What if you find a good target sequence for BLADEN but you can’t find a way to apply relevant selection pressure? To build upon the practicality of BLADEN, we created a hardware concept together with team EPFL that would be able to perform BLADEN autonomically on a larger variety of traits. The main idea behind the CellED is that we want you to be able to optimize any EvolvR-compatible trait in plant cells if you can measure the trait’s efficiency optically. Because of our very wet lab - oriented project, we decided to keep the concept a concept and let our creativity on the loose. Together with team EPFL, we went through many iterations of the device, trying to bring all of its complex components together into a concept worthy of continuation and as an inspiration for other engineers and scientists.
Working principles and components
Continuous directed evolution experiments are characterized by the iterative mutagenesis, selection and amplification of genotypes, where selection pressure is applied through the reduction in growth rate of genotypes that are less fit. Finding a selection method that applies such selection pressure is difficult, or may even seem impossible. Thus, a device that is able to select the fitter genotypes by taking trait relevant measurements would be beneficial in cases for which a selection method can’t be found. The workflow we came up with is shown in diagram 1 and is further conceptualized as shown in figure 2.
Mutagenesis, automatic sampling and amplification
First of all, EvolvR containing cells have to be grown at optimal conditions so that mutagenesis is performed as much as possible. We went for a stir tank bioreactor with an extra touch. By conducting air out of the stirring propellers themselves, we stir the plant cells both horizontally by rotation and vertically by air-flow. We have also integrated the sensor array into the tank design to minimize surface interference for the cells to nest on. Figure 3 shows the bioreactor and its components/requirements to operate successfully.
Using a bioreactor also allows for automatic sampling  of the culture and accurately controlling the culture’s environment, something that may be used for further experimentation. The bioreactor should be operating through multiple iterations of the directed evolution experiment, which in general takes longer than the time the bioreactor can practically sustain the cell culture. For the process to be continuous, cells should bleed out of the bioreactor before the cell population starts declining. This is challenging however, as a very large number of cells have to be sampled and sorted in order to minimize wasting fit genotypes. Cell measuring and sorting should thus be done at a rate that is faster than the bleeding rate of the reactor, while the rate at which cells are discarded should be close to the bleeding rate of the reactor. This requirement will ultimately determine how many cells should be discarded and is important when exploring the fitness-genotype landscape. The CellED uses a perfusion operation  mode in which the CellES acts as an external cell retention device. This perfusion mode does not extract the cells continuously from Bioreactor 1, as the cells that are retained have to be amplified, which is optimally done in a bioreactor as well: Bioreactor 2. Diagram 2 shows a simplified graph that corresponds to the perfusion mode of the CellED.
Cell selection: CellES
Cell measuring and sorting happens in a device we call the CellES: Cell Evolution Selector. The device operates on plant cells and to measure their contents accurately, the cells have to be lysed. To be able to keep cells as intact as possible for the next directed evolution iteration, we have to clone the cells, compartmentalized per genotype, in a reservoir and separate them afterwards while being able to distinguish them on genotype. These requirements may be satisfied by a microfluidics droplet-based system in which single-cell droplets are generated, labelled, cloned, separated, measured and sorted, respectively. Figure 4 shows a schematic overview of the CellES and figure 5 its further conceptualization.
To label the cells per genotype, dyed labels are injected into each droplet [reference paper]. These dyes are later used to distinguish each droplet and label them for their respective genotype. When cloning, the cells stay in the same droplet, preserving their labels which are scattered throughout the droplet. The cloning of individual cells within such small volumes is difficult and growth hormones may be added to increase cloning success. When the droplets are separated, their contents should separate too, such that some cells are lysed while the rest are sorted. To do so, the cells preferably aggregate as little as possible. These mechanisms of cloning and separating have not well been researched and developed. Through the CellED, we create an incentive for such mechanisms to be created.
Before trait related measurements are done, the number of cells per droplet should be measured so we get a normalized trait related measurement for each droplet. This can be accurately done through inductance-impedance measurements using electrodes. Trait related measurements can be done with any microfluidic measuring method, but we choose for a fluorimetry system because it has already been used in directed evolution devices for accurate and fast determination of enzyme efficiency. The bioreactor content data, trait-related fluorescence measurements, the label fluorescence measurements and the inductive impedance cell counting are then sent to a computer which saves the results in a database.
Database / online model
The contents of this database are inputs for an online model. After measurement, the lysed cell content is preferably sequenced, a process we did not integrate in this concept. The sequencing of these cells, together with the trait-related measurements, make up fitness-sequence data which is fed into the model. The model then uses this fitness-sequence data together with machine learning guided directed evolution methods  to explore the fitness landscape as much and as fast as possible. The computer uses the online model to then sort the droplets in which cells are not lysed using a droplet sorting system. This sorting system is also used to refresh the medium of the cells. Fitter mutants go into another bioreactor, where they are amplified and then sent to the main, original bioreactor, ready for the next iteration of the directed evolution process.
The device only lacks development in the field of microfluidics, specifically compartmentalized single cell cloning and plant cell-containing droplet separation. These two mechanisms could serve as foundations for other iGEM projects or further research. For us, working on the hardware conceptualization was a way to see what is still missing for such devices to be developed and we think that with the current technological progression within the field of microfluidics, a device like the CellED might be developed sooner than one might think.
 Miller RA, Shyluk JP, Gamborg OL, Kirkpatrick JW. Phytostat for continuous culture and automatic sampling of plant-cell suspensions. Science. 1968 Feb 2;159(3814):540-2. doi: 10.1126/science.159.3814.540. PMID: 5635158.
 Kropp, Christina & Massai, Diana & Zweigerdt, Robert. (2016). Progress and Challenges in Large-Scale Expansion of Human Pluripotent Stem Cells. Process Biochemistry. 59. 10.1016/j.procbio.2016.09.032.
 Yang, K.K., Wu, Z. & Arnold, F.H. Machine-learning-guided directed evolution for protein engineering. Nat Methods 16, 687–694 (2019). https://doi.org/10.1038/s41592-019-0496-6