Building a Bacterial Computer

1
Living Hardware to Solve the Hamiltonian Path
Problem Davidson College Oyinade Adefuye, Will
DeLoache, Jim Dickson, Andrew Martens, Amber
Shoecraft, Mike Waters, Dr. A. Malcolm Campbell,
Dr. Karmella Haynes, Dr. Laurie Heyer Missouri
Western State University Jordan Baumgardner, Tom
Crowley, Lane Heard, Nick Morton, Michelle
Ritter, Jessica Treece, Matthew Unzicker, Amanda
Valencia, Dr. Todd Eckdahl, Dr. Jeff Poet
Abstract
Modeling a Bacterial Computer
Bacteria Successfully Solve the Problem!
Silicon computers are powerful tools for solving
mathematical problems but are inefficient
parallel processors. For iGEM2007, Davidson
College and Missouri Western State University
have jointly developed a bacterial system capable
of solving a Hamiltonian Path Problem in vivo.
Our system takes advantage of E. colis
exponential growth to address the complexity of
this problem in a way that traditional computers
cannot. We successfully detected solutions to a
Hamiltonian Path Problem through phenotypic
screening.
To determine the feasibility of our system, we
implemented mathematical models to answer to
following questions
The HPP constructs were transformed into E. coli
expressing T7 RNA polymerase, and their
phenotypes were observed. After about two days,
they had visible color. Unflipped colonies
exhibited uniform fluorescent phenotypes, while
colonies exposed to Hin protein varied in their
fluorescence. Fluorescent phenotypes were
observed by eye and confirmed by fluorometer.
Does starting orientation matter? No, after a
large number of flips, all orientations converge
to the same probability.
4 Nodes 3 Edges
Probability of HPP Solution
Can we detect a solution? Yes, for a 14-edge
graph with 7 nodes, growing 1 billion bacterial
computers gives a 99.9 chance of detecting a
solution. We determined this using a cumulative
Poisson distribution. 1 mL of culture contains 1
billion bacteria.
Number of Flips

The Hamiltonian Path Problem
The Hamiltonian Problem (HPP) asks given a
directed graph with designated starting and
ending nodes, is it possible to travel through
the graph visiting every node exactly once?

Cumulative Poisson Distribution

• Are there too many false positives?
• No, using MATLAB, we computed the ratio of true
  positive to total positives and found that even
  for a 14-edge graph, the ratio was above 0.05.
  Combined with PCR screening, this ratio is well
  within detectable limits.

Linear increases in graph complexity yield
exponential increases in the number of possible
paths through a graph. Due to the absence of an
efficient algorithm, parallel processing is
required when attempting to solve complex graphs
in a limited amount of time. The exponential
growth of E. coli computers provides the required
processing power.
Possible Paths through the graph
Extra Edge
False Positive
PCR Fragment Length
of Edges in Graph
True Positive
PCR Fragment Length
of Processors
Building a Bacterial Computer
Flipping was also detected by PCR, using
multiplex primers that bound to three distinct
regions of the HPP-A constructs. Different edge
orientations produce 8 unique PCR product
lengths. Both methods show strong evidence for
flipping of DNA, and PCR shows that some of the
bacteria may have solved the problem.
Cell Division
For our system to work, reporter genes must
maintain functionality despite the addition of 13
amino acids. We were able to perform successful
Design Principles
Hin Hin Hin MW ABC
ACB BAC ABC ACB BAC
As a part of iGEM2006, we reconstituted a hin/hix
DNA recombination mechanism as standard biobricks
for use in E. coli. This system allows for
rearrangement of DNA segments that are flanked by
hixC sites and was implemented in our HPP
computer in the following way
hixC insertions in both GFP and RFP. We therefore
designed the following HPP constructs to test our
system on 2 simple graphs

• Represent each node as a reporter gene
• Represent each edge as a flippable unit of DNA
  flanked by hixC sites
• Flipping generates a random walk through the
  graph
• A solution is screened for by phenotype

Acknowledgements
We thank The Duke Endowment, HHMI, NSF, Genome
Consortium for Active Teaching, James G. Martin
Genomics Program, MWSU Student Government
Association, MWSU Student Excellence Fund, MWSU
Foundation, and the MWSU Summer Research
Institute. Team members Oyinade Adifuye and
Amber Shoecraft are from North Carolina Central
University and Johnson C. Smith University,
respectively. Special thanks to the iGEM
organizers and community. We had a great time!
hixC sites
Edge
Hin-mediated Recombination