What did we get?

We plotted the constructs’ mean of fluorescence across the three measurements of the three biological replicates (Fig.1). As units were arbitrary, we also expressed the ratio between the three promoters (Fig.2) which gives a more relevant information about difference in GFP expression. As showed in the figure, we compared our results with the measured strength of those promoters from the Anderson collection (lien vers la page). While the J23101/J23106 ratio is quite similar to the one measured by Anderson himself (only 1.67 fold difference), J23101/J23117 and J23106/J23117 ratios vary from 27 and 16 fold respectively between our results and Anderson’s. A plausible explanation for such discrepancy could be the variation between measurement instruments. While we used a Flow-Cytometer that allows finer measurements of week constructs such as J23117, it is likely that Anderson used a plate reader or another instrument. Also, the different chassis or protocol used to prepare samples could also impact on GFP expression. Those differences are precisely what the InterLab Study intends to shed light on and we are curious to see the results that other iGEM teams will obtain.

Nevertheless, while waiting for the InterLab results to come out, we found interesting to investigate variations among our own samples or, in other words, to lead an IntraLab study. For that purpose, we reproduced as much as possible the same experimental conditions while making our cultures grow, this way avoiding some unwanted variation due to sample preparation. We first compared technical replicates from a same colony. Each of the three technical replicates had 100’000 individuals measured. After eliminating the noise from our data, we approximated the fluorescence to be normally distributed across single cell measurements due to their large number. In order to determine if technical replicates were significantly different, we calculated confidence intervals for each one of them. Based on the Bonferroni correction (as we are comparing three measurements), we set the confidence level at 99.2% instead of the commonly used 95% level. Despite large samples number per measurement, we were surprised to see that about two third of the technical replicates were significantly different. Figure 3 displays a typical example where we can clearly see how the first two measurements are contained in each other’s CI while the third is clearly out. Unless samples were not sufficiently homogeneous, meaning bacteria were agglomerated in population clusters expressing different fluorescence levels and which seems quite unlikely, we can state that despite their great accuracy, measurements had a significant difference with each other. Increasing the statistical robustness could then be achieved by doing more than three technical replicates and consider each median as a part of a normal distribution itself. This will also allow measuring the proper variation due to the flow cytometer and obtaining a more precise mean of fluorescence for each construct.

Along with the technical variation, we also compared the different biological replicates. And we were blown away by how two colonies with the same exact plasmids could have so different fluorescence expression. Figure 4 shows how wide the gap is between colonies. Unlike Figure 3, it is not possible to clearly see CI bars on this figure since they are extremely small compared to median differences. This result is nevertheless reassuring in a way since it proves technical difference is less than biological one, which should appear as obvious due to the stochastic basis of a living organism. This huge biological variation can also be used to explain technical variation. 100’000 organisms are maybe not sufficient to have a comprehensive sample of the population. This explanation could be assessed by increasing the number of bacteria sampled per measure, for example by a 10x factor. So if the biological variance is so important, what can we expect from the comparison of fluorescence from two different chassis? That’s what we are eager to learn during the Giant Jamboree!

What did it bring to us?

Finally, those results bring up a lot of new insights on or project that aims at making cells possess the circuits of a machine. How could this information be integrated in the design of a system like ours? We can imagine for example applying a selection mechanism to organisms, somehow like it has been done in agriculture and for the cattle in man history. To comply with the rigid structure imposed by logic gates, bacteria must work in the most similar way and of course are expected to be the most efficient ones. This final comment underlines what we wanted to prove at the beginning: measuring is equally important as creating. A powerful system is the product of the strong and bilateral interaction between those two concepts.