Microfluidic Channel Design

For implementation of the SmoColi system, a channel is needed to establish a small molecule gradient (see Circuit Design). However, there were several different possible channel designs, and the final design evolved through an iterative series of design steps and design validations. The first designs were validated based on vast simulations, the final design furthermore by biological experiments in the lab (see Systems Validation).

Microfluidic channel with flow and recycling of the medium

We came up with two different possible microfluidic channel designs, both involving immobilized cells and a flow of medium containing inducer molecules through the channel. By having a flow and degradation, we could obtain a gradient of the inducer molecule. Because of the flow, our cells would also be constantly supplied with nutrients from the medium.

• Variant 1: Microfluidic channel with agarose-immobilized cells in cubic pockets

This channel design consists of two layers: The bottom one is a polydimethylsiloxane (PDMS) plate with regular cubic pockets. These pockets are filled with agarose-immobilized cells by letting the cell-agarose suspension flow over the pockets from the top by gravity. After washing the channel, cells would only be retained in the pockets.

The top layer is the actual PDMS channel, which is several pixels wide and several pixels high. Although this setup is rather complex, it has the advantage of having immobilized cells and thus being more robust, i.e. we could vary flow speeds or even put aerosol in the channel without the cells being washed out. Additionally, cell density can be varied very easily in this design as the cells can be diluted before being immobilized in agarose.

Figure 1: Experimental setup for SmoColi channel, Variant 1.

Video 1: Evaluation of the model.

• Variant 2: Microfluidic channel with cells sitting in pockets inside the channel

This channel design only consists of one part: A PDMS channel that is fixed onto a glass carrier. The PDMS channel contains pockets which "trap" the cells given a constant flow from the direction of the "open end" of the pockets. This channel design does not have the advantages of the above one, i.e. it is not as robust and cell density cannot be reliably varied. However, manufacturing it is a standard process and thus it is easier.

Figure 2: Experimental setup for SmoColi channel, Variant 2. The flow is supplied from the right side of the illustration.

Problems with these design variants:

A problem with both of these designs is that for the AHL-based RFP alarm to work, recycling of the flow back into the channel would be required. AHL-producing cells are only those "after" the GFP band, i.e. those at lower acetaldehyde concentration than the band concentration range. As long as the GFP band has not arrived to the end of the channel, we should make sure that there is AHL everywhere in the channel, so that it inhibits RFP production in every cell. Modeling showed that AHL can not simply diffuse "backwards" against the flow, but by having a recycling all the cells would be supplied with AHL and thus the alarm won't be activated before time.

However, since the tubing and pumps would have very high volumes compared to the channel's volume, the AHL signal would be diluted to the point where no detection is possible anymore. Also, several pumps would be required to accomplish this, further complicating the process design and making it more error-prone.

Improved microfluidic channel without flow

Figure 3: Experimental setup for SmoColi, a tube with no flow, diffusion only

Modeling showed that diffusion and degradation of the inducer are enough to create a concentration gradient in the channel. The experimental validation of this hypothesis was first performed in a 2 mm diameter tube. For proof of concept, we engineered E. coli strain JM101 to express GFP upon IPTG induction. The cells were then immobilized in agarose and the suspension was added to the tube (see Figure 4). One end of the tube was connected to a reservoir (1.5 ml) containing 10 mM IPTG solution. After overnight incubation at 37 °C, an IPTG-inducible GFP gradient was observed (Figure 6 and 7). The experiment confirmed the modeling results. Our cells survived and we concluded that we do not need constant supply of nutrients.

Without a flow there is no need for a moving liquid and thus no need for any of the complex designs above, as one can simply fill the whole channel with cell-agarose solution. We can then wait until the cell-agarose solution solidifies to a gel and then connect one side of the channel to a reservoir with the toxic molecule, while sealing the other. Likewise, we do not need recycling because AHL can diffuse through the whole channel and does not have to diffuse against a flow. Moreover, having a tube instead of a microfluidic device would save us some time that we would need otherwise for the channel construction.

Modeling the system thus had a profound effect on the process design, leading to an extensive reduction of complexity and error-proneness. Additionally, the AHL recycling idea would not have worked in the initial design, and was "saved" by the new channel design we developed by modeling our system.

Figure 4: Photo of the channel in action. The channel (the long thin tube at the right, 2 mm diameter, 7 cm length) is physically attached to a reservoir filled with the sample medium containing the toxic molecule, or in our test system with IPTG (the Eppendorf tube at the lower left). In the case of acetaldehyde, the whole setup would be packed in an impermeable plastic bag to significantly reduce the vaporization of acetaldehyde (not shown).

After the experiment, the agarose gel containing the cells is removed from the tubing. Figure 5: The bald interior of the channel is placed on a petri dish (see picture) and analyzed under a fluorescence microscope.

Fluorescence pictures of the tube showed a clear gradient of the fluorescence signal over approximately 5 cm of the tube. After 5cm, the signal strength dropped under the background noise.

Figure 6: GFP gradient in tube: E. coli with IPTG-inducable GFP were incubated in a tube. GFP expression was assessed under the fluorescent microscope after overnight incubation, with a excitation wavelength of 480 nm and a emission wavelength of 510 nm. The 15 microscope photos were reassembled into one using the Microsoft Research Image Composite Editor.

We quantified the fluorescence signal using a moving average of 80×80 pixel, which moved along the symmetry axis of the tube (see Figure 7), in red you can see the according reaction diffusion model.

Figure 7: Quantification of the gradient in Figure 2: The light intensity of the IPTG-induced GFP signal was quantified by a 80×80 pixel moving average. The peak at around 1.2cm is due to an air bubble in the channel.

The fluorescence distribution of this experiment has a similar shape as the distributions predicted by the model (see modeling section). The difference in the experimental results compared to the simulations can be explained mainly due to the different diffusing molecules: the simulations were obtained for acetaldehyde, whereas the experiments were carried out with IPTG. We expect the different values of the diffusion as well as of the degradation constants of the two molecules to be the main reason for the differences.

Problem with this design variant: No live imaging is possible, but only end point measurements, since the agarose with cells first has to be taken out of the tube and then analyzed under microscope.

Further improvements: Microfluidic Channels with Silicon glue

Since with the tube live imaging is not possible, we returned to microfluidics again and improved our channel design further. In order to image the channel at its top and bottom a glass slide is needed. We tried to build the channel by using a plastic mask and casting the boundaries with different materials. We used a two-component glue, silicon and molten parafilm. The glue turned out to be toxic for our cells. Molten parafilm was not usable with our plastic mask, it stick to the mask. Molten parafilm can be only used with metall mask. For silicon we could avoid sticking of the channel to the mask by covering the mask in oil. All in all a we could make a channel, still we had problem with closing the channel in a tight way to prevent evaporation.

Figure 7: Construction of microfluidic channels with silicon glue

Figure 8: Construction of microfluidic channels with silicon glue

At the same time we tried something more frequently used, Polydimethylsiloxane (PDMS) for constructing the channel, which later turned out to be our final design. For the PDMS as well as for the silicon channel we had problems with evaporation , water seens to evaporate through the PDMS. In our final experimental setup we incubated the channel in a petri dish of water.

Final Design: Microfluidics channel with PDMS

Finally, we decided to construct our channels out of PDMS (polydimethylsiloxane), with a technique called photolithography. Advantages of PDMS are that it is cheap, optically clear and permeable to several substances, including gases (air can quickly diffuse through) [1].

We constructed the channels ourselves, which was a very fun and interesting process. Our final design and the channel construction is explained in details here.

References

[1] http://www.elveflow.com/microfluidic/16-start-with-microfluidic