Team:Oxford/biosensor optimisation


Revision as of 02:15, 18 October 2014 by JHoffman (Talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)


Optimizing the biosensor design

An ideal biosensor would fulfil the following performance criteria:

Fast response to the presence/absence of DCM.
High amplitude of output signal – it must produce enough GFP to generate a distinct signal against background noise.
Sensitive - it must change significantly in low concentrations of DCM. This is vital in order to achieve a response that is as close to binary as possible. The ideal system will have a very sharp decline in fluorescence at a predefined, very low value of DCM. This will ensure that the sensor will clearly indicate when the DCM mixture can be safely disposed of.

(regarding modelling):
Robust - it must be able to cope with variations in ATC concentration without radically altering the behaviour of the system. This is crucial because we cannot ensure that ATC concentrations throughout all the cells will be uniform in the real system.

By modelling the effects of parameters we are able to alter in the biological system, we were able to guide our design process to produce a biosensor that is as close to the ideal as possible without sacrificing any one criterion entirely.

What can we alter in our biological system?
What can we alter in our biological system?

Our biosensor will not be able to meet all ideal criteria because 1) We are limited by biology as to which parameters we can actually change and 2) changing a parameter in a cellular system impacts more than one parameter.
However there are some things we can alter:

The rate of GFP degradation - the cell will degrade GFP, but marking the protein with a degradation tag would increase the rate that this occurs.
The amount of GFP produced per mRNA transcribed – by altering the strength of the ribosome binding site we can alter the efficiency of translation.

By modelling the effects of these we can answer the following questions:

Do we need to include a degradation tag on GFP, or is the turnover of GFP already adequate to give a fast 'off' rate?
What RBS strength should we use to maximise output amplitude or reach a usable signal output?
Will altering one of these to optimise one criterion negatively impact any other of our criteria?
What happens when we change the amount of each input added?
What happens when we change the amount of each input added?

What are these graphs and where did they come from?

Using the bacterial fluorescence models we have built, we predicted the steady-state fluorescence levels of the system in varying levels of DCM and ATC by solving the system of differential equations we produced during the characterization section. The results are illustrated in the 3-dimensional surface plot below.

The two 2-dimensional graphs are slices taken from the 3-D plot. In each of these 'slices' we are effectively holding one variable constant (the amount of either DCM or ATC) while varying the other.

The 3-dimensional plot was produced by plotting the final fluorescence value from many different possible combinations of the two inputs (ATC and DCM). The top graph shows the variation in final fluorescence when DCM is held constant and ATC is varied, the second graph is vice versa.

It is important to understand that these graphs represent the expected steady state level of fluorescence of thousands of different simulations. From this we can select the DCM and ATC concentrations for a specific fluorescence response.

How much of each input should we use to test the biosensor?

Our ideal biosensor must be robust. The top graph demonstrates this nicely. Beyond a certain threshold value of ATC, there is little change in the fluorescence response predicted - it saturates and maintains a constant level. Practically, this means we have to ensure that the ATC concentrations present in our final system must comfortably exceed this threshold ATC value.

From our initial system characterization, we have established that when DCM is not present in the system, there will be no fluorescence response aside from that due to the basal transcription rate. However, the model predicts that when even a small amount of DCM is added and the transient behaviour has stabilized, the fluorescence expressed in the system quickly reaches its saturation value. This corresponds to a highly sensitive biosensor which can effectively only express two fluorescence levels- zero or a predefined maximum. The transition from zero to the maximum saturation value occurs at very low concentrations of DCM.

To summarise, we have established that the inputs to our biosensor should be a constant medium concentration of ATC and a varying concentration of DCM as it is degraded. We should note that the ATC concentration will not value without external influence because the system does not consume ATC and its rate of degradation is negligible.

Should we aim for high or low RBS strength?
Should we aim for high or low RBS strength?
We ran the deterministic model whilst varying the activation rate (see 'where did these equations come from?') of the sfGFP. The response is shown here:

What does this tell us?

As you can see from this graph, increasing the RBS strength only changes the amplitude of the systems response without affecting the response time of the system. This is highly beneficial for the system.

-->Therefore we will aim for as high an RBS strength as possible in our initial design.
Should we aim for high or low degradation rate?
Should we aim for high or low degradation rate?
We ran the deterministic model whilst varying the degradation rate (see 'where did these equations come from?') of the sfGFP. The response is shown here:

What does this tell us?

Changing the degradation rate of the protein is more of a trade-off. As you can see, a higher degradation rate gives a faster response but with a much lower steady state responses.

-->We should aim for a low degradation rate to begin with so that we can ensure a detectable level of fluorescence, and then gradually increase the degradation rate to get a faster response.
How did this inform our design?
How did this inform our design?
Based on the modelling, we could optimise each performance characteristic individually, but to create the best overall biosensor we needed to compromise with what we chose to implement:

RBS strength

Medium RBS strength – our modelling suggested we should use as high an RBS strength as possible. We have used a relatively high strength RBS to try and optimise our signal amplitude without stressing cellular metabolism too much.

GFP degradation

No degradation tag - in this instance the model showed that increasing degradation efficiency of GFP (and thus the speed of response) by utilising a degradation tag would also decrease the signal amplitude. In our first attempt at making a biosensor, we decided it was more important to increase the chance of generating a usable signal than to have a fast off rate. In the future, once our biosensor is made and if we have found it to have very high amplitude, we could add a degradation tag to improve the on/off dynamics at the expense of that excessive signal.

Modelling Summary

The above results demonstrate well the power of modelling genetic circuits. This approach has allowed us to develop our first construct intelligently and to have some trustworthy predictions on which to develop the rest of our system around. However, as always, there are limitations, especially in biological systems.

In an ideal world, we would like to have a very high expression rate (for a high steady state amplitude of fluorescence), a high degradation rate (for a fast responding biosensor) and a high copy number of the plasmid in each cell. Conversely though, optimising these parameters puts metabolic stress on the cells. This leads to the system not actually being as optimal as the model might have predicted. Here we identify the weakness in preliminary models. We will have to actually develop the bacteria and run the experiments in the lab before we will know if our biosensor will respond this well to the DCM. After this, we will work at creating secondary models which should be able to give more reliable predictions. Ideally we would be able to then make more bacteria and the Engineering-Biochemistry cycle would continue.

Oxford iGEM 2014