming speed, acceleration, and maneuverability. To add other realistic effects, successful individuals from one population (computerized or robotic) will be transferred to the other, a process of immigration thought to be a key factor driving large-scale evolutionary change in real biological systems. If the evolution of these artificial fish does not produce vertebrae, it is unlikely that the vertebrae of real fish evolved in response to selection for enhanced swimming performance and navigation. Thus, we use artificial vertebrates, constructed to mimic real vertebrates, to examine the feasibility of different evolutionary scenarios. While computerized organisms have previously been used to test evolutionary scenarios, to our knowledge this is the first time that anyone has (1) evolved complex vertebrate structures and (2) done so by combining computerized and robotic systems.
We build robotic fish as mimics of free-swimming larvae of sea squirts, a form seen by many biologists as a likely candidate for the analogue of the first vertebrates. Swimming at the surface, these robotic fish navigate towards a light source. By using three identical robots, each with a tail of variable vertebral structure, we compete a generation of robotic fish and assign each individual a relative fitness based on swimming and navigation performance. Successive generations are competed, selected, and created until change ceases in the vertebral columns' size, shape, and mechanical properties. What makes possible the creation of variable and realistic vertebral columns is the use of a custom process to build flexible and stiff skeletal parts using biological materials such as collagen. In this way we can create vertebral columns of any size and shape within a biologically realistic range.
We build computerized fish as mimics of free-swimming jawless hagfish, living species that possess no vertebrae in their adult skeleton. We model the mechanical system of the muscles and skeleton interacting with the forces produced by the surrounding water. We vary the structure of the vertebral column, based on biomechanical data from real fishes, and determine how well, compared to real fish motion, the computerized fish swims. Unlike our robotic system, we can generate thousands of generations of computerized fish, allowing us to explore many more possible combinations of vertebral column structure. To create those generations and find locally-optimal structures, we are creating a new set of evolutionary computing tools that decrease, by several orders of magnitude, the time needed to search a large set of possible combinations.
Throughout the four years of this program, we offer 28 undergraduates a two-year combined apprenticeship-mentorship experience in research. Beginning with a summer research apprenticeship, cohorts of students work as teams, under the mentorship of a returning student and one of the principal investigators, to master technique, conduct and analyze experiments, and interpret and present results. By their second year in the program, each student has developed into an independent investigator, conducting a project of their own design for a senior research thesis and returning as a peer mentor for incoming research apprentices. We place particular emphasis on involving students from underrepresented groups, having students present their results at professional scientific meetings and in peer-reviewed scientific papers, and providing support for students to continue in research science at the graduate school level.