How did we achieve this?

The construction of the three devices was achieved using the BioBrick cloning system. Plasmids pSB1C3 containing promoters J23101, J23106 or J23117 were opened using SpeI and PstI enzymes while plasmid pSB1A2 containing I13504 was digested with XbaI and PstI. pSB1C3-J23101, pSB1C3-J23106 and pSB1C3-J23117 were dephosphorylated with antarctic phosphatase in order to prevent their self-ligation. I13504 was finally ligated with each of the open promoter-containing pSB1C3 plasmids using T4 ligase. Constructs were run on a 1.2% agarose gel, purified and transformed in DH5alpha high-efficient bacteria to be finally plated on chloramphenicol LB agar plates. Three colonies per plate were cultured overnight as biological replicates in 5mL LB medium with chloramphenicol in test tubes. Cultures were spun down and pellets resuspended in 1mL PBS. Samples were measured by Accuri c6 Flow-Cytometer (BD) and data were acquired three times, one after the other. They were displayed in arbitrary units. More information is available in our InterLab Protocol, InterLab Worksheet and in our Protocol section.

What did we get?

Fluorescence of constructs was given in arbitrary units by taking the mean of the three measurements of the biological triplicates. Figure 1 shows these results, where the error bars is the standard deviation calculated across replicates. We estimated the promoters to be different from each other and formally assessed this statement by calculating the p-values. They were all less than 10-5 thus proving our assumption. However, it was more relevant to express fluorescence in ratios, thus getting rid of the arbitrary units. We plotted them along with those calculated by Christopher Anderson (Figure 2) and were surprised by the huge discrepancy between the two measurements. J23101/J23106 ratios are close to each other (1.67 fold) but the J23101/J23117 and J23106/J23117 ratios we obtained are much higher than Anderson’s (27 and 16 fold respectively). This first glimpse into inter-laboratory variation is very instructive and shows how challenging reproduce exact measurements can be tough. The gap between ratios involving J23117 and its strong tightness suggest that low fluorescence is harder to calculate and triggers more variation. It is possible that Anderson used a plate reader or another instrument, which could explain this deviation. This can also be the cause of the use of other chassis or protocols. Unfortunately, we were not able to find any further reference about his work. We are then curious to see what results will the other teams provide to confirm our hypothesis.

IntraLab Study

While waiting for those to come out at the Giant Jamboree, we continued to investigate fluorescence variation among our own measurements, carrying out what we called an IntraLab Study. Along with the name, this also implied slightly changing the focus. Instead of different strains, we were interested in comparing different colonies (biological replicates) and in analyzing the variability in measurements from one instrument rather than between instruments.

Each Flow-cytometer measurement computed the fluorescence of 100’000 individuals. Based on this high number and on the shape of the distribution histograms, we assumed the fluorescence to be normally distributed. Doing so, we also approximated the mean by the median. We first performed z tests on technical replicates and biological replicates in order to see any significant difference between them. We used a confidence level of 0.8% instead of 2.5% on the basis of the Bonferonni correction principle. We noticed that p-values for biological replicates were all less than 10-5 assessing the strong difference between them. Fluorescence varied in fact from 20% to 30% among the colonies, constituting an important disparity (Figure 3). This tells us that regardless of the strain, colonies from a same transformation can behave very differently. The major reason is certainly how well the plasmid integration occurred in each of them. As it is high-copy, there are also many distribution possibilities for the number of plasmids in each cell.

We then looked at the variation among technical replicates. This time, we obtained 75% of p-values under 0.08. We were puzzled by this result and even if the variation was only 1% between fluorescence values, we decided to run a more robust Fisher test on our data. On a level of 0.05, p-values for J23101, J23106 and J23117 measurements were 0.065, 0.125 and 0.024 respectively. While the two first indicate no significant difference between measurements, the last suggests it is not the case for J23117. This promoter has a weak fluorescence, which might imply less accurate measurements. Nevertheless, we found those results odd, so we decided to plot measurements next to each other (e.g. Figure 4). We discovered that measurements had a tendency to decrease over time in 80% of measured colonies of each promoter. It is unlikely that the instrument systematically caused a reduction in fluorescence. Rather it was due to a third factor, probably biological or physical. We immediately thought the PBS medium could have affected E.Coli metabolism triggering a loss of GFP expression. But as the variation is only 1% despite this disruptive factor, we can firmly assume the accuracy of the measurements.

What did it bring to us?

The InterLab and IntraLab studies taught us interesting lessons and brought new insights to our project. First, we saw how setting up an experiment requires thorough thinking about multiple parameters. If we had the time, we would have carried out more measurements, to compare how time spent in PBS really impacted on our cells. Secondly, we did not expect colonies with similar DNA contents were to behave so differently. In a project like Bio LOGIC, precision and stability are major features since we want cells to reproduce computer-like transistors. One way of getting around this variation problem would be the selection of clones through many batches of test. In conclusion, those two examples proved the importance of measuring his devices, as said at the beginning. A good system is based on the interaction between design and measurements, constantly improving one another.