Home Team Project Modeling Parts Standard & Protocol Software Tool Human Practice Notebook



The Background

Evolution vs. Design

Evolution and design are two sides of the coin.

Until 150 years ago, people believed that we are designed and created by the God, a god, gods or other intelligent designers. After Charles Darwin's On the Origin of Species in 1859, more and more people began to realize that a god is not necessary. During the process of evolution, we can be created automatically, without any intelligence.

Design and engineering are processes with the need of knowledge. Before we can design a machine, we have to know everything about the possible parts. If any information is unknown, we have to learn it. If nobody knows it, we have to measure it by ourselves. If all the knowledge needed are too much, we have to collaborate with others. Computer-aided design have to be used to accomplish more complex tasks. Space shuttle and microprocessor are seen as the most complex systems engineered, but they are far too simple when compared with the complexity of biological systems created by evolution.

On the other side, evolution is a completely different process. Variation is random, selection is directional based on the fitness to the environment, that is all. However, all the amazing things on our planet is emerged from this simple process, no more thing is needed.

Why? All the complexity is solved by this simple process without any input of knowledge?

Yes. The answer is on the scale. The evolution process is so simple that it can be scaled up infinitely. The complex problem can be solved in the distributed evolution system. Any success in the distributed system can be amplified by the selection process. Therefore, although the variation is random, it will be powerful enough to search for the solutions, when the scale of the system is big enough.

In the design process, the variation is directional based on the knowledge, but the process is not scalable, because of its requirement of intelligence. As a result, it is much more difficult to solve complex problems by design.

Directed Evolution

Although Darwin's theory was published 150 years ago, people have been using the power of evolution for more than ten thousand years. Domestication of plants and animals is done by artificial selection, which is the process of intentional breeding for certain traits, therefore changing the direction of evolution.

Directed evolution is a method using the similar principle to create new biomolecules with desired properties. The targets of directed evolution include enzymes, antibodies, aptamers, ribozymes, biosynthetic pathways, and synthetic genetic circuits. More information can be found on the Frances H. Arnold research group, the Ellington lab, and the Szostak lab.

The directed evolution experiment contains several rounds of 3 steps: variation, selection, and amplification. Variation is the mutation or recombination of the information encoded in the DNA, usually by error-prone PCR and DNA shuffling respectively. Selection is the process of separating the variants with desired phenotypes from others, it can either refer to screening (isolate good variants) or selection (eliminate bad variants), either in vivo or in vitro. Amplification is the replication of the variants after selection, which recovers the population size for the new round of directed evolution.

New function of biomolecules or biological systems is difficult to be rationally designed, but directed evolution is successfully used in solving these problems.

Design & Evolutionary Approaches in iGEM Projects

Flow Chart of Engineering a Genetically Engineered Machine

In the iGEM competition, teams specify, design, build, and test simple biological systems made from standard, interchangeable biological parts. Most projects engineer the systems by modeling and measurement, including this project itself. This engineering approach is proved to be feasible but difficult in the design of biological systems during this years.

Most BioBrick parts have never been characterized, because characterization and documentation of the parts usually takes a lot of time. Further more, the properties of BioBrick parts and devices are very sensitive to the conditions they work in, such as strain, plasmid, other parts used together, culture medium, growth rate, temperature, pH, shaking rate, light, and so on. It is very difficult to characterize all the parameters, so the parts often have to be remeasured in different projects before the behavior of the system can be predicted.

There are also many projects using evolutionary approaches. These approaches use selection or screening to find the needed part, eliminate the steps of measurement or constructing new parts with desired properties.

Some of the teams using evolutionary approaches in their project:
Team Name / People Competition / Lab Project Approach Wiki Link
DavidsoniGEM 2006Solving the Pancake Problem with an E. coli Computerin vivo Recombination; SelectionDavidson
Missouri_WesterniGEM 2006Solving the Pancake Problem with an E. coli Computerin vivo Recombination; SelectionMissouri_Western
AlbertaiGEM 2007Plan B: Building Butanol BioBricksMutagenic Compound; SelectionAlberta
Boston_UniversityiGEM 2007Increasing the Current Output of Microbial Fuel Cells Through the Directed Evolution of Shewanella oneidensisError-Prone PCR; ScreeningBoston_University
Davidson_Missouri_WiGEM 2007Living Hardware to Solve the Hamiltonian Path Problemin vivo Recombination; ScreeningDavidson_Missouri_W
HarvardiGEM 2007Cling-E. coli: Bacteria on targetScreeningHarvard
USTCiGEM 2007Extensible Logic Circuit in BacteriaError-Prone PCR; ScreeningUSTC
Calgary_SoftwareiGEM 2008EvoGEM: An Evolutionary Approach to Designing LifeEvolutionary AlgorithmCalgary_Software
ETH_ZurichiGEM 2008Make yourself simpler, stupid! Or how engineering a self-minimizing cell leads to the Minimal GenomeControlled expression of restriction enzymes and ligases in vivo; SelectionETH_Zurich
IIT_MadrasiGEM 2008StressKit: A BioBrick library of Lac-repressed σ24, σ28, σ32 and σ38 promoters for Escherichia coliScreeningIIT_Madras
IllinoisiGEM 2008Cell-based and in vitro antigenic sensors for medical diagnosticsError-Prone PCR; ScreeningIllinois
Newcastle_UniversityiGEM 2008A Computational Intelligence Approach to Developing a Diagnostic Biosensor: The Newcastle BugBusters ProjectEvolutionary AlgorithmNewcastle_University
Peking_UniversityiGEM 2008A genetic circuit to direct evolution of proteins in vivoin vivo Mutagenesis; ScreeningPeking_University
Tokyo_TechiGEM 2008Coli.Touch – implementation of a pressure-responsive genetic circuit in E. ColiError-Prone PCR; ScreeningTokyo_Tech
WarsawiGEM 2008Bacterial device for creating and production of interactors for any given bait proteinin vivo Mutagenesis; SelectionWarsaw
Jason R. KellyEndy LabScreening plasmidScreeningEndy:Screening_plasmid

Evolutionary Biology, Population Genetics & Evolutionary Algorithm

The process of evolution is as simple as the repeat of variation and selection, but it is also more complex than everything we can imagine. While an organism is evolving to fit the environment, the environment containing other organisms are also evolving. The variation rate of the organism and the selection pressure of the environment themselves are also evolving to keep the survival of the entire ecosystem.

There are many theoretical researches of evolution following the work of Darwin. Evolutionary biology, population genetics and evolutionary algorithm are 3 research fields that are closely related to the research of the dynamics of evolution. The first two fields are on the biology side, while evolutionary algorithm is an engineering approach widely used in optimization problems.

Fitness landscape is often used to characterize evolution. Mountain peaks are high ability to survive, while valleys represent low fitness. Similar genotypes are close to each other, while different ones are far from each other. The population takes an adaptive walk across such a landscape. The paths of evolution is determined by the fitness landscape.

Problems to Be Solved

Evolutionary algorithm have been used for more than 50 years, many strategies have been and are being developed to improve the efficiency of the algorithm. These strategies deal with the control of evolution parameters. There are deterministic control, adaptive control (feedback control), and self-adaptive control (the evolution of evolution).

Directed evolution experiments are usually carried out by hand, the parameters are optimized according to experience. Therefore, it will take a lot of time to optimize conditions, especially for new targets. When the condition is not optimized, different fitness may not be well separated, and the success rate will be low.

If the conditions can be controlled automatically according to theories of evolutionary algorithm, it is expected to greatly increase the success rate and reduce the time, labor and cost. However, automation of the experiments is not practicable under today's technology.

Fortunately, the rapid development of synthetic biology has opened another possibility. Biological systems become programmable now. It is possible to automate the process in the cell, to make an E. coli cell an automatic directed evolution machine.

The Blueprint

The Goal

Automation of Engineering a Genetically Engineered Machine

We designed the E. coli Automatic Directed Evolution Machine (E.ADEM) project.

The ultimate goal is to make E.ADEM a universal framework for evolutionary approaches in synthetic biology. Anything we want in synthetic biology can be automatically created, from promoters, RBS, regulators, receptors, binding partners, aptamers, enzymes and ribozymes, to sensors, logic devices, reporters, metabolic pathways, entire genomes, and even solutions of mathematic problems.

To each evolution object you want it to evolve, a scoring function can be designed to output PoPS as the fitness score to your demand. After that, you can ligate the scoring function device into the E.ADEM plasmid, transform E. coli, culture the cells and wait for them to evolve automatically and robustly, and get what you want at last.

Modules & Flow Chart of the System

Modules & Flow Chart of E.ADEM

Scoring Function

Possible Design:

For Transcription Repressor:

Scoring Function Design for Transcription Repressor

For Device (e.g. Sensor, Logic Gate):

Supervised Learning Scoring Function Design for Device (e.g. Sensor, Logic Gate)

For Enzyme:

If there is a sensor for the product or substrate, use it as the scoring function.

Else, perform evolution for a sensor first.

For Binding Partner:

Use E. coli two-hybrid systems.

Self-Adaptive Controller

Demand Analysis of the Controller

Function: adjust variation rate and selection pressure, base on the fitness score, the population size and the average fitness score calculated by a quorum sensing device.


  • Variation rate
    • Too low: Slow
    • Too high: Not Stable
  • Selection pressure
    • Too low: Not Directional
    • Too high: Die out


  • Control strategies in evolutionary algorithm

See below for our design.

Variation Function

Possible Design:

An Idea for Avoiding Non-Specific Mutation
  • Targeted Mutagenesis
    • Activation induced cytidine deaminase (AID)
      • iGEM 2008 Peking_University
      • iGEM 2008 Warsaw
    • Multiplex Automated Genome Engineering (MAGE) [1]
    • Error-prone DNA polymerase I
    • Bacteriophage
    • Error-prone reverse transcription
  • Recombination
    • Site-specific recombination
      • Including inversion, excision/integration and translocation
    • Homologous recombination
    • Transposition

  1. Wang HH, Isaacs FJ, Carr PA, Sun ZZ, Xu G, Forest CR, and Church GM. Programming cells by multiplex genome engineering and accelerated evolution. Nature 2009 Aug 13; 460(7257) 894-8. doi:10.1038/nature08187 pmid:19633652. PubMed HubMed [Church]
All Medline abstracts: PubMed HubMed

Selection Function

Kill or survive the cell:

Selection Function Design


Reporter Design

The Prototype

What to Do First?

The Self-Adaptive Controller is the core of the machine. We must finish it first in the prototype machine.

The Selection Function is important to test the Self-Adaptive Controller.

The Variation Function can be made later.

The Scoring Function can be made later. Constitutive Promoter Family can be used as Stimulus Signals for the Self-Adaptive Controller.

Constitutive Promoter Family as Stimulus Signals

We choose to use constitutive promoters instead of the conditional operon impressions (represented by IPTG-induced expression) as the stimulus signals to test the system. The stimulus signals in a testing system are supposed to be definite and stable. However, the IPTG-induced signals are susceptible to many environmental factors. The process of inducible expression involves a series of dynamic actions in physical chemistry: the diffusion process of IPTG molecules, and the equilibrium between the attachment and disattachment of IPTG to the promoter. That way, the expression signals would fluctuate in a large scope in experiments and the mathematic analyses would be very complicate.

Comparatively, the stimulus signals based on a series of constitutive promoters of different levels are far stabler since the process are relatively direct.

  • The signals produced by the constitutive promoters can maintain at the steady state during the measurement.
  • Constitutive-promoter expressions can give out several different stimulus signals in one system without any disturbation among them. That is, several constitutive promoters can work independently in one system to produce double or triple stimulus signals. Instead, the IPTG-induced testing system can imput only one signal at a time corresponding to the concentration of the IPTG.
  • Two or more systems with different stimulus signals can grow in the same nurture if the signals are produced by the constitutive promoters. For example, in our project, several kinds of E.coli with different imputs signals can grow in the same culture medium preparing for the screen.
We have characterized the constitutive promoter family in detail and the measurements and the results are described in the registered parts[1].

Design of the Self-Adaptive Controller

Design of the self-Adaptive Controller

The self-adaptive controller consists of two quorom sensing parts one tetR parts and a hybrid promoter.

Two quorom sensing systems

There are two quorom sensing parts in the self adaptive controller. LuxR is expressed by a constitutive promoter and interacts with AHL produced by both LuxI in two parts. The complex then activates hybrid promoter. LuxI in different parts perform different functions. LuxI's expression in Quorom sensing 1 depends on the population density and take effect in evolution in the aspect of cell density. LuxI in part 2, on the other hand, is expressed correspondent to the evolution score directly. Thus it takes part in the evolution process directly.

TetR parts

TetR is expressed correspondent to the evolution score, just like the LuxI from quorom sensing part 2. However tetR represses the hybrid promoter. High score means high tetR level. High tetR level means low death rate.

Principle of Operation

  • Properties of selection system

The selection system consists of hybrid promoter and ccdB. AHL-LuxR complex and tetR input perform contrast effect. We perform several experiments to identify how the hybrid promoter work. The property is as shown below:

  • How different "scores" result in evolution

When high score and low score E.coli are grown seperately:



Low score produce little AHL and little tetR. High score produce more AHL and more tetR at the same time. Two kinds of E.coli will survive in their own way. They will both reach high population density.

What if we mix them together?


At first, two kinds of E.coli were mixed 1:1. Two kinds of E.coli maintain their own density state.


Average AHL level begin to change.Low score E.coli in the high level AHL circumustance will express more ccdB. Then the high score E.coli has a advantage over the low score.More high score ones will survive. The difference between score finally results in the evolution.


Vector & Chassis


PSB1A3: high copy number, therefore less sensitive to mutation.


Top10: common used strain.

MDS™ 42 recA Blue:

Scarab Genomics has bioengineered the Clean Genome® E. coli by deleting over 15% of the E. coli K-12 genome. Using synthetic biology methods, the K-12 genome was rationally designed by making a series of precise deletions, which included the elimination of non-essential genes, recombinogenic or mobile DNA, and cryptic, virulent genes. This genome reduction optimizes the E. coli strain as a biological factory, providing enhanced genetic stability and improved metabolic efficiency. These properties make the Clean Genome® the E. coli strain of choice for a wide spectrum of applications ranging from routine cloning to production of biological material for therapeutic purposes.

Scarab Genomics was established in 2002. The Clean Genome® E. coli is the result of Dr. Fred Blattner's research at the University of Wisconsin, Madison. Scarab Genomics has licensed the patented reduced genome technology from Wisconsin Alumni Research Foundation on an exclusive, worldwide basis.

Assembly Road Map

In order to keep the whole process in perspective, we designed maps to direct our work in wet lab.

Assembly Road Map_1: Milestone1.
Assembly Road Map_2: Milestone2.

The Progress

  • Assembly work

As shown in the roadmap of our work, we finished assemblying most of the designed system except the pNOT parts. We submitted 167 biobricks finally.

  • Measument work : Most our parts are measured to prove their utility and get useful parameters.
    • 6 constitutive promoter are measured and marked with barcode system.
    • Hybrid promoter and related tetR,LuxI,LuxR are proved to work well in our models.
    • Several models were established and fitted our data pretty well.
    • Density control parts(ccdB related) do not work stably and need further improvements.We tried to add LVA to ccdB to improve its stability ,but it did not work.

Beyond iGEM

The E.ADEM project will continue on OWW: